Answers to Selected Exercises

Worked solutions and model analyses for the starred (†) and odd-numbered exercises from each chapter. Underwriting judgment is learned by working real submissions, not by reading conclusions — attempt every exercise, and especially every Underwriting-File entry, before reading its solution.

Chapter 1

Worked solutions to the daggered (†) exercises. (Odd-numbered items not reproduced here are discussion questions whose answers are developed in the chapter text.)

Exercise 1 (definition of insurance)

Insurance is a contract in which an insurer, for a premium, agrees to compensate an insured for a specified, uncertain future loss. The three features: (1) a transfer of risk — the financial consequence moves from the insured to the insurer; (2) pooling — the insurer can bear the risk only because it has accepted many similar risks and pays the unlucky few from the premiums of the many; (3) a promise about the future — the insured pays now for a contingent future payment, which is why insurer solvency and good faith matter.

Exercise 4 (characteristics of an insurable risk)

(1) A large number of similar exposure units — you need a pool for the law of large numbers to work. (2) Definite and measurable loss — the loss must be identifiable in time, place, and amount so it can be adjusted and priced. (3) Fortuitous (accidental) loss — the loss must be a matter of chance from the insured's standpoint, or moral hazard dominates. (4) Not catastrophic to the insurer — losses must not all strike the pool at once (the independence assumption). (5) A calculable chance of loss — frequency and severity must be estimable to set a price. (6) An economically feasible premium — the premium must be small relative to the potential loss, or no one rationally buys.

Exercise 8 (pooling arithmetic)

(a) Expected losses = 50,000 × (1/2,000) = 25 losses per year. (b) Total expected loss = 25 × \$1,000,000 = **\$25,000,000. (c) Pure premium = \$25,000,000 / 50,000 = **\$500 per warehouse (before expenses, profit, and contingencies). The pure premium can also be computed directly as (1/2,000) × \$1,000,000 = \$500.

Exercise 10 (absolute vs. relative volatility)

Doubling a homogeneous, independent book doubles the expected number of losses and increases the absolute standard deviation of the loss count (which grows roughly with the square root of the count). But because the expected count grows faster than its standard deviation, the relative volatility — the standard deviation as a percentage of the expected count — shrinks (roughly as one over the square root of the expected count). The relative figure matters more to a pricing actuary because the premium is set per exposure as a rate; what threatens the insurer is the percentage by which actual results can deviate from the expected rate, not the raw number of losses. A bigger, more diversified book deviates by a smaller percentage and so needs proportionally less capital and can price more keenly.

Exercise 12 (the unprofitable restaurant)

This fails the fortuitous-loss criterion most clearly, and the risk is a speculative risk — one that carries a chance of gain as well as loss. Ordinary commercial success or failure is a business gamble the owner is expected to bear; insuring it would fund bad bets with other policyholders' premiums and create total moral hazard. (It also strains "definite and measurable" and "calculable chance of loss.")

Exercise 16 (the flat-price life insurer)

Step by step: at a single flat price, the coverage is a bargain for high-risk applicants (e.g., those who just received a serious diagnosis) and overpriced for healthy ones. The insurer's marketing attracts precisely the high risks. Healthy applicants, seeing a price that doesn't reflect their low risk, buy less or not at all. Actual losses come in far above the price's assumption; the insurer must raise the price; the remaining healthier applicants now drop out; the pool worsens again; price rises again. This is the adverse-selection death spiral, and it can destroy the product entirely.

Exercise 21 (the deductible's two jobs)

A deductible does a financial job — it reduces the insurer's payout on every covered loss by the deductible amount — and a behavioral job — it keeps the insured exposed to the first dollars of any loss, preserving their incentive to prevent and minimize losses (pushing back on both moral and morale hazard). The behavioral job matters to the whole book because an insured with skin in the game generates fewer and smaller losses; remove the deductible and the insured's loss-prevention incentive weakens, quietly worsening the loss experience of every risk in the pool.

Exercise 26 (open the Harbor Steel file)

The file's question: Should we cover Harbor Steel, and if so, at what price and on what terms? The three things to notice (not decide): (1) the account strains the insurability criteria — the coastal exposure tests "not catastrophic to the insurer," and the fire history tests "fortuitous and calculable"; (2) the prior carrier's non-renewal is itself information to weigh, not simply to trust; (3) the two fires contain a story — about hazard and management, worsening or improving — that must be read before the risk can be priced.

Chapter 2

Worked solutions to the daggered (†) and odd-numbered exercises. (Even-numbered items not reproduced here are discussion questions whose answers are developed in the chapter text and case studies.)

Exercise 1 (define bottomry)

Bottomry is an ancient maritime loan secured by the ship itself, in which the loan does not have to be repaid if the vessel is lost. The charge above ordinary interest functioned as a premium rather than as interest because ordinary interest compensates only for the time value of money and the borrower's general creditworthiness; the extra portion compensated the lender specifically for accepting the risk of loss of the venture — the lender absorbed the loss if the ship sank. That risk-acceptance-for-a-charge is exactly what a premium is, which is why bottomry is a direct ancestor of insurance even though it bundled the protection with a loan.

Exercise 3 (Lloyd's — what it is and is not)

Lloyd's of London is a marketplace and regulated society, not an insurance company. The actual underwriting is done by syndicates (each run by a managing agent) that accept risks on behalf of their members, who provide the capital that backs the risks (historically wealthy individuals, the "Names"; today predominantly corporate capital). A lead underwriter on a syndicate investigates the risk, sets the terms and price, and takes the largest share; other syndicates "follow," each subscribing a portion. Lloyd's provides the infrastructure, rules, and central oversight within which these separate businesses operate.

Exercise 4 (define the tontine)

A tontine is an investment-and-insurance scheme in which subscribers pay into a common fund and the shares of members who die are redistributed among the survivors, so that the last survivors receive the largest payouts. The single dangerous feature is exactly that redistribution-to-survivors mechanism: in its late-19th-century deferred-dividend form it concentrated vast sums of policyholder money under company control with little transparency, created pressure to encourage lapses and forfeitures, and produced conflicts of interest — the abuses that led to major reforms and stronger state regulation. The same feature that made it attractive (large payouts to persistors) made it dangerous (opacity and perverse incentives).

Exercise 5 (actuary vs. underwriter)

The actuary studies the aggregate — builds the mortality tables and the models, sets the class rates and reserves, and measures the experience of the whole book; the actuary prices the class. The underwriter applies that framework to the individual risk — decides whether to accept this applicant, in this class, at this price, on these terms, reading the particulars the aggregate cannot see; the underwriter prices the case. Actuarial science gives underwriting its quantitative spine but does not replace the judgment at its core.

Exercise 7 (what the mortality table estimated)

A mortality (life) table let a life insurer estimate the probability that a person of a given age will die within the year — the age-specific frequency of death. The raw material was the recorded survivorship of a large population by age (e.g., from birth and death registers such as the Breslau records Halley used): by following a large group from birth and recording how many remain alive at each age, you can derive the death probability at any age. That probability is precisely the frequency a life underwriter needs to price a policy.

Exercise 9 (general average as pooling)

General average is risk pooling expressed as a moral and legal mechanism rather than a statistical one. It shares a single loss — a deliberate sacrifice (e.g., jettisoned cargo) — among the small set of parties on one voyage whose property the sacrifice protected, because fairness and common interest demand it. It differs from the law-of-large-numbers pooling of Chapter 1 in two ways: (1) it shares one specific loss among a tiny, known pool (one voyage's participants), not the expected losses across thousands of independent strangers; and (2) it relies on a proportional rule of fairness rather than on statistical predictability. The two are cousins — both spread a loss so no single party is ruined — but one is a contract of fairness and the other a statistical mechanism.

Exercise 11 (timber vs. brick, and adverse selection)

The early fire offices charged more for timber because a timber house was far more likely to burn and to spread fire to its neighbors — the premium should follow the risk. An office that charged timber and brick the same price would have offered a bargain to every timber owner and an overcharge to every brick owner; the timber owners would have bought eagerly and the brick owners would have gone elsewhere, so the office's book would have filled disproportionately with the worst (most flammable) structures. That is adverse selection (Chapter 1) — the pool skewing toward bad risks when one price is offered for unequal risks — and it would have driven the office's losses above its premiums and burned through its capital.

Exercise 12 (underwrite the coffeehouse slip)

As a name in the ~1690s coffeehouse, you would subscribe a share of the £6,000 Prosperous voyage — say £500 or £1,000 at the quoted wartime rate — and let other underwriters subscribe the rest until the £6,000 is filled, rather than taking the whole risk yourself. The reason is survival: a total loss of the ship in wartime (storm or privateer) would be ruinous to a single name's balance sheet, but is survivable if the risk is divided among many risk-bearers, each liable only for their stated share. Two modern mechanisms descend directly from this decision: co-insurance / layering of a large program among multiple carriers (Chapter 12) and reinsurance — insurers spreading catastrophe and volatility among each other (Chapter 27). The founding instinct — no single balance sheet should carry a risk large enough to destroy it — is the root of the whole capital-and-reinsurance apparatus.

Exercise 13 (insuring a house in 1650)

In 1650, before the Great Fire, you would have no good way to write standalone fire insurance on a house because the conditions for insurability did not yet exist. The building stock was a chaotic, non-standardized medieval warren that was hard to assess or compare; there was no large pool of broadly similar structures for the law of large numbers to work on (Chapter 1's first criterion); and fire had not yet been organized into an insurable product or confronted as a catastrophe large enough to create overwhelming demand and the discipline of selection. What had to exist first was a denser, more standardized, better-documented building stock — which the post-fire rebuilding and building regulations created — giving insurers a poolable, priceable set of risks.

Exercise 15 (find the red flag — the tontine)

The red flag is that enormous, easy profits over many consecutive years are being treated as proof the strategy is sound, with no scrutiny of the accumulating liability beneath them. The historian's question the chapter says to ask is: what loss (financial, legal, or reputational) is this practice accumulating that has not arrived yet? For the deferred-dividend/tontine policies, the answer was a large hidden liability of opacity, perverse lapse incentives, and conflicts of interest in the management of vast policyholder funds. The real-world reckoning came in the early 20th century with a famous investigation of the life-insurance industry (the Armstrong investigation in New York), which exposed serious misconduct and led to restrictions on deferred-dividend policies, new transparency requirements, and stronger state regulation.

Exercise 17 (moral-hazard red flags, historical)

(a) A bottomry loan fully forgiven if the ship sinks creates a financial incentive for a desperate or dishonest shipowner to let the ship be lost — to scuttle it or overstate the loss — because the loss extinguishes the debt. (b) An 18th-century life policy taken out by a stranger on another person's life creates a financial incentive for the policyholder to wish, or even hasten, the insured's death, since the policyholder profits from it without suffering any real loss. Both are classic moral hazard (Chapter 1): the arrangement gives someone a financial reason to bring about the very loss being "insured," which is why each was curbed — bottomry by rules on when the debt was discharged and by experience-based pricing, and stranger-life policies by the insurable interest requirement (Chapter 4).

Exercise 18 (predicting the aggregate, not the individual)

You cannot predict when one person will die because an individual death is genuinely unpredictable; but you can predict how many of a large group of people of a given age will die in a year because, in the aggregate, mortality is remarkably stable. This is the law of large numbers (Chapter 1) applied to death: as the number of independent, similar lives grows, the proportion who die in a year converges on the expected rate and its relative fluctuation shrinks, so the insurer can price the rate even though it cannot predict any single death. The mortality table is simply that stable aggregate rate, tabulated by age.

Exercise 19 (the gap between class rate and individual risk)

The statement means that the mortality table (or any rating tool) gives the average risk of a class — here, people at an age — but the individual applicant may be meaningfully better or worse than that class average, and the underwriter's job is to assess where in (or outside) the class the individual actually falls. David Okafor (Chapter 6) illustrates the gap: a 45-year-old with mildly elevated cholesterol and a family history of early heart disease looks, on those factors, worse than the class — but his excellent blood pressure, non-smoking status, normal-range build, and active cycling pull the other way. Whether he is "standard or preferred" cannot be read off the table; it requires the whole-person judgment that lives precisely in the gap between the class rate and his true individual risk.

Exercise 21 (insurance reforms by catastrophe)

The claim is that the insurance industry rarely changes its practices by abstract reasoning; it changes because a disaster — natural catastrophe or scandal — inflicts losses or exposes abuses large enough to force a reckoning, and the resulting reform becomes a permanent rule. Three examples from the chapter: (1) the Great Fire of London (1666) produced fire insurance and the practices of inspection and prevention (§2.2); (2) reckless life-insurance gambling on strangers' lives produced the insurable-interest requirement (§2.4); (3) the tontine/deferred-dividend scandals produced restrictions on those policies, transparency requirements, and stronger state regulation (§2.6). (Also acceptable: insurer failures → state-based regulation; Hurricane Andrew → catastrophe models and percentage wind deductibles.)

Exercise 22 (three modern catastrophes, lasting changes)

• Hurricane Andrew (1992): accelerated the adoption of probabilistic catastrophe models (Chapter 30), made the percentage wind deductible standard (Chapter 12), and transformed the catastrophe-reinsurance market and the flow of capital into it (Chapter 27).
• September 11, 2001: produced an insured loss across an unprecedented combination of lines (property, business interruption, aviation, workers' comp, liability, life) as one correlated event; reshaped reinsurance, drove public terrorism-risk backstops, and added terrorism exclusions and buy-backs to commercial policies.
• Hurricane Katrina (2005): drove change around the contested wind-versus-flood coverage line and the adequacy of catastrophe preparation, and deepened the coastal-insurability crises that persist. (All qualitative; no figures invented.)

Exercise 23 (conduct-driven reform and the social function)

Two examples of conduct-driven (not catastrophe-driven) reform from the chapter: (1) the development of the duty of utmost good faith and the regulation of claims practices (curbing bad-faith claim denials), and (2) the prohibition on unfair discrimination and the long struggle against redlining (Chapters 4 and 35). Each connects to theme six — insurance serves a social function — because insurance holds real power over people's lives (the power to grant or withhold the protection that lets a family survive a fire or a death), and history repeatedly showed that power being abused. The reforms exist because that power carries ethical obligations; the regulation is the scar tissue of the abuses, not an arbitrary burden.

Exercise 24 (memo — wind deductible and cat load) [sample]

To: Underwriting Manager — Re: Why coastal property carries a wind deductible and a cat load. These features are not a punishment for where the insured is located; they are the industry's distilled memory of catastrophes that ruined carriers before us. Hurricane Andrew (1992) inflicted insured losses on a scale no one had modeled, and several insurers failed; the response was permanent and structural — the adoption of catastrophe models to actually quantify correlated storm losses (Ch.30), the percentage wind deductible to keep the insured sharing the catastrophe pain (Ch.12), and the cession of cat exposure to reinsurance so one storm cannot sink us (Ch.27). Katrina (2005) reinforced the lesson. A coastal account genuinely concentrates losses that strike many policies at once — it violates the independence the law of large numbers needs (Ch.1) — so an adequate price requires a cat load and the deductible. The honest caveat: this risk-based pricing has a real social cost (Compliance Corner, §2.2). The same logic that makes coastal coverage adequately priced also makes it expensive or scarce for the people who live there, which is a genuine fairness tension (Ch.35) we should see clearly even as we price the risk correctly. We are not resolving that tension; we are pricing the catastrophe honestly and naming the cost.

Exercise 25 (ethics — risk-based pricing's social shadow) [discussion]

A strong answer argues both sides. For "price accurately and stop there": the underwriter's professional and fiduciary duty is to the pool's solvency; charging a flammable structure (or a coastal home) less than its risk simply transfers cost to safer policyholders and invites adverse selection (Ch.1) — and an insurer that underprices fails everyone, including the very people it tried to help. For "an obligation beyond accurate pricing": insurance is not an ordinary commodity; it is the gateway to financial survival, and a system that prices the poorest out of fire or coastal coverage produces real social harm that the underwriter participates in. The honest resolution (previewing Ch.35) is that the tension between actuarial fairness (price reflects risk) and social fairness (access to affordable protection) is permanent and is usually addressed at the level of policy (residual-market mechanisms, public programs, subsidies, mitigation incentives) rather than by individual underwriters mispricing risk — but the underwriter is obligated to see the shadow clearly and never pretend risk-based pricing is socially costless.

Exercise 27 (Underwriting File historical note) [sample]

Harbor Steel — historical perspective (context only; no pricing/terms). The oldest insurable risk on this account is the steel and fabricated components in transit — a direct descendant of the ancient sea loan and the Lloyd's coffeehouse slip; a broker could have placed marine cover on a steel shipment centuries ago. Among the youngest are the workers'-compensation and products-liability exposures — twentieth-century legal inventions that did not exist for most of insurance history. The account's hardest exposure, the named windstorm that caused the prior carrier to non-renew, is hard for a reason as old as insurance itself: correlated catastrophe violates the independence the law of large numbers needs (Ch.1), and it is exactly the kind of risk early insurers handled worst — the machinery to write it (cat models, percentage deductibles, reinsurance) is a recent, post-catastrophe invention.

Exercise 29 (the arc from ledger to algorithm)

Five stages, in order: (1) the handwritten ledger — premiums in, losses out, recorded by clerks; underwriting was an experienced individual's judgment from the application and inspection. (2) The rating bureau / manual — insurers pooled loss experience and published standardized class codes, base rates, and factors (ISO, NCCI), so even a small office could price using industry-wide data. (3) Statistical methods and the computer — more sophisticated statistics and machines that could process far more policies and analyze loss data at scale. (4) Multivariate models (GLMs) — pricing on many interacting risk factors at once rather than one at a time. (5) Machine learning on new data — models finding patterns in satellite and aerial imagery, telematics, IoT sensors, and vast third-party records, with submissions pre-filled and risks scored in real time. Each stage refines the same ancient task: estimate the risk, set the price.

Exercise 30 (the unchanging relationship)

The unchanging relationship, in one sentence: the tool advises (it estimates the risk and suggests a price from the past and the class), and the underwriter decides (reading the particular, novel, or changing facts the tool cannot see, and overriding it when the case warrants). The Harbor Steel model-override in Chapter 32 — a predictive model scores the account 7/10 and leans to decline; the underwriter, seeing that the two fires were electrical/controllable and that corrective action is underway, writes it at a 6 with a documented override — is "the same move, in principle," as the first life underwriter rating an applicant better than the mortality table because he could see the individual's health and habits that the table's class average could not. The instruments differ beyond recognition across three centuries; the judgment relationship is identical.

Chapter 3

Worked solutions to the daggered (†) and odd-numbered exercises. (Items that are open-ended discussion or memo prompts are answered with a model response or a marking guide rather than a single "correct" answer.)

Exercise 1 (the four carrier structures)

Stock insurer — owned by shareholders (investors who supply capital, bear risk, take profit). Mutual insurer — owned by its policyholders (the customers are the owners). Reciprocal insurer — owned by its subscribers, who insure one another, run by an attorney-in-fact. Lloyd's — not a company but a marketplace, where syndicates backed by members' capital underwrite risk.

Exercise 3 (managing general agent)

An MGA is a specialized intermediary to whom an insurer delegates underwriting authority for a defined class of business — often including the authority to bind coverage, set rates, issue policies, and sometimes handle claims. "Delegated authority" means the MGA is acting as the carrier's underwriter for that book, writing on the carrier's paper and spending the carrier's capital. That is precisely why the carrier must vet, audit, and supervise its MGAs: a delegated book can drift away from appetite or adequacy if it is not watched, and the carrier owns the results.

Exercise 5 (combined ratio formula and interpretation)

Combined ratio = loss ratio + expense ratio. A result below 100% means the company paid out less in losses and expenses than it took in as premium — an underwriting profit. A result above 100% means it paid out more than it took in — an underwriting loss, before any investment income. The figure is read as cents on the premium dollar (e.g., 97% = a 3-cent profit per dollar; 103% = a 3-cent loss).

Exercise 7 (soft vs. hard market)

Soft market: plentiful capacity, fierce competition, falling prices, and loosening terms. Hard market: scarce capacity, receding competition, rising prices, and tightening terms. The whole industry tends to move through these phases together, and they alternate over multi-year cycles.

Exercise 8 (admitted vs. surplus lines)

Rate/form filing: admitted carriers file their rates and forms with the state and are regulated on them; surplus-lines carriers have freedom of rate and form (no filing). Guaranty-fund backing: admitted policies are backed by the state guaranty fund if the insurer becomes insolvent; surplus-lines policies generally are not. Kind of risk: the admitted market writes standard, filed-rate risks; the surplus-lines (E&S / non-admitted) market writes the hard, unusual, high-hazard, and high-catastrophe risks the admitted market won't take at filed rates.

Exercise 9 (rating agency / AM Best)

A rating agency assesses and publishes an insurer's financial strength — its ability to meet its obligations, i.e., to pay claims. AM Best matters specifically in insurance because it is the agency that specializes in insurers, and its ratings are the industry's common currency: many buyers, lenders, and contract counterparties require a minimum AM Best rating (commonly "A- or better"), so the rating acts as a gate on which business will even come to a carrier.

Exercise 11 (same loss ratio, different combined ratio)

Both carriers run a 65% loss ratio, so the difference is entirely in the expense ratio. The direct writer reaches customers through its own captive agents, call center, or website and pays no independent-agent commission, so its acquisition expense is structurally lower — driving its expense ratio to roughly 27% (65% + 27% = 92%). The independent-agency carrier pays independent-agent commission, a larger acquisition cost, so its expense ratio is roughly 33% (65% + 33% = 98%). Identical loss experience, eight points apart on the combined ratio, purely because of the distribution channel (§3.2, §3.4).

Exercise 13 (ownership structure → decision)

The mutual's underwriter answers to policyholder-owners with no quarterly investor, so the institution can tolerate a longer horizon — it may have room to keep a marginal-but-improving account through a rough patch if the long-run book is sound. The stock company's underwriter answers to shareholders who watch the quarter and want growth and a competitive return, which can create pressure to write the account quickly (or, in a soft market, to underprice it to hit a premium number). On identical facts, the mutual might hold the account patiently while the stock company either declines it as not earning its capital fast enough or writes it aggressively under growth pressure — the structure tilts the decision. (The disciplined underwriter accounts for that tilt rather than being unconsciously moved by it.)

Exercise 14 (compute the combined ratio)

(a) Loss ratio = 5,200,000 ÷ 8,000,000 = 65%. (b) Expense ratio = 2,400,000 ÷ 8,000,000 = 30%. (c) Combined ratio = 65% + 30% = 95%. (d) Underwriting result = 8,000,000 − 5,200,000 − 2,400,000 = +\$400,000**, i.e., **\$0.05 of underwriting profit per premium dollar. The underwriting made money.

Exercise 15 (one bad year — the swing)

Hold the expense ratio at 30%; push the loss ratio up 12 points, from 65% to 77%. New combined ratio = 77% + 30% = 107%. Underwriting result on \$8,000,000 of premium = 8,000,000 × (1 − 1.07) = −\$560,000** (a loss). The result swung from **+\$400,000 to −\$560,000**, a **\$960,000 swing — and the only input that changed was the loss ratio. This is the §3.5 lesson in numbers: a ten-ish-point loss-ratio move is the difference between a healthy profit and a serious loss, and the underwriter controls the losses coming in the door more than anyone.

Exercise 17 (waterfall + the broker's 10% cut)

Base case: losses \$0.60 + acquisition \$0.20 + general \$0.13 = **\$0.93 out, so \$0.07 underwriting profit per dollar; combined ratio = 93%. Now the broker wins a 10% rate cut and losses come in exactly as expected. The dollar of losses doesn't change (it's driven by the risk, not the price), but premium falls to \$0.90, so the loss "ratio" rises: \$0.60 of loss ÷ \$0.90 of premium ≈ 67%, and the two expense components, if they scale with premium, fall proportionally to about \$0.18 + \$0.117 ≈ \$0.297 of the reduced premium, i.e., roughly 33% — combined ratio ≈ 67% + 33% ≈ 100%**. The nickel-plus of margin is gone, swallowed by a 10% price cut against unchanged losses. (Even on the simplest reading — knock \$0.10 off premium with the \$0.60 loss and \$0.33 of expense fixed in dollars — you go from a \$0.07 profit to a \$0.03 loss.) The lesson: a price cut comes straight out of the thin profit sliver, not out of some abstract margin.

Exercise 19 (order the cycle)

Starting from a hard market: (iii) prices rise and terms tighten → underwriting becomes profitable → (ii) that profit attracts new capital and capacity → (iv) carriers cut prices to compete for business (the market softens) → (v) the underpriced losses develop over the following years and the combined ratio climbs past 100% → (i) capital exits and capacity shrinks → (back to iii) with less capacity chasing the same risks, prices rise and terms tighten again. Each step causes the next: profit draws capital, capital competes price down, the deferred losses arrive, the losses drive capital out, and scarcity drives price back up. It is a feedback loop, not a series of accidents.

Exercise 21 (real hardening events)

Three from the chapter (qualitative only): the mid-1980s liability crisis consumed capacity as liability losses developed worse than soft-market prices assumed and reinsurers pulled back; September 11, 2001 delivered a sudden enormous loss that consumed capital across property, aviation, and reinsurance; and major catastrophe years (Andrew 1992; the 2004–2005 hurricane seasons; Katrina) consumed property and reinsurance capacity as the surplus backing those lines was depleted. In each, capacity withdrawal is the hinge that snaps price upward.

Exercise 22 (three risks → surplus lines)

Trampoline park — high hazard (frequent, severe injury exposure) the admitted market won't write at filed rates. Coastal chemical plant — high hazard plus catastrophe concentration. Brand-new cyber risk — novelty: too little history to price at a filed admitted rate. The surplus-lines market's freedom of rate and form lets the underwriter set a price and craft terms (high deductibles, tailored exclusions, sublimits) that the filed admitted rate could not accommodate — which is how the difficult risk gets written at all.

Exercise 23 (admitted vs. surplus lines, from the insured's view)

Admitted advantages: (1) guaranty-fund backing if the carrier becomes insolvent; (2) regulated rates and forms, i.e., consumer protections and standardized coverage. Surplus-lines advantages: (1) availability — coverage for a risk no admitted carrier will write; (2) flexibility — terms and pricing tailored to an unusual exposure. The trade is protection-and-standardization (admitted) versus availability-and-flexibility (surplus lines).

Exercise 25 (find the red flag — the "easy" account)

The red flag is the combination the broker mentioned in passing: the expiring admitted carrier is non-renewing and two other admitted markets have already declined. That is strong evidence the risk is not a standard, filed-rate admitted risk — if three admitted carriers won't write it, steering it onto your admitted paper at standard filed rates is probably mispricing it, and it may genuinely belong in the surplus-lines market (freedom of rate and form) where it can be priced and structured for what it is. The "clean, easy" framing is doing work the facts don't support; the path the risk traveled is the tell.

Exercise 26 (find the red flag — growth vs. combined ratio)

The red flag is celebrating premium growth (up 22%) while the combined ratio rose from 96% to 107% — i.e., the book went from a 4-cent profit to a 7-cent loss per premium dollar even as it grew. In combined-ratio terms the growth is bad news because volume multiplies whatever combined ratio you run: 22% more of a 107% book is 22% more loss, not 22% more profit. The cycle dynamic that probably produced it is soft-market growth — the team won volume by underpricing/loosening terms, and the deferred losses are now developing into the rising loss ratio. The right response is not to celebrate but to restore rate adequacy and be willing to shrink the unprofitable portion.

Exercise 27 (memo to a trainee — model response)

Marking guide / model: A strong memo (a) names the combined ratio as the one number that says whether the underwriting made money — below 100% profit, above 100% loss, before investment income; (b) explains that premium growth, market share, and investment returns can all look good while the underwriting loses money, so they are not the test; (c) includes one illustrative figure (e.g., "a book at a 103% combined ratio loses 3 cents on every premium dollar — and writing twice as much of it loses twice as much"); and (d) lands the "we'll make it up on volume" trap — volume multiplies the combined ratio, it does not improve it, and soft-market losses arrive years late. Tone: direct, mentoring, commercially serious.

Exercise 29 (ethics — growth pressure on a stock insurer) — discussion

Marking guide: The tension is between the §3.1 shareholder/growth pressure (real, legitimate to name) and the underwriter's duties to the combined ratio, to policyholders (whose protection depends on solvency), and to the company's long-run financial strength and rating (§3.3). Shaving terms on catastrophe-exposed accounts to book quick premium creates deferred losses (§3.6) and erodes the surplus that backs the rating. A good answer refuses to write inadequate cat risk to hit a quarter, documents the rate-adequacy and accumulation reasoning so the decision is defensible to management and auditors, and proposes meeting growth goals through adequately priced business rather than underpriced volume. The professional move is to make the discipline visible and defensible, not to quietly cave or quietly refuse.

Exercise 30 (ethics — surplus lines outside the guaranty fund) — discussion

Marking guide: For: the surplus-lines market is the industry's pressure-relief valve (§3.7); without freedom of rate and form, hard/novel/cat-exposed risks would be uninsurable, so a higher price outside the guaranty fund is the price of availability, and a sophisticated buyer can choose it with eyes open. Against: the buyer gives up guaranty-fund protection and standardized, regulated coverage, bearing more carrier-insolvency risk — which is troubling if the buyer is unsophisticated or has no real alternative. A good answer lands somewhere defensible (e.g., the surplus-lines option is ethical when the admitted market genuinely won't write the risk, the diligent-search rule is honored, disclosure is full, and the carrier's financial strength/AM Best rating is sound) and notes that the surplus-lines buyer should care about the rating more than an admitted buyer, precisely because the guaranty-fund backstop is absent.

Exercise 31 (Harbor Steel in the market)

(a) Channel: the account arrived through a broker, Meridian Risk Partners, who legally represents Harbor Steel (the insured) and is shopping the account to carriers — so the negotiation ahead is a genuine negotiation with a professional whose duty runs to the buyer. (b) Carrier type: a regional middle-market stock or mutual carrier — this is hand-underwritten, judgment-heavy commercial business, not a Lloyd's-only or MGA-program risk. (c) Open market question: whether it belongs in the admitted or the surplus-lines market. A standard middle-market account would sit in the admitted market, but Harbor Steel's named-windstorm catastrophe exposure, aging roof/original sprinklers, and loss history (bad enough that the prior carrier walked) push it toward the edge. What decides it: whether your admitted carrier can write it at filed rates with the deductibles, roof endorsement, and loss-control subjectivities it needs — or whether the catastrophe exposure requires the freedom of rate and form of the surplus-lines market.

Exercise 33 (no room in the cat zone → forced decline)

Even if Harbor Steel's price and risk grade were perfectly acceptable, the reinsurance structure behind your book (previewed in §3.3, owned by Chapter 27) sets the boundary of how much catastrophe exposure your company can accumulate in the Port Hadley coastal zone. If the cat treaty / zone aggregate has no room left this year, writing Harbor Steel's \$20M named-windstorm-exposed property line would push the company's coastal accumulation beyond what its reinsurance and capital can safely support — a single storm could then threaten solvency. That structural constraint can force a decline regardless of the individual account's merits: the risk is fine; the room is not. It is a clean illustration that capacity, not just price and quality, governs what an underwriter can write.

Chapter 4

Worked solutions to the daggered (†) and odd-numbered exercises. (Items that are open discussion or ethics prompts are answered with the model points an instructor should look for, not a single "correct" answer.)

Exercise 1 (the four special features)

(1) Adhesion — the insurer drafts the take-it-or-leave-it form; the insured adheres to terms they did not negotiate (so ambiguities are construed against the insurer). (2) Aleatory — the values exchanged are unequal and depend on chance; the insured pays a small premium and may collect nothing or a sum hundreds of times larger. (3) Conditional — the insurer's duty to pay is conditioned on the insured first meeting the policy's conditions (premium, notice, cooperation). (4) Unilateral — only the insurer makes an enforceable promise; the insured has already given consideration and need not promise future premiums.

Exercise 3 (insurable interest vs. indemnity)

Insurable interest is the requirement that the insured stand to suffer a genuine financial loss if the event occurs — it is the gate (may you insure this at all?). Indemnity is the principle that recovery restores the insured to the pre-loss position, no better — it is the measure (for how much?). Interest establishes the right to insure; indemnity (plus valuation) sets the amount.

Exercise 5 (representation vs. warranty)

A representation is a statement that induces the contract; it is tested for materiality and substantial truth (it need only be substantially true to the best of the applicant's knowledge, and a false one matters only if it would have changed a prudent underwriter's decision). A warranty is a statement made part of the contract itself; at strict common law it must be literally and exactly true (or exactly performed) regardless of materiality — though modern statutes in many lines soften this to require that a breach be material to the loss. So: representation → materiality/substantial truth; warranty → exact compliance.

Exercise 7 (subrogation)

Subrogation is the insurer's right, after paying a covered loss, to step into the insured's legal shoes and recover from the third party who caused the loss. It serves two purposes: (1) it prevents double recovery — the insured collects once (from the insurer) and may not also keep a recovery from the wrongdoer for the same loss; and (2) it keeps the cost of loss on the party at fault, preserving the deterrent that the careless or negligent party would otherwise escape because the victim was insured.

Exercise 9 (the three-part rate standard)

A rate must not be excessive (unreasonably high relative to the risk and expected costs — consumer protection), not inadequate (unreasonably low, threatening solvency or amounting to predatory under-pricing — the solvency side; note the same law that forbids gouging also forbids under-pricing), and not unfairly discriminatory (not charging similar risks different prices, and not using prohibited classifications).

Exercise 11 (ambiguity in an adhesion contract)

A court will most likely resolve the ambiguity in favor of coverage — against the insurer. The driving feature is adhesion: because the insurer drafted the take-it-or-leave-it form and had the power to make the language clear, the law construes ambiguities against the drafter. The practical consequence is that "we didn't mean it that way" is rarely a defense; the insurer lives by the precision of the words it chose.

Exercise 13 (over-insuring the building)

You should not simply grant the \$30M limit on a \$20M building because indemnity caps a legitimate recovery at restoring the insured's actual loss — an insured cannot be made better off by a loss. A \$30M limit on a \$20M building creates a \$10M reason to be careless (or worse), which is precisely the moral hazard indemnity and insurable interest exist to prevent. The disciplined move is to tie the limit to a defensible valuation (replacement cost), not to the insured's hoped-for collection.

Exercise 14 (key-person life after retirement)

The insurer is wrong. In life insurance, the insurable interest must exist at the inception of the policy, not at the time of death and not continuously. The company had a legitimate key-person interest when it took the policy out; the founder's later retirement does not void the contract. (This is the mirror image of property, where the interest must exist at the time of loss.)

Exercise 15 (indemnity mechanisms)

Subrogation, contribution, salvage, and the bar on collecting twice all enforce indemnity — restoring the insured without enriching them. Two examples: subrogation stops the insured from collecting from the insurer and keeping a recovery from the at-fault third party for the same loss (and shifts cost to the wrongdoer); salvage gives the insurer whatever is left of property it has paid a total loss on, so the insured cannot be paid in full and keep the wreck to sell. Both ensure the insured ends up whole, not ahead.

Exercise 17 (silent withholding of a pending lawsuit)

This is concealment — the silent withholding of a material fact the applicant knows and the insurer does not. The insurer may have a remedy of rescission (voiding the policy from inception). The two elements the insurer must generally show in most modern lines: (1) the fact was material (a prudent underwriter would have decided differently had they known), and (2) the applicant concealed it with the requisite state of mind — knowing it was material and intending to deceive (fraudulent concealment).

Exercise 19 (find the red flag — the vague fire cause)

The disclosure issue: the submission's description of the 2023 fire ("equipment-related") may not match the true hot-work/welding cause the loss-control narrative suggests — a possible gap between what was represented and what is true, going to a material fact (the fire's cause drives both the hazard assessment and the controls you would require). On these facts alone you cannot yet classify it: it could be an innocent representation (the applicant honestly characterized it loosely), a material misrepresentation (a false statement of a material fact), or a concealment (a deliberate hiding of the welding cause). What you do first: get the loss-cause record in writing — the full loss run, the fire-cause determination, the prior carrier's file — and ask the broker directly, before concluding it is any of the three. Document the inquiry. (Chapter 33 develops this as an SIU/disclosure question; the disposition there is "a gap to clarify, not fraud, no rescission issue.")

Exercise 21 (the subrogation path of the dollar)

With an illustrative \$40,000 paid claim: (1) the third-party driver negligently damages the Harbor Steel truck; (2) Harbor Steel files under its own policy and your company pays the \$40,000 (indemnity — Harbor Steel is made whole now, not left waiting for litigation); (3) your company subrogates, stepping into Harbor Steel's shoes to pursue the at-fault driver (or that driver's insurer); (4) you recover up to the \$40,000 you paid. Net effect: (a) Harbor Steel is made whole exactly once (and, under a make-whole rule, its deductible is typically repaid first out of any recovery); (b) the at-fault driver ultimately bears the cost — no free pass for being negligent against an insured victim; (c) your company's true cost of the loss is reduced by the recovery, which over time relieves pressure on the rate.

Exercise 23 (the make-whole rule and deductibles)

In a "make-whole" jurisdiction, when the recovery from the wrongdoer is too small to cover both the insured's deductible/uninsured portion and the insurer's payment, the insured is generally made whole first — the insured's deductible is typically the first money repaid out of the subrogation recovery, before the insurer takes its share. This matters when you set a deductible (Chapter 12) because a higher deductible the insured retains is, in a subrogated loss, the first dollars they will (often) get back — a point worth explaining to an insured weighing a larger retention.

Exercise 24 (rate systems from most to least restrictive)

Prior-approval (most restrictive) — file and wait for affirmative approval before using the rate. File-and-use — file and may begin using immediately (or after a short wait), subject to later review. Use-and-file — use first, file shortly after; review is purely after the fact. Open competition / no-file (least restrictive) — no filing for approval at all; the market disciplines price, with the regulator policing only solvency and unfair discrimination. The systems differ in the timing of the regulator's review: before use, soon after, after, or essentially not at all.

Exercise 25 (the same law forbids gouging and under-pricing)

The two parts of the standard: "not excessive" forbids gouging (consumer protection); "not inadequate" forbids under-pricing that threatens solvency or is predatory. The regulator is, in principle, on the side of rate adequacy because an insurer that under-prices into insolvency cannot pay claims — a solvency failure the regulator must prevent (and the guaranty fund must clean up). The catch: in file-and-use and open-competition states, the regulator polices the excessive side far more actively than the inadequate side, so in practice the discipline against under-pricing is mostly self-imposed (Chapter 11) — the law won't save you from your own soft-market under-pricing until the losses have already arrived.

Exercise 27 (underwrite the admitted-vs-surplus-lines placement)

Whether the catastrophe property stays on admitted paper or moves to surplus lines depends on whether your filed admitted forms and rates can accommodate the risk with the right terms. If your admitted program, with a percentage named-windstorm deductible and an ACV-roof endorsement, can carry the cat property at an adequate filed rate, it stays admitted. If the cat volatility exceeds what your filed rates/forms allow, some or all of the property may need a surplus-lines placement on freedom-of-rate-and-form terms (or a layered placement). What is lost going to surplus lines: state guaranty-fund protection (no state backstop if the carrier fails), plus the need for a licensed surplus-lines broker and a diligent search confirming the admitted market declined it. You cannot decide yet: you need the pricing (Chapter 11), the terms (Chapter 12), and the reinsurance/cat treatment (Chapters 27, 30) first. This chapter only frames the question — flag the cat property as the placement watch-item.

Exercise 29 (the fair-discrimination paradox)

Insurance must discriminate to function because charging each risk a premium reflecting its expected loss — the cure for adverse selection — is, by definition, treating different risks differently (the teen driver pays more than the middle-aged one; the law requires this sorting). That legitimate discrimination becomes unfair when a classification is not justified by a real difference in expected loss — treating same-risk insureds differently, or pricing on a protected characteristic rather than on risk. The line is not between discriminating and not discriminating; it is between discriminating by risk (fair) and discriminating by prejudice or by a proxy for it (unfair).

Exercise 30 (classifications off the table entirely)

Race, color, religion, and national origin are off the table entirely, in every jurisdiction, regardless of any claimed statistical correlation. The two reasons the chapter gives: (1) such correlations reflect historical injustice rather than causal risk — they are not measuring the thing they appear to measure; and (2) the law and basic fairness forbid it outright. The deeper principle: a factor's statistical predictiveness does not by itself make it legal or fair (the proxy-discrimination point) — predictiveness is necessary to be eligible as a rating factor, but not sufficient to make it permissible.

Exercise 31 (ethics — the proxy-laden but accurate model)

Model points an instructor should look for (the prompt is deliberately unresolved): (a) Actuarial accuracy — the model genuinely improves loss prediction; refusing predictive signal can force lower-risk insureds to subsidize higher-risk ones and invites adverse selection. (b) The proxy-discrimination problem — a ZIP-code feature that produces race-correlated prices because of historical segregation is the textbook proxy: race is nowhere in the model, yet the model acts as if it were there; predictiveness is no defense (§4.7 Compliance Corner). (c) Your duty under §4.7 — you are responsible for what the rating plan does, not just what it says; "it's just predicting risk" does not settle the disparate-impact question. (d) The social function (Theme 6) — pricing prejudice (even laundered through math) betrays the people insurance exists to protect. A strong answer does not collapse the tension glibly: it might recommend testing the factor for disparate impact, seeking a less-proxied alternative that captures the genuine risk signal, and escalating the question rather than shipping the model on Gini alone — while acknowledging the real cost of discarding accurate signal. (Chapter 35 is the full treatment.)

Exercise 33 (the legal-frame memo)

A model memo (≈200 words) should hit the three legal facts and the disposition. Example: "Harbor Steel — legal frame (pre-pricing). DISCLOSURE: the submission and all representations are governed by utmost good faith; statements on the building, roof, sprinklers, operations, and especially the five-year loss history are representations we may rely on (materiality/substantial-truth test). Action: obtain the loss runs, prior non-renewal details, and the 2023 fire's cause IN WRITING — that record is the legal basis for any later rescission and a fact to confirm, not yet a problem. INSURABLE INTEREST: Harbor Steel owns the plant (interest at time of loss); the mortgagee has an interest up to its balance and will be named loss payee; any key-person life on the single owner has interest at inception. Indemnity sets amounts once valuations are in — insure the building for replacement cost, not a hopeful number. PLACEMENT: we are admitted in the state; the catastrophe-exposed property is the piece most likely to test whether it rides our filed admitted forms/rates or needs surplus lines (freedom of rate/form, but no guaranty-fund backing, diligent search required). DISPOSITION: legal frame set; nothing priced or bound; cat property flagged as the admitted/E&S decision point; loss-cause record flagged for written confirmation."

Exercise 35 (three documents to get in writing — and the doctrine behind each)

(1) The full five-year loss runs and the fire-cause determinations — because under utmost good faith / the representation doctrine (§4.3), the written record of what the applicant disclosed is the legal instrument any later rescission rests on. (2) A signed statement of values / building valuation — because indemnity (§4.2) sets the limit at the insured's actual loss, and a defensible valuation is what keeps the limit honest and resists over-insurance. (3) Any waiver-of-subrogation obligations and additional-insured demands from customers — because a waiver of subrogation (§4.4) transfers real recovery rights away from your company and must be seen, scheduled, and priced rather than discovered after a loss. (Acceptable alternatives: the prior carrier's non-renewal letter — disclosure; the mortgagee/loss-payee clause — insurable interest.)

Chapter 5

Worked solutions to the daggered (†) and odd-numbered exercises. (Items not reproduced are discussion or short-memo prompts whose answers are developed in the chapter text.)

Exercise 1 (the DICE structure)

The four parts: Declarations — the specifics of the individual contract (named insured, limits, deductibles, policy period, premium, the schedule of forms). Insuring agreement — the insurer's core promise (what loss, from what event, it will pay for and, in liability, defend). Conditions — the duties and rules both parties must follow, especially what the insured must do for coverage to apply. Exclusions — what is carved back out of the grant. DICE is the framework for reading any policy systematically.

Exercise 3 (named-perils vs. open-perils)

A named-perils form covers only the perils it specifically lists; if the cause of loss is not listed, there is no coverage, and the insured bears the burden of proving the loss was caused by a listed peril. An open-perils ("all-risk") form covers loss from any cause except those specifically excluded; if the cause is not excluded there is coverage, and the insurer bears the burden of proving an exclusion applies. On an unexplained loss, the named-perils insured may be unable to prove a listed peril (claim fails), while under open-perils the insurer may be unable to prove an exclusion (claim is paid) — the difference is entirely who carries the burden.

Exercise 4 (three reasons exclusions exist)

(1) To exclude uninsurable or catastrophic risk — perils that are correlated, catastrophic, or non-fortuitous and break the pool (e.g., war, nuclear hazard, flood, earthquake); they need their own programs, pricing, and reinsurance. (2) To exclude risk that belongs in a different policy — keeping the lines from overlapping and double-covering (e.g., a homeowners policy excludes auto liability; a CGL excludes professional services; a property form excludes employee injury). (3) To exclude what the insured should control or expect — losses that are not fortuitous and would create moral/morale hazard (e.g., wear and tear, gradual deterioration, faulty maintenance, intentional acts). Reason (2) means an exclusion is often a signpost to where the coverage lives, not a denial that it exists anywhere.

Exercise 5 (endorsements and precedence)

An endorsement is a document attached to the policy that adds to, deletes from, or modifies the standard form, becoming part of the contract. When an endorsement conflicts with the base form, the endorsement generally controls, because of layered precedence: the specific and later-added provision beats the general, earlier printed language. This is why the endorsements — read last — can rewrite the grant and the exclusions read earlier, and why coverage is always the net of the base form plus every endorsement.

Exercise 7 (the duty to defend)

The duty to defend is the liability insurer's obligation to provide and pay for the legal defense of suits against the insured. It is broader than the duty to indemnify because the insurer must defend any suit that potentially falls within coverage — even if the allegations are groundless, false, or fraudulent, and even if most of the suit is clearly uncovered — whereas the duty to pay (indemnify) attaches only to covered damages actually owed. Defense costs are often paid in addition to the policy limit, so a liability policy's true exposure includes a defense cost that can rival or exceed the indemnity; an underwriter therefore prices the expected defense as well as the expected payout, even on suits the insurer expects to win.

Exercise 9 (grant → exclusion → exception)

Walk the layers. Grant: the insuring agreement covers "bodily injury." Exclusion: it then removes bodily injury "expected or intended from the standpoint of the insured" — which, read alone, would seem to exclude injuring an intruder. Exception: the sub-clause restores coverage for bodily injury "resulting from the use of reasonable force to protect persons or property." So the self-defense scenario falls back inside coverage via the exception. Result: covered (assuming the force was reasonable, a fact question). The exercise tests the §5.5 skill of reading the whole exclusion, including its "this exclusion does not apply to…" exceptions — miss the exception and you'd wrongly deny a covered claim.

Exercise 11 (which declarations govern)

The declarations page is the part of the policy that changes most often — mid-term endorsements that add a vehicle, adjust a value, or change a limit are processed as amended declarations. So the dec page on file today may not be the one in force three months ago at the moment of loss. You should check the effective date of the declarations you are reading and, in a claim, ask for the declarations in force on the date of loss, not merely the latest set on file. The dec page is a snapshot; you must be sure you are looking at the right frame.

Exercise 13 (don't quote from the dec page)

You will not quote or bind from the declarations page alone, no matter the time pressure. The decs give the limits, deductibles, period, the base form (CP 00 10), and the schedule of six endorsements — but the schedule only names the endorsements; it does not say what they do. Some of those six likely restrict coverage and some may broaden it, and the real coverage (and therefore the adequate price) is the net of the base form plus all six. What you do: request the full forms and all six endorsements, read each, net the coverage, and then price. "Matching the expiring" means matching the actual coverage, which you cannot know from the index page. Quoting from the decs risks either under-pricing coverage you didn't know you'd granted or being beaten on terms you didn't know were already restricted.

Exercise 15 (three endorsements on a CGL)

• Additional-insured endorsement (naming a customer): broadens. It adds an insured — extending the policy's protection (and the duty to defend) to the customer for covered claims. This is granted coverage and must be priced.
• Designated-operations exclusion (removing one process): restricts. It carves a specific high-hazard operation out of coverage — taking coverage back, and a basis for a credit / for accepting a risk you otherwise couldn't.
• Protective-safeguards / hot-work warranty: conditions. It makes coverage depend on the insured maintaining a documented hot-work permit program. Before relying on the warranty you must verify the program actually exists and is followed (an inspection / loss-control visit and a renewal question) — an unverified condition is theater (§5.4). Coverage here is the net: broadened by one insured, narrowed by one operation, gated by one warranty.

Exercise 16 (delete an exclusion, model priced the base form)

Deleting a coverage-restricting exclusion grants additional coverage, which raises expected losses and therefore raises the adequate premium above the \$18,000 the model produced for the clean base form. If you charge \$18,000 for the broadened coverage, you are under-pricing by exactly the value of the coverage you added, and the extra losses will surface in a higher loss ratio and a worse combined ratio two or three years later (theme 3/4). "The model approved \$18,000" is no defense because the model priced a different contract — it never ingested the endorsement, so its approval was for coverage you didn't actually sell. Pricing follows risk; you changed the risk after the model spoke, so you owe the re-pricing.

Exercise 18 (why open-perils costs more)

An open-perils grant promises to pay for loss from any cause except those excluded, and puts the burden on the insurer to prove an exclusion. That means it covers the unexplained and the unanticipated — losses a named-perils insured could never prove arose from a listed peril. The insurer is therefore promising more (coverage for causes no one foresaw or can identify) and has a harder job avoiding a claim (it must affirmatively prove an exclusion). More promised + harder to deny = a larger expected loss, which is why the open-perils form is worth more and should cost more. The price reflects the breadth of the promise and the placement of the burden, not just the building.

Exercise 19 (the certificate-vs-policy gap)

The red flag: the certificate of insurance says additional-insured status is provided, but the policy is a standard CGL with no endorsements — and additional-insured status is created by an endorsement, not by the base form or by a certificate. A certificate of insurance is evidence of coverage; it does not itself grant coverage or amend the policy (and standard certificates say so). So the customer who is relying on being an additional insured may, in fact, not be one — the required endorsement was never attached. This is dangerous because everyone (the insured, the customer, the producer) believes coverage exists that the contract does not provide; the gap surfaces only at a claim, when the customer tenders and is refused. The fix: attach the actual additional-insured endorsement to the policy and confirm its terms match the contract requirement (including "primary and non-contributory" if required, which is its own endorsement).

Exercise 21 (the unverified protective-safeguards condition)

The trap is treating a condition as a formality and never using it — granting a protective-safeguards condition requiring a maintained sprinkler system, then building no process to verify it and never asking about it at renewal. When the fire occurs after the sprinklers were quietly shut off, the condition should be a defense — but a condition you never checked is a defense you will fumble, and worse, it lulled you into writing a risk you should have inspected. It is "theater" because its protective value was always contingent on verification: the condition only changes the risk if the behavior it requires actually happens, and you have no idea whether it did. The disciplined move ties each material condition to something real — the sprinkler-maintenance condition to an inspection and a renewal question — so the defense is genuine and, more importantly, the loss it was meant to prevent is actually prevented.

Exercise 23 (coverage-summary memo — model answer)

(150–250 words; a model, not the only correct version.) "Dear [Broker]: To accept this account, we required a designated-operations exclusion on the general liability policy, attached as endorsement [#]. In plain terms, this endorsement removes coverage for liability arising out of [the specified high-hazard operation]. Everything else in the CGL is unchanged — premises and operations liability, products-completed operations for the client's other work, and personal and advertising injury all remain covered at the stated limits, and the duty to defend still applies to covered suits. We did not broaden or remove any other coverage. We required this carve-out because [the specified operation] presents a severity and frequency profile outside our appetite at a standard price; excluding it is what makes the rest of the account writable on competitive terms rather than declined outright. Your client should understand clearly that claims arising out of that operation will not be covered under this policy, and if they need that exposure covered, we should discuss a separately underwritten and priced solution — I would rather they know the edge of this coverage now than discover it at a claim. Happy to walk through the exact wording. — [Underwriter]" (Honest about the carve-out; explains the why; offers a path; does not oversell.)

Exercise 25 (the flood exclusion — both sides + what to do)

Sound design: the flood exclusion walls off a peril that violates the insurability criteria of Chapter 1 — flood is correlated (one event hits thousands of insureds at once), potentially catastrophic, and in many locations closer to expected than fortuitous. A standard homeowners form at a standard price cannot carry it without threatening the solvency that lets the policy pay ordinary fire and wind claims; flood belongs in a separately priced program (the NFIP or private flood). The social cost: the same exclusion is why thousands of uninsured families are devastated after a storm — the protection gap (Chapter 1's second case study). Both are true. What an ethical underwriter/producer does, given the exclusion is defensible: explain the gap clearly and proactively rather than let it surface at a claim; advise the insured of their actual flood exposure and the availability of NFIP/private flood coverage; and, where they sit in a flood-exposed area, treat closing that gap as part of serving the client, not an upsell. The craft is not to pretend the exclusion is unfair, nor to hide behind it, but to make the coverage intentional and the insured informed — which is where theme 6 (social function) meets theme 3 (the discipline that keeps the promise payable).

Exercise 27 (Harbor Steel property: open-perils ⇒ exclusions define the cat treatment)

Because the Harbor Steel property is written on an open-perils (special) cause-of-loss form, coverage is "any cause of loss except those excluded" — so what defines the coverage (and the catastrophe treatment) is the list of exclusions, not a list of covered perils. The two exclusions doing the most work: (1) the flood / water and earth-movement exclusions, which remove the coastal storm-surge and earthquake exposures — exactly why the Chapter 1 catastrophe exposure must be handled separately (specialty coverage, a separate policy, or a government program), and why the named-windstorm peril needs its own deductible treatment; and (2) the wear-and-tear / gradual-deterioration exclusion, which makes the thirty-year-old roof's ordinary aging the insured's problem rather than the policy's (the roof's wind/water failure in a storm is a different question from its wearing out). On an open-perils form, reading these exclusions is reading the catastrophe and maintenance treatment of the account.

Exercise 29 (three Harbor Steel endorsements: broaden / restrict / condition)

• 5% named-windstorm deductible endorsement: restricts (via a higher retained loss). It raises the insured's retention specifically for named-storm losses — taking back first-dollar cat coverage and lowering the insurer's exposure; it supports a lower/adequate price for the cat peril and is a credit/structural control, not an added charge. (Mechanics of percentage cat deductibles: Ch.12.)
• Protective-safeguards condition on the sprinklers: conditions. It gates coverage on the sprinkler system being maintained and on notice if it goes out of service — neither pure broadening nor restricting, but a condition that makes the risk acceptable; price reflects the improved risk only if it's verified.
• Additional-insured endorsements for Harbor Steel's customers: broaden. They extend the policy (and the duty to defend) to named customers for covered claims — granted coverage, which adds exposure and should be reflected in the price. (Net effect: the cat deductible and the safeguards condition help the price; the additional insureds add to it.)

Exercise 30 (what this checkpoint does NOT settle — the four later chapters)

The Underwriting File explicitly defers the open questions to: Chapter 9 (COPE / risk assessment — whether the property is actually well-protected, and the loss-control read of the two fires); Chapter 10 (the math — whether "two fires in five years" is credible signal or small-sample noise); Chapter 12 (terms — what the named-windstorm deductible, AOP deductible, BI period, and ACV-roof endorsement should actually be); and Chapter 13 (the decision — the final accept/decline/modify, quoting with subjectivities). This chapter maps the shape of the contract (the DICE blocks and ISO forms); those four chapters decide whether the risk is any good, what it costs, how it's structured, and whether to write it.

Chapter 6

Worked solutions to the daggered (†) and odd-numbered exercises. (Items not reproduced here are discussion or memo prompts whose answers are developed in the chapter text and the case studies.) All figures are constructed teaching examples.

Exercise 1 (pure vs. speculative risk)

A pure risk has only two possible outcomes — loss or no loss — with no chance of gain (your house burns or it doesn't). A speculative risk carries a chance of gain as well as loss (a stock, a new restaurant, a crop). Insurance addresses only pure risk because the upside an insured would keep on a speculative risk would let them transfer only the downside — a one-sided bet that adverse selection (Ch. 1) would make fatal, and a gamble society wants people to bear, not pool.

Exercise 3 (the four hazard families)

Physical — a tangible condition (worn wiring next to combustibles). Moral — dishonesty/incentive (an owner whose failing warehouse is worth more burned than sold). Morale — carelessness/indifference (a newly-insured driver who stops doing brake checks). Legal — the judicial/regulatory environment (a venue famous for runaway verdicts that raises severity on every liability account written there).

Exercise 5 (exposure unit and bases)

The exposure unit is the standardized unit of risk an insurer measures and charges for — the answer to "per what?" Typical bases: commercial property = per \$1,000 of insured value; workers' compensation = per \$100 of payroll, by class code; general liability = per \$1,000 of sales/receipts (or payroll, or area); commercial auto = per vehicle (adjusted by use, radius, type). Each base is chosen because it grows when the risk grows.

Exercise 7 (the mental move and the controls step)

The sequence is exposure → hazard → controls → frequency × severity → a grade. The controls step is what separates underwriting from mere risk description: it is the active step where the underwriter identifies what is already in place or can be required (a roof replacement, a hot-work program, telematics) to reduce frequency or severity. You then price the residual risk after controls, not the raw risk — and the control is the lever you actually pull to turn a marginal risk into a writable one.

Exercise 8 (peril vs. hazard, by family)

(a) Kitchen fire — peril. (b) Oil-soaked rags by a furnace — hazard, physical. (c) Runaway-verdict jurisdiction — hazard, legal. (d) Owner whose business is worth more burned than sold — hazard, moral. (e) Newly-insured driver who stops brake inspections — hazard, morale. (f) Hailstorm — peril.

Exercise 9 (two identical buildings, different risk)

Both face the identical peril of fire — the same physical process — so the peril cannot explain the difference. The entire difference lives in the hazards: the wiring condition, the housekeeping, the hot-work discipline, the sprinkler maintenance, the combustible storage. One building's hazards make a fire likely and severe; the other's do not. This is why the underwriter studies hazards (this account) rather than perils (physics).

Exercise 10 (legal hazard across states)

Example: the same trucking fleet, same drivers, same maintenance, is a materially worse risk in a state with plaintiff-friendly juries and a history of "nuclear verdicts" (Ch. 23) than in a state where a routine accident produces a modest award — nothing physical changed, only the venue. The chapter treats legal hazard as a peer of the older three families because in modern liability and auto lines the judicial/regulatory environment can drive frequency and severity as powerfully as any physical condition; ignoring it reads half the risk.

Exercise 13 (frequency change arithmetic)

(a) Expected annual collision loss = $4 \times \$18{,}000 = \$72{,}000$. (b) New expected loss = $2.5 \times \$18{,}000 = \$45{,}000$. The fall is $(\$72{,}000 - \$45{,}000)/\$72{,}000 = 37.5\%$. Note that the safety program worked entirely through the frequency dimension; severity was unchanged. This is the signature of a housekeeping/operations control.

Exercise 15 (why model frequency and severity separately)

Because the two have different shapes and respond to different drivers. Frequency is a count — relatively stable, well-behaved, and reasonably predictable from a risk's own short history (it obeys the law of large numbers). Severity is an amount — fat-tailed, with most claims modest and a few enormous, so the average is dragged around by rare large losses. Modeling the total directly would blur two phenomena that move independently; estimating each separately (count vs. amount — Poisson-ish frequency, heavy-tailed severity, as Ch. 32 will make concrete) captures each one's true behavior and lets you act on the right lever.

Exercise 17 (rising severity, flat frequency)

Plausible drivers of rising severity with flat frequency: (i) social/legal inflation — verdicts and repair or rebuild costs climbing while the number of claims holds steady; (ii) valuation drift — the insured amounts or the cost to make whole rising (materials, labor) so each claim costs more. "Raise the deductible a little" is incomplete because a small deductible mostly screens small, frequent losses — it barely touches the large losses that are driving severity. The right response targets size: higher attachment points/limits review, tighter terms, sublimits, valuation correction, and (for catastrophe severity) reinsurance.

Exercise 18 (the three properties of a good base; payroll)

A good exposure base is (1) proportional to expected loss, (2) practical to measure and verify, and (3) hard to manipulate. Payroll satisfies all three for workers' comp: (1) more (and more hazardous) labor means more injury exposure, so loss scales with payroll; (2) the business already tracks payroll for tax and accounting, and the insurer verifies it through premium audit (Ch. 22); (3) an employer cannot cheaply shrink the base without genuinely reducing the workforce (and thus the risk) — you can't fake away payroll the way you could fudge a softer base.

Exercise 19 (sanity-check the building value)

A \$9M value on a 50,000-sq-ft plant implies \$180 per square foot. For industrial/joisted-masonry fabrication construction, defensible rebuild costs (materials + labor + soft costs, varying by region and specification) commonly run well above that, so \$9M looks low — a likely undervaluation. If the building is in fact under-insured, the coinsurance clause (Ch. 12) penalizes a partial loss: the insured recovers only in the proportion that the carried limit bears to the required amount, so a covered partial loss is paid only in part. And a total loss would expose the gap directly — the \$9M limit would fall short of the true rebuild cost. The discipline is to interrogate the value, not accept it.

Exercise 20 (one building vs. fifty buildings)

The claim is false as stated. The exposure base — total insured value — is identical (\$1M either way), so on that single measure they look the same. But the risk is utterly different because of the independence assumption (Ch. 1): fifty buildings in fifty towns are (largely) independent units — a genuine pool the law of large numbers can stabilize — whereas \$1M in one building is a single, undiversified bet. Total value gets the size of the bet right; only accumulation/concentration analysis (§6.4; Ch. 30) captures how correlated the exposure is. The lesson: measure both — the base and the concentration.

Exercise 21 (why GL rates on different bases)

GL rates on whatever best tracks the chance of harming a third party for that operation. Sales/receipts fits a manufacturer or retailer whose third-party exposure grows with volume of product/customers (e.g., a products manufacturer). Payroll fits a contractor or service operation whose exposure grows with the labor it deploys on others' premises (e.g., a general contractor). Area fits a premises-driven occupancy whose exposure grows with the space the public occupies (e.g., a shopping mall or apartment building). The base is always the quantity that scales the injury exposure.

Exercise 23 (find the red flag)

These facts point to the moral hazard family. The specific red flags: (i) values increased 60% in one year; (ii) the business is deeply unprofitable (financial distress — a classic motive); (iii) an unusually high limit on easily-movable "stock"; (iv) urgency to bind before month-end. Individually each is explainable; together they form a pattern consistent with an incentive to cause or exaggerate a loss. Before quoting: order and read the loss runs, require an inspection, verify the values and the financial picture, ask pointed questions about the reason for the increase, and — given the pattern — treat it as a referral and a disclosure/SIU check (Ch. 33), not an ordinary quote. Do not bind on the strength of the cover note's urgency.

Exercise 24 (underwrite this — whole-person classification, David Okafor)

A naive point-counting system tallies two negatives (mildly elevated cholesterol, BMI 28) and reaches for a substandard rate. Whole-person classification reads the combination: excellent blood pressure, confirmed non-smoker, active recreational cyclist, and no early-cardiac family history beyond a father's heart attack at 58 — an otherwise strong cardiovascular profile. Critically, the box-counter misreads the BMI: "28" on an active cyclist often reflects muscle mass, not excess fat, so the number that looks like a negative may be a non-issue or even a positive signal of fitness. On the whole picture David is plausibly standard and quite possibly near-preferred; a competitor's algorithm that summed the two negatives would lose a profitable, fairly-priced risk. (The final class is set in Ch. 17.) The lesson: classification is seeing the whole risk, not summing its parts.

Exercise 25 (why same-rate-per-class only works for similar risks)

A class rate prices every member the same, which is fair and stable only if the members truly share similar expected loss — then the law of large numbers works within the class. If you lump dissimilar risks into one class at one price, the price is a bargain for the worse members and a rip-off for the better ones; the better risks leave or pay too much while the worse risks pile in, and the class's experience deteriorates — this is adverse selection (Ch. 1) operating inside a mis-drawn class. The cure is finer, more homogeneous classification, not a single blended price across unlike risks.

Exercise 26 (underwrite this — Harbor Steel commercial auto, exposure → hazard → controls)

Exposure: a 12-unit flatbed/delivery fleet operating a regional radius; collision and third-party liability, with two minor prior claims. Hazards: PHYSICAL — vehicle condition, securement of heavy steel loads; LEGAL — the auto liability venue and the nuclear-verdict severity tail (Ch. 23); MORALE — driver discipline and maintenance slackening. Controls (requirable): driver selection with MVR review (and removal of any poor-record driver), telematics, a formal maintenance program, and load-securement procedures. Residual risk you'd price: a small, well-managed regional fleet whose frequency is modest and controllable, but whose real exposure is the severity tail from a serious liability claim in a bad venue — which is a limits/umbrella and driver-selection problem, not a housekeeping one.

Exercise 28 (apply the six insurability criteria to Harbor Steel)

Large pool of similar units — yes for the everyday perils (metal fabrication is a known class), no for the catastrophe. Definite and measurable loss — yes; property, GL, comp, and auto losses are all adjustable. Fortuitous — yes; the fires read as accidents made likely by hazards, with no moral-hazard signal on the present facts. Non-catastrophic to the insurer — this is the strained criterion: the coastal, named-storm exposure means one event can strike Harbor Steel and many other insureds at once, so the independence assumption fails for the windstorm peril (and only for it). Calculable chance — largely; a class and a loss history exist, though "two fires in five years" sits at the edge of credibility (Ch. 10). Economically feasible premium — yes, if priced for the risk. Insurability is "a function of the risk plus the machinery" because the same strained risk is writable once you add adequate cat pricing, a percentage windstorm deductible, and reinsurance — the apparatus, not the risk alone, decides whether it can be written.

Exercise 30 (write the memo — risk inventory summary)

Model answer (≈180 words): "Harbor Steel & Fabrication is a coastal metal-fabrication risk requiring a full commercial program. Property (\$20M building / \$8M equipment / \$10M business income): the dominant hazards are the original 1994 roof (wind/water severity) and the aging wiring plus hot-work near combustibles (fire frequency — the source of both prior fires); aging sprinklers raise suppression doubt. General liability / products: fabrication quality and the legal venue, with one pending bracket claim as the severity watch-item. Workers' comp (~\$11M payroll): welding/material-handling injury frequency, plus morale-hazard risk of post-claim safety slackening. Commercial auto (12-unit fleet): modest frequency, but a nuclear-verdict severity tail in a bad venue. Catastrophe overlays all property: a single-county named-storm exposure with no internal diversification — the one criterion under genuine strain. Required controls: roof replacement, hot-work permit program, sprinkler certification, infrared electrical scan, driver MVRs/telematics, return-to-work program. Verdict: controllable, at the right price and with the right requirements. (Price, terms, and decision to follow.)"

Exercise 31 (ethics — proxy and unfair discrimination)

The tension: actuarial fairness says price should reflect measured risk, and the ZIP-code factor "improves accuracy" — it predicts loss. Social/legal fairness, and the law against unfair discrimination (Ch. 4), forbids classifying by protected class. The danger here is a proxy: a permissible-seeming factor (ZIP code) that, after controlling for the legitimate risk variables, still adds price and correlates strongly with a protected class — so the math may be discriminating by protected class through a side door (Ch. 35). "Improves accuracy" does not settle it: a proxy can be statistically predictive precisely because it encodes the protected characteristic. What to do: do not unilaterally keep or kill the factor on the desk. Document the finding, escalate to the actuarial/pricing function, legal/compliance, and the chief underwriting officer; ask whether the residual ZIP-code signal reflects genuine, defensible risk (e.g., theft or weather exposure) that can be captured by a direct risk variable instead, and whether the factor survives the applicable state's standards and the model-governance/bias review (Ch. 35). The honest answer is "it depends on what the residual is really measuring" — which is why it is escalated, not resolved alone.

Exercise 32 (Underwriting-File extension — the open moral-hazard judgment)

This chapter only inventories; it cannot close the moral-hazard question because it has not yet gathered or verified information — it has read the submission, not the account. The two later instruments that will close it: the inspection / risk assessment (Ch. 9), which physically verifies the operation and whether the 2023 fire's corrective actions actually took, and the fraud/disclosure check via the SIU (Ch. 33), which tests the application's representations against the loss runs. One thing you'd want each to confirm: from the inspection — that hot-work discipline and housekeeping are genuinely in place now (not just promised), so the fires read as fixed problems rather than ongoing indifference; from the disclosure check — that the application did not understate or misstate the 2023 fire's cause (a known watch-item), since a material misrepresentation would change the risk from "controllable" to "not as presented."

Exercise 33 ("easy decline" rebuttal)

Model reply: "It's the opposite of an easy decline — it's a writable account if we price and structure it right. The two fires are a frequency story about controllable physical hazards (wiring, hot-work) that we can require fixed, and the hurricane zone strains exactly one insurability criterion — 'non-catastrophic to the insurer' — which we answer with adequate catastrophe pricing, a percentage named-windstorm deductible, and ceding the tail to reinsurance. Your reasoning treats the storm and the fires as automatic disqualifiers; what it ignores is that insurability is a function of the risk plus the terms, price, and reinsurance we bring to it." (The criterion their reasoning ignored: non-catastrophic to the insurer, addressable by machinery rather than by decline.)

Chapter 7

Worked solutions to the daggered (†) and odd-numbered exercises. (Some odd-numbered items are discussion questions whose answers are developed in the chapter text; those are summarized briefly.) All figures are illustrative teaching numbers.

Exercise 1 (define underwriting)

Underwriting is the process of evaluating a risk presented for insurance and deciding whether to accept it, decline it, or accept it on modified terms — and at what price. The word is used in two senses: (1) the whole process — submission through clearance, triage, information gathering, assessment, decision, and implementation; and (2) narrowly, the accept/decline/modify decision at the center of that process. Context tells you which sense is meant.

Exercise 3 (underwriting vs. binding authority)

Underwriting authority is the full set of formal limits within which an underwriter may accept, decline, modify, and bind risks (line, limit, premium, hazard/class, pricing latitude, commitment). Binding authority is one specific dimension of it — the power to commit the insurer to coverage immediately (usually via a binder), putting the company on risk before the policy is issued. It is "the sharpest edge" because a bound risk is a real liability the instant it is bound: if the insured's building burns that night, the company pays, even before any paperwork is finalized.

Exercise 4 (the six authority dimensions)

(1) Line of business — which products you may write at all; (2) limit — the maximum policy limit or total insured value you may bind; (3) premium size — the largest account by premium; (4) hazard or class — how risky a class you may write (with a prohibited list above/around it); (5) pricing latitude — how far you may deviate from manual rate (the credits/debits you may apply); (6) commitment — whether you may bind or only quote. Authority is "an and, not an or" because a risk refers up if it exceeds any single one of these — being within your limit does not authorize a prohibited class, and being in-class does not authorize an over-limit account.

Exercise 5 (appetite tiers)

Target/preferred — business the company actively wants; quote aggressively to win. Acceptable — within appetite but unremarkable; quote on the merits and price to the risk. Restricted/caution — writable only with extra controls, higher pricing, or referral; proceed carefully and document. Prohibited/declined — outside appetite; decline at any price and don't waste anyone's time.

Exercise 7 (the three-cornered triangle)

Underwriting — selects and prices the risk going in; sees the individual account (management quality, the story in the loss run, context a rate table flattens). Claims — pays and reserves the losses; sees the consequences of selection (which classes generate nasty losses, which wordings cause disputes). Actuarial — analyzes aggregate experience and builds rates/reserves; sees the aggregate (trends, deteriorating segments, rate adequacy of the whole book). None is complete alone; the loop is healthy only as an honest conversation.

Exercise 8 (the judgment claim and its qualification)

The theme: underwriting is judgment — data informs and models suggest, but the underwriter decides and must defend the decision. The qualification: this is not "judgment always beats data." For simple, high-volume, well-understood risks, algorithms write faster and more consistently than humans; judgment is irreplaceable specifically for the complex, novel, atypical, and relationship-dependent risks where data is thin or context outweighs the class average. The future belongs to underwriters who do both.

Exercise 9 (upstream failures)

Three examples (any reasonable set): (a) binding off a thin submission because the broker was rushed → you priced and accepted a risk you didn't actually understand, and the unseen hazard becomes a large loss; (b) skipping the inspection on an account that "looked clean" → a physical hazard (bad wiring, an unguarded process) goes unpriced and unmitigated, then ignites; (c) missing a clearance conflict → you quote a risk a colleague already declined for cause, or two of your own brokers compete on one account, wasting work and damaging broker trust. The point: the final decision is usually easy if the upstream steps were done well; when an account goes wrong, the break is almost always upstream of the decision.

Exercise 11 (compressed stages in personal auto)

For a simple personal-auto renewal, clearance, triage, information gathering, assessment, decision, and implementation are largely automated or invisible — the system clears, pulls third-party data (MVR, prior losses, credit-based score where permitted), scores, prices, and renews in seconds. The stage that most often pulls a human in is the decision/assessment when something trips a referral rule — a new serious violation, a large prior loss, a value or limit outside the automated band — i.e., the exception the guidelines (§7.4) route to a person.

Exercise 13 (authority grid routing)

Grid: property limit ≤ \$10M; standard + light-manufacturing classes; credits/debits ≤ ±15%; binding authorized. (a) \$7M standard mercantile at 10% credit → within all dimensions → handle yourself; bind. (b) \$12M warehouse at 5% debit** → exceeds the **\$10M limit → refer up (limit dimension breached, regardless of the modest debit). (c) \$8M light-manufacturing needing a 22% credit → limit and class are fine but the credit exceeds your ±15% pricing latitude → refer (pricing dimension breached). (d) \$6M foundry → a heavy-manufacturing class not on your grid → refer / likely decline (class dimension breached), even though the limit is small. Each illustrates that authority is conjunctive.

Exercise 15 (referring everything vs. stretching)

Referring everything is paralysis — you never decide within your own authority, you clog the referral queue, and you earn a reputation with brokers for being unable to give an answer. Stretching is binding something just outside your grid because it's "basically fine" and the deadline is tight — a genuine authority breach that can put the company on a risk it never agreed to. Both are failures. A good referral is neither: it is you acting decisively within authority and escalating cleanly outside it with a worked file and a recommendation attached — "here's the risk, here's my read, here's what I'd do, do you concur?" That is how you earn more authority over time.

Exercise 17 (guidelines encode the past)

The limit: a manual is a distillation of typical risks, but real submissions are frequently atypical — better than the class in some ways, worse in others, in combinations the rulebook never contemplated. A guideline applied mechanically would decline good business it doesn't recognize and write bad business that happens to fit the boxes. The standard fix is the referral path: the manual handles the cases it was built for and escalates the ones it wasn't, preserving underwriter judgment for exactly the novel case.

Exercise 18 (slavish guideline-following is not safe)

An underwriter who never deviates is "outsourcing their judgment to a document that cannot see the file in front of them." Because guidelines encode the past, slavish application means missing the atypical good risk and accepting the atypical bad one that fits the boxes — both unsafe. What does make an underwriter safe is knowing the guidelines cold and recognizing when you're looking at the case they didn't foresee — then using the referral path and documented judgment rather than the mechanical answer.

Exercise 19 (the three forces; growth-by-underpricing)

The three forces: profit, growth, social responsibility. The easiest way to grow fast is to underprice (cut rate, loosen terms) — and that destroys insurers because the premium is booked now while the losses from the inadequate price arrive two or three years later, after a great deal more of the same business has been written. The damage is invisible and deferred at the moment of the sale, which is exactly what makes it seductive; it becomes visible only when the loss ratio deteriorates well after the growth was celebrated.

Exercise 20 (unconscious soft-market drift)

Step by step: rate falls a little to keep a renewal → schedule credits get more generous to win a new account → marginal risks that would've been declined last year get written "because the market's competitive" → appetite loosens one underwriter at a time. Each step is defensible alone; the sum is a book repriced (say) 15% lower and broadened in risk with no one having decided to do either, surfacing as a worse loss ratio two years later. Two mechanisms that force drift to be a signed decision: (any two of) an explicit, monitored appetite statement; pricing floors; referral thresholds; underwriting audit (Chapter 38).

Exercise 21 (same risk, opposite tiers)

No contradiction. Appetite is philosophy made concrete. A carrier whose philosophy favors steady, low-volatility profit treats coastal wind-exposed property over \$10M as restricted/refer — wrapping it in controls, pricing, and referrals — because it is not organized to carry that volatility cheaply. A carrier that has deliberately built a specialty in coastal wind (with the cat modeling, reinsurance, and pricing to match) treats the identical class as target business it is equipped to win. They differ not on the facts of the risk but on their deliberate stance toward what risk they want and can carry — which is exactly what appetite expresses.

Exercise 22 (triage and route the plastics plant)

Walk it: Clearance — assume no conflict; inland, sprinklered, standard paper, so eligible. Triage / appetite — class is "acceptable"; building is modern (2001), sprinklered, protection class 3, no catastrophe exposure, one small corrected loss — this reads as a clean, acceptable-tier risk. Authority — but the \$14M property limit exceeds your \$10M grid, so on the limit dimension alone you must refer up, even though the risk quality is good. Conclusion: proceed (it's a good, in-appetite risk), but not on your own authority — refer for the limit. A two-line triage note: "Plastics-injection plant, \$14M/\$6M, 2001, sprinklered, PC3, inland, one minor corrected loss — clean acceptable-tier risk. Refer: limit (\$14M) exceeds my \$10M authority; recommend quote." (§7.1, §7.3, §7.4)

Exercise 23 (assign appetite tiers)

(a) 10-yr sprinklered suburban office, no losses → target/preferred (clean, well-spread, low-hazard). (b) Wood-frame nightclub with pyrotechnics → prohibited (combustible construction + ignition source + assembly occupancy = severe life-safety/fire hazard most conservative carriers won't touch). (c) Standard auto-parts retailer → acceptable (ordinary mercantile; unremarkable). (d) Coastal seafood processor, ammonia refrigeration, three freeze losses → restricted/caution (cat exposure + a hazardous refrigerant + an adverse loss pattern → writable only with controls, pricing, and likely referral).

Exercise 25 (authority math)

Pricing latitude ±20%; manual premium \$80,000. (a) Lowest you may quote yourself = \$80,000 × (1 − 0.20) = **\$64,000. (b) The broker needs \$60,000. That is \$80,000 × (1 − x) → x = 1 − (60,000/80,000) = 25% off manual, which is \$4,000 below** your \$64,000 floor and 5 percentage points beyond your ±20% authority. Therefore you cannot quote \$60,000 on your own authority — you must refer the requested deviation up the chain (with your recommendation), or hold at \$64,000.

Exercise 27 (override accountability)

Run the comparison: take the 40 overridden accounts and compare the actual loss/underwriting results of the underwriter's overrides against what the model's recommended action would have produced (and/or against the book's overall results). Vindication: the overridden accounts run at or better than the model's recommendations would have — the underwriter is seeing signal the model can't, and the override is real judgment. Indictment: the overrides run worse than the model's recommendations — the underwriter is systematically rationalizing risks they wanted, and the "judgment" is preference. This is the discipline that keeps override a power rather than a license. (§7.7)

Exercise 28 (find the red flag — the thin file)

At least three problems: (1) no loss-run analysis — there's no evidence the history (and its story) was ever read, so the company can't show the risk was assessed; (2) no inspection — a physical-hazard read was never done, which on a property risk is a serious omission; (3) no pricing rationale — nothing shows the rate was adequate for the risk; (4) the subjectivity isn't recorded or tracked — a condition the broker mentioned may never have been met, leaving a hole in coverage. Now that the \$900K fire has hit, each omission costs: with no documented assessment/pricing/reasoning, even a good decision looks negligent in front of management, an auditor, or a court — "I assessed it carefully" is worth nothing without the file showing it; and the untracked subjectivity may mean a required control was never in place. A defensible decision in a thin file looks indistinguishable from a rubber stamp. (§7.5)

Exercise 29 (the problematic decline note)

"Neighborhood's going downhill" is a serious problem in a legal record because it reads as an impermissible basis for the decision — it can be construed as declining on the character of an area rather than on the assessed risk of the specific property, which edges toward unfair discrimination / redlining (Chapters 4 and 35) and can become evidence against the company in a bad-faith or discrimination action. A defensible version must tie the decline to specific, risk-based, guideline-consistent facts about the account itself — e.g., the construction, the protection class, the documented loss history, a named hazard, or a specific guideline the risk fails — language a deposition would vindicate because it is risk-based, consistent, and honest. (§7.5; §7.2 Compliance Corner)

Exercise 31 (the Harbor Steel referral memo) — model answer (~200 words)

TO: Senior Underwriter, Middle-Market Commercial RE: Referral — Harbor Steel & Fabrication, Inc. (new business, via Meridian Risk Partners)

Referring for authority. Harbor Steel is a custom metal-fabrication/structural-steel plant in Port Hadley (Gulf Coast, named-windstorm zone) seeking a full commercial program: property \$20M building / \$8M equipment / \$10M business income, GL with products, workers' comp on ~\$11M payroll, a 12-unit fleet, and a \$10M umbrella. Single 50,000 sq ft plant, built 1994, original roof and sprinklers; ~180 employees; ~\$45M revenue. Loss history: two fires in five years (~\$180K 2021, ~\$1.2M 2023), several WC claims, two minor auto, one pending products claim. Prior carrier is non-renewing for cat exposure + losses.

Why referred: the property limit, the named-windstorm catastrophe exposure, and the adverse loss history each independently exceed my grid. Appetite: I've triaged this as restricted/caution — not target, not an automatic decline.

Open questions: information not yet gathered (loss runs, inspection, SOV, financials, MVRs); risk not yet assessed; no pricing or terms proposed. Recommendation: proceed to information-gathering and assessment under senior oversight; the question to resolve is whether controls and adequate pricing can make a caution-tier, cat-exposed account one we want. Please advise on appetite and authority.

(Note: correctly does not price or set final terms — those are Chapters 11–13.)

Exercise 33 (documentation alone is not enough)

No — documentation alone does not make the override defensible. A "plausible-sounding rationale" attached to an override driven by the account's size, the broker friendship, and the premium target is exactly "preference wearing judgment's clothes": the reasons that actually moved the decision are impermissible (wanting the business), and the written rationale is post-hoc cover. The difference between judgment and preference is not whether a justification exists but whether the justification is the real, risk-based reason and whether it would survive scrutiny. What tests which one it was: hold the override accountable to results (Exercise 27) — do this underwriter's overrides actually run better than the model's recommendations? Judgment is vindicated by outcomes and reviewable reasoning; preference is exposed by them. (§7.7)

Exercise 35 (consistency ≠ fairness)

A consistently applied automated rule can still be unfair if the rule itself relies on a factor that proxies for a protected class or has an unjustified disparate impact — e.g., a rule that systematically surcharges or declines based on a characteristic correlated with race or national origin rather than with risk. Applying such a rule consistently makes the outcome worse, not better, because now the unfair effect is produced systematically across every applicant. Consistency prevents disparate treatment (handling similar risks differently for bad reasons) but does not by itself guarantee the rule being applied is risk-based and lawful; that is a separate test (Chapters 4 and 35). (§7.4, §7.7)

Exercise 37 (why the non-renewal isn't an automatic decline)

The prior carrier's non-renewal is information to weigh, not a verdict to adopt. A carrier may non-renew a perfectly writable account for reasons that have nothing to do with the risk's merits — its own appetite or strategy shift, a pullback from the entire coastal zone, a reinsurance or capital constraint, portfolio concentration limits, or a book-wide repricing. So Harbor Steel triages as restricted/caution, not prohibited: the features that worried the prior carrier (cat exposure, two fires) are real but potentially manageable with controls and pricing — which is what makes it an interesting account rather than an automatic no. The thing you'd still want to understand before going further: why, specifically, the prior carrier walked — a strategic/portfolio exit is benign; a refusal after seeing something specific in the controls, the management, or the claims is a genuine red flag worth chasing. (§7.4, The Underwriting File)

Chapter 8

Worked solutions to the daggered (†) and odd-numbered exercises. (Discussion-only items whose answers are developed in the chapter text are noted briefly rather than reproduced in full.)

Exercise 1 (the application and its three jobs)

The application (submission) is the formal request for coverage in which the applicant describes the risk and answers the insurer's questions. Its three jobs: (1) describe the exposure — what is insured, where, how big, used for what; (2) ask the underwriting questions — the specific loss-predicting questions the insurer has learned to ask for this class (prior losses, prior non-renewals, roof age, hot-work practices, open claims); (3) create the representations — the applicant's answers, on which the insurer relies. The third is a legal matter because those representations invoke utmost good faith (Ch.4): a material misrepresentation can give the insurer grounds to rescind the policy (Ch.33). The signed, dated application is the legal record of what the insured told you.

Exercise 3 (third-party data)

Third-party data is information about a risk obtained from an independent source rather than from the applicant (MVR, CLUE, credit-based insurance score, public records). Its independence is what makes it valuable: it is not filtered through the applicant's interest in your "yes," so it can confirm or contradict the self-reported application. You read it before fully believing the application because the application is self-reported, a snapshot, and unverified.

Exercise 4 (the three third-party products — what each measures)

• MVR — measures the official driving record: license status, moving violations, and at-fault accidents over a state look-back. (Limit: what was cited, not what was done.)
• CLUE — measures filed property/auto claims across all carriers over a multi-year look-back. (Limit: filed claims only; not out-of-pocket losses; a claim is not the same as the applicant's fault.)
• Credit-based insurance score — measures a statistical correlation between credit-history elements and future insurance loss, as a class signal. (Limit: it is about the group, not the individual; contested and restricted in some states.)

Exercise 5 (what the inspection does that the application cannot)

Three functions: (1) it verifies the physical facts — measures and photographs the building, confirms construction, sprinkler coverage, roof condition, fire-protection class (the COPE inputs of Ch.9); (2) it surfaces hazards the application never mentioned, because the inspector sees what the applicant has stopped noticing (combustible scrap, makeshift wiring, blocked egress); (3) it begins loss control by producing recommendations that become the conditions/subjectivities on a quote (Ch.13).

Exercise 7 (the three FCRA obligations)

(1) Permissible purpose — you may pull a consumer report only for an allowed purpose; underwriting a policy the consumer applied for qualifies. (2) Adverse-action notice — if you deny, cancel, non-renew, or charge a higher premium based in whole or part on a consumer report, you must notify the consumer, identify the reporting agency, and disclose dispute and free-copy rights (and, for credit scores, the key factors). (3) Accuracy and dispute rights — the consumer may see, dispute, and have the agency reinvestigate the information.

Exercise 8 (reading the property loss run)

(a) Primarily a frequency story early, with one severity event in Year 4. Years 1–2 show a drip of small, mixed-cause claims (a frequency/management signal); Year 4's \$280K fire is a severity event. Both elements are present, but the standout question is severity. (b) The single most useful question about the Year 4 fire: what was the cause, and what corrective action followed? Cause tells you whether it is a recurring hazard or a one-off; the corrective action tells you whether the risk has been changed. (c) The clean Year 5 is mildly reassuring but not conclusive — one clean year is a thin sample. Before relying on it, confirm what changed after the fire (a control installed, management change, process fixed) and that the run is complete and currently valued. A loss run shows the loss, not the response.

Exercise 9 (two fires "fire" vs. "electrical" and "hot work")

Two fires both coded "fire" tell you only that the building burned twice. Coded "electrical" and "hot work," they tell you the building has two distinct hazards, each requiring a separate control (an electrical upgrade/IR scan for one; a hot-work permit program for the other). Repeating causes signal a hazard never fixed; differing causes signal a breadth-of-hazard problem. The cause-of-loss field is the most important column on the run and the one most often skimmed.

Exercise 10 (the stale valuation date)

A run valued fourteen months ago is a problem because reserves on open claims move: the \$150K reserve on the open WC claim may have developed substantially since (up or down) as the claim matured. Pricing off a stale reserve risks under- or over-stating the true loss. Ask for a currently-valued run (within ~90 days), and specifically the current status and reserve on that open claim. Treat the open claim's eventual cost as a real but uncertain number, not as the stale figure and not as zero.

Exercise 11 (the "loss-free" summary letter)

A one-page summary is not a substitute for carrier-produced loss runs because it is the applicant's (or broker's) characterization, not the underlying detail — and "essentially loss-free" hides exactly the kind of claim that drives a decline. You request the actual five-year, all-lines, currently-valued carrier runs, and you treat the offer of a summary in place of the runs as itself a small red flag worth noting. The discipline to demand the real history is part of how you avoid writing the account everyone else declined.

Exercise 13 (what the industry does and does not claim about credit scores)

The industry does not claim that poor people cause more losses, or that credit causes accidents or fires. The accurate claim is purely statistical: credit-based insurance scores are correlated, across large populations, with future loss frequency and severity, and the correlation persists after controlling for other rating factors. It is a correlation-based class factor — a statement that applicants who look like this person, as a group, have run higher loss ratios — not a judgment that this individual is careless. Stating it as "bad credit means careless" is both wrong and the caricature that poisons the fairness debate.

Exercise 15 (a clean MVR doesn't prove low risk)

Two reasons: (1) the MVR shows what was cited, not what was done — a driver may simply not have been caught; (2) a short or sparse record may reflect low recent mileage or a recently issued license, not safe driving. The MVR is authoritative about citations, not about the underlying behavior, which is exactly the gap telematics (Case Study 2) tries to close.

Exercise 16 (vetting a new alternative-data product)

At least one accuracy/provenance question and one legality question, e.g.: (1) Provenance/accuracy — where does the data come from, how is it sourced and updated, what is its error rate, and can a consumer see and dispute it? (2) What does it actually measure — is it a direct risk component or a proxy correlated with something else? (3) Legality/fairness — is it permitted for this line in the states I write; does it function as a proxy for a protected class; does its use trigger FCRA-style adverse-action obligations? The governing principle (Case Study 1): predictiveness is the start of the analysis, not the end.

Exercise 17 (order the information — the distributor)

You have only the ACORD application and three years of property runs. The information order should add: five years of loss runs across ALL lines (GL, commercial auto, plus the missing two property years), currently valued — to see the full claim story; MVRs on all nine fleet drivers — auto is a listed exposure; three years of financial statements — to test health and confirm the revenue/sales exposure base; inspections of both warehouses — to verify construction, protection, and housekeeping for two real properties; the SOV behind the property values; public records (registration, OSHA if applicable, liens, news). Each line gets one reason. You also reconcile the revenue/sales discrepancy (see #18) before quoting.

Exercise 19 (how much verification a risk gets)

You might write the accounting office on the documents alone because it is small, low-hazard, low-premium, and homogeneous — a class risk where the cost and delay of an inspection are not justified by the exposure. You would never write the \$20M fabrication plant without an inspection because the exposure is large, the hazards (hot work, electrical, aging roof, named-storm) are serious and verifiable only on site, and the loss history demands it. The governing principle: match the depth of verification to the size and hazard of the risk — spend your information budget where the exposure is.

Exercise 20 (build the risk picture — the plastics manufacturer)

Four things to verify and their sources: (1) the prior-loss claim ("no prior losses") → order five-year, all-lines loss runs and a CLUE-style history; (2) the building/physical condition ("state- of-the-art equipment") → on-site inspection; (3) the financial health/exposure base → financial statements + public records; (4) the timing and substance of the new safety committee → ask the broker for the written program and its formation date. The most concerning absence: the glowing application provides no loss runs ("available on request") while asserting no losses — the missing independent history on a hazardous class is the loudest silence in the file, and you do not quote until it is produced.

Exercise 21 (find the red flag — four items)

(a) Blank prior-losses field → red flag (a missing answer on the loss-predicting question); source: the broker, and independent loss runs/CLUE. (b) Runs covering years 2–5 of a five-year request → the missing Year 1 is the concern; source: the carrier's full run. (c) Two of three years of financials → the missing year; source: the company's complete statements. (d) No mention of hot-work permits for a welding shop → omission of the obvious control on the obvious hazard; source: ask the broker for the written procedure.

Exercise 22 (the suspiciously polished submission)

An experienced underwriter slows down because a submission engineered to pre-address every concern is a curated artifact — smoothness can be the broker's skill, or it can be cover for a gap. The move is to verify the load-bearing claims independently (loss runs, inspection, financials) rather than accept the narrative, and to ask specifically: when was the new safety program formed (after a loss?), and has it actually changed outcomes? Polish is not evidence; verification is. Never mistake the absence of a visible red flag for the presence of a clean risk.

Exercise 23 (the news-search fire not on the runs) — discussion

Possible explanations: the loss was paid out-of-pocket (never a filed claim, so not on the runs or CLUE); it occurred under different ownership; it is on a line/period not yet provided; or it was omitted. It stops the quote because an unexplained loss is a material gap — you resolve which explanation is true before pricing, since each has very different risk implications.

Exercise 24 (the FCRA adverse-action walkthrough)

You declined in part on a CLUE report and in part on a credit-based insurance score — both consumer reports, both contributing to an adverse action. The FCRA requires you to send an adverse-action notice. It must: (1) state that an adverse action was taken; (2) identify each consumer-reporting agency that supplied a report used (so the consumer can request a copy); (3) disclose the consumer's right to dispute the accuracy/completeness of the information and to obtain a free copy of the report; and (4) for the credit-based insurance score, disclose the key factors that adversely affected the score. The trigger is "based in whole or in part" — partial reliance is enough.

Exercise 25 (charging more is adverse action) — discussion

Incorrect. Under the FCRA an adverse action includes charging a higher premium, not only declining. If the consumer report moved the price up, the notice is owed. This is the single most common FCRA error on the underwriting desk; the fix is to automate the notice so it fires on any less-favorable outcome.

Exercise 26 (ethics dilemma — the predictive-but-proxy factor)

The genuine tension is actuarial fairness (price should reflect risk; the factor predicts loss) versus social fairness (access to affordable coverage; pricing on a proxy for a protected class entrenches disadvantage). Proxy discrimination (Ch.35) is using a permitted, neutral-looking factor that stands in for a protected characteristic, producing a discriminatory result — and "it's filed and it predicts" is not a complete defense, because a factor can be predictive because it correlates with a protected class. The responsible underwriter neither uses everything that predicts nor refuses all contested data; instead: confirm it is permitted for this line/state, document why it is a legitimate risk signal, test for and escalate the disparate-impact concern, and treat the fair/unfair line as a live constraint — bringing it to compliance/ actuarial review rather than quietly applying it on a manager's say-so. (Previews Ch.35.)

Exercise 27 (pulling a report with no permissible purpose) — discussion

Pulling a CLUE report on a non-applicant violates permissible purpose: the FCRA permits a consumer report only for an authorized reason (here, underwriting a policy the consumer applied for). Curiosity is not a permissible purpose. The rule exists to protect consumers' privacy from unauthorized access to sensitive information.

Exercise 28 (the information-request memo) — model answer sketch

A strong memo: addresses the broker by name; lists each item with specificity (5 years, all lines, currently valued (≤90 days) loss runs; on-site inspection; 3 years financials; MVRs on all listed drivers; SOV; relevant supplements); names the gaps identified (e.g., missing loss-run years, no hot-work program documented, exposure-base discrepancy); states plainly that you cannot quote until the file is complete; and keeps a collaborative tone ("to get you the best terms, I need to underwrite the real risk"). Grading: specificity, completeness, the explicit can't-quote-yet condition, professionalism.

Exercise 29 (explain to a sales colleague) — model answer sketch

A good answer, jargon-free: we order independent loss runs and a CLUE report because the applicant's own summary reflects what they remember and what helps their case — and the accounts most eager to switch carriers are sometimes the ones with a history they'd rather not feature. Independent records show what actually happened, so we price the real risk and don't end up subsidizing hidden losses with everyone else's premiums. (That last clause is adverse selection, explained without the term.)

Exercise 30 (the Harbor Steel information order)

The six categories and the key thing each tells you: (1) 5-yr all-lines currently-valued loss runs — the actual claim story (cause + trajectory of the two fires, the WC drip, the open products claim); (2) on-site inspection — the true condition of the roof/sprinklers/fire class and the hot-work practices; (3) MVRs on all 12 drivers — the fleet's real driving records; (4) 3 years of financials — the company's health and the real revenue/payroll exposure bases; (5) the SOV — verified property values behind the \$20M/\$8M/\$10M; (6) public records — OSHA history, liens/judgments, and any loss the runs missed.

Exercise 31 (why grading/pricing now would be a mistake) — discussion

Grading or pricing now would mean deciding a risk you have not finished seeing. The decisive facts are still missing: the causes and corrective actions behind both fires, whether a hot-work program exists, the financial trajectory, the inspection's verification of the roof and protection, and the MVRs. Any grade built on the submission alone is a grade of the applicant's self-portrait, not the risk. The discipline is to order, flag the gaps, and wait — the assessment is Chapter 9.

Exercise 32 (the two sources most likely to change your view of the fires)

The inspection/loss-control report and the complete, cause-detailed loss runs. The inspection can reveal whether the 2023 fire's corrective actions were actually implemented, whether a hot-work program now exists, and the true state of the electrical and roof — the response the dollars don't show. The detailed runs can reveal the precise causes, whether the electrical issue behind 2021 recurred, and the trajectory. Together they convert "two fires, \$1.38M" into a story about hazard and management — the difference between a decline and a writable, controllable risk.

Exercise 33 (the flagged gaps and who closes them) — discussion

Gaps and likely closing chapter: causes/corrective actions of the fires → Ch.9 (risk assessment) and the inspection; whether the losses are credible signal or small-sample noise → Ch.10 (credibility); the written hot-work program and roof-replacement contract → Ch.13 (subjectivities) and Ch.39 (broker delivers them); the open products claim → Ch.21 (GL/products); the catastrophe treatment → Part V (Ch.27, 30); the model's view of the enriched data → Ch.31–32. (The point: the order opens questions that later chapters answer.)

Exercise 35 ("the data tells you what is there; the underwriter notices what is not")

It means a model scores the fields it is given but has no concept of the field that should be present and isn't. Concrete example: a fabrication plant's submission that never mentions a hot-work permit program — when hot work is the obvious hazard and caused one of its fires — is informative by omission; the underwriter notices the missing field and goes to get it, while the model cannot ask a question it doesn't know it needs to ask. This is theme 5 (technology augments but does not replace): the machine is better at weighing present variables consistently; the human is still required for the absent one, the suspiciously smooth narrative, and the follow-up question.

Chapter 9

Worked solutions to the daggered (†) and odd-numbered exercises. Section references point back to index.md. Numbers in the worked examples are illustrative/constructed teaching figures.

Exercise 9.1

COPE = Construction (what the building is made of / how it behaves in a fire), Occupancy (the use inside), Protection (what fights a fire — public fire protection class + private sprinklers/alarms), and Exposure (neighbors + the natural/catastrophe environment). Occupancy chiefly drives frequency (the use brings the ignition sources and processes); Construction chiefly drives severity (it governs how a fire spreads and whether the building survives). (§9.2)

Exercise 9.3

Bare structural steel does not burn, so it adds no fuel — but it is not fire-resistive. At fire temperatures steel loses strength rapidly and can soften, sag, and buckle, leading to structural collapse — sometimes faster than heavy timber, which chars on the surface and keeps its core strength for a while. "Non-combustible" describes what the material contributes (no fuel); "fire-resistive" describes the rated ability to withstand fire for a defined period (e.g., protected steel or concrete rated 2+ hours). A metal building can be non-combustible and still suffer a structural total loss. (§9.2)

Exercise 9.4

Fire protection class is a numerical grade (commonly 1–10, 1 best) of a location's public fire protection. It summarizes three things: the responding fire department (staffing, training, equipment, response), the water supply (hydrants, available flow, distance to the risk), and emergency communications (how alarms reach the department). It is primarily a severity lever — it does not stop a fire from starting, but a strong department with ample water determines whether a fire is contained or becomes a total loss. (§9.2)

Exercise 9.5

External exposure is the risk to the insured property from sources outside it. Neighboring-property example: the insured building shares a wall with a solvent-using dry cleaner or a fireworks distributor — a fire or explosion next door crosses the property line. Natural/catastrophe example: the building sits in a named-windstorm and storm-surge zone (Harbor Steel) or a wildfire interface — the surrounding environment threatens the risk regardless of how clean the insured's own housekeeping is. (§9.2)

Exercise 9.7

Risk quality / grade is the underwriter's overall comparative opinion of how good a risk is relative to others of its class. It is comparative because it places the risk within its class ("a good fabricator" vs. "a poor fabricator"), not against the universe of all risks — a fabricator is not "bad" for being more hazardous than an office. It is conditional because the grade should reflect the risk as you propose to write it, including the controls you will require: the same account can grade below-average as-is and acceptable as-proposed. (§9.7)

Exercise 9.9

A reasonable COPE read of the masonry-walled restaurant: - C: masonry exterior walls (resist spread) but likely combustible roof/interior — joisted masonry; moderate severity. - O: commercial kitchen — open flame, hot grease/deep-fryer, grease-laden exhaust; high frequency for fire. This is the driver. - P: class-3 public protection (good) and monitored alarms, but no sprinklers — a serious gap; the kitchen especially needs suppression. - E: shares a wall with a dry cleaner — solvents/heat next door; a real neighboring exposure. The letter to worry about most is P (no sprinklers) interacting with O (high-frequency kitchen ignition): a likely ignition source with no automatic severity control. The recommendation writes itself — require kitchen fire suppression (and address the dry-cleaner exposure). (§9.2)

Exercise 9.11

A bare-steel building can still suffer a structural total loss, because unprotected steel weakens and buckles in a serious fire even though it adds no fuel — non-combustible is not fire-resistive (Exercise 9.3). More important, construction is only one COPE letter. Even if construction were strong, the occupancy (what's done inside) drives how often a fire starts, the protection determines whether it is contained, and the exposure can threaten the building regardless. Calling construction "a non-issue" both misunderstands steel's fire behavior and ignores three-quarters of the assessment. (§9.2)

Exercise 9.13

A loss-control report can be silent on a real hazard for two innocent reasons (the inspector did not look at it, or looked but did not write it down) and one dangerous reason (the hazard is bad and went unrecorded). The principle: absence of a finding is not a finding of absence. A welding shop always carries an electrical hazard and a hot-work hazard, so silence on the panels, the hot-work procedures, and the sprinkler test date is a gap to close, not a clean bill of health. What you do next: put pointed questions back to the broker (current sprinkler test certificate? written hot-work procedure? panel condition / IR scan?), or order a fuller loss-control survey, before grading or binding. (§9.3)

Exercise 9.15

Housekeeping is a leading indicator because it reflects the current discipline of the operation, which predicts losses that have not happened yet — whereas the loss runs are lagging, recording only what already occurred. A consistently messy plant (combustibles near ignition sources, oil-soaked rags piling up, blocked aisles, clutter) signals an operation that is not managing its hazards day to day, and that predicts elevated future frequency and impaired emergency response even if the past five years happen to be clean. A clean loss run on a messy plant is luck that has not run out yet. (§9.3)

Exercise 9.17

(a) Moral hazard — over-insurance creates an incentive tied to a loss; response: obtain an independent valuation, set limits to value, weigh against the financials and price. (b) Morale hazard — indifference/ deferred care once insured; response: a deductible to restore skin in the game + a loss-control requirement to restore the protection. (c) Moral hazard — coverage timing that fits a known loss too neatly; response: scrutinize timing/intent, treat as a question to resolve before binding (red-flag detail is Ch. 33's). (d) Morale hazard — the "we're covered, why worry" attitude; response: deductible + documented loss-control conditions. (§9.4)

Exercise 9.19

The paranoia vs. gullibility trap: a paranoid underwriter sees fraud in every loss and declines good accounts that simply had bad luck, starving the book and alienating brokers; a gullible one takes every application at face value and writes the structurally bad risk because the building looked fine. Both are errors. The disciplined move is to treat each signal as a question to resolve, not a verdict: over-insurance → get the valuation; financial distress → read the statements and weigh against controls and price; a murky loss → request the full claim detail. "Questions to resolve" differs from both extremes because it neither convicts nor ignores — it investigates to a defensible answer, then decides whether the questions resolved well enough to write the risk at protective terms. The underwriter is not the judge; they are deciding whether to put capital behind the risk. (§9.4)

Exercise 9.21

Frequency controls (reduce how often a loss occurs): hot-work permit program, infrared electrical scans, housekeeping discipline. Severity controls (reduce how bad it is): automatic sprinklers, fire walls, monitored fire detection. A strong risk needs both because you cannot drive frequency to zero — people weld and wires age, so some fire is eventually likely — which means you must also ensure that the fire which does occur is held to a contained, survivable loss. Frequency controls lower the expected number of events; severity controls cap the cost of the events you can't prevent. (§9.5)

Exercise 9.23

A working automatic sprinkler is uniquely cost-effective because it acts locally, automatically, at the fire, without waiting for human detection or fire-service arrival, and it turns the overwhelming majority of structure fires into small, contained events. What it does not do is stop a fire from starting (it is a severity control, not a frequency control) — and it does nothing against perils it isn't designed for (it won't stop a hurricane, an explosion that outpaces it, or a flood). That gap is covered by frequency controls (hot-work programs, electrical maintenance) and by other severity controls (fire walls, cat hardening, business-continuity planning). (§9.5)

Exercise 9.24

For the catastrophe peril, frequency is set by geography — you cannot lower the chance that a hurricane strikes a coastal location through any on-site control. So cat "loss control" is almost entirely about severity: physical hardening (storm shutters, roof tie-downs and upgrades, elevating critical equipment above surge, drainage) reduces how much a given storm costs, and financial/structural tools manage the rest. The chapters that carry those tools are Chapter 12 (terms — percentage windstorm deductibles that share the loss and align incentives) and Chapter 27 (reinsurance — ceding the catastrophe exposure so one storm cannot sink the carrier), with Chapter 30 (cat modeling) quantifying the exposure. On-site controls manage what they can; terms and reinsurance manage what they cannot. (§9.5)

Exercise 9.25

A defensible assessment of the furniture-manufacturing plant: - Frequency read: high. Heavy wood dust (a combustible-dust explosion/fire hazard) plus a spray-finishing booth (flammable vapors and overspray) are two of the highest-frequency fire hazards in manufacturing; the recent dust-collector fire confirms the exposure is live. - Severity read: high. Frame construction (adds fuel, spreads fire), class-6 protection (mediocre), and — critically — a partial sprinkler system that does not cover the finishing area, i.e., no automatic severity control over one of the two worst hazards. - As-is grade: below-average, leaning toward decline as it stands — a high-frequency and high-severity combination with a protection gap exactly where the hazard is worst. - Three controls to require (conditions of binding): (1) extend automatic sprinkler protection to the finishing/booth area; (2) a combustible-dust management program (housekeeping, dust collection maintenance, ignition-source control per recognized dust standards); (3) proper design/maintenance of the spray booth and its exhaust, with documented procedures. - Verify before binding: current sprinkler test certificate and confirmation of the extended coverage, the dust-collector's post-fire corrective actions, and the booth's compliance documentation. With those controls the risk moves from "decline-leaning" toward "writable below-average with debits." (§9.2–§9.7)

Exercise 9.27

Red flags and classification: - The sprinkler system was shut off for renovation and never recommissioned — a physical hazard (the building's single most important severity control is currently absent, not merely old) and arguably a morale signal (no one restored it). This alone can move the grade toward decline until corrected. - A prior "incidental" fire was not disclosed — a disclosure problem (and a potential misrepresentation issue; rescission/material-misrepresentation mechanics are Chapter 33's). The application said "no losses in 5 years," which is now false. - "Fully sprinklered, alarmed, no losses" on the application vs. the inspection findings — the gap between what the insured claimed and what is true is itself the most important red flag, and it taints the reliability of the rest of the application. What each does: the impaired sprinkler makes the as-is severity unacceptable (require recommissioning + certification as a condition, or decline); the non-disclosure both reopens the loss-history analysis (get the full claim detail) and lowers your trust in every other representation (verify, don't assume). Next step: do not bind on the current submission — require sprinkler recommissioning and certification, obtain full loss detail and a corrected application, and re-grade on verified facts. (§9.3, §9.4)

Exercise 9.29

A four-band scale with actions: - Above-average / preferred → write; consider a schedule credit. - Average / standard → write at class terms. - Below-average / substandard → write with debits and required controls, or decline. - Unacceptable / decline → decline cleanly, with the reason documented. Placing the three risks: (a) the sprinklered, fire-resistive warehouse with clean losses and strong housekeeping = above-average — strong on all four COPE letters and on management; credit-worthy. (b) the hot-work shop with two explained fires and controls available = below-average but controllable — write with debits and the required controls (this is Harbor Steel). (c) the derelict frame building, no protection, repeated unexplained losses, owner in distress = decline — severe hazard, absent and unfixable protection, plus a moral-hazard cluster (distress + unexplained losses). (§9.7)

Exercise 9.31

Arguments against making the sprinkler certification and hot-work program post-binding "recommendations": - Underwriting: the whole reason Harbor Steel is writable is that you priced and graded it as proposed — i.e., with those controls. Bind without them and you have bound the as-is risk (impaired/unverified protection, uncontrolled hot work) at the as-proposed price. You are now underpriced for the risk you actually hold. - Combined ratio (theme 3) and pricing-follows-risk (theme 4): the unpriced frequency/severity you just absorbed shows up as losses two or three years out — the classic soft-market mistake of writing for the growth number and paying for it later. The combined ratio tells that truth eventually; the growth number flatters it only this quarter. - Good faith / documentation: your file now says you identified controls as necessary and then waived them under pressure — a record that is indefensible to an auditor and damaging if a loss follows. A defensible alternative that still tries to win the account: keep the controls as conditions but make them easy to satisfy fast — issue a quote with subjectivities and a short, firm compliance window (e.g., bind contingent on the hot-work procedure and sprinkler certification being delivered within a set number of days, IR scan to follow), and have the broker (Meridian) pull the documents forward. You can also bind narrowly with an interim protective term (e.g., a higher deductible or a fire-related sublimit) until the controls are verified. This protects the pool and the combined ratio and gives the broker a live quote — which is how you actually keep the account without giving away the underwriting. (§9.5, §9.7)

Chapter 10

Worked solutions to the daggered (†) and odd-numbered exercises. (Discussion-only items whose answers are developed in the chapter text are not all reproduced.) All figures are constructed teaching numbers.

Exercise 1 (pure premium, two ways)

Pure premium = expected frequency × severity (e.g., 0.5 claims/yr × \$120,000 = \$60,000 per exposure), and equivalently = total losses ÷ exposures. They are the same because frequency = claims ÷ exposures and severity = total losses ÷ claims, so their product is (claims ÷ exposures) × (total losses ÷ claims) = total losses ÷ exposures — the claim counts cancel. Keeping the frequency × severity form preserves the diagnostic information (is this a frequency or a severity risk?) that the collapsed ratio throws away.

Exercise 3 (trend vs. development)

Trend corrects for change over time in the cost and frequency of losses between the data period and the period being priced (the inflation-like adjustment — rebuild costs, rising verdicts). Development corrects for the immaturity of losses: open claims whose reserves will grow and claims incurred but not yet reported (IBNR) that will appear later. Trend = "restate to future conditions"; development = "grow immature losses to their ultimate value."

Exercise 4 (credibility, full, partial)

Credibility ($Z$) is the weight, between 0 and 1, given to a risk's own experience when predicting its future, versus a class benchmark. Full credibility ($Z = 1$): the experience is large enough to be used on its own. Partial credibility ($0 < Z < 1$): the experience is blended with the class. $Z = 0$ instructs you to ignore the risk's own experience and price off the class; $Z = 1$ instructs you to price off the risk's own experience alone.

Exercise 8 (expected loss and pure premium)

(a) Expected annual loss = frequency × severity = 0.5 × \$120,000 = **\$60,000 per shop. (b) The pure premium per \$1,000-of-value unit = expected loss per shop ÷ units per shop, but here we are given the class totals: a class pure premium = total class losses ÷ total class units. If the class's expected losses scale to, say, \$60,000 per shop across the shops that make up 300,000 units, you need total class losses ÷ 300,000. Assumption: if the class total expected loss is \$4,500,000 across 300,000 units, pure premium = \$4,500,000 ÷ 300,000 = \$15.00 per \$1,000 of value**. (The exercise rewards stating the assumption that lets you go from a per-shop expected loss to a per-unit loss cost — the key idea is pure premium = total losses ÷ exposures.)

Exercise 9 (frequency risk vs. severity risk) — discussion

Account A (many small claims, no large ones) is a frequency risk; a higher deductible helps most, sweeping the attritional small claims back to the insured. Account B (zeros punctuated by one seven-figure loss) is a severity risk; a deductible does little (the insured cannot retain a seven-figure loss), and the defense is the policy limit and the reinsurance behind you. Same expected dollars, opposite structures.

Exercise 10 (median vs. mean)

Claims: \$3K, \$5K, \$8K, \$12K, \$22K, \$1,150,000. Six values → median = average of the 3rd and 4th = (\$8,000 + \$12,000) ÷ 2 = \$10,000. Mean = (3,000 + 5,000 + 8,000 + 12,000 + 22,000 + 1,150,000) ÷ 6 = 1,200,000 ÷ 6 = **\$200,000**. The mean is 20× the median, dragged up entirely by the one \$1.15M loss. Quoting "average claim \$200,000" misrepresents the *typical* claim (~\$10K); but ignoring the \$1.15M is just as wrong, because that tail event is the real exposure — it is a severity story that needs limits and reinsurance, not a frequency story.

Exercise 11 (one clean year) — discussion

Frequency is a low-count, lumpy distribution: an account averaging well under one claim a year expects most years to be zero, so a single clean year is the most likely outcome even for an average or below-average risk — it is weak evidence. Look instead at a multi-year pattern, the class experience, and the qualitative loss-control read.

Exercise 12 (the loss ratio four ways)

Written = \$2,000,000; earned = \$1,500,000; incurred = \$975,000; paid = \$500,000. - paid/written = 500,000 ÷ 2,000,000 = 25% - paid/earned = 500,000 ÷ 1,500,000 = 33.3% - incurred/written = 975,000 ÷ 2,000,000 = 48.75% - incurred/earned = 975,000 ÷ 1,500,000 = 65% ← the honest figure. The honest loss ratio is 65% (incurred over earned). The other three understate it by using paid losses (ignoring reserves and IBNR) and/or written premium (which hasn't earned).

Exercise 13 (permissible loss ratio) — discussion

Expenses ~28% + profit/contingencies 6% = 34% load, so permissible loss ratio = 100% − 34% = 66%. Writing an account you expect to run above 66% is, by definition, planned underwriting loss on that account — sometimes justified for a relationship or growth, but never by accident.

Exercise 14 (the too-good young book)

A 35% paid loss ratio on a first-year workers'-comp book is almost certainly illusory: the premium is freshly written and only partly earned (denominator inflated), and the losses are paid to date on a slow line, so the big payments still sit in reserves and IBNR not yet counted (numerator understated), and the immature losses have not developed toward ultimate. Judge the book on incurred-and-developed losses over earned premium — which would land far higher — before tripling it. Tripling on the reported number is how the book blows up (Case Study 2).

Exercise 16 (pure premium → rate)

(a) Pure premium = \$9,000,000 ÷ 600,000 = **\$15.00 per exposure unit. (b) Charged rate ≈ pure premium ÷ permissible loss ratio = \$15.00 ÷ 0.65 = **≈ \$23.08 per unit before credits/debits. (The division by the permissible loss ratio "grosses up" the pure premium to cover expenses and profit.)

Exercise 17 (trend and develop) — discussion

\$300,000 × 1.30 (develop) = \$390,000; × (1.05)^3 ≈ × 1.1576 (trend three years at 5%) ≈ \$451,500. That exceeds the raw \$300,000 by about 50% — the raw loss-run figure understates the true future cost by half.

Exercise 18 (recent year: relevant but unreliable)

A recent accident year best reflects current conditions (most relevant), but it is the most immature — open claims unsettled, IBNR not yet reported, development factors largest — so it is the least reliable single year. Implication: weight recent and older years sensibly; do not price off the latest year alone, and lean harder on older, more-developed years (and the class) the slower the line pays.

Exercise 20 (square-root rule)

With N = 1,000: (a) Z = √(10/1000) = √0.01 = 0.10; (b) Z = √(90/1000) = √0.09 = 0.30; (c) Z = √(250/1000) = √0.25 = 0.50. "Quadrupling the data only doubles the credibility" because √4 = 2: going from n to 4n multiplies Z by 2, not 4 — credibility grows with the square root of the data, steeply at first then flattening.

Exercise 21 (credibility weighting) — discussion

Weighted = 0.20 × 105% + 0.80 × 68% = 21% + 54.4% = 75.4% ≈ 75%. The account's own (bad) 105% experience earns only a small share — 20% — because at Z = 0.20 the class carries 80% of the answer; the bad sample is too thin to fully believe, so the estimate sits much closer to the 68% class than to 105%.

Exercise 22 (the two opposite errors)

Over-reacting to a risk's own bad experience: pricing a shop as a 95% loss ratio because of two claims, charging a punitive premium, and losing a basically-class account to a competitor. Ignoring a risk's own experience: pricing every account at the class and missing the genuine signal in one that has run eight claims a year for three years (real, higher-credibility deterioration). Credibility weighting holds both off by listening to own-experience exactly as much as it deserves — Z × own + (1−Z) × class.

Exercise 23 (the Bühlmann question) — discussion

Plain English: "Is the difference I see between this risk and the class a real, persistent difference (signal), or the random bounce of a small, skewed sample (noise)?" Credibility is high when between-risk variance (VHM) is large and within-risk noise (EPV) is small (K small → Z high); low when risks are basically alike (small VHM) and each is noisy year to year (large EPV).

Exercise 24 (life applicant, zero own-credibility) — discussion

A life applicant's death is a one-time future event; they have no own loss experience for it (you find out only once), so own-experience credibility ≈ 0. The underwriter relies on the class — mortality tables built from millions of lives, sliced by age, sex, smoking, build — and the whole task is placing the applicant in the right class. General principle: the less an individual's own experience can tell you, the more carefully you must classify it, because classification is how you borrow the credibility of the class.

Exercise 25 (underwrite the credibility call)

Own raw loss ratio 92%, class 60%, Z = 0.25. (a) Weighted = 0.25 × 92% + 0.75 × 60% = 23% + 45% = 68%. (b) Defensible expected loss ratio carried into pricing: ≈ 68% (not the 92% the raw history screamed, not the 60% class). (c) Override the blend upward only if a qualitative finding shows a changed or live hazard — e.g., the \$900K loss was caused by a process or condition that is still present and likely to recur (the credibility math, blind to causation, would wrongly shrink that signal).

Exercise 27 (build a pure premium and a rate)

(a) Pure premium = \$4,500,000 ÷ 250,000 = **\$18.00 per \$1,000 of payroll**. (b) Charged rate ≈ \$18.00 ÷ 0.65 = ≈ \$27.69 per unit before credits/debits. (c) A shop with 1,000 units of payroll → indicated manual premium ≈ 1,000 × \$27.69 = **≈ \$27,690** before experience or schedule adjustment.

Exercise 29 (the trend you forgot)

Two years at ~6%/yr severity trend: (1.06)^2 ≈ 1.1236, so the true current cost is about 12.4% higher than your un-trended pure premium. If your premium did not move, your expected loss ratio rises by roughly the same proportion — e.g., a planned 65% loss ratio drifts toward ~73% — quietly turning an adequate account into an underwriting loss. Pricing the past under-prices the future.

Exercise 31 (pricing on two years of own experience)

The error is over-trusting a low-credibility sample (treating thin own-experience as if fully credible). Two years of a mid-size account carries, on any reasonable full-credibility standard, only a modest single-digit-to- low-double-digit Z. The underwriter should have credibility-weighted the account's own experience against the class — Z × own + (1−Z) × class — letting the class carry most of the price.

Exercise 33 (the memo) — model answer

A 150–200-word memo to the manager should hit: (1) we are not pricing off raw two-fire history because two fires is a low-count, severity-skewed sample; (2) the credibility (square-root rule) of two claims is single digits, so the class loss cost for metal fabrication deserves most of the weight, and the credibility-weighted expected number is far below the naive losses-÷-years figure; (3) the one exception is the hot-work severity signal from the 2023 fire — a live, recurring hazard, not random noise — which we are handling through subjectivities and terms (hot-work permit program, deductible/roof terms) rather than spiking the base rate; (4) conclusion: math sets how much the data may move the price; judgment handles the part the math is blind to.

Exercise 35 (ethics dilemma) — discussion

The genuine tension: the class/territory loss cost is actuarially real — losses in that class or territory genuinely run higher, and charging for that is risk-based pricing, which the law permits (fair discrimination, Chapter 4). But the owner's concern is the social-fairness one — that pricing on neighborhood-level class data, when the individual's own record is clean, can entrench disadvantage and may act as a proxy for protected characteristics (the proxy-discrimination and redlining problem owned by Chapter 35). The legal line: insurance may price by risk but may not discriminate by protected class, and several jurisdictions restrict or scrutinize factors (and proxies) that produce disparate impact. The honest answer names both sides, does not pretend the class loss cost is fake, and does not pretend risk-based pricing is automatically fair — and points to Chapters 4 and 35 for where the legal and ethical line is drawn.

Exercise 37 (extend the file) — model answer

The two-sided disposition: (a) Math — two fires is low credibility; credibility-weighting Harbor Steel's own (large, scary) experience against the metal-fab class lands the defensible expected number much closer to the class than to the ~\$276K naive losses-÷-years figure, so the fire frequency should not stampede the price. (b) Judgment — the 2023 hot-work fire is a live, recurring severity hazard that the credibility shrink would wrongly dampen, so it must be addressed through terms/subjectivities, not the base rate. What this layer does not settle: the actual rate (Chapter 11), the deductible/roof terms that handle the severity (Chapter 12), and the final accept/modify decision (Chapter 13). Disposition: losses partially credible; hot-work is the watch-item.

Exercise 39 (why no decision yet) — discussion

This chapter only establishes how much Harbor Steel's own history may move the price; it does not produce a rate, terms, or a decision. Still missing: the built rate with loads, experience and schedule rating (Chapter 11); the deductibles, limits, and endorsements that make the severity acceptable (Chapter 12); and the accept/ decline/modify decision with subjectivities (Chapter 13). The math is the raw material; those three chapters turn it into a quotable, bindable deal.

Chapter 11 — Answers to Selected Exercises

Worked solutions to the daggered (†) and odd-numbered items. All dollar figures are constructed teaching examples. Section references point back to the chapter.

Exercise 11.1 †

The three blocks are: (1) pure premium — pays for the expected loss (frequency × severity, from Chapter 10); (2) expense load — pays for the cost of running the business (commissions, underwriting and servicing, premium taxes, overhead); (3) profit and contingencies load — pays for the insurer's target profit and a cushion for being wrong. The smallest is the profit and contingencies load (often around 5% of premium), which is exactly why a small pure-premium error is dangerous: it can exceed the entire margin. (§11.1)

Exercise 11.3 †

A rating factor (relativity) is a multiplier on the base rate that adjusts price for a risk characteristic, reflecting how much that characteristic moves expected loss relative to the baseline. 1.00 = baseline (no effect). 1.25 = 25% more expected loss than baseline, so a 25% price increase (a debit). 0.90 = 10% less expected loss, so a 10% price decrease (a credit). The factors chain by multiplication. Note it is expected loss, not probability of loss. (§11.3)

Exercise 11.5

A rate is adequate when it covers the expected losses, the expenses, and a reasonable profit/contingency margin for the risk accepted. The three-word statutory standard: a filed rate must not be inadequate (protects solvency), excessive (protects consumers), or unfairly discriminatory (price differences must reflect risk, not protected class). (§11.7, §11.2; Ch. 4)

Exercise 11.7 †

Retrospective rating is a loss-sensitive plan in which the final premium is adjusted after the policy period based on the insured's actual losses during that period, bounded by a minimum and a maximum. Structural difference from a prospective plan: a prospective plan (manual/experience/schedule) sets the price before the period, and the insurer bears the risk that losses exceed expectation. A retro plan sets the final price largely from the losses themselves, so the insured retains much of the risk (within the min/max band). (§11.6)

Exercise 11.9 †

(a) Permissible loss ratio = 1 − (expense % + profit %) = 1 − (0.28 + 0.06) = 0.66 (66%). (b) The rate is set so the pure premium fills exactly that 66%: rate = pure premium ÷ permissible loss ratio = \$3.20 ÷ 0.66 = \$4.85 per \$1,000 (rounded). Check: 66% of \$4.85 ≈ \$3.20 of expected loss, leaving 34% for expenses and profit. (§11.1)

Exercise 11.11 †

(a) Insurer A: \$2.50 × 1.45 = **\$3.625 per \$1,000**. Insurer B: \$2.50 × 1.70 = \$4.25 per \$1,000. (b) The single most likely reason A is cheaper is a lower expense structure — a smaller loss cost multiplier reflecting lower commissions and operating costs (e.g., a direct writer vs. a higher-commission agency carrier). It is not that A judges the risk safer: both start from the same loss cost, i.e., the same expected loss. The difference is entirely in what each loads on top of that expected loss, not in the expected loss itself. (§11.2)

Exercise 11.13 †

The rate is missing the entire expense load and profit load. An NCCI loss cost is the expected-loss component only; quoting \$1.40 = \$1.40 means selling workers' comp with zero expense and zero profit built in. The first-year combined ratio: even if losses come in exactly as expected (a ~100% loss ratio against a rate that is all pure premium), the expense ratio (commissions alone often 10–20%, plus operating costs and taxes) pushes the combined ratio well above 100% the day the policy binds — before a single claim, the account is structurally unprofitable. The fix: multiply the loss cost by the carrier's loss cost multiplier. (§11.2, §11.1)

Exercise 11.15 †

Deriving relativities one factor at a time (univariate) double-counts when characteristics are correlated. Example: newer buildings tend to also be better protected (modern sprinklers, better wiring). A univariate analysis of construction would credit newer construction for loss reductions that are really the effect of the protection that tends to accompany it — and a separate univariate protection analysis would do the same in reverse, so the two together overstate the combined credit. The modern fix is the generalized linear model (GLM) (Chapter 32), which estimates all relativities simultaneously and isolates each factor's true independent effect. (§11.3; preview of Ch. 32)

Exercise 11.17 †

(a) Modified estimate = Z × (own rate) + (1 − Z) × (class rate) = 0.35 × \$1.10 + 0.65 × \$0.75 = \$0.385 + \$0.4875 = **\$0.87 per \$100** (rounded). (b) It is a **debit** relative to the \$0.75 class rate (the modified estimate is higher). (c) With Z = 0.70: 0.70 × \$1.10 + 0.30 × \$0.75 = \$0.77 + \$0.225 = \$0.995 ≈ \$1.00 per \$100. It has moved up, closer to the risk's own (worse) experience, because a larger risk's own loss history is more credible and so is weighted more heavily — exactly the credibility principle from Chapter 10. (§11.4; Ch. 10)

Exercise 11.19 †

Beyond pricing accuracy, experience rating reduces moral and morale hazard by making the insured pay for its own losses through next year's premium. Under experience rating, a loss this year raises the insured's own premium next year (e.g., a higher X-mod), so the insured — not just the insurer — bears the cost. That gives the insured a direct financial incentive this year to prevent losses: fund the safety program, run return-to-work, keep training current. A flat class rate, identical regardless of the individual's behavior, supplies no such incentive. Experience rating doesn't just price the risk more accurately; it changes the risk. (§11.4; Ch. 1)

Exercise 11.21 †

A defensible schedule-rating worksheet for the woodworking shop:

  category                      credit/(debit)   documented basis
  ─────────────────────────────────────────────────────────────────
  Management / cooperation       (5%)            owner runs documented daily clean-up & tool maintenance
  Equipment / protection         (4%)            dust-collection system present and well maintained
  Premises / housekeeping        +6%             combustible sawdust accumulating in back room (fire load)
  Employee selection/training    +3%             no formal safety training program
  ─────────────────────────────────────────────────────────────────
  net schedule modification      0% (≈ neutral)  credits and debits roughly offset

The net is roughly neutral — the genuine credits for the owner's program and the dust system are offset by debits for the housekeeping fire load and the absence of formal training. Acceptable variations exist (the exact percentages are judgment within filed maxima), but every entry must carry a documented, risk-based basis, and the back-room sawdust must be a debit, not waved away. The key skill demonstrated: crediting what is genuinely controlled while debiting what is genuinely hazardous, on the same account. (§11.5)

Exercise 11.23 †

There is no contradiction because the two modifications answer two different questions. The debit experience modification is retrospective: it reflects that the account's past losses ran worse than the class expected. The credit schedule modification is prospective: it reflects present conditions and controls that make the account's future better than its past. Harbor Steel is the exact example — it carries a debit experience effect (its workers'-comp losses and fire history) and schedule credits available for the improving management and the new controls. The experience mod says "you have lost money before"; the schedule credit says "but you have fixed the thing that caused it." A good underwriter prices both truths into one premium. (§11.4, §11.5)

Exercise 11.25 †

The trade-off is expected cost vs. volatility of cost. A guaranteed-cost (fixed prospective) program gives the insured certainty — it pays a set premium and transfers the risk to the insurer, which loads the price for bearing that risk. A retrospective plan gives the insured a lower expected cost (it stops paying the insurer to bear risk it can bear itself, and is rewarded directly for good loss experience) in exchange for more volatility — its final cost swings with its actual losses, between the minimum and the maximum. The single most important question to answer before recommending a retro plan: can this insured financially survive a maximum-premium (bad-loss) year? A retro plan only makes sense for an insured strong enough to absorb the downside it is agreeing to retain — a credit-and-character judgment, not just a pricing one. (§11.6)

Exercise 11.27

The account pays the \$500 minimum premium**, not the calculated \$140. This is not unfair discrimination against a good risk, because the minimum premium reflects a real cost, not a judgment about risk quality: it costs roughly the same to underwrite, issue, and service any policy, and below some premium those fixed expenses (from the §11.1 expense load) exceed what the rate collects. The minimum ensures every policy at least covers the cost of being a policy. The excellent risk is charged the minimum because of the economics of servicing it, applied consistently to all small accounts — which is exactly what makes it fair rather than discriminatory. (§11.6, §11.1)

Exercise 11.29

"The punishment for inadequate pricing is delayed by two or three years" because insurance sells a promise about the future: you bind the policy and collect the premium now, but the losses that reveal whether the rate was adequate arrive later — the fire, the suit, the claim that develops over years. In an ordinary business, pricing below cost shows up immediately in the receipts; you sell the thing and you are short the same day. In insurance, an inadequate rate produces benign current numbers (premium in, no losses yet) and the deficiency only surfaces when the accident years mature. That delay is why a whole market can underprice for years before correcting, and why rate adequacy is uniquely hard to enforce: the feedback that would stop the behavior is deferred past the point of doing it again and again. (§11.7)

Exercise 11.31

Charging an adequate rate is an ethical act because an insurer that prices inadequately and becomes insolvent protects no one. The chain: an inadequate rate does not cover the expected losses → losses eventually erode surplus → the carrier weakens or fails → its promises evaporate, its claimants go unpaid, and its insureds scramble for replacement coverage in a market it has helped destabilize. The premium is the fuel that pays claims; a rate that cannot cover expected losses is a promise the carrier cannot keep. So adequacy is the precondition for the protection being real — an obligation owed to every future claimant who will need the carrier to still exist, not merely a profit target. Underpricing is not generosity to the insured; it is selling a promise that will not be honored. (§11.7; Ch. 1)

Exercise 11.33 †

The three pricing steps applied to Harbor Steel's property line: 1. Manual rate from the class — start from the base rate for the building's class and apply the published relativities for construction (joisted masonry/metal frame), protection (fire-protection-class 4), and the metal-fabrication occupancy (a genuine, defensible hot-work debit). This sets the class price per \$1,000 of value, applied to the \$20M building. Contribution: the legible, defensible starting price from the law of large numbers applied to the class. 2. Experience rating — Harbor Steel's two fires in five years are thin, shock-driven, single-location history: low-credibility on future frequency, so this moves the property price only modestly on its own. Contribution: a small, credibility-bounded adjustment for the risk's own losses (the more formal experience effect lives in the WC X-mod, Chapter 22). 3. Net schedule modification — a net debit: debits for the end-of-life roof and the housekeeping/hot-work-fire history outweigh credits for improving management and maintained sprinklers, with the hot-work-program credit held contingent on verification. Contribution: prices the conditions the numbers miss, netting to an above-manual price.

What pricing does not settle (explicitly stated in the chapter): the terms that make the price work (named-windstorm deductible, ACV-roof endorsement, BI period) — Chapter 12 — and the decision itself — Chapter 13. The rate is built; whether and how to bind it is still ahead. (§11.3, §11.4, §11.5, The Underwriting File)

Exercise 11.35

The hot-work-program credit is verification-contingent because a promised-but-uninstalled control is not yet a fact about the risk — it is an intention. A schedule credit must reflect a condition that actually reduces expected loss now; crediting a control that does not yet exist would underprice the risk as it currently stands and would not survive an audit (the §11.5 documentation rule). What would make it applicable: verified installation of the hot-work permit program (an inspection or documented proof). This connects directly to the subjectivities of Chapter 13: the correct mechanism is to attach the program as a condition to binding (a subjectivity), price the account on its current merits, and apply the credit (or adjust the rate at the next term) once the program is confirmed in place. (§11.5, The Underwriting File; preview of Ch. 13)

Chapter 12

Worked solutions to the daggered (†) and odd-numbered exercises. (Even-numbered items not reproduced here are discussion or memo prompts whose answers are developed in the chapter text.)

Exercise 1 (define deductible; the three jobs)

A deductible is the portion of a covered loss the insured agrees to pay before the insurer pays anything — the first dollars of every claim, retained by the insured. Its three jobs: (1) it sheds the small, frequent, high-friction losses the insurer would rather not administer; (2) it lowers the price, because the insured retains expected losses the rate no longer has to cover; and (3) it preserves the insured's skin in the game — the incentive to prevent and minimize loss (the moral/morale-hazard answer from Chapter 1). The third is the one a junior underwriter most often forgets — treating a deductible only as a discount rather than as a behavior lever.

Exercise 3 (percentage/catastrophe deductible)

A percentage (catastrophe) deductible is expressed as a percentage of the insured value (not a flat dollar amount) and applies only to a named catastrophe peril, usually wind or earthquake. Example: a 5% named-windstorm deductible on a \$20M building means the insured retains the first \$1,000,000 of a named- storm loss. It is required where a flat deductible fails because catastrophe is the correlated peril that breaks the law of large numbers (Chapter 1): one storm hits thousands of insureds at once, so the insurer cannot afford first-dollar (or near-first-dollar) exposure. The percentage deductible forces the insured to retain a meaningful slice of the correlated loss, prices the tail the insurer keeps, and gives the insured a reason to invest in mitigation.

Exercise 5 (per-occurrence vs. aggregate limit)

A per-occurrence limit caps the insurer's payout for any single loss event; an aggregate limit caps the total for all covered losses across the policy period. The aggregate bites hardest on liability, because the same ongoing operation (a product in distribution, a contractor's completed work, a polluting process) can generate many claims in one year, any of which could exhaust the annual aggregate. Property is usually written per-occurrence with no annual aggregate (each fire is its own event and a building can only burn so many times).

Exercise 6 (sublimit; two reasons to cap rather than exclude)

A sublimit is a limit inside the policy limit — a cap on a particular coverage, peril, or category of loss set lower than the overall limit. Two reasons to sublimit rather than exclude: (1) the underwriter is willing to provide some of the coverage (it is a real insured need) but not to expose the full limit to a peril whose severity is hard to bound (flood) or whose data is thin (property in transit); (2) sublimiting the one exposure dragging up the price lets the rest of the policy stay at full limit and a competitive price — a yes, but only this far instead of a decline.

Exercise 7 (coinsurance in one sentence)

The coinsurance clause requires the insured to carry insurance equal to a stated percentage of the property's full value, and if it carries less, the insurer pays only a proportional share of any loss — even a partial one. "The insured becomes a coinsurer of its own loss" means the under-insured insured bears a share of every loss proportional to its under-insurance, not just losses that exceed the limit.

Exercise 9 (large-deductible program vs. SIR; the line)

A large-deductible program uses a deductible (the insurer pays the claim from the first dollar, then bills the insured back the deductible portion), whereas an SIR has the insured handle and pay claims within the retention directly, with the insurer genuinely off the risk below it. Large deductibles are most common in workers' compensation (also commercial auto and GL), where the statutory obligation forces the insurer to pay the worker first-dollar regardless of the deductible — which is exactly why the insurer carries the insured's credit risk and demands collateral.

Exercise 11 (two-deductible computation) †

Building insured for \$10,000,000; \$25,000 AOP deductible; 5% named-windstorm deductible. - (a) \$70,000 fire — a fire is an "all other peril," so the AOP deductible applies. Insured retains \$25,000; insurer pays \$70,000 − \$25,000 = **\$45,000. - (b) \$2,000,000 named-storm loss** — the **named-windstorm** deductible applies: 5% × \$10,000,000 = \$500,000. Insured retains \$500,000; insurer pays \$2,000,000 − \$500,000 = \$1,500,000. The point: the percentage deductible (5% of value = \$500,000) is far larger than the flat AOP deductible, because it is sized to the catastrophe exposure, not the working loss.

Exercise 13 (large-deductible: good for both; the insurer's risk)

Two reasons it is good for both: (1) the insured stops paying the insurer's expense load and profit margin on the working-layer losses it can comfortably self-fund, and keeps the cash-flow benefit of paying losses over the long tail rather than as upfront premium; (2) the insurer earns a fee for fronting and servicing while taking real insurance risk only on the catastrophic layer above the deductible. The single biggest risk for the insurer is the insured's credit: the policy pays statutory/third-party claims from the first dollar and relies on the insured to reimburse the deductible, so an insured insolvency mid-program leaves the insurer holding unreimbursed losses — hence the collateral requirement.

Exercise 14 (the censored loss history) †

On a large-deductible account, only the claims that pierced the deductible ever appear in the loss runs; everything below the deductible is invisible to the insurer. So the underwriter is reading a censored (truncated) distribution, and a clean-looking loss run can hide a worsening frequency of sub-deductible claims. The two questions the loss runs cannot answer: (1) how many claims below the deductible is the insured actually generating, and is that frequency trending up? (obtainable from the third-party administrator the insured uses); and (2) is the insured's balance sheet strong enough to fund a bad year and reimburse the deductible across the runoff? Both are judgment the model cannot supply.

Exercise 16 (GL limit exhaustion)

Policy: \$1,000,000 per occurrence / \$2,000,000 aggregate. Occurrences: \$900K, \$1.3M, \$700K. - 1st (\$900K): within the per-occ limit, aggregate available → insurer pays **\$900,000. Aggregate remaining: \$2,000,000 − \$900,000 = \$1,100,000. - 2nd (\$1.3M): capped by the **per-occurrence** limit at \$1,000,000, and \$1,100,000 of aggregate remains → insurer pays \$1,000,000**. Aggregate remaining: \$1,100,000 − \$1,000,000 = \$100,000. - 3rd (\$700K): only **\$100,000 of aggregate is left → insurer pays \$100,000, and the insured is uncovered for the remaining \$600,000. - Total insurer payment: \$900,000 + \$1,000,000 + \$100,000 = **\$2,000,000 (the aggregate, fully exhausted). The gap: \$600,000 on the third occurrence — discovered at the worst possible moment. Note also the second occurrence's \$300,000 excess over the per-occ limit is the insured's, separate from the aggregate issue.

Exercise 18 (percentage deductible vs. sublimit — the design choice) †

A percentage deductible forces the insured to retain the bottom slice of a peril's loss while leaving the full policy limit available above it — use it for wind/quake on a property you are willing to insure to full value but need the insured to share the correlated first slice of (Harbor Steel's 5% wind deductible). A sublimit caps the insurer's top exposure to a peril below the policy limit — use it for flood/transit/valuable papers, exposures you will write but not to the full limit because severity is unbounded or data is thin. Rule of thumb: cap the top with a sublimit when you fear the worst case; retain the bottom with a deductible when you want the insured to share the loss and keep mitigation incentives.

Exercise 19 (coinsurance computation) †

Full value at loss \$8,000,000; 80% coinsurance; \$50,000 deductible; carried \$4,800,000; loss \$1,000,000. - Insurance required = 80% × \$8,000,000 = \$6,400,000. - Penalty ratio = carried / required = \$4,800,000 / \$6,400,000 = 0.75. - Loss payment = \$1,000,000 × 0.75 − \$50,000 = \$750,000 − \$50,000 = \$700,000. - Coinsurance penalty = \$1,000,000 − \$750,000 = \$250,000 (before the deductible). The insured collected less than \$1,000,000 even though the loss was far below its limit because coinsurance penalizes under-insurance relative to value, not relative to limit: carrying \$4.8M where \$6.4M was required means carrying 75% of the requirement, so the policy pays 75% of the loss.

Exercise 21 (the value-shaving trap; both sides) †

For the insured: shaving declared value from \$10M to \$7M saves a little premium but, under (say) 80% coinsurance, the requirement is 80% × \$10M = \$8M at the time of loss, so carrying \$7M means a penalty ratio of 7/8 = 0.875 on every partial loss — the insured silently under-collects 12.5% (less the deductible) on every claim all year, a far worse trade than the small premium saved. For the underwriter: accepting the \$7M value means you *rated a \$10M building as a \$7M building — you collected too little premium all year (pricing did not follow risk, Chapter 11) — and the coinsurance penalty does not make you whole; it transfers the shortfall to the insured at claim time, where it becomes a bad-faith complaint and a lost account. The fix is to verify value up front (appraisal/cost-estimator) and, where you want certainty, write agreed value*.

Exercise 23 (restrictive vs. broadening endorsements)

(a) ACV settlement on an aging roof — restrictive; carves out the worn-out-roof hazard so the insurer does not fund a new roof through a storm claim on a roof at the end of its life. (b) Additional-insured status for a customer — broadening; extends liability coverage to a party the insured is contractually required to protect (table stakes on commercial accounts). (c) Exclusion of a discontinued product line's completed operations — restrictive; removes a specific past-products exposure that generated claims. (d) Ordinance-or-law coverage on an old building — broadening; pays the extra cost of rebuilding to current code (and should itself be priced/sublimited, as it is real added exposure, not a free add).

Exercise 24 (find the structuring move) †

The move is a restrictive endorsement excluding the discontinued product line's products-completed- operations exposure going forward. The rest of the account is clean, so you do not have to decline the whole risk or price it as if that product were certain to keep generating claims — you carve out the one hazard. The one-sentence explanation to the broker: "We'll write the account, but the [discontinued product line] is excluded from products and completed operations — that line is the source of the past claims, it's no longer in production, and excluding it lets us write the rest of the operation at a competitive price." (Note the contrast with the model: the model recommended a decline because it scored the historical product claims; the underwriter's structuring move turns a decline into a qualified yes — judgment the model cannot supply.)

Exercise 25 (manuscripting: powerful and dangerous)

Manuscripting is powerful because it lets you write a one-of-a-kind risk no standard form fits — you define the covered exposure, triggers, exclusions, and limits exactly. It is dangerous because every word is new language with no litigation track record, and an ambiguity in it is construed against the drafter — the insurer. The contract-law doctrine (from Chapter 4) is contra proferentem: ambiguity in a contract is construed against the party that drafted it. Standard bureau forms are "standard" precisely because they have been litigated into clarity; manuscripting abandons that clarity, so it should be conservative, narrow, and reserved for risks the standard forms genuinely fail.

Exercise 27 (the binder omission) †

The broader coverage applies — replacement cost on the roof — because the binder binds exactly what it says. A binder is a real, temporary contract of insurance; if the ACV-roof restriction was omitted, the binder granted full (replacement-cost) coverage, and the insurer cannot retroactively narrow it after a loss by pointing to the restriction it "intended" to include. The lesson: the most expensive errors in a binder are the conditions and restrictions left off it; structure the policy brilliantly and omit one subjectivity from the binder, and you have given away exactly the protection you built. Every restriction and subjectivity must appear on the binder.

Exercise 29 (ethics: matching the mispriced flat wind deductible)

Case for holding your structure: a flat \$50,000 deductible on a coastal wind peril is the §12.1 trap — a rounding error against a hurricane that hits your whole coastal book at once; matching it means writing a correlated, under-priced catastrophe exposure that will surface in a bad-storm year and damage the combined ratio (Chapter 3). You owe your own book rate adequacy and your reinsurers honest cat exposure. Case for matching: losing the whole account (and the broker relationship) over one term has real cost, and a single account's wind exposure may be small relative to your aggregate. What to actually do: hold the percentage wind deductible — it is not negotiable on a coastal property — but look for other ways to win the account (price the working layers keenly, broaden a coverage the insured values, improve service/turnaround). If the broker will only place it at a flat wind deductible, let it go and be glad a competitor wrote the mispriced catastrophe risk. Matching a structurally unsound term to win business is exactly how soft markets manufacture the losses that arrive two or three years later.

Exercise 31 ("rarely declines" vs. rate adequacy)

The two reconcile through adequacy: restructuring is the right move when a different structure makes the risk acceptable at an adequate price — a higher deductible, a sublimit on the bad peril, an exclusion of the bad hazard, all of which let you charge a rate that genuinely covers what you keep. "I'll structure around it" becomes a bad account when the structure is just a rationalization for writing an inadequate price — matching a mispriced flat wind deductible (Exercise 29), giving a deductible credit larger than the retained expected loss, or sublimiting a peril but still under-pricing the rest. The test: after restructuring, does the price follow the risk you are actually keeping (Chapter 11)? If yes, you have structured a good account; if you are only talking yourself past an inadequate rate, you have not.

Exercise 32 (Harbor Steel terms inventory) †

Exercise 33 (the ACV roof as the fair move)

ACV on the roof until a warranted replacement is fairer than either alternative because: (a) declining over the roof throws away an otherwise-good account over a single, fixable hazard; (b) writing the roof at replacement cost and pricing as if a full roof loss were certain overcharges the insured for a loss that is not certain, and funds deferred maintenance through the policy. ACV settles the roof at depreciated value — fair to the insurer (it is not buying a new roof through a storm claim on a worn-out one) and fair to the insured (it pays the actual value of what was lost) — while leaving a clear path back to full coverage. What earns the insured back replacement cost is completing the warranted roof replacement, after which you endorse the roof back to replacement cost.

Exercise 35 (drafting terms vs. deciding to offer them)

The four subjectivities that will condition the Harbor Steel quote in Chapter 13: (1) roof replacement within 12 months (ACV until then); (2) a hot-work permit program; (3) sprinkler certification; and (4) an infrared electrical scan. "Drafting the terms" (this chapter) is the design work — deciding the deductibles, the wind deductible, the roof endorsement, the valuation basis, the sublimits. "Deciding to put them on the table" (Chapter 13) is a separate underwriting step because it requires judging whether the risk, at these terms, fits appetite and price, attaching the conditions precedent to binding, and — for an account exceeding the line underwriter's authority — taking it to referral. You can design a perfectly sound structure and still, on reflection, decline the account, refer it, or condition it further; the design and the decision are genuinely distinct.

Chapter 13

Worked solutions to the daggered (†) and odd-numbered exercises. (Even, non-daggered items are discussion or applied-judgment prompts whose reasoning is developed in the chapter text.)

Exercise 1 (the three outcomes)

The three possible outcomes of an underwriting decision are accept, decline, and modify. Accept — agree to insure the risk on the stated terms (communicated and made effective by binding). Decline — refuse to offer coverage (a real decision requiring a risk-based reason and documentation). Modify — decline the risk as presented and offer to write it on different terms, price, or coverage (a counter-offer, typically carrying subjectivities).

Exercise 3 (subjectivity vs. condition of the policy)

A subjectivity is a condition precedent — something that must be satisfied before coverage is bound, or within a defined window after, failing which the quote is void or coverage does not attach (e.g., a satisfactory inspection, an IR scan, a signed roof-replacement contract). A condition of the policy is an ongoing duty the insured owes throughout the term (pay premium, maintain protective safeguards, report claims, permit inspection). The difference is timing and failure mode: an uncleared subjectivity is an underwriting problem that bites now (you may be on risk for something unverified); a breached policy condition is a coverage problem that surfaces later, at claim time.

Exercise 4 (counter-offer)

A counter-offer is a quote on terms, price, or coverage other than what was requested. It is, legally, a rejection of the original request and a new offer in its place — which is why the original ask is off the table and no contract exists unless the broker/insured accepts the new terms. This matters because a modify is not a tweak to the requested coverage; it puts a different deal on the table, and the underwriter is insuring the risk as restructured, not as submitted.

Exercise 5 (referral vs. peer review vs. escalation)

• Referral — upward movement of a decision to someone who holds the authority to make it (trigger: over-limit, restricted class, mandatory-referral rule).
• Peer review — lateral movement for a second, independent opinion (trigger: size threshold, a close call, or self-request); it does not transfer authority — its value is that the reviewer is not anchored by your reasoning.
• Escalation — upward movement of a problem to be resolved (trigger: a dispute, a stuck referral, an ethical concern). Referral moves a decision up; peer review moves it sideways; escalation moves a problem up.

Exercise 7 (declination and its three directions)

A declination is the underwriter's decision not to offer coverage. Its three directions of consequence: business (how you decline shapes whether the broker sends their next, good submission — decline the risk, not the broker); legal (the reason must be risk-based, never a protected characteristic; a consumer-report basis triggers an adverse-action notice); file (it must be documented unambiguously — that it was declined, when, to whom communicated, and why — or it becomes an E&O exposure).

Exercise 8 (the marginal feature)

This hits gate 3 (terms). The risk is in appetite and the base price is roughly adequate, so it does not fail gate 1 or gate 2; what makes it marginal is a specific, fixable feature (the aging roof in a wind zone). Gate 3 asks whether terms — a percentage wind deductible, an ACV roof endorsement, a roof-replacement subjectivity — can convert the marginal risk into an acceptable one. When terms can close that gap, gate 3 produces the modify outcome (a counter-offer), which is why modify is the most common result on accounts worth thinking about.

Exercise 9 ("I get a feel for it")

The "feel" approach is harder to do well because it tangles the three distinct questions — appetite, price, terms — into one holistic impression, so the underwriter can miss that a risk fails on appetite while being seduced by a good price, or accept an inadequate price because the risk "feels" good. It is harder to defend because the reasoning was never separated or recorded; when the auditor (or a court) asks "why did you write this?", there is no gate-by-gate record, only a reconstruction built after the fact. Deciding on the page, in sequence, makes the file reconstructable (the reconstruction test) — and the defense becomes simply the record of how you actually thought.

Exercise 11 (decline disguised as a quote)

"In appetite but cannot be priced adequately" means gate 1 passed (the company wants the kind of risk) but gate 2 failed (the only premium the market will bear is below the indicated rate the risk requires). Writing it anyway is a decision to take the risk at a price you know is inadequate — i.e., to write a loss — because rate adequacy (Ch.11) means the premium must cover expected losses plus expenses plus profit, and an inadequate rate guarantees the losses arrive uncovered, on schedule, two or three years later. The professional move is to recognize that such a "quote" is really a decline, and decline it (or hold the adequate rate and let the broker walk).

Exercise 13 (the over-authority bind request)

You do not bind it. The \$20M line exceeds your \$10M authority, so it is not yours to bind — binding it would create an obligation the company never authorized (the failure direction of §13.2). The correct path is to refer it now, as a complete recommendation, to the underwriter who holds the authority for that limit, and let them decide before the deadline if they can. "Bind today, refer tomorrow" is backwards because binding is what creates the obligation; once you have bound past your authority, the company may be on risk (apparent authority) for an exposure it never agreed to, and the referral "tomorrow" cannot unwind a bind that already attached. Authority gates the bind, not the reverse.

Exercise 15 (find the red flag — the silent condition)

The error is a silent condition (an "uncleared subjectivity treated as met"): the underwriter intended to require sprinkler certification but bound clean, with no such condition in the binder or policy and no tracked subjectivity. A condition that lives only in the underwriter's notes — or only in their intention — is not a condition at all; the company is on risk for an unverified protection, and if a fire occurs there is nothing conditioning coverage on the certification. What should have been done: attach the sprinkler certification as a documented subjectivity (condition precedent), with an owner, a deadline, and a tracked status, and not record a clean unconditional bind until it cleared — and/or write a protective-safeguards endorsement (Ch.5) making coverage contingent on the system being maintained and in service.

Exercise 16 (write the declination — the trucking fleet)

A model answer (tone and content; adapt the specifics):

Dear [Broker], Thank you for the opportunity on [Insured], the 30-unit long-haul fleet. After a full review, we are unable to offer terms on this account. The decision is based on the loss experience: three at-fault losses over the past two years, including one with a serious injury, place the frequency and severity outside our current appetite for this class, and no premium the market will bear would be adequate for that exposure. This is not a reflection on you or on [Insured] — it is a risk-selection decision on the loss history. If the operation implements a documented telematics and driver-management program and runs a clean year, we would be glad to look again at renewal; a nonstandard market may also be able to write it in the interim. Thank you again, and please keep sending us your submissions. — [Underwriter]

Note the four features: prompt and in writing; a clear risk-based reason (loss frequency/severity vs. appetite, gate 2 inadequacy); relationship-protective ("not a reflection on you… keep sending"); and a constructive door left open where genuine.

Exercise 18 ("the neighborhood is bad")

This reason is dangerous because it is geographic and demographic rather than property-specific: declining based on the character of the neighborhood rather than the risk of the property is the historical practice of redlining (Ch.35) and can constitute unfair discrimination (Ch.4) if it stands in for a protected characteristic. A legitimate risk-based reason attaches to the property or operation itself — its construction, its protection class, its specific loss history, its measured catastrophe exposure — facts about this risk, not about who lives nearby. The two legal concepts in play are unfair discrimination (pricing or selecting on a prohibited basis rather than on risk) and redlining (its geographic form). Decline for the property's own risk characteristics, documented, or do not decline on this basis at all.

Exercise 21 (repricing vs. restructuring)

Repricing adjusts the premium for the risk as-is — e.g., a schedule debit on a coastal property for the aging roof, or an experience load for the loss history. It says "the risk is what it is; here is what it costs." Restructuring changes the risk itself through terms — e.g., a 5% named-windstorm deductible that transfers the most correlated layer back to the insured, a sublimit capping an exposure you can't fully price, or a required control. It says "I won't insure the risk you brought; I'll insure this modified risk." Restructuring can make a risk writable when repricing alone cannot because some exposures have no adequate price (too uncertain, market too soft) but can be made acceptable by structurally removing or capping the worst layer — the deductible that takes out the attritional/correlated loss is the classic example.

Exercise 23 (forgotten subjectivity)

A forgotten subjectivity "is worse than none" because it creates the appearance of a controlled risk while the control never materialized: the underwriter (and the file) believe the IR scan / hot-work program / roof contract is handled, so the risk is treated as the improved version — when in fact it is the original, uncontrolled risk, now bound and mispriced for a control that doesn't exist. Honest acknowledgment that nothing is yet in place at least keeps the risk visible. The three things every subjectivity must have: an owner (who is responsible for clearing it), a deadline (by when), and a tracked status (open / received / satisfied / waived) — and any waiver is itself a documented (often referred) underwriting decision.

Exercise 25 (the Harbor Steel subjectivities)

• (a) Infrared (IR) electrical scan addresses the 2021 electrical fire and the inspection's flag on the wiring; it verifies the panels are not running hot before you go on risk, converting an unverified electrical hazard into a verified (or remediated) one.
• (b) Hot-work permit program addresses the 2023 hot-work/welding fire — the larger loss and the signal that management's welding-safety controls were inadequate; requiring a documented permit program changes the behavior that caused the loss, not just the price of it (the morale-hazard lever from Ch.1).
• (c) Roof-replacement contract within 12 months addresses the 30-year-old original roof in a named-windstorm zone — the wind/water loss driver; the warranty commits the insured to remove the hazard, and the ACV roof endorsement until then keeps the insurer from paying replacement cost on a roof at the end of its life. In each case the subjectivity pushes the risk toward the version you are actually willing to insure.

Exercise 26 (the complete decision file)

The decision-file checklist (from §13.5): submission (application, broker note, date, the ask); information (loss runs, inspection/loss-control, MVRs, financials, SOV); assessment (the risk grade and its COPE/hazard/control basis); analysis (loss math, credibility, indicated rate build-up); the decision (accept/decline/modify, dated, with the decision-maker); the rationale (the why — the three gates answered); terms (limits, deductibles, sublimits, endorsements); subjectivities (each with owner, deadline, status); authority (whose authority bound it; any referral/peer review/escalation); communication (what was sent to the broker, when). The rationale (the why) is the most important and most neglected because systems capture what (bound, \$20M, 5%) well and why (the appetite call, the rate gap, the hazard the terms address) poorly — and the why is exactly what a hard question (from an auditor, manager, or court) demands and what the reconstruction test checks for.

Exercise 27 (contemporaneous vs. reconstructed documentation)

Contemporaneous documentation is written at the moment of decision, before the outcome is known; it is evidence of what the underwriter actually thought. Reconstructed documentation is written later — often after a loss — and inevitably reflects hindsight and self-interest. Auditors and courts treat them differently because a contemporaneous note is reliable evidence of the decision's quality on the information then available (the standard underwriting is judged by), whereas a post-loss reconstruction looks exactly like what it is: a justification assembled to defend an outcome. The practical rule: write the why in the file before you communicate the decision to the broker.

Exercise 29 (a complete-recommendation referral)

A model paragraph:

Referral — [Insured], commercial package. I recommend we write this account, modified, at the debit-rated indication: agreed-value property (\$20M building) with a 5% named-windstorm deductible and an ACV roof endorsement, GL with products, WC at a debit X-mod with a return-to-work credit, commercial auto with mandatory telematics, a \$10M umbrella, and a modest cyber add-on — subject to five subjectivities (inspection, IR electrical scan, sprinkler certification, hot-work permit program, and a signed roof-replacement contract with replacement within 12 months). Full analysis, COPE, loss-math, and pricing are in the file. I need your authority for the \$20M property line and your sign-off on the Port Hadley catastrophe-zone aggregate. One open question: should the IR-scan condition be a strict condition precedent, or a 30-day post-bind warranty? — [Underwriter]

The point: the senior underwriter is now deciding, not redoing the analysis — the recommendation, rationale, requested authority, and open question are all on the page. That is what makes referral fast.

Exercise 31 (the model override rationale)

A defensible override rationale, for the file:

Model score: 7/10 (decline-leaning). Decision: write at an effective 6, modified, with subjectivities. Override basis: the model's score is driven by the two fire losses and the aging roof, which it reads as a level. Reading the file, both fires point to controllable hazards — one electrical, one hot-work — and the broker has delivered (or committed to) the corrective measures the subjectivities require (IR scan, hot-work program, roof-replacement contract). The model cannot see the management/operational change behind the loss dates, nor the counter-offer terms that transfer the worst exposure (the 5% wind deductible, ACV roof). With those controls required as conditions precedent, the residual risk is consistent with a 6. Referred to senior UW for the property-line authority and cat-aggregate sign-off.

What makes it defensible rather than "a good feeling": it names the specific features of this account that fall outside what the model could see (the trajectory the model read as a level; the corrective contracts; the restructured terms), and it ties the new grade to conditions (subjectivities) rather than optimism. A defensible override is more rigorous than the model, not less — documented, reasoned, grounded in the file — because overriding a validated model means claiming your judgment beats its statistics on this risk, and the only honest way to claim that is to show your work.

Exercise 33 (defending to a pushy broker without folding)

You defend on the logic of the structure in the client's interest, not on apology or capitulation: explain that the necessary terms are what make the account writable at all given its risk ("the wind deductible isn't us being difficult — without it there is no market here, not a cheaper one"; "the roof subjectivity is what lets us hold this price instead of loading it further"), so the client's real alternative to your terms is not better terms elsewhere but no adequate coverage. You stay warm, responsive, and collaborative on everything that doesn't touch the risk — turnaround, service, explanation — while holding the adequate rate (Ch.11) and the necessary terms, because those are the discipline that keeps the account (and the book) profitable. Brokers respect the underwriter who explains and holds and learn to roll the one who folds; folding to the push is how a soft market degrades a whole book.

Exercise 34 (Underwriting-File extension — Harbor Steel through the gates)

Gate 1 (appetite): coastal metal-fabrication with a fire history is at the edge of appetite, not outside it — but the Port Hadley catastrophe-zone aggregate is a senior-authority question (flag, don't resolve). In appetite, subject to the cat-zone sign-off. Gate 2 (price): the debit-rated indicated premium is achievable post-non-renewal. Priceable adequately. Gate 3 (terms): the marginal features (roof, two fire hazards, unverified protection) are fixable with terms + subjectivities. Terms close the gap. Decision, one sentence: Modify — issue a counter-offer for agreed-value property with a 5% named-windstorm deductible and an ACV roof endorsement at the debit-rated price, conditioned on five subjectivities, and refer the decision to senior underwriting for the \$20M property-line authority and the cat-zone aggregate. The five subjectivities, each with its open status fields (owner / deadline / status):

  1. Satisfactory inspection / loss-control survey      owner: ___  deadline: ___  status: open
  2. Infrared (IR) electrical scan, no critical finding owner: ___  deadline: ___  status: open
  3. Sprinkler certification (inspected, tagged)        owner: ___  deadline: ___  status: open
  4. Hot-work permit program documented/implemented     owner: ___  deadline: ___  status: open
  5. Roof-replacement contract signed; replace ≤12 mo   owner: ___  deadline: ___  status: open
     (ACV roof endorsement applies until replacement is warranted)

(Do not pre-empt the final bound disposition — that is the Chapter 40 capstone. A "quote-with-conditions, referred" is the complete, defensible Chapter 13 decision.)

Exercise 35 (why Harbor Steel must be referred)

It must be referred because the \$20M property line exceeds the limit a line underwriter is authorized to bind, and the Port Hadley catastrophe-zone exposure is a mandatory-referral trigger (writing into the zone affects the carrier's accumulation and reinsurance, which are senior-authority concerns). Referral makes it neither a worse account nor a slower one: the line underwriter still does the full analysis, pricing, terms, and subjectivity package and proposes the decision; referral simply joins the right authority to that work, so the bigger, more correlated risk gets more-experienced judgment and a check on the cat aggregate. Authority is layered on purpose; using it correctly is discipline, not deficiency.

Exercise 37 (why "quote-with-conditions, referred" is a complete Ch.13 decision)

Because the underwriting decision — accept/decline/modify — has been made (modify), structured (the counter-offer terms and the five subjectivities), documented (rationale through the three gates), and sent to the right authority (the referral). Coverage is not yet cleanly bound only because (a) the subjectivities are conditions precedent that must clear and (b) the senior authority must sign off — both of which are the correct state for a risk of this size at this stage. The Chapter 13 skill is reaching and defending the decision, not finalizing the bind; the bind itself comes after the conditions clear and the broker accepts (Ch.39), and the capstone (Ch.40) assembles the whole file and states the bound disposition. A decision can be complete and defensible while coverage is still conditional.

Chapter 14

Worked solutions to the daggered (†) and odd-numbered exercises. All dollar figures are constructed teaching examples. Section references point back to index.md.

Exercise 14.1 (†)

The four principal coverage parts of the PAP: (1) Liability (Part A) — pays for bodily injury and property damage the insured is legally responsible for causing to others; (2) Medical payments / personal injury protection (PIP) — pays medical expenses (and, under PIP, lost wages and more) for the insured and passengers regardless of fault; (3) Uninsured / underinsured motorist (UM/UIM) — pays the insured for injuries caused by an at-fault driver with no or insufficient insurance; (4) Physical damage — collision (impact) and comprehensive (theft, fire, flood, hail, glass, animal strike), paying for damage to the insured's own vehicle. The part the state requires is liability (financial-responsibility laws). The first-party parts are medical payments/PIP, UM/UIM, and physical damage (they pay the insured); liability is third-party (it pays those the insured harms). (§14.1)

Exercise 14.3

Driver class — the rating segmentation capturing the household's operators and their loss-relevant characteristics (historically age, marital status, gender where permitted, and the driving record). Vehicle symbol — a code classifying a make/model/year by its loss characteristics (repair cost, theft attractiveness, damageability), used as the physical-damage relativity for the car. Nonstandard auto — the market segment writing higher-risk drivers (serious/multiple violations, at-fault accidents, suspensions, DUI, lapses, no prior insurance) at higher prices, backstopped by the residual market. (§14.2, §14.6)

Exercise 14.4 (†)

Telematics is the data: the collection of vehicle-operation information (mileage, time of day, braking, acceleration, cornering, speed, distraction) via plug-in device, embedded system, or smartphone app. Usage-based insurance (UBI) is the product: the offering that prices, and sometimes underwrites, using that telematics data instead of (or in addition to) the static proxies. Telematics is the input; UBI is the "pay how you drive" policy built on it. (§14.4)

Exercise 14.5

The driving record is a weak predictor for two reasons. (1) It is a low-frequency, lagging signal: serious violations and at-fault accidents are rare events, so a clean record is the default — most drivers have one, including most drivers who will have an accident next year. (2) It records only what a driver was caught doing, not how they actually drive day to day; the continuous behavioral signal (speed, braking, distraction, mileage) is invisible to the record and only becomes visible through telematics. So a clean record discriminates poorly between two otherwise different risks. (§14.2)

Exercise 14.7

(a) The insured rear-ends another car and injures its driver → liability (third-party: bodily injury + property damage to others). (b) Hail dents the insured's own roof → comprehensive / other-than-collision (first-party). (c) An uninsured driver runs a light and injures the insured → uninsured/underinsured motorist (UM/UIM) (first-party). (d) The insured's own car is stolen → comprehensive (first-party). (§14.1)

Exercise 14.9

Three legitimate reasons a renewal premium rises with no change in the insured's own driving: (1) a territory trend — rising theft, traffic density, weather (comprehensive) losses, or a worsening injury- claim/litigation climate in the garaging territory; (2) a vehicle-cost trend — the cost to repair or replace the insured's model rose (e.g., sensor-laden parts, higher used-car values lifting total-loss settlements); (3) a book-wide severity trend — medical, repair, and litigation inflation pushing the liability and physical-damage loss costs up across the whole pool, recovered through an approved rate increase. None of these is the insured's record. (§14.1, §14.7)

Exercise 14.11 (†)

Multivariate rating is necessary because young-driver and high-performance-vehicle factors are correlated — young drivers disproportionately drive high-symbol cars. A one-factor-at-a-time (univariate) analysis estimates each factor's apparent effect while the other factor is also varying in the background, so it double-counts the overlap: it credits the young-driver factor with risk that is really the car's, and the vehicle factor with risk that is really the driver's. A GLM (Chapter 32) estimates each factor's effect holding the others constant, so the plan does not penalize the overlapping risk twice. This is why a gut sense that "this single relativity is too high" is usually wrong — the relativity was fit in the presence of all the others. (§14.2; Ch. 32)

Exercise 14.13

The univariate intuition is usually wrong because each filed relativity was estimated in the presence of every other factor (multivariate / GLM fitting, Chapter 32), not in isolation. When the broker mentally "isolates" one factor and judges it too expensive, they are imagining a univariate world that the plan does not live in: the factor's value reflects its marginal effect holding the others constant, and the correlations with other factors have already been accounted for. So the act of isolating the factor — the very move that makes it "look" too high — is the thing that introduces the error. The disciplined response is to explain that the relativity is not a standalone price tag but a coefficient in a joint estimate. (§14.2)

Exercise 14.15 (†)

(a) B is the higher expected loss, on multiple factors: younger driver class (steep age-related crash frequency), higher vehicle symbol (high-performance coupe — costly repairs, theft, harder driving, worse injury outcomes), higher-cost rating territory (dense, high-theft urban), more miles (greater exposure), a coverage lapse (behavioral + selection signal), and minimum prior limits (correlates with a worse risk). A is the steadier risk on every one of these. (b) Yes — both are writable in the standard market, but at very different prices; the matching clean records should not flatten a real difference in expected loss. (c) Neither application reveals how either actually drives day to day — the continuous behavioral signal (speed, braking, distraction, real mileage) is invisible to the record and the application; only telematics (§14.4) would surface it. (§14.2, §14.4)

Exercise 14.17 (†)

Pricing the two "nonstandard" applicants the same is an error because nonstandard is a wide distribution, not a single risk. The first applicant (a five-year-old DUI, clean since, employed, continuous coverage, modest vehicle) is an improving risk well below the segment's average; the second (three at-fault accidents in two years, current lapse) is a deteriorating risk well above it. Charging both the same does one of two harmful things: it over-prices the improving driver, who then shops and leaves — taking the better-than-average risk out of your pool and leaving you adversely selected (Chapter 1) — or it under-prices the deteriorating driver, whose higher losses then arrive on schedule. The whole craft of nonstandard is classifying within the segment and re-rating as the risk changes; treating it as a monolith guarantees you keep the wrong half. (§14.6; Ch. 1)

Exercise 14.18 (†)

Relativities are multiplicative, applied to the base rate. (a) Premium = \$800 × 1.45 × 1.30 × 1.20 × 0.95 × 0.90. Step through: \$800 × 1.45 = \$1,160; × 1.30 = \$1,508; × 1.20 = \$1,809.60; × 0.95 = \$1,719.12; × 0.90 = \$1,547.21** (≈ \$1,547). (b) The single largest upward contributor is the driver-class factor (1.45)**, which here is pricing the operator's loss-relevant characteristics (e.g., a younger or higher-risk driver class) — the steep age/experience-related crash-frequency mechanism. (The vehicle symbol at 1.30 is close behind, pricing repair/theft/damageability.) (§14.2)

Exercise 14.19

(a) Premium = \$800 × 0.85 × 0.90 × 0.80 × 1.00 × 0.90. Step through: \$800 × 0.85 = \$680; × 0.90 = \$612; × 0.80 = \$489.60; × 1.00 = \$489.60; × 0.90 = \$440.64** (≈ \$441). (b) Ratio to the Exercise 14.18 applicant ≈ \$1,547 / \$441 ≈ 3.5×. Two drivers with the same clean record differ by roughly three-and-a-half times because the price is built almost entirely from factors other than the record: driver class (age/experience), vehicle symbol (the car), territory (where it's garaged), and use. The record is the factor they share; the price is everything else. This is the chapter's title in arithmetic. (§14.2)

Exercise 14.20 (†)

(a) Loss ratio = (\$7,200,000 + \$600,000) / \$10,000,000 = \$7,800,000 / \$10,000,000 = 0.78 (78%). Expense ratio = \$2,600,000 / \$10,000,000 = 0.26 (26%). Combined ratio = 0.78 + 0.26 = 1.04 (104%). (b) No — it is unprofitable on underwriting: the carrier paid \$1.04 in losses and expenses for every \$1.00 of premium earned, a 4-cent underwriting loss on the dollar (before any investment income). (c) To reach a 98% combined ratio at the same 26% expense level, the loss ratio must fall from 0.78 to about 0.72 — roughly a 6-point improvement (which rate increases aim to produce). The force pushing the other way: the loss-cost trend (medical, repair, and litigation severity inflation) is simultaneously lifting the loss ratio, so the rate increase must overcome the trend just to hold ground. (§14.7; Ch. 3)

Exercise 14.21

Manually discounting the filed \$1,640 to \$1,300 because the owner is a valued commercial client is wrong because the personal-auto rate must reflect the filed, risk-based plan for that household — and the commercial relationship is not a risk characteristic of the household's driving. Deviating from the filed rate for a reason unrelated to the risk is unfair discrimination (Chapter 4): charging two similarly-situated insureds different prices for a non-risk reason. It also exposes the carrier to a market-conduct finding (a regulator looks for exactly this) and, if it became a pattern, undermines rate adequacy. The right move is to charge the household its filed price and pursue the account relationship through legitimate, filed multi-policy credits that apply to everyone who qualifies. (§14.7, The Underwriting File; Ch. 4)

Exercise 14.23

Two distinct errors. (1) "Credit must be junk." The factor was not found unpredictive — it was restricted on fairness grounds despite being predictive (§14.3). Banned ≠ not predictive; the state made a social-fairness judgment that the cost of using a predictive-but-contested factor outweighed its benefit there. (2) Inferring the competitor "uses credit" from price patterns — and, worse, contemplating matching it — risks using credit (or a proxy for it) in a state that prohibits it, which is a compliance violation that can cost the carrier its license. The disciplined posture is to hold both facts at once (predictive and restricted) and to rate strictly within what the state permits as filed. (§14.3, §14.5)

Exercise 14.25 (†)

A model adverse-action explanation to the applicant:

"Your premium is higher than our base rate in part because of a credit-based insurance score — a number drawn from your credit history and calibrated specifically to predict insurance losses. Your state permits us to use it, and it is one of the factors in your price. We're required to send you this notice because credit information affected your premium, and you have the right to a free copy of the credit report it was based on and to dispute any errors in it. If a major life event — a divorce, a serious medical episode, identity theft — has temporarily harmed your credit, our 'extraordinary life event' provision lets you ask us to set the credit factor aside with documentation. The score is only one part of your rate; your vehicle, where the car is garaged, how it's used, and your coverage history all factor in as well."

The tone is plain and non-defensive; it states why credit was used (permitted + predictive), discloses the applicant's FCRA rights, and surfaces the extraordinary-life-event remedy. (§14.3)

Exercise 14.27

A disciplined answer to the market-conduct examiner: "Our rate increase reflects the loss-cost trend in this state — rising medical, repair, and litigation severity that has outrun our prior rates — and is what rate adequacy requires to keep the line from running at an underwriting loss. The increase was filed and [approved/under review] per the state's rate-approval process; it is not a growth or margin target but the rate the forward loss trend supports." It is grounded in trend and adequacy (Chapter 11), not in volume or market share. (§14.7; Ch. 11)

Exercise 14.29

Low-cost auto programs for low-income drivers and a deliberately high-priced residual market together express the "insurance serves a social function" theme (Chapter 1): auto is mandatory, so the state has a public interest in keeping the mandate from pricing low-income people out of legal driving (and often out of getting to work) — hence subsidized low-cost programs — while still pricing the insurer of last resort high enough that it remains a last resort, not a destination that drains the voluntary market. But these are policy mechanisms layered on top of the risk; they do not change the underwriting math on any individual risk — the expected losses are what they are. Society chooses to absorb or redistribute some of that cost for a social end; the underwriter still prices each risk to its expected loss within the rules. (§14.6; Ch. 1)

Exercise 14.31 (†)

The Underwriting File — account rounding on the owner's personal auto. (a) Different underwriting logic. The commercial fleet (Chapter 23) is underwritten as an operation: you assess the drivers as a qualified pool, the radius and cargo, DOT compliance, the fleet's own loss runs, and the nuclear-verdict exposure, then structure terms (telematics, removing a high-risk driver). The owner's personal auto is underwritten by classifying and rating the household into the filed plan: the driver classes (including the young adult driver), the vehicle symbols, the rating territory of the home, the use and mileage, the prior-insurance history, and — subject to the state's rules — credit and a telematics option. The commercial risk is assessed and structured; the personal risk is classified and rated. (b) Why charge the filed price. The household must pay the price its risk earns under the filed plan, because the commercial relationship is not a risk characteristic of the household's driving. Discounting the personal auto because the owner is a valued commercial client would be unfair discrimination (Chapter 4) — a deviation from the filed rate for a non-risk reason — and a market-conduct exposure. The personal umbrella (Chapter 16) will sit over both the personal auto and the home, and its underlying-limit requirement reaches back into this placement: the auto liability must be written at limits high enough to satisfy the umbrella's required underlying limit, or there is a gap the umbrella won't fill. So the personal-auto limit decision is not standalone — it is set partly by what the umbrella above it demands. (§14.2, The Underwriting File; Ch. 4, Ch. 16)

Chapter 15

Worked solutions to the daggered (†) and odd-numbered exercises. All figures are illustrative teaching examples. Section references point back to index.md.

Exercise 1

HO-3, HO-5, HO-6. HO-3 (the workhorse) covers the dwelling on an open-peril basis and the contents on a named-peril basis — the split that defines it. HO-5 (the premium form) covers both dwelling and contents on an open-peril basis. HO-6 is the condominium unit-owner's form, covering the unit interior, contents, liability, and loss assessment, and is distinguished by who the insured is (a unit owner who does not own the building shell) rather than by peril basis. (§15.1)

Exercise 3

The four COPE elements, with homeowners examples: Construction — frame vs. masonry, roof age/shape/ covering, hurricane straps, defensible-space materials. Occupancy — owner-occupied (preferred) vs. tenant-occupied, seasonal/vacant, or home-business use. Protection — fire protection class (responding department, water supply, hydrant/station distance), central-station alarms, community wildfire response. External exposure — the brush-covered hillside above, the river behind, the proximity to the coast — for homeowners, mostly where the catastrophe peril enters. (§15.2)

Exercise 4

A catastrophe peril produces severe, correlated losses across many insureds from a single event (hurricane, wildfire, earthquake, flood). The law of large numbers stabilizes the aggregate only when exposures are independent — when one loss tells you nothing about the next, as with kitchen fires. A catastrophe violates independence: one hurricane strikes thousands of insureds at once, so the losses are not many independent draws but a single event wearing thousands of costumes. A large book of correlated catastrophe exposures is therefore not a pool but one enormous bet, which is why catastrophe needs its own machinery (models, reinsurance, capital). (§15.5)

Exercise 5

Covered by the standard HO policy: hurricane wind and wildfire (both as forms of covered perils — windstorm and fire). Excluded: earthquake (separate policy/endorsement) and flood (always, including storm surge). Note the trap embedded here: the hurricane's surge is excluded even though its wind is covered. (§15.5)

Exercise 7

A named-storm/wind deductible applies specifically to loss from a named storm or windstorm and is expressed as a percentage of the dwelling limit (e.g., 5% of Coverage A), whereas the all-other-perils deductible is a flat dollar amount (e.g., \$1,000). The percentage structure both shifts a meaningful, scaling share of every catastrophe loss back to the insured and grows the retained amount with the value at risk. (§15.7)

Exercise 9

Open-peril (special form): the loss is covered unless the insurer can point to an exclusion — the insurer bears the burden of proving an exclusion applies. Named-peril: the insured must prove a listed peril caused the loss. HO-3 uses open-peril on the dwelling because the structure is the high-value, broad- exposure asset where the insured should not have to prove the cause of every odd loss; it uses named-peril on contents because contents losses are more varied, more theft- and mysterious-disappearance-prone, and cheaper to cover on a defined list — and the difference is also a price/selection lever (HO-5 upgrades contents to open-peril for the best risks). (§15.1)

Exercise 10

Warn the collector that Coverage C (personal property) is typically a default percentage of Coverage A (say 50%), which may be far too little for a serious collection, and that high-value categories — jewelry, fine art, firearms, furs — hit special sublimits that cap theft recovery at a few thousand dollars regardless of the overall Coverage C limit. The fix is to schedule the items (scheduled personal property / a personal-articles floater — Chapter 16), which insures each listed item to an agreed value, on a broader (often open-peril, worldwide) basis, outside the sublimit. The cheapest form is the wrong economy for this household. (§15.1)

Exercise 13

Roof replacement cost new = \$28,000; age 12 of a 20-year life → 60% of life remaining, 40% depreciated … wait, check: 12/20 = 60% used, so 40% remains. ACV = replacement cost × remaining-life fraction = \$28,000 × 0.40 = **\$11,200. Replacement-cost settlement = \$28,000 (no deduction). Under ACV the insured funds \$16,800** themselves; under replacement cost they fund **\$0** (beyond the deductible). The larger the depreciation, the larger the gap the insured absorbs — which is exactly why the valuation basis on an aging roof is an underwriting decision, not a clerical one. (§15.2)

Exercise 15

Coinsurance recovery on a partial loss = loss × (coverage carried ÷ coverage required). Required = 80% × \$600,000 = **\$480,000. Carried = \$420,000. Ratio = 420,000 ÷ 480,000 = 0.875. Recovery = \$120,000 × 0.875 = **\$105,000 (before the deductible). The homeowner absorbs \$15,000 of a loss they believed was fully covered, and experiences the coinsurance penalty as a betrayal even though it was in the contract — which is why the disciplined underwriter verifies Coverage A against a rebuild estimate up front. (§15.4)

Exercise 17

Demand surge is the spike in the local cost of labor and materials that follows a catastrophe, when an entire region's worth of damaged homes must be rebuilt at once and bids up the price of construction. It makes a Coverage A limit that was adequate at inception inadequate after the event: the same house now costs more to rebuild than the policy limit, so even an accurately-valued home can fall short exactly when the loss is widespread. This is why limits must be revisited at renewal (especially in inflationary periods) and why extended/guaranteed replacement cost endorsements exist — to protect the adequately-insured homeowner against the surge. (§15.4)

Exercise 19

Location is the most powerful homeowners rating-factor family, and it is more powerful than its personal-auto equivalent because it drives two things at once: the everyday losses (through protection class, local rebuild costs, theft and weather frequency) and the catastrophe exposure, which can dwarf everything else and which personal auto does not face in the same way. Two identical houses, one inland and one on a barrier island, are not the same risk at any price — the second carries a hurricane exposure the first does not. That catastrophe dimension, riding entirely on location, is what makes homeowners structurally harder to write than auto. (§15.3)

Exercise 20

Red flags: (1) Coverage A of \$240,000 on a 2,800-sq-ft custom home badly understates rebuild cost — a classic ITV failure that will trigger the coinsurance penalty on a partial loss and under-collect premium; require a credible replacement-cost-estimator output, not the purchase price. (2) Three water claims in four years is a pattern (not noise) pointing to plumbing, maintenance, or filing propensity; pull the CLUE history, ask about the plumbing, and weight it in selection. (3) A 1998 roof is at/near end of life; endorse it to ACV (or schedule it) rather than writing full replacement cost on wear that's already coming. Together they describe a risk that is writable but only at corrected valuation, with a roof endorsement, and with the claims pattern understood — not an automatic decline, but not an as-submitted accept either. (§15.2, §15.3, §15.4)

Exercise 21

A pattern of small, frequent, non-weather claims is signal — it predicts future claims, because it reflects something durable about the property (plumbing, wiring, maintenance) or the insured (a propensity to file). A single large weather claim is mostly noise — the storm hit, the insured had no control over it, and it says little about next year. This is the credibility logic of Chapter 10: weight experience by how much genuine signal it carries. For a single dwelling, even a "pattern" is a tiny sample, so you lean hard on the class rate and treat the dwelling's own record as a modifier, not the whole story. (§15.3; Ch. 10)

Exercise 22

Tell the colleague that credit-based insurance scores are permitted in most states but restricted or banned in several (California, Maryland, and Massachusetts are among those that limit them for property lines), so a multi-state book cannot apply the factor uniformly — it must follow each state's rules. Where the factor is used, the FCRA requires an adverse-action notice to the applicant whenever an unfavorable score raises the premium or results in a denial, disclosing that a consumer report was used and how to obtain it. And the deeper caution (Chapter 35): even where legal, the factor raises a genuine proxy-discrimination concern because credit correlates with income and through it with protected characteristics. (§15.3; Ch. 4, Ch. 35)

Exercise 23

Hurricane — covered (wind); lever: percentage wind deductible + accumulation control (surge is excluded flood). Wildfire — covered (fire); lever: selection (defensible space, ignition-resistant materials, community protection) and availability (non-renewal in the worst zones). Earthquake — excluded; lever: separate policy/endorsement with a large percentage deductible. Flood — excluded always; lever: require a separate flood policy (NFIP or private) and read the flood zone. (§15.5)

Exercise 25

Flood fails several insurability criteria from Chapter 1 about as completely as any peril: it is severe (a flood often totals the lower floors), correlated (one event inundates an entire flood plain at once), concentrated (the people who need it live in known, mapped flood zones), and historically hard to model. A private insurer offering ordinary flood coverage would have attracted exactly the homes most likely to flood — the textbook adverse-selection setup — at a price the market could not bear, and the death spiral would have followed. So in 1968 the federal government created the NFIP to make flood coverage available where the private market would not, mapping the zones, tying coverage to floodplain management, and mandating it on federally backed mortgages in high-risk areas. (§15.6; Ch. 1)

Exercise 27

Decision: quotable, not as submitted — conditioned on flood coverage. The wind exposure prices cleanly and the three-year-old hip roof is favorable (hip sheds wind better than gable; recent roof reduces wind/water loss). But the home sits in a designated high-risk flood zone, and storm-surge flood is the most probable total-loss scenario — and it is excluded from the homeowners policy. Condition: confirm a separate flood policy (NFIP, or private flood given the \$700,000 value the NFIP cap may under-protect) is in place before binding, and document that the gap was identified and communicated. Also apply a percentage named-storm/wind deductible appropriate to the zone. The policy I am binding covers the wind; the peril most likely to destroy the home is the water, which my policy does not cover — so the flood policy is the condition that makes this a responsible write rather than a half-done one. (§15.5, §15.6, §15.7)

Exercise 29

Four hardening-market levers: (1) Raise deductibles, especially the percentage named-storm/wind deductible — shifts catastrophe loss back to the insured and reduces net exposure. (2) Narrow coverage via exclusions and endorsements (ACV roofs, cosmetic-damage exclusions) — pulls the promise back toward fortuitous loss. (3) Tighten selection / restrict new business (roof-age maximums, mandatory mitigation, ITV verification, writing fewer homes per zone) — improves quality and manages accumulation. (4) Non-renew or withdraw — the most drastic, removing exposure entirely. (§15.7)

Exercise 30

A percentage named-storm deductible does two jobs at once: it shifts a meaningful share of every catastrophe loss back to the insured (reducing the carrier's net catastrophe exposure and reconnecting the insured to the loss — the skin-in-the-game logic of Chapter 1), and it scales the retained amount with the value at risk (a bigger home retains more). On a \$450,000 dwelling with a 5% named-storm deductible, the homeowner absorbs \$22,500 before coverage responds. (§15.7; Ch. 1)

Exercise 31

(Ethics dilemma — model answer sketch.) The genuine tension: letting rates rise to the risk-based level preserves the carrier's solvency and willingness to write (so coverage stays available), reflects the true catastrophe risk, and avoids cross-subsidy — but prices some coastal homeowners out of coverage they can afford. Holding rates down preserves affordability and access for existing homeowners and recognizes that many cannot simply move — but, held below the risk, it drives carriers to non-renew and withdraw, so the coverage does not get cheaper, it disappears, and the risk lands on the residual market and ultimately on taxpayers. There is no costless answer because the catastrophe cost is real and must be borne by someone: suppress the price and you do not eliminate the cost, you relocate it (to taxpayers, to surviving insurers via assessments, or to the uninsured homeowner after the storm). The honest position holds both halves and refuses to pretend either is free. (§15.7; Ch. 3, Ch. 35)

Exercise 33

A swelling FAIR Plan / residual market means the private market has retreated — carriers are non-renewing and declining, and homeowners who cannot find private coverage are falling back on the insurer of last resort. It is a safety net (it keeps homeowners covered when no one else will write them) and a warning sign (it concentrates catastrophe risk on a public/quasi-public entity that is least diversified to bear it, with the cost ultimately on taxpayers or on surviving insurers through assessments, and it signals a market whose price and risk have come apart). A healthy market is one in which the residual market is small. (§15.7)

Exercise 35

The gap is flood (storm surge): the home's HO-3 covers hurricane wind but excludes the water, and on that coast the surge is the most probable total-loss scenario — the lower floors filling in a major landfall. It can be filled two ways: an NFIP flood policy, or a private flood policy (worth considering given that a \$600,000 home may exceed the NFIP coverage cap). It is the most probable total loss because surge inundates the structure completely in a way wind damage often does not, and because the home sits in a coastal zone where surge is the defining hazard — which is exactly why the chapter insists the flood policy is not optional account-rounding but the coverage that addresses the peril most likely to destroy the home. (The Underwriting File, §15.6)

Chapter 16

Worked solutions to the daggered (†) and odd-numbered exercises. All dollar figures are constructed teaching examples. Section references point back to index.md.

Exercise 16.1 (†)

A personal umbrella is a personal liability policy that provides an additional layer of coverage above the liability limits of the underlying auto, home, and other personal policies, responding only after those underlying limits are exhausted (and, for certain offenses the underlying excludes, dropping down to act as primary). Excess liability differs from primary in when it responds: the primary policy pays first, from the first dollar of covered loss (subject to a deductible); the excess/umbrella layer pays only the portion of the loss that exceeds the underlying limit, up to its own limit. The umbrella is a second layer, not a standalone policy — its value depends on the layer beneath it. (§16.1)

Exercise 16.3 (†)

Scheduled personal property is high-value items individually listed (scheduled) and valued on the policy or a separate personal-articles floater, insured for their scheduled amount. Scheduling does four things blanket homeowners contents coverage cannot: (1) removes the category sublimit (the scheduled \$15,000 ring is covered for \$15,000, not the homeowners form's ~\$2,000 jewelry cap); (2) broadens the perils to open-peril / all-risk, including mysterious disappearance and accidental breakage (so a ring lost down a drain, not stolen, is covered); (3) usually waives the deductible for scheduled items; and (4) can fix the value in advance on an agreed-value basis, avoiding a post-loss valuation dispute. (§16.6)

Exercise 16.5

An umbrella "drops down" when it covers a claim the underlying policy excludes entirely — so there is no underlying limit for the umbrella to sit above. Instead of attaching above an underlying limit, the umbrella acts as primary for that claim: the insured pays a self-insured retention (commonly a few hundred to a few thousand dollars), and the umbrella pays the rest from the first dollar above the retention, up to its limit. Crucially, drop-down reaches only claims the umbrella form affirmatively covers (e.g., a defamation "personal injury" offense) — not everything the underlying excluded. (§16.3)

Exercise 16.7

Agreed value (as applied to personal valuable articles) means the insurer and insured agree, at the time the policy is written and usually with a professional appraisal, on the item's value — and that agreed amount is what the insurer pays on a total loss, with no depreciation argument and no post-loss valuation fight. A unique painting is best insured this way because it has no meaningful replacement cost (it is irreplaceable), so a replacement-cost or actual-cash-value basis leaves the value to be argued after the loss, when the item is gone — a recipe for dispute. Agreed value converts an unanswerable question ("what was this irreplaceable thing worth?") into a contractual fact settled in advance. (§16.6; note Ch. 19 owns "agreed value" as a commercial-property term — same idea, applied here to personal articles.)

Exercise 16.9 (†)

The stack, with \$500,000 underlying auto and a \$2,000,000 umbrella against a \$3,000,000 judgment: - (a) Underlying auto pays \$500,000 — its full limit (the primary layer / attachment point). - (b) Umbrella pays \$2,000,000** — its full limit; it covers the band from \$500,000 up to \$2,500,000. - (c) The insured personally pays \$500,000** — the judgment is \$3,000,000; coverage reaches \$2,500,000 (\$500K primary + \$2M umbrella); \$3,000,000 − \$2,500,000 = \$500,000 lands on the insured's own assets.

  $3,000,000 judgment
  umbrella    $2,000,000   (covers $500K → $2.5M)
  ───────────────────────  ceiling at $2.5M
  primary     $500,000     (covers $0 → $500K)
  INSURED owes $500,000    (everything above $2.5M)

The umbrella raised the ceiling enormously, but a ceiling still exists. (§16.1)

Exercise 16.11 (†)

A \$1,000,000 umbrella over a \$300,000 homeowners-liability limit provides \$1,300,000 of total coverage for a covered home-liability claim because the two layers stack: the homeowners pays the first \$300,000, and the umbrella pays the next \$1,000,000 on top (the band from \$300,000 to \$1,300,000). The umbrella does not include the underlying \$300,000 — it sits *above* it. The \$300,000 level is the attachment point (the required underlying limit the umbrella attaches above). (§16.1)

Exercise 16.13 (†)

Umbrella requires \$500,000 underlying auto; insured actually carries \$250,000; a \$1,500,000 judgment: - Underlying auto pays \$250,000 (its actual limit). - Umbrella pays \$1,000,000** — but it attaches at the **required \$500,000, "as if" that underlying limit were in force; it covers the band from \$500,000 to \$1,500,000. - The gap is the \$250,000-to-\$500,000 band — \$250,000 — which nobody's policy pays: the auto is exhausted at \$250,000 and the umbrella does not attach until \$500,000. - The insured personally bears that \$250,000 gap, plus nothing else here (the umbrella covered the top). Total paid to the claimant from insurance: \$250,000 + \$1,000,000 = \$1,250,000; the insured owes the \$250,000 gap. This is the gap problem, caused by the insured carrying less underlying than the umbrella requires. (§16.2)

Exercise 16.15 (†)

Correction: "Verifying the underlying limits at issuance is necessary but not sufficient — the gap almost always opens later, when the insured lowers an underlying limit, lets a required underlying policy lapse, or moves coverage to a carrier writing lower limits, all without telling us. So you must re-verify the underlying limits at every renewal rather than assume they still match the umbrella's requirement; an umbrella file that was correct three renewals ago and has never been re-checked is a gap waiting to be discovered." (§16.2)

Exercise 16.17 (†)

Underwrite the umbrella. The exposures that should change the underwriting, in order of severity concern: - The 16-year-old new driver — the dominant umbrella auto-severity exposure; the catastrophic liability claim most often originates in auto, and a brand-new teen driver is the sharpest version. - The unfenced in-ground pool with a diving board — drownings are among the largest personal-lines liability claims; unfenced + diving board elevates the severity materially. - (Secondary) the Rottweiler (breed + the household's other risk features) and the boat.

Structural changes (not a decline) to write it more safely — at least three: 1. Raise the required underlying auto limit from \$250,000 to \$500,000 (and verify the insured carries it). This both reduces umbrella claim frequency and forces real primary protection where the exposure is worst (the teen driver). It also closes the latent gap (the schedule currently shows only \$250K underlying auto — below where you'd want the umbrella to attach for a teen-driver household). 2. Require the pool to be fenced (and consider removing the diving board) as a condition — a genuine loss-control improvement that reduces the severity exposure. 3. Require a scheduled underlying watercraft liability policy for the 28-foot powerboat before the umbrella will sit above it — and consider a higher self-insured retention or a modest surcharge for the overall exposure profile.

The gap by construction: the 28-foot powerboat with no underlying watercraft policy on the schedule — there is a boat exposure but no underlying boat liability for the umbrella to attach above, so a boating liability claim falls into a gap unless underlying watercraft coverage is put in place. (§16.2, §16.4)

Exercise 16.19 (†)

The retired couple's low-activity profile (no pool, boat, dog; one home; clean records) genuinely lowers the frequency of a serious liability claim — but the umbrella's risk is severity, and severity is driven by the "high-target" exposure (§16.4): a household with significant assets is a more attractive and more lucrative litigation target, so the same accident (say, an at-fault auto crash with a catastrophic injury) generates a larger demand against them than against a household with nothing to take. So a \$1 million umbrella may be inadequate despite the quiet lifestyle. Recommendation: size the limit to the assets at risk and the target status, not the activity level — likely \$2 million to \$5 million (or more, layered, if their assets are substantial), over adequate underlying limits. Low activity does not mean low limit. (§16.4)

Exercise 16.21 (†)

A standard cost-per-square-foot valuation is dangerous on a \$12 million custom home because the standard table is built from comparables — and a custom home with imported stone, hand-plastered walls, and irreplaceable millwork has none; its true replacement cost can be far higher than the table estimates, leaving the home badly under-insured to value (the insurance-to-value problem Chapter 15 owns). An undervalued unique home is catastrophic in a total loss. The HNW carrier instead appraises the home individually and frequently writes it on a guaranteed or extended replacement cost basis — paying to rebuild even if that exceeds the stated limit — a promise that is only safe to make because the home was properly appraised first. The appraisal is the underwriting control that makes the coverage grant solvent. (§16.5; Ch. 15.)

Exercise 16.23 (†)

Price/structure this risk (fine-art collection). - Itemized scheduling — advantage: maximum precision; each piece (\$5,000–\$120,000) is described, valued, and covered for its agreed amount, which matters most for the high-value individual pieces and for proving the loss. - Blanket scheduled coverage — advantage: a single agreed limit for "the collection" with itemization only above a threshold; far less administrative burden for a growing, frequently-changing collection, so new acquisitions are covered without re-issuing the schedule each time. - Recommendation: a hybrid — itemize/schedule the high-value pieces (e.g., the \$120,000 and other large items) on agreed value, and write the balance of the collection on a blanket scheduled limit with itemization above a threshold, plus periodic re-appraisal of the scheduled pieces (the collection appreciates). This captures the precision where the value is concentrated and the flexibility where the collection changes — exactly the structure HNW carriers use. (§16.6)

Exercise 16.25

Rule of thumb: schedule an item when its value materially exceeds the homeowners special limit for its category, when it is unique enough that agreed value matters, or when the named-peril/deductible limitations of blanket coverage would cause a meaningful loss. - (a) \$400 watch — leave in the blanket (below any special limit; not worth scheduling). - (b) \$15,000 ring** — **schedule** (far exceeds the ~\$2,000 jewelry sublimit; agreed value useful). - (c) \$30,000 painting — schedule on agreed value (unique; no meaningful replacement cost). - (d) \$200,000 coin collection changing monthly — blanket scheduled** coverage (a single agreed limit with itemization above a threshold), because itemizing a monthly-changing collection is impractical. (§16.6)

Exercise 16.26 (†)

Find the red flag. Two red flags on the renewal schedule: 1. Underlying auto liability dropped from \$500,000 to \$250,000. If the umbrella requires \$500,000 underlying auto, this opens a gap (§16.2): a serious auto judgment now leaves the \$250K–\$500K band uncovered, and the insured would bear it. Exposure: any large auto-liability claim. 2. The watercraft policy is gone but the boat remains (still listed under household exposures). There is now a boat with no underlying watercraft liability for the umbrella to attach above — a gap by construction. Exposure: any boating-liability claim.

What you do before renewing: contact the insured/agent and require the underlying limits be restored to the umbrella's requirements (auto back to \$500,000; underlying watercraft liability re-established for the boat) as a condition of renewal; do not renew the umbrella over a schedule that no longer supports it. This is the renewal re-verification discipline in action. (§16.2, §16.4)

Exercise 16.27

The legitimate underwriting issue is the stale appraisal on appreciating property: an eight-year-old appraisal on a watch model known to have appreciated sharply means the \$45,000 scheduled value is probably low, leaving the insured under-insured on the very item they scheduled to be protected — the fix is to require a current re-appraisal and update the scheduled value. What is not a fair concern: the insured's questions about "how claims on scheduled items work" are not, by themselves, evidence of fraudulent intent (moral hazard, Ch. 1) — a household with a valuable, appreciating item asking how its coverage works is behaving reasonably, and treating ordinary diligence as suspicion is both unfair and bad service. Separate the real issue (stale value → re-appraise) from the unfounded suspicion (questions ≠ fraud). (§16.6; Ch. 1.)

Exercise 16.29 (†)

Memo (illustrative).

To: [Agent] Re: [Household] personal umbrella and the home catering business

The household has asked whether their personal umbrella will respond to a lawsuit arising from the small catering operation they run out of their home. It will not. The personal umbrella covers the household's personal liability; it contains a standard business/professional activities exclusion, and a catering operation is a business pursuit — even a small, home-based one. A claim arising from the catering work (a foodborne-illness suit, an injury to a customer or helper, property damage at an event) falls squarely within that exclusion, and neither the umbrella nor the homeowners policy is designed to cover it.

What the household actually needs is commercial coverage for the catering business: a commercial general liability policy (and, depending on the operation, product liability for the food and possibly commercial auto for deliveries). Once that commercial program is in place, the personal umbrella can remain what it is — protection for the family's personal exposures — and the business is protected by the coverage built for it. I'd recommend we review the catering operation's scope so we can place the right commercial program. Happy to walk through it.

(Clear, names the exclusion, identifies the real need, non-condescending.) (§16.3)

Exercise 16.31 (†)

Ethics dilemma (breed-based surcharge). The tension: a breed-based rule is defensible risk classification when it rests on genuine, credible loss experience that the breed (as a class) is associated with higher dog-bite frequency or severity — insurers lawfully price by risk. It shades into pricing by reputation rather than conduct when it is driven by a breed's image rather than data, or when it ignores the individual risk features that actually bear on this household's exposure: a fenced yard, no bite history, a well-managed household, a clean account. What to check before applying or waiving it: (1) whether the carrier's breed rule is supported by actual loss data or is a reputation-based blanket; (2) the individual mitigating facts (fencing, bite history, training, household management); (3) the state rule — some states restrict or prohibit breed-based underwriting (confirm where you write); and (4) whether a conduct-based alternative (bite history, control measures) would classify the risk more fairly. Connection to the fair-vs-unfair-discrimination line (Ch. 4, Ch. 35): the lawful basis for any factor is the risk it predicts, applied consistently; a factor that has become a proxy for reputation rather than measured risk, or that a state has judged off-limits, is where fair risk classification turns into unfair (or unlawful) discrimination. The disciplined move is to price the exposure (this dog, this yard, this history), not the reputation, and to operate inside the state's line. (§16.4; Ch. 4, Ch. 35.)

Exercise 16.33 (†)

Underwriting-File extension (Harbor Steel's \$10M commercial umbrella). - (a) What it sits above: the umbrella is excess liability over Harbor Steel's underlying liability coverages — the commercial general liability (premises/ops and the products-completed-operations coverage, important given the pending bracket claim), the commercial auto liability on the 12-unit flatbed fleet, and the employer's-liability portion of the workers' compensation. (Exact required underlying limits are set by the Part IV chapters that own those lines.) - (b) Why insist on a full \$1M CSL on the fleet: the umbrella attaches above the required underlying auto limit, and a flatbed fleet hauling heavy steel carries serious bodily-injury and nuclear-verdict exposure (Ch. 23); a low underlying auto limit would (i) drop more claims into the umbrella (raising its frequency) and (ii) risk a gap if the underlying limit fell short of the requirement. Insisting on \$1M combined single limit keeps the umbrella attaching where it was priced to attach and forces real primary protection where a catastrophic fleet claim is most likely. - (c) New condition added: the account must maintain the required underlying liability limits on the GL, auto, and employer's-liability lines for the umbrella to sit above them — a subjectivity/condition tracked alongside the others from Ch. 13 (roof replacement, hot-work program, sprinkler cert, IR scan, telematics). (No pricing; no capstone pre-emption.) (§16.2; The Underwriting File.)

Exercise 16.35

The two exposures from the frozen file that most threaten even a \$10M umbrella limit, and why each is a severity (not frequency) concern: 1. The commercial auto fleet (12 flatbed/delivery units). A flatbed fleet hauling heavy steel in a litigious environment is exposed to the nuclear-verdict trend (Ch. 23): a single catastrophic multi-vehicle bodily-injury accident can generate a demand reaching into the very high millions — the size set by the injury and the courtroom, not by how often the fleet crashes. Severity, not frequency. 2. The products-liability exposure (the allegedly failed fabricated bracket). A structural-steel component that fails can cause catastrophic injury or property damage, and a serious products-completed-operations claim is a high-severity, courtroom-determined loss — exactly the kind a high umbrella limit exists to absorb, and exactly the kind whose size the umbrella underwriter cannot bound in advance. (§16.4; The Underwriting File.)

Exercise 16.37 (†)

Account rounding and the combined ratio. The two mechanisms: 1. Better expense ratio through retention. Retention rises sharply with the number of policies a household holds with one carrier; longer retention amortizes the (expensive) acquisition cost over many more years, directly lowering the expense ratio. Acquisition/churn cost is the silent killer of personal-lines profitability, and a rounded account starves it. 2. Better loss ratio through favorable selection and visibility. A household willing to consolidate, carry an umbrella, and schedule its valuables is, on average, a more engaged, risk-aware, stable insured (favorable self-selection, the inverse of adverse selection, Ch. 1) — improving loss experience; and writing the account whole lets you see the catastrophe accumulation and liability exposure, preventing you from unknowingly stacking correlated risk.

The condition under which rounding would worsen the combined ratio: if the added policies are not adequately underwritten — i.e., you round up the risk by writing poor exposures just to grow the account. More premium from bad risk worsens the loss ratio. Rounding helps only when each added policy is itself soundly underwritten and priced. (§16.7; Ch. 3.)

Chapter 17

Worked solutions to the daggered (†) and odd-numbered exercises. All dollar figures and table positions are constructed teaching examples. Section references point back to index.md.

Exercise 17.1 (†)

Mortality is the rate at which people with a given set of characteristics die within a given period (expressed per thousand lives, as $q_x$, or as a percentage of the standard life). Life underwriting prices timing, not occurrence, because the insured event — death — is certain; what is uncertain is when it happens, and specifically whether it falls inside the policy's term. A 20-year term policy is a bet the insured outlives the term; a whole-life policy funds a claim that will eventually come. So the underwriting question is never "will a loss occur?" but "at what rate, over this horizon, relative to the standard lives the price was built on?" (§17.1)

Exercise 17.3

The standard life evidence stack, in roughly the order it enters the file, and what each can do: application (frames the whole risk and tells you what else to order); APS (confirms/denies/deepens the application against the real medical record); paramedical exam + labs (the insurer's own objective snapshot — catches undisclosed tobacco via cotinine, undiagnosed diabetes, silent impairments); MIB (reveals inconsistency with what the applicant told other insurers and flags multiple recent applications); prescription (Rx) history (fast, cheap, highly predictive map of conditions via the drugs prescribed). (§17.2)

Exercise 17.4 (†)

The APS is the medical record obtained, with authorization, from the applicant's own treating physician or facility. It is the gold standard because it is contemporaneous (written at the time of care), professional (by a clinician), and not created for the insurer (so it is not shaped by the applicant's incentive to look healthy). Its two chief drawbacks are time (it can add days or weeks) and cost — which is exactly why accelerated underwriting (§17.7) works so hard to identify which applicants genuinely need one rather than ordering it reflexively. (§17.2)

Exercise 17.5

From best to worst: Preferred Plus → Preferred → Standard Plus → Standard → Substandard (Table 2, 3, 4 …) → Decline/Postpone, with the whole ladder typically split into parallel nonsmoker and smoker versions. A table rating is a graded surcharge for above-standard mortality in which each "table" represents a fixed increment of extra mortality above standard (commonly ~+25% per table) — the mechanism that lets the insurer write an impaired life at a price reflecting its expected mortality rather than declining it. (§17.3)

Exercise 17.7

Simplified issue abbreviates the evidence to an application plus knockout health questions (and often instant database checks) with no exam, and prices higher to cover the lives that slip through. Guaranteed issue does little or no health underwriting at all. The structural device that lets a guaranteed-issue product survive without medical underwriting is the graded death benefit: natural-cause death in the first two-to-three years returns only premiums (often with interest), not the full face amount — which removes the incentive for someone who knows they are terminally ill to buy a full benefit and die immediately (the adverse-selection death spiral). (§17.6)

Exercise 17.8 (†)

Accelerated underwriting uses third-party data (Rx history, MVR, MIB, public/behavioral records) and predictive models to reach a life decision quickly — often instantly — for low-risk applicants who would traditionally have needed an exam and fluids. An applicant whose instant data raises questions is routed to traditional full underwriting (the exam and the APS), not declined. "Routed, not declined" matters because it preserves the access benefit (you do not lose a writable life over a data flag) while still putting human/clinical evidence on the files that genuinely need it — the exam is targeted, not abolished. (§17.7)

Exercise 17.9

Ordering by typical weight, with the governing condition for each: age (the axis — built into the base rate, steepens with years; not a debit but what everything is read against); tobacco (largest modifiable factor; splits every class; governed by the cotinine test and an honest declaration); build (graded by the chart but only meaningful in context of the labs); blood pressure (treated and controlled ≪ untreated — the APS decides which); family history (weighs only when disease is early and in close relatives; late disease is discounted). (§17.4)

Exercise 17.10 (†)

Untreated hypertension is a clear debit because elevated pressure drives cardiovascular mortality. But treated and well-controlled hypertension — documented in the APS as stable on therapy with good current readings — means the risk driver has been brought back down, so the residual mortality impact is much smaller, often a minor debit or none. The point: what matters is whether the applicant is controlled, not whether they are medicated — and you cannot know control from the application alone; the APS and the labs establish it. Reflexively debiting "on blood-pressure medication" without reading control is exactly the box-checking error the chapter warns against. (§17.4, §17.2)

Exercise 17.11

The sibling's cardiac event at 47 carries underwriting weight; the parent's heart disease at 82 carries little. The rule is that family history matters when it shows early disease in close relatives, because early disease in close relatives signals a heritable or shared-environment predisposition that may bear on the applicant. Disease at 82 is largely the ordinary mortality of old age — almost everyone's relatives develop something eventually — so it has little predictive value for the applicant's excess mortality, and the disciplined underwriter discounts it. (§17.4)

Exercise 17.12 (†)

For two applicants with identical height and weight, the three pieces of evidence that would most change the classification are: (1) the lipid/glucose/blood-pressure labs (do the numbers reveal a metabolic cluster that makes the build a debit, or are they pristine, suggesting the weight is benign?); (2) the avocation / activity level and body composition cues (a competitive athlete's "overweight" number is often muscle — a credit context — while a sedentary applicant's is more likely metabolic fat); (3) family and personal history (does the build sit on top of cardiac/diabetic risk, compounding it, or an otherwise clean profile?). Each can push either way: pristine labs + athletic build + clean history pull the same number toward a credit; a metabolic cluster + sedentary + cardiac family history pull it toward a debit. The build chart alone sees only the number, not what it is made of. (§17.5)

Exercise 17.13

Two independent strikes are added; one metabolic picture is read as a single, coherent elevated risk. "High cholesterol + BMI 28" treated as independent strikes double-counts when the two are mechanistically linked — high build, high lipids, high glucose, and high blood pressure are often one underlying metabolic condition expressing itself through several readings. Summing the manual blindly over-penalizes the correlated case because it charges separately for what is really one thing; the skilled underwriter rates the condition, not each of its symptoms. (Conversely, if the two findings are genuinely unrelated and sit in an otherwise excellent profile, they are each mild and well-offset.) (§17.3, §17.5)

Exercise 17.14 (†)

Risk class: a strong candidate for Preferred Plus / Preferred. The single most important reason is that every independent source agrees on a favorable, low-mortality picture — lifelong nonsmoker confirmed by a negative cotinine, excellent build (BMI 22), pristine labs and blood pressure, clean MIB, no medications, no history, and an active life. There is no impairment to debit and no inconsistency to chase. APS: not needed — nothing in the file approaches a class boundary or raises a question an APS would resolve, so ordering one would add cost and delay for no underwriting value. This is the textbook accelerate candidate. (§17.2–§17.5)

Exercise 17.15

Walk the file: the positive cotinine does two things at once — it makes this a smoker-class risk (materially higher mortality, so smoker rates), and it contradicts the "former smoker, quit recently" representation, raising a good-faith/disclosure question (handled under §17.2's utmost-good-faith frame; a clear, material misrepresentation can route toward Chapter 33). On the medical side, treated hypertension at 138/88 is a modest debit (read control via the APS), and type 2 diabetes with an elevated HbA1c plus a heavy build is likely one metabolic cluster — a real, ratable debit. Net: this is most likely a substandard table rating at smoker rates if the diabetes is controlled and uncomplicated; it becomes a postpone if the diabetes is poorly controlled or newly destabilized (not yet calculable), and a decline only if complications (renal, vascular) push expected mortality beyond what a table can price. The disclosure issue is handled separately from the rate. (§17.2, §17.4, §17.6)

Exercise 17.16 (†)

Box-checking read of David Okafor: two negatives — total cholesterol mildly elevated, BMI 28 — tallied as strikes → offer standard or worse, possibly deny preferred outright. Whole-person read: the nonsmoker status holds (cotinine-negative); BMI 28 on a competitive cyclist is very likely muscle, not metabolic fat, corroborated by excellent blood pressure, strong HDL, and normal glucose — i.e., no metabolic cluster; the mildly elevated cholesterol is a single mark inside an otherwise preferred cardiovascular profile; → place at preferred or near-preferred, APS only at the class boundary. The one finding that earns its debit weight in both readings is the father's heart attack at 58 — a genuine family-history debit precisely because it was early. (That it is modest, not disqualifying, is why David lands preferred-or-near rather than substandard.) (§17.5, The Underwriting File)

Exercise 17.17 (†)

Standard premium = \$1,200. Using the convention "each table ≈ +25% of standard mortality ≈ +25% of premium": - Table 2 (+50%) → mortality ratio 150% → premium ≈ \$1,200 × 1.50 = **\$1,800. - Table 4 (+100%) → mortality ratio 200% → premium ≈ \$1,200 × 2.00 = **\$2,400. - Table 6 (+150%) → mortality ratio 250% → premium ≈ \$1,200 × 2.50 = **\$3,000**.

(Real plans differ in exact mechanics — some apply the table factor to the mortality cost only, not the whole premium — but the teaching point is the graded, proportional relationship between extra mortality and extra price.) (§17.3)

Exercise 17.19 (†)

Simplified-issue rates are higher than fully underwritten rates for the same face amount and age even for a genuinely healthy buyer because the price reflects the pool, not the individual. By skipping the exam and fluids, the insurer accepts that some impaired lives it would have caught will slip through the knockout questions, so the expected mortality of the simplified-issue block is higher than a fully screened block — and everyone in it, healthy or not, pays for that. The healthy simplified-issue buyer is paying for speed and convenience (no exam, fast issue) and is effectively subsidizing the less-screened pool they have chosen to join. (§17.6)

Exercise 17.20 (†)

Red flags: (1) "no medications" contradicted by Rx fills of a cardiac drug and an anticoagulant — this is evidence of an undisclosed cardiac/clotting condition, though it does not by itself prove the diagnosis (Rx cannot show why a drug was prescribed); (2) two other applications in six months on the MIB — this is a pattern consistent with adverse selection or stacking coverage, though it proves nothing on its own (a pointer, not a diagnosis). Next two actions: order an APS to establish the actual cardiac history and current status, and clarify the discrepancy with the applicant/producer (and weigh the large \$2M amount and the application pattern in deciding scope). Resolve the medical picture before pricing; treat the misstatement under the good-faith frame. (§17.2)

Exercise 17.21

The single positive cotinine on a "nonsmoker" application creates two distinct problems: a pricing problem — the applicant is, on the evidence, a tobacco user and belongs in the smoker class with its materially higher mortality and rate — and a disclosure problem — the "nonsmoker" answer is a material misrepresentation, handled under utmost good faith (and, if egregious, Chapter 33's rescission/SIU territory). The two are handled separately: the rate moves to smoker class regardless, and the misstatement is documented and addressed on its own terms. (§17.4, §17.2)

Exercise 17.23 (†)

Substance of the adverse-action note (draft): "Thank you for your application. We are able to offer coverage, but not at the preferred rate you requested; based on our review you qualify at a substandard (Table 4) rate. This decision was based in part on information obtained from a consumer-reporting agency, including a prescription-history report, and from medical records (the attending physician statement). Under the Fair Credit Reporting Act you have the right to obtain a free copy of the consumer report used and to dispute the accuracy or completeness of any information in it; you may also submit additional medical information for us to reconsider. Here is how to contact the reporting agency and how to provide further records." Note what it does: discloses the basis and the FCRA rights, names the dispute/supplement path, and does so without dumping more clinical detail than is appropriate. (§17.2, §17.7; Ch. 4 FCRA)

Exercise 17.25 (†)

The tension: the model is actuarially accurate (it predicts mortality well on held-out data), but it achieves that accuracy through a non-medical proxy (a credit-based attribute correlated with neighborhood) that may stand in for a protected characteristic — colliding with social fairness and the law. The specific legal concept at risk is proxy discrimination / disparate impact (Chapter 35 owns it): a facially neutral variable producing a protected-class effect. Concrete governance step (any one): test the model's outcomes for disparate impact across protected groups; remove or constrain the offending proxy and re-fit; require a documented fairness review and sign-off before deployment; and ensure FCRA adverse-action handling on any decline/rating it drives. Accuracy is necessary but not sufficient — fairness must be tested, not assumed. (§17.7; preview of Ch. 35)

Exercise 17.27

Both can be true, and the chapter holds both. The graded death benefit is a necessary anti-selection device: without it, a guaranteed-issue product (no health questions) would be bought disproportionately by people who know they are dying and would collapse — so the two-year graded period is what makes coverage available at all to a group that could not otherwise be underwritten. Because the product relies on a feature that pays less than buyers may expect in the early years, the disclosure obligation is especially heavy: it is ethical to sell only if the elderly buyer genuinely understands that early natural-cause death returns premiums, not the full face. The defense of the device raises, rather than lowers, the bar on clear disclosure. (§17.6)

Exercise 17.29

The one question Chapter 35 will press on David: how far may data-driven (accelerated) life underwriting go in using genetic or genetically-inflected information to classify a life like his — given the GINA gap and the unsettled state law. His preferred-or-near classification here is what makes the question sharp because it was reached by a careful whole-person read (and would be reached by a well-built model on ordinary data); the moment you add genetic predisposition data — say, a variant linked to the very cardiac risk his father's early MI hints at — you confront whether a person who is currently an excellent risk may be re-rated on DNA they did nothing to cause and may never express. David is the ideal test case precisely because he is otherwise good: the genetics question is hardest when it would worsen an applicant the rest of the evidence rewards. (§17.5, §17.7)

Chapter 18

Worked solutions to the daggered (†) and odd-numbered exercises. (Even-numbered items not reproduced here are discussion or memo prompts whose answers are developed in the chapter text and the case studies.) All dollar figures are illustrative teaching numbers.

Exercise 1 (guaranteed issue + community rating)

Guaranteed issue = the insurer must offer coverage to any eligible applicant regardless of health status (no declines, no medical underwriting, no condition exclusions). Community rating = the insurer prices for the whole community rather than the individual's health; under the ACA's adjusted form, premiums may vary only on age (capped), tobacco (capped), area, family tier, and plan tier — not health. They must be enacted together because each alone can be evaded through the lever the other closes: guaranteed issue without community rating lets a carrier "decline" a sick applicant by quoting an impossible price; community rating without guaranteed issue lets a carrier charge one price but simply refuse the sick. Closing one lever just pushes the carrier's selection instinct to the other.

Exercise 3 (the four moves)

Before the ACA, an individual health underwriter facing a serious chronic condition could: (1) decline the application outright; (2) rate up — issue at a surcharged premium reflecting the higher expected claims; (3) exclude the condition — issue with a pre-existing condition exclusion permanently carving out the named condition; (4) impose a waiting period — cover the condition only after a defined period during which related claims are excluded. The ACA abolished all four for compliant individual/small-group plans.

Exercise 4 (specific vs. aggregate stop-loss)

Specific (individual) stop-loss caps the employer's exposure to any one claimant — it reimburses the employer once a single covered person's claims exceed a per-person attachment (e.g., \$200,000). It defends against the single catastrophic claimant (the transplant, the premature infant, the late-stage cancer). Aggregate stop-loss caps the employer's exposure to the whole group's total claims — it reimburses once total paid claims exceed a threshold (conventionally ~125% of expected claims). It defends against many moderate claims piling up (a bad flu season, a cluster of surgeries) that individually never pierce the specific deductible but collectively blow the budget. A sound self-funded plan usually carries both, because they protect against the two distinct failure modes of a small pool.

Exercise 5 (why stop-loss is sold to the employer)

Stop-loss insurance reimburses a self-funded employer when the employer's own claims payments exceed a defined threshold. It is sold to the employer, not the employees, because in a self-funded plan it is the employer — not a carrier — that bears the claims risk and pays claims directly (usually through a TPA). The employees have ordinary plan coverage; stop-loss is the employer's own insurance against its own retained risk, structurally the same idea as a self-insured retention (Ch. 12) or an excess-of-loss reinsurance treaty (Ch. 27).

Exercise 7 (ACA rating factors)

An ACA adjusted-community-rated individual premium MAY vary on: age (within a capped older-to-younger ratio), tobacco use (within a capped surcharge), geographic rating area, family size/tier, and plan/metal tier (actuarial value). It may NOT vary on (any three): health status / medical history, claims experience, gender, pre-existing conditions, or duration since last covered. The load-bearing point: health is on the forbidden side.

Exercise 8 (what the MLR rule caps)

The medical-loss-ratio (MLR) rule requires a health insurer to spend a minimum share of premium on actual medical care and quality improvement, rebating the shortfall to policyholders. In the Chapter 11 premium build-up (pure premium + expense load + profit/contingency load), the MLR rule is a cap on the expense-and-profit load — equivalently, a floor on the share of premium that must go to claims. It is a public ceiling on the underwriting/administrative margin itself.

Exercise 9 (insured vs. self-funded — who bears the risk)

In a fully insured plan the employer pays a premium to a carrier and the carrier bears the claims risk (keeps or loses the difference between premium and claims). In a self-funded plan the employer keeps the money, pays employees' claims itself (usually via a TPA), and the employer bears the claims risk — purchasing stop-loss only to cap the catastrophic tail.

Exercise 10 (abolishing underwriting ≠ abolishing adverse selection)

Mechanism: guaranteed issue + community rating removes the carrier's defense against adverse selection (underwriting) while leaving the buyer's incentive to wait-and-select fully intact. A healthy person can buy community-rated, guaranteed-issue coverage the day they get sick and not before; if enough do, the pool fills with the sick, the community rate rises, the remaining healthy leave, and the death spiral (Ch. 1) begins. The four scaffolding pieces erected to hold the pool up: (1) the coverage incentive / individual mandate; (2) premium subsidies (buying the good risks back into the pool); (3) limited enrollment windows; (4) the back-office machinery — risk adjustment, the MLR rule, and transitional reinsurance.

Exercise 11 (correct the friend)

The ACA did not solve adverse selection — it relocated the job of managing it. It removed individual underwriting (the carrier's tool against selection) and replaced it with public machinery: the mandate, subsidies, and enrollment windows keep the healthy in the pool, and risk adjustment + the MLR rule + (early) reinsurance manage the consequences across carriers. Adverse selection is now managed by the market design and the regulator, not by the underwriter — and when a strut weakens (e.g., the mandate penalty going to zero), the selection pressure reasserts itself exactly as Chapter 1 predicts.

Exercise 13 (abolish the enrollment windows)

Step by step: with no enrollment windows, a healthy person waits, uninsured, until they are diagnosed, then buys guaranteed-issue, community-rated coverage and drops it once treated. The carrier collects little or no premium from the healthy and pays claims for the sick on demand. Average claims soar above the community rate; the rate must rise; the few remaining healthy buyers drop out; the rate rises again. This is the adverse-selection death spiral, and removing the enrollment windows would accelerate it dramatically — which is why the windows are the structural descendant of the old waiting period, moved from the policy to the calendar.

Exercise 14 (underwrite the stop-loss)

(a) Both a specific and an aggregate attachment — specific to cap the single catastrophic claimant, aggregate to cap the total budget against an accumulation of moderate claims. A 600-life self-funded pool is small enough to need protection against both failure modes. (b) The known \$1,400,000 claimant is not a fortuitous risk — it is a near-certain expense (Ch. 1) — so you would laser that individual: set their specific attachment far above the \$200,000 that applies to everyone else (e.g., \$1,000,000), or load the rate, pricing the known cost honestly rather than insuring it. (c) You must state the laser in plain language in the quote (the named claimant's higher attachment and the employer's resulting retained exposure), so the employer is not surprised at claim time. (Optionally note the employer can buy a no-new-laser guarantee at a price.)

Exercise 15 (the 45-life group wants experience rating)

You would decline to fully experience-rate a 45-life group because its own one-year experience is not credible (Ch. 10) — at that size a single good or bad year is mostly small-sample noise, not reliable signal about future cost. A 45-life group is also below the small-group line and is, in ACA-compliant terms, community-rated, not experience-rated, for that reason. What you do instead: rate it on the manual/book rate for its class and area (community rating), giving its own experience little or no credibility weight.

Exercise 16 (which market permits underwriting?)

(a) Forbidden — individual ACA marketplace plan: guaranteed issue + community rating (§18.2). (b) Permitted — a Medigap applicant outside a guaranteed-issue window may generally be medically underwritten (§18.5). (c) Permitted — a 2,000-life employer is large-group and may be experience-rated (§18.3). (d) Permitted — an individual critical-illness policy is a supplemental product outside the ACA's major-medical reforms and is medically underwritten (§18.5).

Exercise 17 (the 1,200-life blend)

Blended expected claims = Z × own + (1 − Z) × manual = 0.65 × \$500 + 0.35 × \$440 = \$325 + \$154 = **\$479 PMPM**. You did not simply use the group's own \$500 because at 1,200 lives the group's experience is substantially but not fully credible (Z = 0.65); blending toward the manual rate guards against the chance that the \$500 reflects a one-off shock claim rather than the group's true expected cost.

Exercise 18 (build the blend)

(a) Blended expected = 0.80 × \$610 + 0.20 × \$540 = \$488 + \$108 = \$596 PMPM. (b) Indicated premium = \$596 × 1.12 = **\$667.52 PMPM** (≈ \$668). (c) The carrier blends toward the manual rate because even an 80%-credible group's own number can be distorted by a single large claim or a one-time event; the manual rate anchors the price to the broader class so the premium reflects sustainable expected cost, not a lucky or unlucky year — the credibility discipline of Chapter 10 and the rate-adequacy discipline of Chapter 11.

Exercise 19 (aggregate attachment)

(a) Aggregate begins to pay at ~125% × \$3,600,000 = **\$4,500,000 of total group claims. (b) It protects the employer against a bad-budget year — many moderate claims accumulating past the total threshold (a bad flu season, a surge of surgeries). It does not protect against a single catastrophic claimant whose claims pierce the specific attachment but whom the group's total still stays under \$4.5M — that is the specific stop-loss's job. The two covers protect against different failure modes.

Exercise 20 (specific attachment trade-off)

As the specific attachment rises from \$150,000 to \$300,000, the stop-loss premium falls (the carrier insures a thinner, less likely tail) and the employer's retained risk rises (it now eats the first \$300,000 of any claimant itself). As it falls, premium rises and retained risk shrinks. A very low attachment defeats the purpose of self-funding: if you transfer almost everything to the stop-loss carrier, the stop-loss premium approaches the cost of just buying fully insured coverage, and you have given back the savings (the carrier's risk margin and premium taxes) that were the reason to self-fund. The right attachment is the largest single-claim loss the employer can genuinely absorb.

Exercise 21 (read the renewal — the laser)

(a) The device is a laser — a higher specific deductible applied to a single named, already-known high-cost claimant. (b) It is not improper underwriting: the named claim is a known, ongoing cost, not a fortuitous risk, and Chapter 1 says you cannot insure a loss that is already happening — you can only pre-fund it. The carrier is pricing correctly. (c) The danger is to the employer: it may believe it has \$200,000 protection on every life when in fact its largest known claim is carved back to \$1,000,000 of retained exposure — a six-figure surprise hidden in a schedule. The employer could have bought a no-laser or no-new-laser guarantee (at a price) to cap this, and at minimum should have had the renewal walked through line by line.

Exercise 23 (diagnosis coding completeness)

Under risk adjustment, a carrier is paid (via transfers) according to its enrollees' documented diagnoses, so a more completely coded chronic condition raises a member's risk score and the carrier's transfer. Devoting resources to "coding completeness" is therefore a rational response — it captures transfers the carrier is entitled to for the risk it actually holds. But it is not underwriting judgment: it is data capture after enrollment, not a decision about whether to accept or how to price a risk. It draws regulatory scrutiny because the line between accurately documenting real conditions and aggressively coding to inflate scores is thin, and inflated coding shifts transfer dollars without reflecting real risk — the relocated "incentive to win" that every automated replacement for judgment in this book carries.

Exercise 25 (write the recommendation — see also even items)

(Daggered memo prompt; model paragraph.) "For a healthy 250-life group like ours, I recommend moving from fully insured to self-funding with stop-loss. The single biggest upside is that we keep the savings of our own good claims experience instead of paying a carrier's risk margin and premium taxes on it. The single biggest risk is variance: in a bad year, a few catastrophic claimants could spike our costs well above budget. To cap that, I would require two stop-loss features — a specific attachment set at the largest single claim we can absorb (so no one claimant can ruin us), and an aggregate attachment around 125% of expected claims (so a pile-up of moderate claims can't blow the budget) — and I would insist the contract basis and any lasers be spelled out in plain language so there are no surprises at claim time."

Exercise 27 (plain-language laser disclosure)

(Model.) "Please note: while the specific stop-loss attachment is \$175,000 for all other covered persons, one named employee currently in active treatment has a specific attachment of \$1,000,000 — meaning the plan (the company) will pay the first \$1,000,000 of that person's claims this year before stop-loss reimburses anything. Budget for this retained exposure, and consider a no-new-laser guarantee at renewal to prevent future surprises of this kind."

Exercise 29 (good underwriting, social failure)

Pre-ACA individual underwriting was good in the narrow sense that each accepted applicant was priced for their own expected cost and known-future-cost applicants were screened out, protecting the pool's loss experience and the carrier's combined ratio — textbook risk selection. It was simultaneously a social failure because that same selection underwrote a large share of the sick out of coverage entirely: people were left uninsurable at any price, locked into jobs, and exposed to financial ruin from illness. A selection system can be actuarially flawless and still produce an outcome a society judges intolerable — which is exactly the actuarial-fairness vs. social-fairness collision (§18.7, Ch. 35).

Exercise 31 (Medigap underwriting and unfair discrimination)

It is not unfair discrimination (Ch. 4) for an insurer to medically underwrite a Medigap applicant outside their guaranteed-issue window. Unfair discrimination means pricing or declining on a protected class (race, religion, national origin) or otherwise treating like risks unlike; medical underwriting prices on risk (health status), which is lawful where permitted. The genuine ethical debate is not about legality but about fairness of access: whether permitting health underwriting outside the window strikes the right balance between giving people a protected chance to enroll when first eligible (access) and preventing them from gaming the system by waiting until sick (anti-selection). That is a social-fairness judgment, not a protected-class one.

Exercise 33 (three names for one idea)

A self-insured retention (Ch. 12), an excess-of-loss reinsurance retention (Ch. 27), and a stop-loss specific attachment (this chapter) are the same structural idea: a party that bears risk keeps the predictable, affordable layer of loss itself and transfers only the layer above a defined threshold to someone else. The SIR is the insured keeping the bottom layer; XOL is the insurer keeping the bottom layer and ceding the top to a reinsurer; stop-loss is the self-funded employer keeping the bottom layer and ceding the catastrophic top to a stop-loss carrier. In every case the retention preserves skin in the game (Ch. 1) and the transfer caps the ruinous tail.

Exercise 35 (synthesize: "underwriting moved")

(Daggered synthesis.) The chapter's thesis is that underwriting did not disappear after the ACA; it moved. It moved up-market to large-group experience rating, where a pre-formed pool can be priced on its own credible claims; it moved inside the employer to self-funded stop-loss, where the catastrophic tail is still medically underwritten; and it persisted in Medicare Supplement (outside the window) and the supplemental products the reform never touched. What stayed behind in the individual market to do underwriting's old job is the machinery of §18.6 — risk adjustment, the MLR rule, enrollment windows, subsidies, and the mandate — which manage adverse selection at the level of the whole market because no carrier may any longer manage it at the level of the single life. The first underwriting decision in health is therefore diagnostic: which market am I in — does the law want my judgment, or has it replaced it?

Chapter 19

Worked solutions to the daggered (†) and odd-numbered exercises. All figures are constructed teaching examples. Section references point back to index.md.

Exercise 19.1

The Building and Personal Property Coverage Form insures three things, written as separate limits: (1) Building — the structure plus permanently installed fixtures and equipment; (2) Your Business Personal Property — the contents the insured owns (machinery, inventory, stock, furniture); and (3) Personal Property of Others — property of others in the insured's care, custody, or control. A submission's schedule of values typically breaks down into these buckets, by location. (§19.1)

Exercise 19.3

Business income (BI) pays two things: (a) the net income (profit or loss) the business would have earned had no loss occurred, and (b) the continuing normal operating expenses, including payroll the insured elects to continue. It insures the consequence of a property loss — the lost earnings during the shutdown — not the property. Extra expense differs in that it pays the additional costs the insured incurs to avoid or reduce the interruption (temporary space, leased substitute equipment, expedited shipping). BI replaces lost earnings; extra expense funds the scramble to get earning again, shortening the interruption. (§19.3)

Exercise 19.4 (†)

The period of indemnity is the length of time business income coverage will pay, running from the date of loss until operations are or reasonably could be restored to the condition that would have existed had no loss occurred. It is governed by the longest critical-path element of the recovery — not the building rebuild — because recovery is only as fast as its slowest dependent step. A generic structure may be rebuilt in six months, but if production depends on a specialized machine with a twelve-month manufacture-and-delivery lead time, the business cannot resume until that machine arrives and is installed; the period of indemnity is therefore twelve months, not six. Setting it off the building timeline misses the long-lead equipment, the recertification, and the rebuilt customer relationships. (§19.4)

Exercise 19.5

A highly protected risk (HPR) is a property built and protected to the highest loss-prevention standards. Roughly four things qualify it: (1) superior construction (typically fire-resistive or non-combustible); (2) full automatic sprinkler protection engineered to the occupancy's hazard; (3) a robust water supply (and fire pumps/alarms as needed); and (4) strong management, housekeeping, and active loss-control engineering. Harbor Steel is not an HPR: a 1994 joisted-masonry/metal-frame plant with its original sprinkler system, a hot-work (welding) occupancy, and public fire protection class 4 is a solid but ordinary — non-HPR — commercial risk. (§19.5)

Exercise 19.7

A statement of values (SOV) is a schedule, location by location, of every insured property and its values (building, business personal property, business income, by address) that together form the total insurable value of an account. Three things built off it: (1) the premium (the rate is applied to the values on the SOV); (2) the limits (set from the values); and (3) the catastrophe accumulation (the geographic spread of values is what cat models aggregate — Chapter 30). The coinsurance and agreed-value tests also run against it. (§19.7)

Exercise 19.8 (†)

Equipment breakdown insures sudden, accidental physical damage to covered equipment from internal mechanical or electrical forces — a boiler rupturing, a transformer shorting out, a compressor or motor failing from within — plus the resulting damage and the business income lost while the equipment is down. The commercial property Special form covers loss from external perils (fire, wind, impact) but excludes loss caused by internal mechanical/electrical breakdown, so equipment breakdown fills exactly that gap. The exposure is easy to miss on inspection because it is an exclusion on the main form rather than a visible hazard — the inspector sees a working machine, not the excluded peril hiding inside it. (§19.6)

Exercise 19.9 (†)

(a) With accumulated depreciation of roughly 30% of a \$20,000,000 replacement cost, an ACV policy would pay on a total loss approximately \$20,000,000 − (0.30 × \$20,000,000) = about \$14,000,000 (the exact figure depends on the appraiser's depreciation schedule; ACV = replacement cost minus depreciation). (b) An insured prefers replacement cost anyway because it makes them genuinely whole — they can actually rebuild new, rather than receiving a depreciated payment that leaves a ~\$6M gap to fund themselves. The insurer's offsetting concern is that replacement cost is a larger exposure and is more vulnerable to under-valuation as construction costs rise — which is exactly why you would put a known end-of-life component (like Harbor Steel's roof) on ACV while keeping the sound parts on replacement cost. (§19.2)

Exercise 19.11 (†)

(a) The brand-new warehouse the owner intends to rebuild → replacement cost (the insured wants to be made whole to new condition, and the value is current and easy to verify). (b) The 1920s ornate-masonry building → functional replacement cost (reproducing the original masonry exactly would be wildly expensive and pointless; insure a functional, modern equivalent). (c) The 30-year-old roof at end of life → actual cash value (ACV) (you do not want to insure a worn-out component to brand-new standard; ACV aligns the payout with its real depreciated value, and you require replacement as a subjectivity). (§19.2)

Exercise 19.13 (†)

The \$2M limit is wildly inadequate. A total loss idles the single plant for about twelve months (driven by the 9–12 month lead time on the specialized line), and a year of gross earnings is about \$9–10M — so the income exposed to loss over the real period of indemnity is roughly \$9–12M**, not \$2M. A defensible limit is set by deriving it: gross earnings (≈ \$9–10M/year) × period of indemnity (≈ 12 months) ≈ **\$9–12M, which you would round to a credible limit (say \$10–12M) and possibly extend with an extended-period-of- indemnity endorsement for the ramp-back of lost customers. The building value is irrelevant to this calculation; BI must be derived from earnings and time. (§19.3, §19.4)

Exercise 19.15 (†)

A representative critical path for a plant's recovery from a total fire loss: 1. Debris removal & site cleanup (~1 month) 2. Design, permits, bidding (~2–3 months) 3. Rebuild the structure (~5–6 months) — the element everyone sees 4. Order + receive the specialized production line (~9–12 months) — the real driver 5. Install, test, recertify the line (~1–2 months, but cannot start until step 4 arrives) 6. Rehire/retrain staff and requalify customers (~1–2 months)

The single element that most often makes the period of indemnity longer than the building rebuild suggests is the long-lead specialized equipment (step 4): installation cannot even begin until the machine arrives at month 9–12, so the period of indemnity is governed by that chain, not the six-month structure rebuild. (§19.4)

Exercise 19.17

Scaling BI at "50% of the building limit" ignores that business income has no fixed relationship to the building value — it is set by earnings × time to restore, not by the structure's cost. The rule will be wrong in both directions: a business in a cheap building with high earnings (e.g., a data-dependent operation in a plain warehouse) will be under-insured — the dangerous direction, because the company can fail for lack of income while the limit runs out — and a business in an expensive building with modest earnings will be over-insured (wasting premium). The dangerous error is the under-insurance, because it is invisible until the loss and then it is fatal. (§19.3)

Exercise 19.18 (†)

(a) Required limit = 80% × \$20,000,000 = **\$16,000,000. (b) Coinsurance ratio = limit carried / limit required = \$12,000,000 / \$16,000,000 = 0.75. (c) Policy pays = loss × ratio − deductible = (\$4,000,000 × 0.75) − \$50,000 = \$3,000,000 − \$50,000 = \$2,950,000. (d) Beyond the deductible, the insured absorbs the coinsurance shortfall: \$4,000,000 − \$3,000,000 = \$1,000,000** (the deductible is separate). And no — the \$12M limit is not exhausted; the penalty applied even though the loss (\$4M) was far below the limit, because coinsurance tests the limit against value, not against the loss. (§19.4)

Exercise 19.19

If the insured had carried the full required \$16,000,000, the coinsurance ratio would be \$16,000,000 / \$16,000,000 = 1.0, so the policy pays the full \$4,000,000 (less the deductible) — no penalty. This shows the purpose of the coinsurance clause: it is not a profit device but an enforcement mechanism for honest valuation. Insure to at least the required percentage of value and you are paid in full; under-insure and you bear a proportional penalty. The clause exists to stop insureds from carrying a low limit (and paying a low premium) while expecting full payment on the common partial loss. (§19.4)

Exercise 19.20 (†)

Explanation to the frustrated insured: "Your policy carried an 80% coinsurance clause, which required you to insure the building to at least 80% of its full value. Your building's true replacement cost had grown to [\$X], so you needed to carry [0.8 × \$X]. You carried less than that, so by the terms of the policy the loss is paid at the ratio of what you carried to what you should have carried — and that ratio applies to every loss, including this partial one, which is why the penalty hit even though your limit wasn't used up." What the underwriter could have structured at binding to prevent it: write the building on an agreed-value basis backed by a current appraisal, which suspends the coinsurance clause entirely (no test, no penalty) — or, at minimum, require an updated valuation at each renewal and document in writing that the coinsurance clause was live so the insured understood the exposure before the loss. (§19.4)

Exercise 19.21

Agreed value is right when you have a verified, current valuation (appraisal or credible replacement-cost estimate) you are willing to stand behind, and you want to protect the insured from a coinsurance penalty — common on larger, well-documented accounts like Harbor Steel. Choosing it transfers the valuation responsibility from the insured to the underwriter: there is no coinsurance test to catch an under-valuation, so if the agreed number is wrong, you (the carrier) absorb the consequence rather than the insured. You would refuse agreed value when the value is stale, unverified, or implausibly high (e.g., a broker pushing a \$24M agreed value on a building you estimate at \$19M) — because agreed value on an unverified number is simply a pre-committed under- or over-insurance, and an inflated one edges into the indemnity/moral-hazard problem. (§19.2, §19.4)

Exercise 19.22 (†)

COPE on Harbor Steel: - Construction — a 1994 joisted-masonry/metal-frame plant: adequate, ordinary construction, but not the fire-resistive/non-combustible standard of an HPR. - Occupancy — welding, cutting, and metal fabrication: a hot-work occupancy with ignition sources designed into the daily workflow (the 2023 welding fire is the evidence). - Protection — original wet-pipe sprinklers and public fire protection class 4, with the nearest hydrant ~600 ft: ordinary, not superior; the original system is at the age where reliability is a question. - Exposure — coastal, named-windstorm zone, with a storm-surge zone nearby: a real catastrophe exposure looming over the fire picture. Taken together — ordinary construction, a hazardous occupancy, aging original protection, and a catastrophe exposure — this is plainly a non-HPR (ordinary) commercial risk, and must be priced as one, not at an HPR rate. (§19.5)

Exercise 19.23

Red flags that this is not an HPR account: (1) construction — joisted masonry/metal frame is ordinary, not the fire-resistive/non-combustible standard HPR requires; (2) original (end-of-life) sprinkler system — HPR demands sprinkler protection engineered to and maintained for the occupancy's hazard, not a thirty-year-old original system of unverified reliability; (3) hot-work occupancy — welding builds ignition sources into the workflow, the opposite of the controlled, low-ignition environments typical of HPR; and (one could add) the fire protection class 4 and the loss history (two fires in five years) are inconsistent with HPR loss experience. Writing it at an HPR rate would price a genuinely higher-expected-loss risk as if it were an elite, low-loss risk — guaranteeing the premium is inadequate for the risk accepted, which surfaces as a deteriorating loss ratio two or three years later (the rate-adequacy failure of Chapters 3 and 11). (§19.5)

Exercise 19.25

The COPE factor most readily improved through a subjectivity is Protection — the private fire protection the insured controls. On Harbor Steel, the requirements that do exactly this are the sprinkler certification, the hot-work permit program, and the infrared electrical scan (and, for the roof and wind exposure, the warranted roof replacement) — each a condition precedent to binding that changes the protection and thereby the risk, rather than merely re-pricing it. Construction and occupancy are largely fixed; exposure (the coastal peril) cannot be requirement-ed away; protection is the lever. (§19.5)

Exercise 19.26 (†)

(a) The compressor damage from an internal electrical fault → NOT covered by the property Special form (internal mechanical/electrical breakdown is excluded); covered by equipment breakdown. (b) The spoiled inventory resulting from the breakdown → covered by equipment breakdown as consequential damage (the property form would not respond, because the proximate cause is the excluded breakdown, not a covered property peril). (c) The lost income for the week closed → covered by the business-income element of equipment breakdown (triggered by the breakdown). If the closure had instead been caused by a covered property peril like a fire, the property policy's BI would respond — but here the cause is internal breakdown, so equipment breakdown is the coverage that responds throughout. (§19.6)

Exercise 19.27

Inland marine responds. The fabricated steel is property in transit on a public road, away from the described premises — exactly the gap inland marine exists to fill (it descends from ocean marine's coverage of property that moves). The commercial property policy does not respond because it covers property at the described premises, and the steel has left the premises. (Commercial auto covers the truck and the insured's liability arising from the accident — Chapter 23 — but the cargo's own physical damage is an inland-marine, not an auto-physical-damage, exposure.) (§19.6)

Exercise 19.29 (†)

High-excess catastrophe capacity is cheap per dollar of limit because the high layers are reached only by rare, severe events — low frequency — so each dollar of limit is touched seldom and the expected loss per dollar is small. Primary capacity is expensive per dollar of limit because the primary layer absorbs the frequent, attritional losses — the ordinary fires and water damages — so each dollar of primary limit is exposed to loss constantly and carries a high expected loss. In frequency-×-severity terms (Chapter 6): the primary layer faces high frequency (and moderate severity); the high-excess layer faces very low frequency (but extreme severity). Price follows the expected loss the layer actually sees, so the primary costs more per dollar of limit even though the catastrophe layer covers the scarier event. (§19.7; Ch. 6)

Exercise 19.31 (†)

Ethics dilemma — the \$24M agreed value on a \$19M building. The problems: insurance is built on the principle of indemnity (Chapter 4) — it exists to restore the insured to their pre-loss position, not to profit them. Insuring a building for materially more than its replacement cost (a) violates indemnity by making the insured better off after a total loss than before, and (b) creates moral hazard — the building becomes worth more destroyed than standing, which is precisely the incentive the whole system guards against. Agreeing a \$24M value you have independently estimated at \$19M also means you have pre-committed to paying \$5M of phantom value with no coinsurance test to catch it. What you will do: write the building on an agreed value backed by your verified ~\$19M (or a current independent appraisal if the broker contests yours), explain that you cannot insure above replacement cost because indemnity forbids it, and offer to revisit the value with a new appraisal at renewal if costs genuinely rise. How you keep the relationship: you frame it as protecting the client — an inflated value wastes their premium and would be challenged at claim time anyway, whereas a verified agreed value gives them full, penalty-free recovery on a real number. You hold the line on the number while solving the client's actual worry (being fully covered), which is what a broker who brings good business will respect. (§19.7; Ch. 4)

Exercise 19.33

A shorter (6-month) lead time on the upgraded fabrication line shortens the period of indemnity: the longest critical-path element of the recovery has been cut roughly in half, so a realistic recovery now runs nearer to the building-rebuild timeline than the old 9–12 month equipment chain. That reduces the business-income limit required (the income exposed over a ~6-month period of indemnity is smaller than over 12 months) and therefore the BI premium at the next renewal — a genuine, defensible reduction you would recalculate on an updated business-income worksheet. It does not change the building valuation: the structure's replacement cost is unaffected by the speed of equipment delivery (and indeed a newer, more valuable machine might raise the equipment value even as it lowers the period of indemnity). This is a clean illustration that the building and the income are independent exposures that move for different reasons. (§19.3, §19.4, The Underwriting File)

Chapter 20

Worked solutions to the daggered (†) and odd-numbered exercises. (Even, non-daggered items are discussion or memo prompts whose answers are developed in the chapter text and the rubric notes in the instructor fragment.)

Exercise 1 (define the BOP; three core coverages)

A business owners policy (BOP) is a pre-packaged commercial policy that combines, in one simplified contract at one premium, the core coverages a small, standardized, low-hazard business needs. The three core coverages it bundles: (1) property (building if owned, and business personal property — contents, inventory, equipment), (2) business income and extra expense, and (3) commercial general liability (premises/operations, products-completed operations, personal and advertising injury). Workers' comp, auto, and professional/specialty lines are not in a BOP.

Exercise 3 (define STP; the distinguishing phrase)

Straight-through processing (STP) is the fully automated handling of a submission from intake to bound policy with no human underwriting intervention. The distinguishing phrase is "the algorithm binds" (or "no human underwriting intervention"): STP makes and executes the underwriting decision itself — issuing a binding offer — rather than merely advising an underwriter who then decides. A model that recommends a price for a human to approve is augmentation (Ch. 31–32), not STP.

Exercise 5 (what algorithmic underwriting relocates)

Automated/algorithmic underwriting does not eliminate underwriting judgment; it relocates it. It moves the judgment (a) from the individual file to the rule set — the eligibility rules, the classifications, the rating plan, the score thresholds, the referral triggers — and (b) from the moment of binding to the moment the rules were written. The machine executes, at machine speed, the thinking an underwriting team did in advance. The underwriter "writes" a million policies a year without seeing one; they see the rules.

Exercise 7 (three layers of the STP pipeline)

Between intake and bind: (1) the rule engine — checks eligibility (eligible class, within size limits, in appetite) and applies knockout rules (hard stops that prevent auto-bind: ineligible class, cat-zone cap, claims-history flag, data gap); (2) rate and score — applies class rating and a predictive risk score to price the risk and decide whether it is clean enough to auto-bind or belongs in the grey band; (3) the decision routing — auto-bind the clean/in-tolerance risk, decline the knocked-out risk, refer the grey-band/uncertain risk to a human. (Intake/pre-fill builds the submission before the engine; bind/issue executes after it.)

Exercise 8 (the upside-down pyramid)

Small commercial is the broad base of the commercial market by policy count (millions of small businesses, each needing modest, standardized coverage) and the narrow tip by premium per policy (each pays only a few thousand dollars or less). The pyramid points opposite ways depending on what you measure. The forced consequence: you cannot afford an individual underwriting review on a tiny premium — the review would cost more than the policy earns — so small commercial must be underwritten by class and rule (standardized product, class rating, straight-through processing), not one file at a time, and its profitability is decided by the expense ratio (the cost of touching a policy) as much as by losses.

Exercise 9 (the eligibility line as underwriting)

"Auto repair — eligible up to 5,000 sq ft, no body/paint operations" is an underwriting judgment frozen into a rule: an underwriting team decided once that small auto-repair shops are acceptable risk but that larger ones, and the higher fire/paint-fume hazard of body and paint work, are not. That single decision is then applied automatically to every auto-repair submission — thousands of them — without a human re-deciding each time. The power is that a sound decision is made at scale; the danger is that if the rule (or the classification feeding it) is wrong, the error is systematic, repeated across the whole class.

Exercise 11 (BOP business-income standardization — help and harm)

The BOP often includes business income automatically (commonly on an actual-loss-sustained basis for a standard period), which helps the typical small business by partly solving the Chapter 19 trap — the business-income limit nobody calculated — since the coverage is simply present rather than depending on the agent deriving a limit. It under-serves a business with an unusual income exposure: a small operation with a very long recovery time (long-lead specialized equipment, a single irreplaceable location, a contract-penalty exposure) or an income exposure far larger than its class suggests gets the standardized default, which may be far too short or too low. Standardization is right for the standard business and dangerous for the exception — which is one signal the account may belong in a package, not a BOP.

Exercise 13 (underwrite the "marketing consultant")

The referral logic should stop the auto-bind and refer this to a human (or knock it out for inspection). The signal that should catch it is a data conflict / unusual combination: the class ("marketing consultant") is inconsistent with the address (a light-industrial park), the \$400,000 of business personal property (far more than a consultant carries), and a loading dock (a consultant does not ship). Any one of those is a grey-band flag; together they are a clear "this is not what it says it is" signal. What is probably going on: the business is misclassified — it is likely a light-manufacturing, distribution, or assembly operation coded as a benign office class, the classic §20.2 trap. Bind it as a "consultant" and you have applied an office rate to a manufacturing exposure. Refer, verify the operation, reclassify, and re-route (likely out of the BOP lane).

Exercise 15 (the machine shop through the STP pipeline)

Intake & pre-fill: the agent enters the basics; pre-fill confirms the address, the good protection class, and a light-metalworking industry code. Rule engine (eligibility + knockouts): the class (light metalworking, no welding) is within an eligible/near-eligible band; 4,000 sq ft is under the size cap; it is not in the coastal cat zone, so no catastrophe knockout fires; the clean five-year history trips no claims knockout; no data gap. Rate & score: class rating prices it off the light-metalworking class and a few modifiers; the model scores it cleanly (no adverse signal), within auto-bind tolerance. Bind & issue: nothing requires a human, so it auto-binds in seconds and issues. It should auto-bind because it is exactly the standardized, in-appetite, data-clean, no-story risk STP is built for — the salon's machinery applied to a small fabricator. (Contrast Harbor Steel, which fails the eligibility filter on class, size, complexity, and catastrophe.)

Exercise 17 (route three submissions)

(a) Ineligible class → decline (or refer) via a hard knockout rule — the eligibility filter is the signal family. (b) Clean, standardized, in-appetite office → auto-bind — it is the broad-middle case STP is for; no signal should fire. (c) Eligible class, marginal score, unusual coverage request → refer to a human via the grey-band (the marginal score) and the catch-all ("a coverage/limit request outside the standardized package," "an unusual combination") — the model is uncertain and the request departs from the standard, so a person decides.

Exercise 18 (price this risk — the economics)

(a) \$300 / \$1,600 = 18.75% of premium consumed by the fixed handling cost alone, before any claim. (b) On a \$250,000 account, \$300 / \$250,000 = 0.12% of premium — trivial. (c) The same fixed cost of human handling is negligible on a large account and crushing on a small one, so small commercial must automate to drive the cost of touching a policy low enough that a low-premium product can run an expense ratio the combined ratio survives. (All figures illustrative.)

Exercise 19 (combined ratio of the small-commercial book)

(a) Combined ratio = loss ratio + expense ratio = 62% + 41% = 103%. Above 100%, so the book is losing money on underwriting before any investment income (Ch. 3). (b) The most available lever in small commercial is the expense ratio (41% is very high), because the line's defining problem is the cost of touching a low-premium policy — automation, pre-fill, and straight-through processing attack exactly that. The chapter emphasizes the expense side because, unlike on a large account, the expense ratio here is large enough to be the difference between profit and loss; the loss ratio still matters, but the expense ratio is the battleground this line uniquely fights on.

Exercise 20 (the five-point class inadequacy at volume)

A five-point inadequate class rate bound automatically across 100,000 policies is 100,000 under-priced policies — the inadequacy is not isolated to one account but multiplied across the whole class at once, and it binds automatically, with no underwriter feeling the discomfort that would make a person slow down. On a single large account a five-point error is one negotiation, one file, caught or corrected in isolation; at small-commercial volume it is a systematic leak that emerges as losses across the book before anyone reacts. It illustrates the theme "pricing follows risk" (Ch. 1, Ch. 11) failing at industrial scale — the volume that makes the law of large numbers work also amplifies any pricing flaw. (Kept qualitative; no dollar figure invented.)

Exercise 21 (efficient but unprofitable)

Example: a digital small-commercial carrier achieves a very low expense ratio (say the high teens) through excellent automation — fast, cheap, frictionless. But its eligibility filter is too loose and its class rates are inadequate, so it binds a large volume of under-priced and marginally-selected business automatically; its loss ratio climbs past the point where the low expense ratio can save the combined ratio, and the book runs above 100%. What went wrong: it confused efficiency (low cost of handling) with profitability (adequate price and sound selection). Automation lowered the expense ratio but did nothing for the rate and selection discipline; the volume then amplified the rate leak. Efficiency is necessary but not sufficient.

Exercise 23 (the competitor tightens; your volume jumps)

This may be a red flag of adverse selection (Ch. 1), not a win. When a looser competitor tightens its filters, the marginal and misclassified risks it stops writing flow to the next-loosest filter in the market — possibly yours. A sudden jump in auto-bind volume concentrated in certain classes, with no change in your own pricing or marketing, suggests you may now be the cheapest door for risks others are turning away. What to check: the mix and score distribution of the newly-bound business (is it shifting toward the lower-quality end of each class?), the classes where the jump concentrated, early loss emergence in those classes, and whether your filters are now the loosest in the relevant segment. If so, tighten before the book turns.

Exercise 25 (the memo — route Harbor Steel to a package)

Sample (≈180 words):

Re: Routing — Harbor Steel & Fabrication. Harbor Steel should be underwritten as a commercial package policy, not a BOP, and should never enter the automated lane. Class: metal fabrication with a hot-work (welding/cutting) occupancy is a classic ineligible BOP class — the very hazard that drove the 2023 fire. Size: a 50,000 sq ft plant, ~\$45M revenue, ~\$11M payroll, ~180 employees is far over any small-commercial size cap. Complexity: the account needs coverages and a structure the BOP cannot provide — agreed-value property with a 5% named-windstorm deductible and an ACV roof endorsement, a calculated \$10M business-income limit on a 12-month period of indemnity, GL with a real products exposure, workers' comp, a 12-unit auto fleet, a \$10M umbrella, and inland marine. Catastrophe: the coastal, named-windstorm exposure exceeds any auto-bind cat cap. Routing principle: the BOP exists for risks with no story worth reading; Harbor Steel is all story — build it, do not pull it off the shelf.

Exercise 27 (note to the agent — referred, not auto-bound)

Sample: "Thanks for the submission — this one fell just outside our instant-quote criteria, so it's with an underwriter now and you'll have an answer by [time]; nothing's wrong, it just has a detail the system flags for a quick human look. Everything standard from you still binds in seconds, and I'll turn this one around fast." (Preserves the speed expectation and the relationship while being honest that a human is involved.)

Exercise 29 (resisting the pressure to over-automate — the ethical posture)

The discipline is to make the clean risks instant while holding the line on the knockouts and grey-band referrals, and to treat the speed you win on the easy risks as what buys you the right to be careful on the hard ones (§20.6). It is an ethical posture, not merely commercial, for two reasons: (1) insurance serves a social function (Ch. 1) — binding a risk the system cannot competently price, or declining one it should have referred and examined, harms real businesses and pollutes the pool; and (2) the carrier remains responsible for every automated decision (the algorithm is no defense — §20.4 Compliance Corner), so loosening filters to chase speed at the expense of sound selection is an abdication of that responsibility. Speed and care are not opposites here; speed on the standardized middle is precisely what funds the careful human attention the exceptions deserve.

Exercise 31 (Underwriting-File extension — the machine shop as a Read-the-Submission block)

Sample (constructed teaching example):

FIGURE 20.x — "The small shop that binds itself"        [constructed teaching example]
  THE SUBMISSION   A 4-person custom machine shop seeks a BOP: business personal property ~$600K
                   (manual lathes, a drill press, tooling), business income, and general liability,
                   in a 4,000 sq ft leased unit.
  THE CONTEXT      Good protection class; clean 5-year claims history; light metalworking, NO welding
                   or spray finishing; inland (NOT in the coastal cat zone); modest receipts.
  WHAT IT SHOWS    A standardized, low-hazard, in-appetite small risk with no story in its history —
                   an eligible (or near-eligible) class within the size cap, needing only the BOP's
                   standard coverages.
  WHAT IT DOESN'T  It does not (by itself) confirm the occupancy has not drifted — a light shop can
                   quietly add welding or finishing; the class flag and any pre-fill cannot see that.
  THE DECISION     AUTO-BIND. Nothing requires a human: no knockout fires, the model scores it clean,
                   the price is the class rate plus a few modifiers — bound in seconds.
  THE LESSON       This is the risk STP is FOR; the only thing separating it from Harbor Steel is scale,
                   hazard, complexity, and catastrophe. Route by those four, and verify the class.

Exercise 33 (the strong "STP is better than human judgment" claim)

The claim holds under tight conditions: the risk is standardized (its class is a reliable guide), the exposure is well-understood (the rules were built for it), the data is rich and accurate, and the account is within appetite with nothing departing from the pattern. In that broad middle, a model reading the behavior of the whole class beats a human "investigating" a risk they can only glance at for a tiny premium — the model is faster, more consistent, and free of fatigue and variability, while the human's judgment on a risk they cannot afford to examine is mostly guessing. The claim fails at the edge: the misclassified risk, the business doing something the class does not capture, the local hazard no variable encodes, the account changed since the data was gathered, the novel exposure the training data predates — the model is confident where it has seen the pattern and blind where it has not. Small commercial is the line where automation comes closest to replacing the individual decision because its risks are the most standardized in commercial insurance and the least able to bear individual investigation — the two conditions that most favor a class-and-rule machine over a human file-by-file.

Exercise 35 (the limit that Chapters 31–32 will revisit)

The limit this chapter names is that automated underwriting is confident where it has seen the pattern and blind where it has not — it cannot, on its own, catch the misclassified, the novel, or the changed risk, which is why the referral logic (the human backstop) is essential. Chapters 31 (data-driven underwriting) and 32 (predictive modeling) revisit exactly this: better data and better models widen the range of risks the machine can handle well, but they do not close the gap, because the gap is about judgment — knowing when the model is outside its competence and when to override it — rather than about data. More data sharpens the model inside the patterns it knows; it does not teach the model to recognize the case it has never seen, and that recognition is the underwriter's irreplaceable contribution (the model-override anchor, which pays off when the model scores Harbor Steel a 7 and the underwriter overrides to a 6 in Chapter 32).

Chapter 21

Worked solutions to the daggered (†) and odd-numbered exercises. (Odd-numbered discussion items not reproduced in full are developed in the chapter text and the case studies; the responses below give the load-bearing points an instructor would look for.) All numbers are illustrative/constructed.

Exercise 1 (definition of CGL)

Commercial general liability is the standard commercial policy under which an insurer agrees to pay the sums an insured becomes legally obligated to pay as damages to third parties for bodily injury, property damage, and personal & advertising injury arising out of the insured's premises, operations, products, and completed work. The two structural features: (1) it is third-party coverage — it protects the insured against claims by others, not against the insured's own losses (that is property insurance, Chapter 19); and (2) it carries a duty to defend that is typically broader than the duty to pay (the insurer must defend any claim that could be covered), with defense costs usually paid in addition to the limits.

Exercise 3 (premises/ops vs. products-completed ops)

Premises/operations is liability from the insured's ongoing activities and the condition of its premises — the harm happens while the insured is operating (the customer falls today, the trench caves in today). It is the short-tail half: observable, inspectable, pricable from the record. Products-completed operations is liability from the insured's products and finished work after the insured has parted with them — the harm happens later, off-site, outside the insured's control (the bracket fails in year nine; the finished roof leaks two winters on). It is the long-tail half because the latency between the act and the claim can run years, which is where the real underwriting risk in CGL usually lives.

Exercise 5 (exposure base + bases by class)

An exposure base is the variable that measures the size of an insured's loss exposure and to which the rate is applied to compute premium; it is chosen to track the exposure and must be auditable. Standard liability bases, one class each: payroll (contractors, service firms); gross sales/revenue (manufacturers, retailers, restaurants); area/square footage (landlords, lessors of premises); admissions/headcount (theaters, events); units or total cost (some installation contractors).

Exercise 6 (products-completed operations aggregate)

It is a separate aggregate cap on all products and completed-operations claims in the policy period, distinct from the general aggregate. It is separate because products/completed-ops is the long-tail, high-severity half of Coverage A; giving it its own aggregate lets the insurer cap and price that exposure deliberately rather than letting it consume the same limit as the everyday premises claims. On a products account, that limit is doing real work, and setting it (and pricing for it) is a conscious decision, not a default.

Exercise 7 (additional insured)

An additional insured is a party other than the named insured added to the policy — usually by endorsement — as an insured for liability arising out of the named insured's operations, products, or premises. A contract typically demands one because an upstream party (a general contractor, an owner, a lender) wants the protection of the downstream party's coverage for claims arising out of that downstream party's work, paid for by the downstream party's premium.

Exercise 9 (the four management/professional lines and the gaps they fill)

E&O / professional liability fills the professional-services exclusion — negligent advice or service (the accountant, architect, consultant). D&O covers the personal liability of directors and officers for wrongful acts in managing the company (fiduciary breaches, mismanagement, shareholder suits), which the CGL (covering the company's third-party injury liability) does not reach. EPL covers employment-related wrongful acts (discrimination, harassment, wrongful termination) — the CGL excludes employee injury and doesn't reach these employment torts. Cyber covers data-breach and network-event liability and costs — an exposure the modern CGL largely excludes.

Exercise 10 (the two failed-bracket scenarios)

Scenario (a) — the only loss is that the bracket itself is now worthless: excluded by the "your product" business-risk exclusion. The CGL is not a warranty on the insured's own product; it will not pay to repair or replace the insured's own defective work. Scenario (b) — the bracket's failure drops a mezzanine and injures a worker: the resulting third-party bodily injury and property damage is covered under products-completed operations. The line: the CGL does not insure the quality of the insured's work; it insures the consequences to third parties when that work goes wrong.

Exercise 13 (duty to defend; defense outside limits)

The duty to defend obligates the insurer to defend any claim that could fall within coverage — even one that is ultimately groundless. Because litigation costs money regardless of outcome, a frivolous suit the insured wins still generates real defense expense (attorneys, experts, court costs). Defense costs are usually paid outside (in addition to) the policy limits so that defending the insured does not erode the indemnity available to pay a covered judgment or settlement — meaning the insurer's true cost on a claim is indemnity plus defense, and on many liability accounts defense is a large share of the total.

Exercise 15 (why the industry moved to claims-made in the 1980s)

Under an occurrence form, an insurer owns the latency between harm and claim: a policy written in the 1950s–60s accepted asbestos exposures whose diseases manifested in the 1980s+, and pollution contamination litigated decades later (given new force by the 1980 CERCLA/Superfund regime). Those claims landed on old policies priced with no knowledge of them, leaving an open-ended, nearly un-reservable tail (the incurred-but-not-reported problem). These long-tail catastrophes helped drive the mid-1980s liability insurance crisis. The claims-made trigger solved it for the insurer by closing the tail: the insurer's exposure for a policy year is fixed once that year's claims-reporting window shuts, making the liability reservable and pricable again. (See Case Study 1.)

Exercise 17 (frequency vs. severity in the two halves)

Premises/operations is mostly a frequency game: many small, predictable losses (slips, minor incidents) — manageable through loss control and routine pricing. Products-completed operations is mostly a severity game: rare, but capable of a single catastrophic loss (a structural collapse, a multi-state contamination) that exhausts the limit and reaches the umbrella. You underwrite them with different instincts — tighten controls and pricing against frequency, but underwrite survivability against severity — because a clean frequency record tells you nothing about the rare severe claim that decides the account.

Exercise 18 (split classification of a maker-and-installer)

A custom manufacturer that both makes products (exposure scales with dollars of product sold) and installs them at customer sites (exposure scales with labor/payroll) presents two different exposures. Rating the whole account on sales alone under-captures the installation injury exposure (a sales base doesn't reflect how much field labor creates third-party injury risk); rating it on payroll alone under-captures the products exposure (payroll doesn't reflect how much product is out in the world able to fail). Proper classification splits the operations and applies the right base — and the right class code — to each part, which is exactly why the application asks for operational detail.

Exercise 19 (audit mechanics and the understated projection)

(a) The premium increases: it was quoted on estimated sales of \$25M but the audit found \$40M of actual exposure, so the audit collects the additional premium for the exposure actually presented. (b) The adjustment is possible because the exposure base is auditable — gross sales is measurable and verifiable from the insured's records, so the carrier can true the estimate up to the actual figure at year-end. (c) A deliberately understated projection is a red flag (Chapter 33) because it lowers the up-front premium while the insured presents the full exposure during the term; if the projection is lower than last year's audited figure despite a growing order book, the discrepancy suggests either error or an attempt to defer premium, and warrants a question before binding.

Exercise 21 (underwrite: decorative vs. structural fabricator)

The decorative-railing fabricator is the materially better products risk: its components are non-load-bearing, so the products-completed operations severity is far lower — a failure is cosmetic or minor, not a collapse — and it has no pending claim. The structural-bracket fabricator carries load-bearing components (catastrophic failure mode), long latency (slow steel failure modes), a litigious construction field, and a live claim proving the exposure is real. Questions for the decorative shop: confirm nothing it makes bears a load, confirm installation scope, check the field's litigation history. Questions for the structural shop: the pending claim's facts (design vs. fabrication defect, isolated vs. pattern), QA/weld records, where its steel is installed, and contract/indemnity terms. Pricing: rate both on sales, but the structural shop gets a higher products loss cost, a debit for the pending claim and thin controls, and a deliberately-set (likely higher) products aggregate; the decorative shop can be written nearer the class rate. The disciplined choice is the decorative risk at target loss ratio — or the structural risk only at a products rate that respects the tail, never priced off its clean premises record.

Exercise 23 (find the red flag)

Three red flags: (1) products rate ~25% below comparable accounts — suggests the products exposure was under-rated or priced like a premises-only risk (§21.4), the classic long-tail underpricing. (2) a three-year loss run with "nothing on products" — on an occurrence long-tail line this is expected and proves nothing; the worst products claims may simply not have been reported yet (§21.2), so "clean" recent experience is being misread as a clean risk. (3) a revenue projection lower than last year's audited figure despite a growing order book — an understated exposure base that lowers the up-front premium and a possible disclosure issue (§21.3; Chapter 33). Together they describe an account that looks cheap and clean precisely on the dimensions where a long-tail risk hides its cost.

Exercise 24 (additional insured + contractual liability)

(a) Naming the general contractor as an additional insured means Harbor Steel's CGL must now defend and indemnify the GC for claims arising out of Harbor Steel's work: the per-occurrence limit is shared among more parties, the coverage may (depending on endorsement edition) reach the GC's own negligence, and Harbor Steel's carrier loses subrogation against the GC (you can't subrogate against your own insured). (b) The CGL by default covers tort liability (what the law imposes when the insured injures someone); contractual liability is liability the insured assumes by contract (a hold-harmless/indemnity covering another party's liability) that it would not otherwise have had. The "insured contract" exception gives back coverage for a defined category of business agreements, but an indemnity broader than that definition is an uninsured assumption the insured carries naked.

Exercise 27 (price the tail — illustrative)

Given (constructed): base products loss cost \$3.00 per \$1,000 of sales; 15% debit modifier; projected sales \$45,000,000. (a) Modified products loss cost = \$3.00 × 1.15 = **\$3.45 per \$1,000 of sales. (b) Indicated products loss contribution = \$3.45 × (\$45,000,000 ÷ \$1,000) = \$3.45 × 45,000 = \$155,250 (before expense and profit loads). (c) This figure could still be inadequate even if it looks fine against this year's reported losses because the products exposure is long-tail: the worst completed-operations claims occur this term but won't be reported for years, so reported experience understates ultimate losses, and a loss cost calibrated to immature reported data systematically under-funds the tail. (See §21.2, §21.4; Chapter 10 on development.)

Exercise 29 (umbrella as a severity question; nuclear verdicts)

Umbrella pricing is "almost entirely a severity question" because the umbrella, by design, is the layer that pays the catastrophic claim — it only attaches after the primary limit is exhausted, so its losses are the rare, large ones, not the everyday frequency claims the primary absorbs. The nuclear-verdict trend (Chapter 23) — liability verdicts reaching once-unthinkable sizes — therefore presses hardest on the excess layers: a routine verdict stays within the primary and never touches the umbrella, but an outsized verdict blows through the primary and lands squarely in the excess layer, which is exactly where the severity inflation is felt.

Exercise 30 (the memo)

A model response (≈170 words): "Harbor Steel — CGL within the package. The premises/operations exposure is routine: a standard fabrication-shop slip-and-fall and forklift profile, clean loss runs, pricable from the record. The risk that decides this account is products-completed operations: Harbor Steel fabricates load-bearing structural steel installed in third-party buildings — catastrophic failure mode, long latency, litigious construction field — and there is one pending products-liability claim alleging a fabricated bracket failed. I'm classifying the products exposure on a gross-sales base (~\$45M, auditable), splitting any field-installation work to its proper class, and setting the products-completed operations aggregate deliberately with the tail in mind — not pricing this like a premises-only risk. The pending claim is documented and on watch; I'll need the claim file, QA/weld records, and the engineering to judge whether it's isolated or a pattern. Expect additional-insured demands from Harbor Steel's GCs — I'll control the endorsement edition and price for the limit-sharing. Net: writable, with the products tail respected in the rate and the aggregate."

Exercise 31 (ethics dilemma)

The case against "rate the products like the premises": it deliberately underprices a known long-tail exposure, violating rate adequacy (Chapter 11). Because completed-operations losses surface years later, the loss ratio will look fine this year and next — the "we'll re-rate at renewal" logic — but the carrier is accepting the products exposure now, on an occurrence form, at a rate that doesn't fund it; when the tail develops, the carrier faces several accident years of under-reserved business and a combined ratio (Chapter 3) driven past 100% on losses set in motion years earlier. What you owe the future policyholders of the pool: a rate adequate to the risk accepted, so that this account's late-arriving losses are funded by this account's premium rather than subsidized by tomorrow's insureds (the social-function and pricing-follows-risk themes). The honest line: compete aggressively on coverage, service, terms, and genuine risk-quality credits — but do not cut a products rate below the level the known tail requires. The discipline to hold that line in a soft market, with the broker holding competing quotes, is the chapter's central lesson.

Exercise 33 (open Harbor Steel's GL page)

Sort the exposure: premises/operations — routine shop exposures (slips, forklift, visitors), short-tail, pricable from the record. Personal & advertising injury — minimal; not an advertising-driven business. Products-completed operations — the watch-item: load-bearing structural steel installed in third-party buildings, long latency, litigious field, with one pending bracket-failure claim that proves the exposure is live. Settled this chapter: the GL is assessable and writable; the premises piece is routine; the products piece is the real risk; additional-insured and contractual exposure is typical of a fabricator and must be controlled and priced. Unsettled for later chapters: whether the pending claim is isolated or a pattern (needs the claim file, QA records, engineering); the final products rate and terms (Chapter 11); the umbrella's interaction with the CGL (Chapter 16/40); reinsurance treatment (Chapter 27). Running disposition: GL assessed; products-completed operations is the watch-item.

Exercise 35 (additional-insured demands on Harbor Steel)

Because Harbor Steel supplies general contractors, its customer contracts will require it to name those GCs (and often owners/lenders) as additional insureds and to indemnify them. The effect on this account: Harbor Steel's CGL limit becomes shared among Harbor Steel and those upstream parties; depending on the endorsement edition, coverage may reach the GCs' own negligence; and the carrier loses subrogation against them. Before granting, require of the broker (Chapter 39): a sample of Harbor Steel's standard customer contract, confirmation that the additional-insured and indemnity language is within what the carrier's form supports, control of the endorsement edition (current, tightly-worded), and disclosure of the volume of contracts Harbor Steel signs — then price the account knowing its CGL funds upstream parties' risk on top of its own tort exposure. (See Case Study 2.)

Chapter 22

Worked solutions to the daggered (†) and odd-numbered exercises. (Discussion-only items are summarized; all figures are illustrative.)

Exercise 1 (define workers' compensation; the grand bargain)

Workers' compensation is a statutory, no-fault system requiring employers to pay defined benefits (medical, indemnity, disability, death) to employees injured or made ill by their work, regardless of fault. The grand bargain: the worker gives up the right to sue the employer in tort (and the chance at a large jury verdict) in exchange for prompt, certain, no-fault benefits for every covered injury without proving negligence; the employer gives up its common-law defenses in exchange for a bounded-per-claim liability and immunity (exclusive remedy) from most injury suits by its own employees.

Exercise 3 (the four benefit categories)

Medical (full reasonable-and-necessary treatment, typically uncapped, no worker deductible); indemnity (partial lost-wage replacement, a percentage of average weekly wage up to a state maximum); disability (scheduled/unscheduled payments for permanent impairment); and death benefits (payments to dependents plus a burial allowance). The largest controllable cost is usually the indemnity tail — the lost-wage payments that accrue for every week the worker is off work — which is precisely what return-to-work attacks (§22.7).

Exercise 4 (NCCI class code; governing class; standard exception)

An NCCI class code groups employers by the type of work their employees do and carries a published loss cost per \$100 of payroll. The governing class is the basic classification that best describes the business as a whole — in practice, usually the largest non-standard-exception payroll — and it anchors how the risk is rated. A standard exception (clerical office employees, outside salespersons, drivers in some states) is genuinely different, lower-hazard work that is split out and rated separately rather than folded into the governing class. You cannot move welders into clerical or sweep office staff into the welding class; the bureau's rules decide which payroll goes where.

Exercise 5 (one-line definitions)

X-mod — a filed multiplier on manual premium reflecting an employer's own multi-year losses vs. its class (debit >1.00, credit <1.00), weighting frequency. Premium audit — the post-term count of actual payroll/records to settle the final premium against the deposit. Monopolistic state fund — a state-run insurer that is the only lawful WC source in that state (private carriers barred). Employer's liability — Part Two of a WC policy, a limited liability coverage for injury suits outside the no-fault bargain.

Exercise 7 (the two coverage parts; which carries limits)

Part One — Workers' Compensation: the statutory, no-fault benefit with no dollar limit. Part Two — Employers' Liability: a liability coverage with dollar limits. Part One has no limit because the statute imposes none; Part Two has limits because it responds to ordinary liability suits (third-party-over, dual-capacity, consequential) that fall outside the statutory bargain.

Exercise 8 (classification arithmetic) †

Manual premium per class = (payroll ÷ 100) × loss cost. - Fabrication (governing): (6,000,000 ÷ 100) × \$5.00 = 60,000 × 5.00 = **\$300,000. - Drivers: (1,200,000 ÷ 100) × \$3.50 = 12,000 × 3.50 = **\$42,000. - Clerical: (900,000 ÷ 100) × \$0.30 = 9,000 × 0.30 = **\$2,700. - Total manual premium = \$344,700** (before mod, schedule rating, and loads).

Exercise 9 (apply the mod)

Using the \$344,700 total from Exercise 8: - At a 1.20 mod: \$344,700 × 1.20 = **\$413,640. - At a 0.85 mod: \$344,700 × 0.85 = **\$292,995. The mod is applied to the total manual premium (after summing classes), before schedule rating and loads.

Exercise 10 (the two dangers of misclassification) †

(1) To the premium: moving \$1,000,000 of fabrication payroll into a lower-rated class under-collects the rate difference — at, say, \$4.50 vs. \$0.30 per \$100, that is (1,000,000 ÷ 100) × \$4.20 = \$42,000 of premium never charged for exposure that is genuinely there, and the losses will still arrive. (2) To the file/downstream readers: it mis-states the risk to everyone who relies on the file — pricing, the audit (which will reclassify and create a refund/collection fight), reinsurance, and portfolio management — so the whole chain reasons from a false picture. Classification is the foundation the price and the file stand on.

Exercise 11 (the erection question)

Treat classification as a question of fact to be verified, not negotiated. A documented fall-from-height at a customer job site is direct evidence the firm performs (or recently performed) steel erection — a far more hazardous class than shop fabrication. If the underwriter accepts "no erection anymore" without verification, the premium audit will classify what the records actually show; if erection payroll is found, the insured owes additional premium at the much higher erection rate, producing an unhappy insured and, if undocumented, possibly the highest-rated applicable class by rule. Verify at inspection and in the loss runs.

Exercise 13 (why the plan weights primary over excess) †

Each claim is split into a primary portion (the first slice, up to a split point) and an excess portion (the rest); the plan weights the primary portion far more heavily. The underwriting truth it is built around: frequency is more controllable and more predictive than severity. How often workers get hurt reflects the employer's safety culture (which it controls); how badly a given injury turns out is partly the lottery of circumstance. By weighting primary (frequency) dollars, the mod holds the employer sharply accountable for the pattern it can change and only partly for the severity it often cannot — which also makes the mod largely a frequency signal, the leading indicator of the severe claim not yet incurred.

Exercise 15 (direction of the mod; credibility and size) †

Actual losses below expected → mod below 1.00, a credit. A larger employer's mod swings further from 1.00 on the same relative experience because its own loss history is more credible (Ch. 10): with more exposure, the plan trusts the company's own data more and the class average less, so the same good (or bad) relative result moves the mod more. A small employer's experience is thin and less credible, so its mod stays closer to 1.00 — the class average dominates.

Exercise 17 (the mod costs work, not just premium)

A company's X-mod is portable (it follows the company) and widely used outside insurance: general contractors and owners routinely require a mod below 1.00 to bid or to be allowed on a job site, and some write a maximum mod into contracts. So a debit mod can lock a company out of work, not merely raise its premium. The incentive you can harness: because a sub-1.00 mod wins business, the better employers are strongly motivated to drive frequency down — exactly the behavior you reward and verify with return-to- work and loss-control credits (§22.7).

Exercise 18 (underwrite the WC) †

Posture: modify (quote with conditions), not decline. The risk is a controllable frequency problem, not a severity catastrophe: many small back-and-laceration claims, a rising frequency trend, a 1.18 debit mod, no large losses. Conditions/credits: require and schedule-credit a formal return-to-work program (welding shops have abundant light duty — inspection, layout, fixturing); require a documented safety/loss- control program targeting the recurring back-injury cause (lifting/ergonomics); price at the debit mod the data earns; consider a modest schedule debit if controls are weak. Defending it to the broker: "The 1.18 mod is your client's own three-year experience, weighted toward the frequency that's still rising — not my opinion. The absence of a big claim isn't safety; it's luck that hasn't run out. I'll reward real improvement: put the return-to-work and ergonomics programs in place, drive the frequency down, and the mod (and the price) follow it down over the next three years."

Exercise 19 (how return-to-work flows through)

(a) Indemnity cost: falls — workers back on light duty quickly accrue far fewer lost-wage weeks, and moderate claims stop drifting into permanent disability. (b) X-mod over three years: drifts down — shorter, smaller claims (especially keeping medical-only claims from becoming lost-time claims) reduce the primary/frequency dollars the mod weights most, so actual-to-expected improves. (c) Loss ratio: falls — lower incurred losses against the same premium; and the better mod can win contracts (more, better business). All three move together because they are driven by the same thing: getting workers back fast.

Exercise 21 (price this risk) †

Manual premium by class: - Governing: (10,000,000 ÷ 100) × \$4.60 = 100,000 × 4.60 = **\$460,000. - Drivers: (1,000,000 ÷ 100) × \$3.40 = 10,000 × 3.40 = **\$34,000. - Clerical: (1,000,000 ÷ 100) × \$0.30 = 10,000 × 0.30 = **\$3,000. - (a) Total manual = \$497,000 (the class loss cost × payroll — the expected-loss core per class). - (b) Apply 1.22 mod: \$497,000 × 1.22 = **\$606,340 (the company's own experience moving the price). - (c) Apply 10% schedule credit: \$606,340 × 0.90 = **\$545,706 (documented loss controls the mod doesn't capture, per Ch. 11). Indicated premium before expense and profit loads ≈ \$545,706.**

Exercise 23 (the low deposit)

At audit, actual payroll is counted; because the true payroll was 20% above the estimate, the insured owes substantial additional premium. The problem during the term: a deliberately low deposit means the insurer is under-collateralized all year — it is on risk for the full statutory benefit on the real payroll while holding premium for only 80% of it, a cash-flow/credit exposure (and audit leakage if the additional premium is later disputed or uncollectible). A sane deposit should track a believable payroll estimate, sanity-checked against revenue and prior audits.

Exercise 25 (clean severity, rising frequency)

It is a red flag because the X-mod — and good judgment — is built around frequency, which is the leading indicator of severity. Many small injuries signal a safety-culture problem; that same culture is what eventually produces the catastrophic claim. A clean severity record with rising frequency is not a good risk getting lucky on big losses — it is a deteriorating risk that has not had its bad day yet. Price and condition on the frequency trend and the loss-control response, not on the comforting absence of a large claim.

Exercise 27 (coverage recommendation memo) †

Model answer (≈180 words). "Recommendation: bind WC at a debit X-mod with a return-to-work credit and loss-control conditions. Governing class: structural-steel/metal fabrication (confirm at inspection whether field erection payroll exists — a materially higher class). The account's own three-year experience produces a debit mod (>1.00), driven by a recurring pattern of back-strain and laceration claims; I am pricing the mod the data earns, not discounting it. The single most important loss-control condition is a formal return-to-work program, credited on the schedule: light duty is abundant in a fab shop, so it will shorten the indemnity tail on the back injuries that dominate the losses and bend the mod down over three years. I am also requiring an ergonomics/lifting program targeting the loss cause. File line for the auditor: 'Priced at the filed X-mod (a debit) reflecting the insured's own loss experience; return-to-work and ergonomics programs required as conditions and credited per the schedule rating plan; classification verified at inspection.' The catastrophic single-claim tail is addressed through reinsurance, not the working-layer price."

Exercise 29 (ethics: correcting mod data vs. abuse) †

Correcting erroneous loss data that inflates a mod (e.g., a claim coded to the wrong employer, a reserve never taken down after a claim closed cheaply) is a legitimate practice — the mod is supposed to reflect the company's true experience, and an error that overstates losses overcharges the insured and misprices the risk. The insured (or its broker) is entitled to have the data be accurate; fixing it makes the mod more correct, not less. This is categorically different from abuse: misclassifying payroll into a lower class to dodge premium, hiding operations, or splitting a company to dilute experience are misrepresentations of fact that defeat the rating system and shift cost onto honest insureds. The line is truth: correcting data toward reality is fair; distorting reality to lower the number is not.

Exercise 31 (illegally uninsured workers — the fairness problem)

Beyond the employer's legal exposure (loss of exclusive remedy, statutory penalties): the workers bear the risk of an injury with no certain benefit and an employer that may be unable to pay; the honest employers in the pool are undercut by a competitor who illegally avoids the cost of coverage and may underbid them; and your own loss experience is distorted because exposure that should be priced and reserved is invisible until a claim surfaces. It is the adverse-selection/fairness problem (Ch. 1) in statutory form — the uninsured employer free-rides on the system that protects everyone else.

Exercise 32 (Harbor Steel — classify and price) †

Governing class: structural-steel/metal fabrication (welders and fabricators carry the bulk of payroll and hazard); drivers split into a driver class; office into clerical. X-mod direction: a debit (>1.00) — the file's several WC claims (back injuries, a serious laceration near-miss) are the frequency pattern the plan weights heavily, so the company has run worse than its class. Loss-control lever: a return-to-work program (credited) — light duty is plentiful in a fab shop, shortening the indemnity tail and bending the mod down. Classification to verify at inspection: whether Harbor Steel erects steel at customer job sites — field erection is a far more hazardous class and must be classified (and priced) as such if meaningful payroll belongs there.

Exercise 33 (the catastrophic tail)

You do not price the catastrophic single-claim possibility (a young worker with a life-altering injury) into the working layer because it is a low-frequency, very-high-severity event that would distort the guaranteed-cost price and is better handled by spreading it: the tail is ceded to reinsurance (Ch. 27). The working layer prices the frequent, predictable losses; reinsurance absorbs the rare, volatile, statutorily-uncapped severe claim — the same logic by which catastrophe property risk is ceded rather than fully retained.

Exercise 35 (extend the file)

Settled this chapter (any three): the WC governing class (fabrication) is identified; the X-mod direction (a debit) and its driver (frequency) are read; the pricing posture (debit mod + return-to-work credit) is set; the erection classification question is flagged for inspection. Still open (any two): the premium audit exposure on the ≈\$11M payroll estimate; reinsurance treatment of the catastrophic single-claim tail (Ch. 27); and the judgment question the mod can't answer — whether the safety culture is actually improving (read the open claims, verify the program at renewal).

Chapter 23

Worked solutions to the daggered (†) and odd-numbered exercises. All figures are constructed teaching examples. Section references point back to index.md.

Exercise 23.1 (†)

Commercial auto is the line that insures vehicles used in a business, most commonly written on the business auto policy (BAP). The single most important risk difference from the personal auto policy (Chapter 14): severity. A commercial truck at fault in a serious bodily-injury crash presents a corporate defendant and a commercial policy — exactly the target the plaintiff's bar pursues — so the line's underwriting problem is the rare, enormous loss, not the frequency of ordinary crashes. (§23.1)

Exercise 23.3

Radius of operations is how far from its home base a vehicle typically travels. The three common classes are local (short radius, often under ~50 miles), intermediate (roughly 50–200 miles), and long-haul (beyond that). As radius grows, so does time on the road, highway speed, driver fatigue, distance from home support, and hours-of-service pressure — a correlated bundle of severity drivers, which is why the rate climbs with the band. (§23.4)

Exercise 23.4 (†)

Hired & non-owned auto (HNOA) is liability coverage for vehicles the insured uses but does not own. Hired autos are vehicles the insured rents, leases, hires, or borrows — e.g., a rented box truck used to cover a delivery when a company unit is in the shop. Non-owned autos are vehicles owned by others used for the insured's business — most importantly an employee's personal car driven on a company errand. Both can make the company (and its insurer) liable even though the company never owned the vehicle. (§23.5)

Exercise 23.5

A nuclear verdict is an exceptionally large jury award — conventionally tens of millions of dollars or more, far above the economic damages. This line is where they cluster because a commercial truck crash combines catastrophic injuries with a "deep pocket" corporate defendant and a commercial policy, which is precisely the profile an aggressive, well-financed plaintiff's bar is built to pursue. (§23.6)

Exercise 23.7

Fleet rating treats the whole fleet as one rated unit and leans on the fleet's own loss experience (experience rating, Chapter 11), blending it with the class rate; individual rating builds the premium vehicle by vehicle off the manual (this vehicle, this driver, this territory, this use). Fleet rating takes over once the fleet crosses a size threshold (set by the plan, often around five vehicles but variable) — the point at which the fleet generates enough of its own loss history to be partially credible (Chapter 10, the law of large numbers applied to one account). (§23.2)

Exercise 23.9 (†)

Same headcount, different risk. Account A (fifteen local sedans) is a frequency risk with modest severity — lots of small claims, predictable, priced near the class with attention to the drivers and territory. Account B (fifteen heavy flatbeds hauling steel on an intermediate radius) is a severity risk first: heavy units plus dense, shifting cargo plus public highways is the nuclear-verdict profile. Your attention and your rate should shift accordingly — far more diligence on B's drivers, limits, attachment, and controls, and a severity-aware rate that does not let B's (likely clean) frequency history argue you out of pricing the tail. The single characteristic driving the gap is vehicle weight (and the cargo that rides on it): weight is destiny in auto severity. (§23.2)

Exercise 23.11 (†)

A forty-unit fleet generates a multi-year loss history large enough to be partially credible (Chapter 10): the plan blends the fleet's own experience with the class rate, so a clean larger fleet earns a credit its size has justified and a loss-heavy one can no longer hide inside the class average. A three-unit fleet generates too little experience to be credible, so it is rated essentially off the class. The underlying principle is the law of large numbers (Chapter 1) applied to a single account — more exposure units mean the fleet's own average is a more stable estimate of its true loss cost — which is exactly what credibility (Chapter 10) formalizes. (§23.2; Ch. 1, Ch. 10)

Exercise 23.13 (†)

• Single speeding violation, 31 months old: noise. Note it and move on; one stale minor is not a pattern.
• Two minor violations plus an at-fault accident in 18 months: a material negative. A pattern is emerging on a recent window. Weigh the at-fault accident's severity and context; at minimum this driver earns a debit and counseling, and on a heavy unit you scrutinize hard.
• DUI eight months ago on a heavy flatbed: disqualifying — remove the driver from the policy as a condition of coverage. This is not a "rate-up" because the math doesn't allow one: when a single at-fault serious-injury claim can exceed the fleet's lifetime premium, there is no rate that adequately prices a recent major violation on a heavy unit. You price the frequency you can see; you cannot price the severity a DUI driver on a flatbed threatens, so you remove the exposure instead of charging for it. (§23.3)

Exercise 23.15 (†)

The MVR shows convictions, not behavior, so three things it cannot tell you: (1) how the driver actually drives day to day (a constant speeder never caught looks clean; one unlucky ticket looks worse than the driver is); (2) the conditions the driver faces — routes, hours, dispatch pressure, fatigue; (3) recent or current behavior, since it is backward-looking and varies in completeness by state. Telematics (§23.7) addresses the gap by observing the whole of driving behavior directly and in real time — harsh braking, following distance, speed, hours — rather than inferring it from the small, biased subset of behavior that produced a citation. The MVR is necessary; telematics is what makes it sufficient. (§23.3, §23.7)

Exercise 23.17 (†)

Two explanations for "local" on the application with crashes 250 miles out: (1) misclassification — the account genuinely runs farther than "local" and the radius field is simply wrong; you fix the classification and the rate to reflect the true (intermediate or long-haul) exposure. (2) misrepresentation — the account understated its radius to lower the premium; this is a disclosure problem governed by Chapter 33 (material misrepresentation), and it requires you to clarify, document, and decide whether the application can be relied on at all. The discipline is to test the radius claim against the loss run rather than take it on faith, because radius is one of the easiest fields to understate and most expensive to get wrong. (§23.4; Ch. 33)

Exercise 23.19

You reconcile them by trusting the independent record over the self-description: a carrier that says it runs a tight, safety-first operation but whose public FMCSA profile shows an elevated out-of-service rate and a high unsafe-driving percentile is contradicting itself, and the chapter is explicit that the public data is usually the more honest witness. Practically, the FMCSA profile becomes a subjectivity-generator: you press on the discrepancy, require corrective evidence, and price (or decline) on what the record shows, not what the application claims. (§23.4)

Exercise 23.20 (†)

The exposure: employees using personal cars for company deliveries creates a real, occasionally catastrophic non-owned auto liability. The doctrine that reaches the employer is respondeat superior — an employer can be liable for an employee's acts in the course of employment, regardless of who owns the vehicle. The employee's personal auto policy responds first, but at a personal-sized limit; when injuries are severe and the verdict is commercial-sized, the plaintiff looks past the employee to the company, and the company looks to you. HNOA written for a nominal premium without questions is a trap because the exposure here is genuinely large (frequent deliveries by personal car) and is being priced and assessed as if it were trivial; you must ask how many employees drive personal vehicles for work, how often, under what supervision, and whether the company verifies those employees carry sane personal-auto limits. (§23.5)

Exercise 23.21

A gap in the primary auto's HNOA coverage becomes the umbrella's problem because the \$10M umbrella (Chapter 16) is expected to sit over the auto liability including the hired-and-non-owned exposure. If the primary is written on a narrow symbol that excludes hired/non-owned autos, the umbrella either inherits the gap (no underlying coverage to sit over) or you get a dispute about whether the umbrella must "drop down" to fill it. This is why coverage architecture (Chapter 5) is not abstract: the symbol on the primary declarations decides whether a real claim is paid and whether the tower above it attaches cleanly. (§23.5; Ch. 5, Ch. 16)

Exercise 23.22 (†)

Model response to the broker (holds the rate): "I hear you — the loss history really is clean, and the program sounds well run. But a fleet's loss run records the crash frequency I can see, not the severity tail I can't, and on this line the money is in that tail. A spotless fifteen-year fleet and a fleet that draws a thirty-million-dollar verdict next year can be the same risk today — the difference is one bad intersection in the wrong venue, which no clean history rules out. I'm not pricing for the fender-benders; I'm pricing for the catastrophic loss that severity inflation has made both more likely and more expensive. That's the rate." (§23.6)

Exercise 23.23

Social inflation (severity inflation) is the rise in claim costs driven by the legal and social environment rather than by more accidents or ordinary economic inflation. Three contributing factors the chapter names: a more aggressive, better-financed plaintiff's bar (including third-party litigation funding); "anchoring" / the normalization of very large damages, which pulls awards upward over time; and jury attitudes toward corporate "deep pocket" defendants. It pushes risk into the tail rather than the middle because it inflates the largest awards specifically — the rare catastrophic verdict — rather than the cost of routine claims, so the line's loss distribution grows a long, expensive right tail that does not average out on a single fleet. (§23.6)

Exercise 23.24 (†)

Even though no rate or control can price the nuclear verdict away, the disciplined underwriter does four things, which compound: (1) manage limits and attachment — be careful how much primary limit you offer, make sure the umbrella attaches at the right point, and know where in the tower the catastrophic loss lands, so the carrier's net exposure to the tail is controlled (accomplishes: caps the downside). (2) charge an adequate rate held against the broker's push (accomplishes: funds the tail you can't predict). (3) select hard on drivers (accomplishes: reduces the number of serious crashes that could go nuclear — you can't predict which crash, but you can reduce how many). (4) require controls — telematics and cameras (accomplishes: lowers frequency and produces evidence that defends the claims that do happen). The pattern: you cannot remove the tail, so you reduce its frequency, fund it in the rate, cap it with structure, and arm yourself to defend it. (§23.6, §23.7)

Exercise 23.25 (†)

The first question — "Do you have telematics?" — is the trap, because a "yes" tells you almost nothing: a black box nobody reviews improves no risk. The second question — "What do you do with the data?" — is the one that matters: a fleet that reviews scores weekly, coaches its worst drivers, documents improvement, and has actually removed a driver the data condemned is running a closed-loop safety program, and that is the credit-worthy risk. The second answer determines whether a credit is warranted because the schedule-rating credit (Chapter 11) belongs to the behavior change, not the hardware — you price the program, not the purchase. (§23.7)

Exercise 23.27

Two genuine limits of telematics: (1) it does not prevent the nuclear verdict driven by venue, injury, and the litigation environment — factors outside the fleet's control (§23.6); it reduces serious-crash frequency and improves defense, which is meaningful but not total. (2) it can be gamed or ignored — a fleet can install the hardware to win a credit and quietly disable the coaching, and it raises real privacy/labor questions the insured must navigate. These limits do not change the conclusion that telematics, required and acted on, is the best loss-control lever this line has, because the alternative (no direct visibility into driver behavior, relying only on the conviction-only MVR) is strictly worse — the limits argue for underwriting the program, not for abandoning the tool. (§23.7)

Exercise 23.28 (†)

Net modification: a 20% debit partially offset by a 10% credit nets to a 10% debit. Indicated liability premium = \$120,000 × 1.10 = **\$132,000. Why you would not simply net them to a small modification and move on: the debit and the credit are doing different jobs and respond to different facts — the debit prices the elevated risk of the aging roster and weak process (a condition you'd like to fix), while the credit is contingent on a telematics-and-coaching program you are requiring and monitoring. If the program isn't actually run, the credit must come off (raising the price), and the debit may need revisiting as the roster changes. Collapsing them into one number hides the levers you are actually pulling and the conditions the price depends on. (§23.2, §23.7; Ch. 11)

Exercise 23.29

Dropping the 10% telematics credit while keeping the 20% debit gives a straight 20% debit: \$120,000 × 1.20 = \$144,000** — \$12,000 higher than the \$132,000 with the program. The higher number is the better-underwritten outcome because the lower price was earned by a control that genuinely reduces frequency and builds evidence; remove the control and the risk is worse, so the rate should be higher. The account is asking to pay less for a riskier version of itself. Charging \$144,000 for the no-telematics version is pricing following risk (Chapter 11); discounting to \$132,000 without the program would be giving away a credit for a control that doesn't exist — the classic soft-market mistake. (§23.6, §23.7; Ch. 11)

Exercise 23.30 (†)

Four red flags and the required fix for each: 1. Liability on symbol 7 — silently excludes hired and non-owned autos. Fix: rewrite liability on symbol 1 (any auto) so the HNOA exposure is actually covered, and price for the breadth (§23.1, §23.5). 2. "Local" radius boxed but crashes 200+ miles away — the radius claim is contradicted by the loss run. Fix: reclassify to the true radius and re-rate, and consider whether it's misrepresentation (Chapter 33) (§23.4). 3. No hired-auto coverage on a delivery fleet — a real, likely exposure left uncovered. Fix: add hired/non-owned coverage and ask the diagnostic questions about personal-vehicle use (§23.5). 4. A driver with a recent major violation still on the schedule — disqualifying on this fleet. Fix: remove the driver as a condition of coverage (§23.3). (Telematics "installed" is a fifth flag — you'd ask what they do with it before crediting it; §23.7.) You could not quote until each is addressed. (§23.1, §23.3, §23.4, §23.5, §23.7)

Exercise 23.31

Auto claims clustering between 4:30 and 6:30 a.m. suggests a driver-fatigue / early-dispatch hypothesis: drivers starting very early, possibly under-rested, in low-light conditions. You would ask the insured about shift-start times, dispatch scheduling, hours-of-service practices (§23.4), and whether the same drivers or routes recur in those crashes. The control that most directly addresses it is telematics with active coaching and scheduling review (§23.7) — using the hours and behavior data to identify fatigue patterns and adjust dispatch — paired, if needed, with hard limits on early-start scheduling. (The loss run's timing is signal that no single application field would reveal — the §23.2 point about reading the story in the losses.) (§23.2, §23.7)

Exercise 23.32 (†)

Coverage-recommendation memo (model). To: UW Manager — Re: Harbor Steel & Fabrication, commercial auto line. "Recommend we write the 12-unit fleet, conditioned. The fleet is heavy flatbeds hauling structural steel on an intermediate radius plus a few light units — a severity risk, not a frequency one. MVRs are mostly acceptable; the loss run shows only two minor claims (no catastrophic history, which I read as the absence of a tail event, not proof of safety). I'm attaching two conditions: (1) removal of the one driver with a recent major violation — disqualifying on a heavy unit, not rate-able; and (2) mandatory telematics with active driver-coaching and forward-facing cameras, monitored as a standing condition. Liability is written broad (symbol 1) so the modest hired/non-owned exposure is covered and the \$10M umbrella attaches cleanly. Biggest residual risk I cannot price away: the nuclear-verdict tail — heavy units on public roads in the current litigation environment — which I've addressed through driver selection, the controls, and adequate limits/attachment, not by pretending a rate can cover it. Rate defense: it is severity-aware for heavy haulers on an intermediate radius; I did not let the clean two-claim history shade it." (The Underwriting File; §23.3, §23.6, §23.7)

Exercise 23.33

Model message to the broker: "We're glad to write the Harbor Steel fleet, and we can get you good terms — but we need one thing first. One driver on the schedule has a recent major violation, and on heavy flatbeds that's a risk we can't rate around; we'll need that driver removed from the policy as a condition of coverage. It's not negotiable on equipment this size, and frankly it protects the insured as much as us. Everything else looks workable — let's get that change confirmed and move ahead." (Firm on the substance, warm on the relationship — the Chapter 39 posture.) (§23.3)

Exercise 23.35 (†)

Underwriting-File extension — the clean-MVR, bad-telematics driver. Note for the file: "Telematics shows a driver with a clean MVR posting the fleet's worst harsh-braking and following-distance scores, not improving despite coaching. This is exactly the case the chapter built toward (§23.3): the MVR measures convictions, the telematics measures behavior, and here the behavior is the truer signal. Recommend escalating coaching with a documented improvement deadline; if the scores don't move, reassign the driver off the heavy routes or remove from the fleet, consistent with how we'd treat a major MVR violation in effect. Because removal/keep is a risk-quality decision on an in-force account, confirm whether it sits within line authority or needs a referral (Chapter 13/38). This illustrates 'you're insuring the process, not the roster': the account stays a good risk only because the telematics program is being acted on — the value was never the hardware, it was the closed loop that catches a dangerous driver the MVR called clean." (The Underwriting File; §23.3, §23.7)

Chapter 24

Worked solutions to the daggered (†) and odd-numbered exercises. (Even, non-daggered items are discussion or self-test prompts whose answers are developed in the chapter text; a few are summarized briefly.)

Exercise 1 (define E&O; standard of care)

Errors & omissions (E&O), or professional liability, is coverage for the financial harm a policyholder causes a third party through a negligent act, error, or omission in rendering professional services. It insures a standard of care rather than a result because a professional is not liable simply for a bad outcome — only for work that fell below the degree of skill and care a reasonably competent member of that profession would have exercised in the same situation and that caused a quantifiable financial loss. The garage that floods in a thousand-year storm is not malpractice; the garage that floods because the engineer ignored the published drainage code is.

Exercise 3 (define EPL; plaintiff; vs. workers' comp)

Employment practices liability (EPL) covers claims by an organization's employees and applicants (and sometimes third parties) alleging wrongful employment conduct — wrongful termination, discrimination, harassment, retaliation, failure to promote. The plaintiff is someone in the employment relationship. It differs from workers' compensation (Ch. 22), which covers physical injury on the job on a no-fault basis; EPL covers the dignitary and economic harms of how someone was hired, managed, or fired, and it is fault-based and contested.

Exercise 4 (cyber first- vs. third-party coverages)

First-party (the insured's own losses): e.g., incident response/forensics (the experts who investigate and contain the breach) and business interruption (lost income while systems are down — the dominant ransomware loss). Also data restoration, cyber extortion/ransom, notification & credit monitoring. Third-party (liability to others): e.g., privacy liability (suits by individuals whose data was exposed) and regulatory defense & penalties (investigations and fines where insurable). Also network-security liability and sometimes media/content. The key skill is keeping the two halves distinct: first-party pays the insured's own costs; third-party pays what the insured owes others.

Exercise 5 (define tail/ERP; what it extends and doesn't)

The tail / extended reporting period (ERP) is an endorsement or right that extends the time to report claims under a claims-made policy for wrongful acts that occurred during the now-expired policy period. It extends only the reporting window; it does not cover new acts committed after the policy ended. It is the device a retiring professional, a dissolving firm, or an acquired company needs so that a claim for old work arriving after coverage ends does not fall into a gap.

Exercise 7 (why claims-made for these lines)

These lines are written claims-made because the gap between the negligent act and the resulting claim can be years — an architect's design error may not surface for a decade, a board's decision may not be challenged until the company collapses, a breach may not be discovered for months. On an occurrence basis, the insurer would carry uncertain "incurred but not reported" exposure on old policies across a peril whose severity keeps changing — an actuarial nightmare. Claims-made closes the books nearer the policy period, so the insurer knows its exposure much sooner, which is what makes a volatile long-tail line priceable at all.

Exercise 8 (traditional vs. miscellaneous professional liability)

Traditional professional liability covers the classic licensed professions — lawyers, doctors/hospitals, accountants, architects/engineers, insurance agents — where the standard of care is well-defined by the profession itself. Example: legal malpractice for a law firm. Miscellaneous professional liability (misc E&O) covers the explosion of service businesses without a licensed box — IT consultants, marketing agencies, staffing firms, property managers, home inspectors. Example: a marketing agency sued for a botched campaign. The standard of care is clearer for the licensed professions (defined codes, established norms) and fuzzier for misc E&O, where underwriting depends on understanding exactly what the insured does and for whom.

Exercise 9 (two agencies, same revenue, different E&O)

Both have \$5M revenue, but the E&O exposure is the size of the harm a single error transmits to the client, not the size of the firm. Agency A (logos/brochures) can produce an ugly design — annoying, cheap to fix. Agency B manages multimillion-dollar ad budgets and targeting — a single error (mis-targeted spend, a campaign that runs wrong) can vaporize a client's marketing budget and cost real money, which the client will sue to recover. You would read the service contracts to see the scope, the client size, and the dollar value of the decisions the agency actually controls — none of which shows up on a revenue figure.

Exercise 11 (return-of-fees and intentional-acts exclusions)

The return-of-fees exclusion exists because E&O covers negligence in rendering services, not being made to refund money you should not have charged — refunding your own fee is not an insurable third-party loss, it is disgorgement. The intentional-acts exclusion exists because insurance covers fortuitous (accidental) loss (Ch. 1, Ch. 6); deliberate wrongdoing is not fortuitous, and covering it would create unlimited moral hazard. You are covered for negligently giving bad advice, not for deliberately defrauding a client.

Exercise 12 (IPO as a high-risk D&O event)

An IPO is one of the highest-risk D&O events because the moment a company sells stock to the public it becomes exposed to securities class actions: if the stock later drops, plaintiffs' firms allege the offering documents and subsequent statements were materially misleading. A private company's manageable D&O risk converts overnight into a public-company securities exposure with enormous defense-and-settlement costs. The disclosures in the offering materials become a litigation target, and the litigation climate for newly- public companies is unforgiving — IPO D&O is priced and structured as its own hard problem.

Exercise 13 (public vs. private D&O; why private packages with EPL)

Public-company D&O is dominated by securities class actions — severity-driven, sensitive to the broad litigation climate, heavily reinsured. Private-company D&O has a broader, lower-securities claim mix: disgruntled minority shareholders, competitors, creditors, vendors, and — very commonly — employees. Because employee-driven claims loom so large for private companies, private D&O is frequently sold packaged with EPL and fiduciary liability as a "management liability" suite, with a shared limit the underwriter must size for the worst-case year.

Exercise 15 (D&O vs. tech E&O for a startup)

They cover different exposures. Tech E&O covers harm the company's product/service causes its customers — e.g., the software fails and a client loses money. D&O covers harm the founders' and directors' management decisions cause the company's investors, creditors, and the entity itself — e.g., a funding round goes bad and investors sue the board for breach of duty. A claim from an investor over a management decision will not touch the E&O policy, and a claim from a customer over a product defect will not touch the D&O policy.

Exercise 16 (two retailers, same revenue, EPL exposure)

Retailer Y (400 seasonal employees, high churn) has the larger EPL exposure despite identical revenue, because EPL scales with headcount and turnover, not revenue. Every employee and applicant is a potential plaintiff, and every termination is a potential wrongful-termination or discrimination claim; high churn means many more hiring and firing events, each a claim opportunity. Revenue is the wrong base because a capital-intensive business can post huge sales with few employees (little EPL exposure), while a labor-intensive one generates many claims at modest revenue.

Exercise 17 (the machinery of employment)

Want to see, with how each lowers loss: (1) a current, lawyer-reviewed employee handbook — sets clear, defensible rules; (2) documented anti-harassment/anti-discrimination policies with real training — both prevents conduct and provides a defense; (3) a defined complaint-and-investigation procedure — catches and resolves issues before they become suits and shows good faith; (4) consistent, documented performance management — the single best defense against wrongful-termination claims is a paper trail showing the termination was performance-based; plus HR competence proportionate to headcount. Good machinery genuinely lowers loss, exactly as a hot-work program lowers fire loss.

Exercise 19 (EPL jurisdiction sensitivity)

EPL is acutely jurisdiction-sensitive because employment law is a patchwork: some states and cities have employee-friendly statutes, expanded protected classes, lower thresholds for liability, and plaintiff- friendly courts. The identical termination can be far more expensive where statutes add protected categories, allow broader damages, or where local courts and juries favor employees. An underwriter must weigh where the insured's workforce sits, not just how many people it employs.

Exercise 20 (why Harbor Steel still has cyber exposure)

Harbor Steel holds little sensitive customer data, so its privacy exposure is small — but its operational exposure is real and rising: the plant runs on networked systems, it emails invoices and accepts payment instructions (a social-engineering/funds-transfer exposure), and it stores customer drawings and production data. For a manufacturer, the dominant cyber loss is business interruption from ransomware — the days of lost production and missed shipments if its systems are locked. That exposure exists regardless of whether a customer is demanding coverage.

Exercise 21 (cyber breaks all three LLN requirements)

Independent: cyber losses are correlated — one vulnerability in widely-used software, or one prolific ransomware group, triggers losses across thousands of insureds at once (like a hurricane hitting many homes). Similar: the threat landscape is so heterogeneous that two superficially identical companies have very different real exposure based on configuration details no application captures. Stable: the data is non-stationary — attackers, tools, and tactics change continuously, so last year's experience poorly predicts next year's. Correlation makes cyber a catastrophe peril; non-stationarity makes it a moving- target peril; together they are the hardest combination in insurance.

Exercise 22 (find the red flag — partial MFA, untested backups)

Red flag 1: partial MFA. MFA on the VPN but not on email and not on administrator accounts leaves the two highest-value targets exposed — email is the most common attack entry point, and admin accounts are what an attacker needs to deploy ransomware network-wide. "Yes, we have MFA" was true and misleading. Red flag 2: untested backups. "Nightly backups never test-restored" means there is no evidence the backups can actually be restored under attack conditions — and attackers now hunt and destroy reachable backups first. Before binding: require MFA extended to email and all privileged/admin accounts, and require a successful test restore of offline/immutable backups (and confirmation the backups are segmented from the network) — as subjectivities, exactly like a roof-replacement condition on a property risk.

Exercise 23 (why tested offline/immutable backups are decisive)

From the ransomware economy's perspective, the criminal's leverage is that the victim cannot operate without its data. Tested offline or immutable backups remove that leverage: the insured can restore its systems from clean copies, refuse the ransom, and suffer a far shorter business interruption — turning a company-ending event into an expensive inconvenience. That is precisely why sophisticated attackers now try to find and destroy the backups first, and why the words "offline or immutable" and "tested" carry the whole value — an online, reachable backup gets encrypted along with everything else.

Exercise 25 (claims-made: switching vs. closing)

(a) Switching: set Carrier B's retro date to match the original date (here, eight years back), preserving the firm's prior-acts coverage so none of its earlier work falls into a gap. Sloppily setting the retro date to your inception leaves eight years of work uninsured the moment they leave Carrier A. (b) Closing/retiring: the firm does not need a new policy — it needs a tail (ERP) on the expiring policy, long enough to cover the statute-of-limitations window for its old projects. A cheap new policy is the wrong answer because a new claims-made policy covers claims for acts during its period; as the practice winds down there are no new acts to cover, and the new policy would not respond to a claim for old work — only a tail extends the reporting window for that old work.

Exercise 27 (why states regulate claims-made)

States regulate claims-made policies — mandatory ERP offers of a minimum length, plain-language disclosure of the trigger and retro date — because an unsophisticated buyer can easily end up with a gap without understanding it. The consumer harm to prevent: a professional who does not realize that "this policy only covers claims reported while it is in force" can let coverage lapse, switch carriers, or retire and discover too late that an old claim is uncovered. The ERP offer and retro-date disclosure are regulated duties, not courtesies, designed to ensure the insured understands the trigger.

Exercise 28 (underwrite the IT MSP)

Read: a serious risk. An IT managed-services provider that configures and monitors the backup systems several hospitals rely on sits at a catastrophic transmission point — if its work fails, hospitals could lose the very data they depend on, and the downstream harm is enormous. Largest exposure: professional-liability (tech E&O) for a systemic failure of the backup/monitoring service affecting multiple hospital clients at once (a correlated, high-severity loss). Two things to require/want: (1) limitation-of-liability clauses in the client contracts (their absence is a major red flag — the MSP has accepted unlimited downstream liability); (2) the firm's own security and operational controls (their MFA/EDR/tested-backup posture is strong, which is good, but you also want evidence of change-management and monitoring rigor on the client systems). Lean: modify — quotable, but conditioned on (or debited for) the contractual liability gap, with careful limit-setting given the correlated multi-hospital exposure, and possibly a sublimit on systemic events. Strong internal controls and one well-handled prior claim are encouraging, but the contract terms and the severity profile drive the decision.

Exercise 29 (price/size two cyber limits — qualitative)

Harbor Steel: modest data, real operational/BI exposure → a modest limit sized to the contract requirement and the business-interruption exposure; coverage emphasis on first-party (BI, ransomware/ extortion, incident response); binding conditions = the core controls (MFA, tested backups). Tindall Stores: large payment-card and personal-data trove plus operational exposure plus a prior breach → a larger limit; coverage emphasis spans both first-party (BI, extortion) and substantial third-party (privacy, regulatory) because a retailer's breach exposes many consumers; binding conditions = the full controls plus detailed remediation evidence from the prior incident. Same line, very different exposure profiles → different limits, different first-/third-party emphasis, different subjectivities. (No precise dollar figures — the profile drives the structure.)

Exercise 30 (find the red flag — post-breach renewal, no remediation)

The red flag is the absence of remediation detail: "Incident resolved. No further issues" tells you nothing about whether the company actually fixed anything. Questions to ask: Did you engage incident responders/ forensics? What was the root cause/entry point? Did you rebuild systems clean or just decrypt? Is MFA now on email and all privileged accounts? Are backups now offline/immutable, segmented, and test-restored? Did you get an independent post-incident security assessment? Are you certain the attacker no longer has a foothold? The fact of the prior breach is not the concern — a genuinely hardened survivor can be a better risk than a complacent never-victim (Ch. 1 adverse selection). What matters is the evidence of what they did, and silence on remediation is itself the warning.

Exercise 31 (write the memo — Harbor Steel)

Model answer: "On the cyber/E&O question for Harbor Steel: I do not recommend professional liability (E&O). Harbor Steel fabricates and installs steel — if a component fails, that's a products/general-liability claim (bodily injury or property damage), not negligent professional advice causing economic loss, so E&O isn't the right tool unless they move into design services. I do recommend a modest cyber policy. As a manufacturer their privacy exposure is small, but their operational exposure is real: their dominant cyber loss would be business interruption from ransomware — lost production and missed shipments if their systems are locked — exactly what happened to a regional competitor last winter. I'd place a standalone cyber policy sized to your customer's contract requirement, emphasizing first-party business-interruption, extortion, and incident-response coverage. I'm conditioning the quote on confirmation of multi-factor authentication on email and remote access and tested offline backups — standard controls that are the price of admission and that genuinely reduce the loss. Happy to walk the owner through it."

Exercise 33 (ethics — overriding a stale security scan)

Overriding is required, not merely permitted, because the model is wrong on the facts: the vulnerability it penalizes has been patched, so the "decline" is based on stale data, and binding a fair decision on bad data would be both unfair to the insured and a poor underwriting decision. You must document: the scan's finding, the evidence the vulnerability was remediated (date patched, confirmation/rescan), your reasoning, and your decision — so the file defends itself (Ch. 7). This is the book's central theme in miniature: the model is a tool, the underwriter decides, and a documented override of a demonstrably-wrong model output is exactly the judgment the profession exists to exercise. (See also the Model-vs-Judgment callout and Ch. 32.)

Exercise 35 (Tindall Stores remediation evidence; adverse selection)

Remediation evidence to demand: independent post-incident forensic/assessment report; confirmed root cause and entry point; evidence of a clean rebuild (not just decryption); MFA deployed on email and all privileged accounts; EDR/MDR deployed and monitored; backups now offline/immutable, segmented, and successfully test- restored; patching brought current; confirmation the attacker's access was eradicated; and security-awareness training. Why the prior breach alone tells you almost nothing: by adverse selection (Ch. 1), the insureds most eager to buy are often those who most expect to need coverage — a post-breach applicant could be a hardened, lesson-learned risk (now better than a complacent never-victim) or a checkbox buyer who changed nothing (attackers possibly still inside). The breach is just a fact; only the remediation evidence discriminates between the two, and that evidence — not the breach — is the underwriting.

Chapter 25

Worked solutions to the daggered (†) and odd-numbered exercises. (Even, non-daggered items are discussion prompts whose reasoning is developed in the chapter text or in class.) Section references point to index.md.

Exercise 1 (define a surety bond; the three parties)

A surety bond is a three-party contract in which one party (the surety) guarantees to a second party (the obligee) that a third party (the principal) will fulfill an obligation, and if the principal fails, the surety performs or pays the obligee and then seeks reimbursement from the principal. The three parties: the principal performs (or is supposed to perform) the obligation; the obligee is the protected party who receives the obligation; the surety guarantees the principal's performance to the obligee. (§25.1)

Exercise 3 (what bid, performance, and payment bonds guarantee)

• Bid bond: that if awarded the contract, the contractor will enter into it at the bid price and furnish the required performance and payment bonds.
• Performance bond: that the contractor will complete the work according to the contract (on time, to specification); penal sum commonly 100% of the contract price.
• Payment bond: that the contractor will pay its subcontractors, laborers, and material suppliers. (§25.2)

Exercise 4 (contract vs. commercial surety; two examples each)

Contract surety guarantees a contractor's obligations on a construction or service contract (examples: a performance bond on a highway project; a payment bond on a school). Commercial surety guarantees a broad miscellany of other obligations to governments, courts, and parties (examples: a contractor's license bond; an executor's fiduciary/probate bond — or a motor-vehicle dealer bond, an appeal bond, a notary bond). The unifying thread is the three-party credit structure; the difference is what is guaranteed. (§25.2, §25.3)

Exercise 5 (the three C's and the question each answers)

Character — will they perform honestly and stand behind the job? Capacity — can they actually do this work, at this size and volume? Capital — can they absorb cash strain and survive a bad job? Traditional weighting: character > capacity > capital. (§25.4)

Exercise 7 (the General Indemnity Agreement; why it is the spine)

A General Indemnity Agreement (GIA) is a contract — signed before any bond is written, typically by the principal and its owners personally (often their spouses too) — under which they agree to indemnify the surety for any loss, cost, and expense it incurs under any bond, and which grants the surety rights to collateral, to the contract funds, and to the contractor's books on trouble. It is the spine of the business because it is the legal mechanism that makes surety credit: it lets the surety, having paid an obligee, reach back to the principal's and owners' personal assets for reimbursement. Waive or weaken it and surety stops being credit. (§25.7)

Exercise 9 (correct "a bond is just insurance for the owner")

The statement misses the two features that make surety not insurance. (1) Three-party structure: a bond protects the obligee (owner) against the principal's (contractor's) failure, with the surety guaranteeing — three parties, not two, and the party that pays the premium (the principal) is not the party protected (the obligee). (2) Reimbursement: unlike an insurer, which absorbs the loss it pays, a surety that pays the obligee pursues the principal and its personal indemnitors to be made whole. So a bond is a guarantee (credit), economically like a co-signature on a loan, not a transfer of risk to a pool. (§25.1)

Exercise 11 (same 62% loss ratio, opposite verdict)

For the auto line, a 62% loss ratio leaves enough of the premium dollar to cover expenses and a profit — healthy, because auto premium is priced to fund expected losses. For the surety line it is a catastrophe, because surety underwrites toward a near-zero expected loss: the "premium" is a small fee for the use of the surety's credit, not the price of an expected loss, and the line's whole structure assumes losses are rare and largely recovered through indemnity/salvage. A 62% surety loss ratio means defaults are happening at a rate the fee never contemplated — years of fees wiped out. (§25.1)

Exercise 13 (prequalification and the social function)

An obligee that requires a bond is, in part, outsourcing a credit-and-competence check to the surety industry: a contractor that can obtain the bond has been vetted by a party putting its own money behind the judgment, and one that cannot has effectively failed the check. The social function of that gatekeeping is to keep unqualified or undercapitalized contractors off serious work — protecting the public's money (on public projects) and the subcontractors' livelihoods — before any default can occur. The bond both backstops the obligee and screens out the firms most likely to fail. (§25.1)

Exercise 14 (the bid-bond scenario)

The bid bond responds. The contractor, awarded as low bidder, refused to sign; the owner awarded to the next bidder at \$300,000 more. The bid bond pays the owner the difference between the defaulting contractor's bid and the cost of the next acceptable bid — here \$300,000 — up to the penal sum. The penal sum is 10% of the \$5M bid = \$500,000. Since \$300,000 is within \$500,000, the bond pays the full \$300,000 extra cost. (If the gap had exceeded \$500,000, the bond would cap at \$500,000.) (§25.2)

Exercise 15 (the unpaid supplier on public work)

The payment bond protects the supplier. The supplier cannot lien the public building because public property generally cannot be liened — so on public work the payment bond is the subcontractors' and suppliers' principal recourse for nonpayment. The supplier makes a claim against the contractor's payment bond for the \$200,000 it is owed (subject to the bond's claim procedures and penal sum). (§25.2)

Exercise 17 (match each bond to its obligee)

(a) contractor's license bond → a state licensing agency; (b) executor's fiduciary bond → a probate court; (c) appeal/supersedeas bond → an appellate court / the judgment creditor (required by the court, protecting the party that won the judgment); (d) performance bond on a city water project → the project owner (the city). (§25.2, §25.3)

Exercise 19 (underwrite the program — the fundable contractor)

Character: 18 years in business, clean payment record — strong. Capacity: steady backlog around its historical \$16M volume shows it can manage that volume; *but* the requested \$4M single-job limit exceeds its largest completed job of \$3.5M — only modestly, so the real question is incremental, not disqualifying. Capital: \$1.2M adjusted working capital, \$2.8M net worth, an unused \$1.5M bank line — solid liquidity and a real backstop; clears the floor for a program of this size. Decision: approve the program, essentially as requested, but consider holding the single-job limit close to its demonstrated experience (e.g., near \$3.5–4M) and raise it incrementally as it completes jobs at the new level; take a full personal GIA. This is a fundable credit: the floor is cleared and character/capacity carry it. The only discipline point is not letting the single-job limit run far ahead of proven experience. (§25.4, §25.5, §25.6)

Exercise 21 (why "just charge a higher fee" is wrong for contractor B)

A higher bond fee cannot offset a default. One defaulted job on B can produce a net loss of many times the fees ever collected across the whole program, so no feasible fee prices the risk — you would simply be collecting a slightly larger fee while remaining exposed to a catastrophic, largely uncoverable loss. The right response is selection and capacity control: decline the requested \$4M/\$10M program; if you write anything, offer a much smaller program — a single job well within B's demonstrated \~\$2M experience and an aggregate near its historical volume — and require stronger indemnity and possibly collateral. You manage the risk by not writing it, or by limiting how much of it you write, not by pricing it. (§25.4)

Exercise 23 (working-capital arithmetic and the program ceiling)

(a) Reported working capital = current assets − current liabilities = \$4.0M − \$2.6M = \$1.4M. (b) Adjusted working capital: remove the \$0.5M related-party receivable and the \$0.3M of receivables aged past 120 days (discounted to zero): adjusted current assets = \$4.0M − \$0.5M − \$0.3M = \$3.2M; adjusted working capital = \$3.2M − \$2.6M = \$0.6M. (c) At a rough ceiling of 10× adjusted working capital, the implied aggregate program is 10 × \$0.6M = **\$6M — far below what the reported \$1.4M (which would imply \$14M) suggested. It is a starting point, not the decision, because the multiple is only a rule of thumb: a judgment then adjusts it for character, capacity, the quality of the backlog, the bank line, and the type and size of work. The headline-vs-adjusted gap (\$14M vs. \$6M of implied capacity) is itself the warning. (§25.5)

Exercise 25 (what the bank line tells a surety)

The bank line reveals the contractor's access to liquidity beyond its own balance sheet — and, crucially, what another sophisticated credit grantor with its own money at risk thinks of the firm. A strong, long-standing relationship with an unused line is a powerful positive: a liquidity backstop for a bad job plus an independent vote of confidence. A contractor fully drawn on its line with no other liquidity has no cushion behind its working capital and no independent vote — and it puts the surety in the position it least wants: being the only credit grantor standing behind the firm. So such a contractor is worse than its balance sheet alone suggests, because the balance sheet does not show that its only liquidity source is exhausted. (§25.5)

Exercise 27 (find the red flags)

The three most worrying are (b) the requested single-job limit nearly triple its largest completed job (a capacity red flag — classic overreach into unproven size), (c) a backlog roughly double its historical annual volume (overreach in volume, straining capital and management), and (e) a fully-drawn bank line with no unused capacity (no liquidity backstop, no independent credit vote). The others are neutral-to-positive: (a) 20 years in business, (d) audited percentage-of-completion statements, (f) an unused \$2M line, and (g) a clean lien/judgment history are all reassuring. (§25.5, §25.6)

Exercise 29 (the small fiduciary bond's real exposure)

The \$25,000 penal sum understates the exposure. A fiduciary/probate bond's penal sum may be small relative to the assets the executor controls, and the temptation and opportunity to defalcate scale with those assets, not with the bond. So a dishonest executor over a large estate is a real exposure even under a small bond, and the bond can be exhausted by the first claimant, leaving later heirs unprotected. Before issuing on a credit score alone, check the size and nature of the estate, the executor's relationship to it and track record, any co-fiduciary or court controls on the assets, and the obligee statute governing how the bond can be called. Size the exposure, not the penal sum. (§25.3)

Exercise 31 (holding the line on capacity without harming the relationship)

A model answer, in the underwriter's own voice: "I want to support your growth, and I think you've earned a bigger program — which is exactly why I want to get there the right way. A job 50% larger than anything you've built is a different animal: the cash demands, the staffing, the schedule risk all step up. Let's raise your single-job limit deliberately as you complete the next jobs at the top of your current range, so you're never overextended and I'm never the reason a good job goes wrong. Bring me the next opportunity and the plan to staff and finance it, and we'll size it together." The move: empathize, name the specific capacity step-up, offer an incremental path, and frame the discipline as protecting the contractor, not just the surety. (§25.6)

Exercise 33 (the broker's push to drop personal indemnity)

You would be giving up the core protection of the line. The personal General Indemnity Agreement is what lets the surety reach the owners' own assets for reimbursement — without it, the surety's recovery is limited to a corporate shell that, in a default, is often worthless, and surety stops being credit and becomes an uncompensated guarantee. Response: decline to waive it, explain (to the broker and the principal) that personal indemnity is standard and non-negotiable precisely because the whole economics of surety rest on reimbursement; if a "competitor will waive it," that competitor is mispricing a guarantee it cannot recover on, and that is not a contest worth winning. If the owner balks at standing behind their own firm, that reluctance is itself an underwriting signal worth weighing. (§25.7)

Exercise 35 (why the surety decision on Harbor Steel is a different decision)

It is "a different decision, by a different underwriter, on different evidence" because the surety is making a credit decision about Harbor Steel as a principal — will it perform a bonded obligation, and is it good for the money — on the evidence of its financial statements, backlog, bank line, capacity, owner character, and personal indemnity, underwritten toward zero loss. The property-casualty underwriter is making an insurance decision — pricing the probability and severity of physical and liability losses — on the evidence of the roof, the fires, the fleet, and the cat exposure. Adding a surety relationship would not change the insurance disposition, because the bonds are separate credit instruments outside the submitted program; Harbor Steel's package terms, pricing, and conditions stand exactly as the property-casualty file builds them. The two decisions live in different houses on different evidence. (§25.1, The Underwriting File)

Chapter 26

Worked solutions to the daggered (†) and odd-numbered exercises. Numbers in these problems are illustrative teaching figures. (Even-numbered, non-daggered items are largely discussion prompts whose reasoning is developed in the chapter text.)

Exercise 1 (define ocean marine)

Ocean marine is insurance covering vessels, their cargo carried over water, and marine liabilities. Its four classic coverages: hull (physical damage to the vessel itself); cargo (loss of or damage to the goods in transit by sea); protection & indemnity (P&I) (the shipowner's liability for injury, collision, pollution, and wreck removal, written through specialized mutual P&I clubs); and freight (the shipowner's loss of earnings when a voyage is interrupted). It is the oldest line, and the doctrines of general average (Ch.2) and utmost good faith (Ch.4) are at their most demanding here.

Exercise 3 (define aviation insurance; why low-frequency, high-severity)

Aviation insurance covers aircraft hull (physical damage to the aircraft) and aviation liability (injury and damage to passengers and third parties from aircraft operation). It is "low-frequency, high-severity" because serious aviation losses are rare but, when they occur, are catastrophic — a hull worth tens or hundreds of millions of dollars plus liability for everyone aboard and on the ground — so there is no large pool of small, frequent claims to stabilize the math.

Exercise 5 (define parametric; vs. indemnity)

Parametric insurance pays a pre-agreed amount when a measurable, objective trigger event occurs (a wind speed, a quake magnitude, a rainfall threshold), regardless of the insured's actual loss. It differs from indemnity insurance, which reimburses the insured's actual, adjusted loss after the fact: parametric keys the payout to a parameter, not to a claims adjuster's assessment of damage — trading precision for speed.

Exercise 6 (define program business; the MGA's role)

Program business is the packaging of a specialized, homogeneous book (a single niche — e.g., tow trucks, pet groomers) and the delegation of its underwriting to a niche specialist, typically a managing general agent (MGA) (defined in Ch.3), under a binding-authority agreement on a carrier's paper. Unlike an ordinary distributor, which only brings risks to the carrier, an MGA with binding authority underwrites on the carrier's behalf — it selects risks, sets terms, and binds coverage, all within an authority the carrier grants.

Exercise 7 (define specialty/niche line; the four conditions)

A specialty/niche line is any line that cannot be written with standardized forms and broad statistical credibility, and so demands deeper expertise, more judgment, and often a distinct market. The four conditions that push a line into the specialty world: (1) small or heterogeneous pools; (2) correlated catastrophe; (3) unusual exposures; (4) thin data. Any one can be enough; specialty lines usually have several.

Exercise 9 (the property-vs-inland-marine rule of thumb)

The rule of thumb is "movement, mobility, or instrumentalities of transportation/communication." Property that sits in a fixed building is property (Ch.19); property that moves (cargo, contractors' equipment), is held for others (bailee), bridges or transmits (bridges, radio/TV towers, power and phone lines), or is by nature mobile and high-value (fine arts, medical equipment on loan) is inland marine. Applied: a radio broadcast tower is an instrumentality of communication → inland marine. A warehouse full of finished goods sits in a fixed location → commercial property.

Exercise 11 (general average and utmost good faith in ocean marine)

General average (Ch.2) — where all parties to a sea voyage share a loss voluntarily incurred to save the whole venture (e.g., jettisoning cargo to save the ship) — is a doctrine native to ocean marine and shapes how marine losses are apportioned. Utmost good faith (Ch.4) is especially load-bearing in ocean marine because the underwriter cannot inspect a vessel or cargo halfway around the world and must rely almost entirely on the insured's complete and honest disclosure; the information asymmetry is extreme, so the duty of disclosure is correspondingly strict.

Exercise 12 (why aviation ≠ personal auto)

Personal auto (Ch.14) has tens of millions of similar, independent vehicles and millions of small claims a year, so the pool is enormous and the law of large numbers (Ch.1) makes the math stable. Aviation has a small worldwide fleet, an even smaller fleet per insurer, and rare-but-catastrophic losses — there is no large pool of similar, independent risks. Critically, any one operator's loss history has near-zero credibility (Ch.10): there is too little exposure for "no claims in eight years" to be statistically meaningful, so you cannot rate on it. Aviation is therefore written by a small expert global market that pools internationally, prices on engineering and severity, and reinsures heavily.

Exercise 13 (four aviation-specific factors)

(1) Pilot experience — total hours, hours in type, ratings, recent currency, training; an under-experienced pilot in a high-performance aircraft is the biggest general-aviation red flag. (2) The aircraft — make, model, age, equipment, maintenance, single- vs. twin-engine — drives the hull exposure. (3) The use — pleasure, business, instructional, agricultural application, charter, medical transport — reshapes the liability. (4) War/terrorism/hijacking — a distinct, separately reinsured exposure (the September 11, 2001 lesson) that the ordinary market cannot hold alone.

Exercise 14 (the loss-free aviation record)

"Eight years with no claims doesn't earn a steep credit here — it can't. Severe aviation losses are rare, so a single operator's loss-free record has almost no statistical credibility (Ch.10); the calm stretch tells us little about the severity tail. We price aviation for the catastrophe that hasn't happened yet, the way we price any cat-exposed risk, and we let the engineering and the experienced market set the floor — not a lucky streak." The trap is mistaking a quiet stretch for a low risk in a low-frequency, high-severity line.

Exercise 15 (why a refinery is uninsurable by ordinary means)

Three structural problems, each with its device: (1) No pool — there is essentially one of this exact refinery, so no "similar risks" to credibility-weight → engineering-led underwriting (the risk-engineering survey) replaces actuarial credibility. (2) Enormous value — no single insurer can hold a billion-dollar risk → layered and shared across a tower of many insurers (Ch.19, §19.7). (3) Exotic, correlated perils — explosion, fire, machinery breakdown, named windstorm offshore, plus huge BI → heavy reinsurance (Ch.27) and PML-based sizing.

Exercise 16 (the central energy document; PML over TIV)

The central document is the risk-engineering survey — a specialist inspection of process safety, spacing, fire protection, maintenance, and management systems — which replaces the loss run because there is no credible loss history for a unique plant. Probable maximum loss (PML) (Ch.30) — the largest loss reasonably expected from a single event — drives capacity and price rather than total insured value, because no realistic single event destroys the entire plant: you need enough capacity to cover the worst plausible event, not the full TIV, and pricing the PML (not the TIV) is how the market sizes the real exposure.

Exercise 17 (why energy is reinsurance-dependent)

A single energy loss can exhaust dozens of insurers' capacity at once, so the primary market spreads the risk by sharing and layering the tower (Ch.19) — the primary-market analogue of reinsurance. At the carrier level, any insurer fronting a slice of a refinery almost always cedes most of it to reinsurers (Ch.27), because no balance sheet of normal size can absorb a billion-dollar event. The shared/layered tower and the carrier's cession are the same risk-spreading logic at two levels; energy needs both.

Exercise 18 (why crop fails independence)

Crop risk fails the independence assumption because a single peril — a regional drought, an early freeze — damages many farms across a region simultaneously, rather than striking one farm independently of the next. The losses are massively correlated: one peril hits the whole pool at once, exactly the condition the law of large numbers (Ch.1) cannot absorb. Historically this wiped out private-only crop insurers because a single bad weather year produced losses across their entire book at the same time — a pool of "many independent farms" was really a single regional bet, and no private balance sheet could survive it.

Exercise 19 (how MPCI inverts the underwriter's job)

In ordinary commercial lines the underwriter builds the rate (Ch.11) and selects the risk freely (Ch.13). In MPCI the rates, forms, and eligibility are set federally (USDA-RMA), so the underwriter generally cannot decline an eligible farmer or re-price the risk. The competitive craft relocates to: distribution and service (winning and keeping agents and farmers); loss-adjustment quality and integrity; data and technology (yield, satellite, and weather modeling); and portfolio-and-reinsurance management under the Standard Reinsurance Agreement (deciding how much of which states/crops to retain vs. cede). Any one of these is a valid answer.

Exercise 21 (crop and the social-function theme; the protection gap)

Crop insurance "serves a social function" because it exists by a deliberate societal choice: a nation decided that a stable food supply and a survivable farm sector were worth a public backstop, and built a subsidized, publicly-reinsured program to make a privately-uninsurable, correlated-catastrophe risk survivable. This connects to the protection gap (Ch.30) — the shortfall between economic losses and insured losses for catastrophe perils. Where private markets cannot or will not bear a correlated catastrophe, a gap opens, and society's response is often a public–private structure (crop, the NFIP for flood, state wind pools) rather than leaving the exposure uninsured.

Exercise 22 (define basis risk; both directions)

Basis risk is the risk that a parametric policy's trigger and the insured's actual loss diverge — the trigger is a proxy for the loss, and a proxy can always be off. Large loss, no payout: a hurricane passes just outside the trigger radius (or registers just below the trigger wind speed) yet devastates the insured, who collects nothing because the parameter was not met. Windfall, little loss: a triggering event occurs (the parameter is met) but the insured happened to suffer little damage, and collects the full payout anyway. Basis risk is the price the insured pays for speed and simplicity.

Exercise 23 (the three indemnity weaknesses parametric attacks)

(1) Speed — indemnity claims take months/years to adjust; the trigger is objective, so a parametric policy pays within days (nothing to adjust). (2) Basis of loss — indemnity covers only insurable (usually physical) loss; parametric can pay on the event, capturing economic loss with little physical damage (a resort empty after a nearby storm). (3) Certainty/simplicity — the trigger is pre-agreed and objective, so disputes nearly vanish.

Exercise 24 (the resort with revenue exposure)

Recommend a parametric hurricane structure keyed to a measured trigger (storm intensity within a defined proximity), paying a fixed sum on the event. What it can do: deliver fast liquidity to offset hurricane-driven revenue loss even where physical damage is minor — exactly the resort's exposure. What it cannot do: guarantee the payout matches the actual revenue loss — basis risk means a near-miss storm that still empties the resort could pay little or nothing, and a triggering storm with full bookings could pay a windfall. The doctrine that must still hold: the buyer must have an insurable interest (Ch.4) in the triggering event (the resort plainly does), or the contract is a wager rather than insurance; a loss attestation is often added to keep it on the insurance side of the line.

Exercise 25 (parametric as complement, not replacement)

Parametric is a complement because the very feature that makes it fast — paying on a trigger instead of adjusted loss — is the feature that creates basis risk, so it cannot reliably pay a buyer's exact loss. Sophisticated buyers therefore layer it with indemnity cover and reserves: parametric supplies fast liquidity in the first days, while indemnity addresses the precise, surveyed loss over the following months. It clearly shines (a) where speed of payout matters most (sovereign disaster response) and (b) where the loss is economic rather than physical (hurricane-driven revenue loss with little building damage).

Exercise 26 (the MGA program's structural moral hazard)

The structural moral hazard (Ch.1) is a misalignment of decision and consequence: the MGA makes the underwriting decision (selects, prices, binds) and, paid on volume of premium written, profits from writing more business — while the carrier bears the loss if that business is underpriced. The party making the call is not the party that pays for a bad one. It matters because the MGA's incentive tilts toward growth and looser selection, and the losses surface on the carrier's combined ratio two or three years later (Ch.10), after the MGA has earned its commissions.

Exercise 27 ("great loss ratio — let's grow it")

In a delegated-authority program, a low reported loss ratio in year one or two often means the losses have not developed yet (Ch.10's trend-and-development point), not that the underwriting is sound — especially in long-tail lines like liability or auto, where claims take years to mature. The MGA, paid on volume, has every incentive to grow the book fast before the true ultimate loss ratio emerges. Growing on an immature reported ratio can therefore mean pouring premium into an unprofitable book; the disciplined carrier prices and judges the program on developed/ultimate losses, not the reported figure.

Exercise 28 (the five program controls)

(1) A tight, written binding-authority agreement — prevents the MGA from exceeding the scope of its pen. (2) Clear underwriting guidelines the MGA must follow — prevents off-appetite or under-controlled risks being bound. (3) Caps on limits and aggregate — prevents a single large loss or runaway accumulation from the program. (4) Aligned incentives (a profit-share or a retained-risk slice) — prevents the volume-only moral hazard by making the MGA share the downside, not just the upside. (5) Regular, rigorous underwriting audits with file pulls (Ch.38) — prevents drift between what the MGA promised and what it actually wrote, by verifying compliance directly.

Exercise 29 (underwrite the food-truck program)

Decision: modify (interested, not as proposed). The niche expertise and ready book are genuinely valuable, but the structure is dangerous: a 52% loss ratio over only 18 months is immature (property/GL losses, especially liability, are not fully developed — Ch.10), and a commission-only, no-retained-risk MGA profits from volume regardless of ultimate loss (the §26.6 moral hazard). Conditions to attach: (a) an aligned-incentive structure — a profit-share or a retained-risk slice so the MGA shares the downside; (b) caps on per-risk limits and program aggregate; (c) mandatory adherence to your property/GL guidelines with defined referral triggers; (d) quarterly underwriting audits with file pulls (Ch.38); (e) data/reporting standards. Pricing: price for developed, not reported, losses — load the immature 52% toward a prudent ultimate, and do not let the projected doubling drive the rate down. Approve a controlled, capped entry, not the open-ended proposal.

Exercise 30 (price the parametric trigger)

(a) Pure premium = annual probability × payout = 0.02 × \$10,000,000 = \$200,000. This is the expected annual cost of the trigger payout, before expenses and profit (the pure-premium method, Ch.11). (b) Loading for expenses and profit at 40% of pure premium: indicated premium = \$200,000 × 1.40 = \$280,000. (c) The residual basis risk: even at this price, the buyer can suffer a severe loss from an event that falls just outside the trigger zone or just below the magnitude threshold and collect nothing — the premium buys a payout on the parameter, not on the buyer's actual loss.

Exercise 31 (structure the inland-marine transit limit)

Set the per-conveyance limit nearer the maximum single load (~\$400,000) than the typical load (~\$120,000)** — for example, \$400,000–\$450,000 — because the policy must respond to the worst single shipment that could be lost in one event, not the average; a limit set at the typical value would leave the largest loads dangerously underinsured. Verify first with the broker the true maximum value ever moved in a single conveyance (and whether multiple high-value loads ever travel together), since a single truck/trailer carrying more than \$400,000 would breach the limit. (Connects to limits, Ch.12.)

Exercise 32 (find the red flag — marine cargo program)

Three serious red flags: (1) High-theft cargo on high-theft routes — \$2M average loads of electronics through known cargo-theft corridors is a severity-and-frequency time bomb; theft attractiveness is the core cargo hazard. (2) An immature 30% year-one loss ratio taken as proof of profitability — cargo theft and claims develop, and one year is far too green (Ch.10). (3) A commission-only MGA with \$5M binding authority and no carrier referral — the §26.6 moral hazard at maximum, with a high per-conveyance cap and no check on the pen. Require before binding: aligned incentives (profit-share/retained risk); a hard, lower per-conveyance cap with carrier referral above it; route/security/tracking requirements and loss-prevention standards; pricing for developed losses; and quarterly audits with file pulls (Ch.38).

Exercise 33 (the referral memo)

A model answer (structure, not exact words): "Recommend conditional entry into the [delegated] program. Niche expertise and distribution are strong, but the proposed commission-only structure misaligns incentives and the early loss ratio is immature. Non-negotiable condition: an aligned-incentive structure (profit-share or retained-risk slice) so the MGA shares the downside, paired with limit/aggregate caps, our guidelines, and quarterly audits. Without incentive alignment, we are funding someone else's volume with our capital and should decline. With it, the program is attractive. Pricing to developed, not reported, losses." The single non-negotiable is incentive alignment, because it is the one control that addresses the structural moral hazard rather than just its symptoms.

Exercise 34 (ethics — parametric basis risk and disclosure)

Your obligation is full, plain disclosure of basis risk before the sale: you must show the buyer the scenarios in which the cheaper structure pays little or nothing despite a severe loss (the near-miss storm), not bury them to "keep the pitch simple." The social-function value of fast disaster cash (§§26.4–26.5, theme 6) is real and worth pursuing — fast liquidity saves lives after a catastrophe — but it does not license misrepresenting the product. Reconcile the two by being honest about the limits and helping the buyer manage them: offer a tighter (if costlier) trigger, a multi-parameter or modeled-loss index, or a small basis-risk/indemnity layer on top, and let the buyer choose with eyes open. A product the buyer understands builds trust; one sold by hiding its gap destroys trust — and the buyer's relief funding — the first time it fails to pay. Honesty about limits is the profession's duty (the same "state what it can and cannot do" posture the whole book demands).

Exercise 35 (Underwriting File — Harbor Steel starts importing by sea)

(a) The coverage now required is ocean cargo (marine), which belongs to the ocean marine family — distinct from the inland-transit floater that covers domestic over-the-road movement. (b) Three new ocean-voyage exposures the inland floater never faced: (i) marine perils — heavy weather, vessel sinking, stranding, and the prospect of general average contributions (Ch.2) if cargo is sacrificed to save the voyage; (ii) far longer transit and custody chains — multiple handlers, ports, loading/unloading, and storage, with the goods out of Harbor Steel's control for weeks; (iii) jurisdictional and documentary complexity — bills of lading, Incoterms (who bears risk at which point), customs, and cross-border liability. (c) Utmost good faith (Ch.4) becomes more demanding because the underwriter cannot inspect a vessel or cargo at sea and must rely almost entirely on the insured's complete, honest disclosure of the cargo, route, vessel, and packing — the information asymmetry is at its widest.

Exercise 37 (the one problem; the different solutions; automation implication)

The single problem: each line denies the law of large numbers (Ch.1) one or more of its requirements (large pool / similar / independent) — small pools, heterogeneous risks, or correlated catastrophe. The different solutions: marine → manuscript flexibility and a specialist market for the unusual; aviation → a small global expert market that pools internationally and reinsures; energy → engineering-led underwriting with shared/layered towers and reinsurance; crop → a federal public–private backstop reinsuring the correlated tail; parametric → a measurable trigger transferring risk to capital markets; programs → delegating the pen to a niche specialist under controls. Automation implication: because the common thread is thin data plus unique risk, specialty is the least automatable corner of underwriting (Ch.36) — the algorithm cannot price what it has too few analogues for, so expert human judgment remains irreplaceable here.

Exercise 39 (two specialty lines vs. BOP over the cycle)

Example with aviation and energy vs. small-commercial BOP (Ch.20): BOP is high-volume, short-tail, and well-diversified, so its combined ratio (Ch.3) is relatively stable year to year — the law of large numbers smooths it. Aviation and energy are low-frequency, high-severity and catastrophe-exposed, so they can post excellent combined ratios for several quiet years and then surrender a decade of profit in a single hull loss or refinery fire — their results are lumpy. "Judged over the cycle, not the quiet year" matters more in specialty because a short window of good results is not evidence of adequate pricing; only the long-run experience, which must include the rare large event, reveals whether the line is truly profitable. (Either of two other lines from the chapter is acceptable.)

Exercise 40 ("specialty is just regular underwriting with bigger numbers")

The view is wrong because specialty differs in kind, not just scale. (1) Energy isn't "bigger property" — with essentially one of each plant there is no pool to credibility-weight (Ch.10), so you underwrite by engineering, not data: a different method. (2) Crop isn't "bigger property" — the government sets the rates and forms and reinsures the tail, inverting the underwriter's job from rate-builder to program operator: a different role. (3) Parametric isn't "bigger indemnity" — it pays on a trigger, not on adjusted loss, introducing basis risk that has no analogue in standard lines: a different product mechanism. (Aviation and programs reinforce the point: a different market, and a delegated pen.) The common thread is that specialty lines defy the law of large numbers and require different machinery — expertise, engineering, backstops, triggers, delegated authority — not merely larger limits.

Chapter 27 — Worked Solutions to Selected Exercises

Solutions are provided for the daggered (†) exercises and the odd-numbered exercises. All figures are illustrative teaching numbers; the point is the method.

Exercise 1 (†)

Reinsurance is insurance purchased by an insurer: a contract under which a reinsurer, for a premium, indemnifies a ceding insurer for all or part of the losses on the policies the cedent has issued. The three jobs: capacity (write limits larger than the carrier could hold alone), catastrophe protection (survive the correlated event that defeats the law of large numbers), and volatility / earnings smoothing (protect surplus and steady year-to-year results).

Exercise 3

Treaty covers a whole class or book; facultative covers one specific risk. Treaty is automatic and obligatory (cedent must cede, reinsurer must accept every qualifying risk); facultative is optional for both sides. Treaty is negotiated once, at renewal, and the risk is reinsured the instant it is bound; facultative is negotiated per submission, before or while the risk is bound.

Exercise 4 (†)

Quota share cedes a fixed percentage of every risk (premium and loss) from the first dollar — best at relieving surplus strain and funding growth (it reduces premium-to-surplus leverage across the whole book). Surplus share cedes only the part of each risk above the cedent's retention, in multiples called lines — best at solving the large-line problem (and homogenizing the net book, since the cedent keeps a similar-sized net line on every risk).

Exercise 5

Excess of loss (XOL): the reinsurer pays the part of a loss above the cedent's retention (attachment point), up to the reinsurer's limit. "\$4M xs \$1M" means the reinsurer pays losses above \$1M up to \$4M more — i.e., it covers the band from \$1M to \$5M. The cedent keeps the first \$1M; a \$5M-or-larger loss produces the full \$4M recovery from this layer.

Exercise 7

Retrocession is reinsurance bought by a reinsurer, transferring part of its assumed risk to another reinsurer. Ceding commission is the allowance a reinsurer pays the cedent on proportional business to reimburse the cedent's acquisition and overhead expenses on the ceded premium.

Exercise 8 (†)

Gross is the risk as written — the full limit, full premium, full potential loss, before any reinsurance. Net is what the company actually keeps after cessions — the retained limit, premium net of reinsurance cost, loss net of recoveries. Gross is what you offered the insured; net is what your company is actually betting.

Exercise 9 (odd)

25% quota share on a \$2,000,000-limit risk. (a) Cedent retains 75% = \$1,500,000**; reinsurer takes 25% = **\$500,000. (b) A \$600,000 loss: cedent pays 75% = **\$450,000; reinsurer pays 25% = \$150,000. (c) The \$8,000 premium splits the same way: cedent keeps **\$6,000, reinsurer gets \$2,000 (before any ceding commission, which the reinsurer would pay back to the cedent out of its share).

Exercise 11 (†)

\$1M-retention, 4-line surplus-share treaty; a \$6,000,000 risk. Treaty capacity = retention + lines = \$1M + 4 × \$1M = \$5,000,000 total, of which the treaty itself provides \$4,000,000 (the 4 lines) above the \$1M retention. A \$6M risk therefore leaves **\$1,000,000 above the \$5M capacity with nowhere to go in this treaty. Two ways to place the remainder: (1) retain it on the cedent's net account (raising the net line above the intended retention), or (2) place it facultatively** (or into a second surplus/XOL treaty layer if one exists). The disciplined choice is usually facultative, so the net line stays at the intended retention.

Exercise 13 (odd)

Quota share relieves surplus strain because ceding a percentage of premium written lowers premium-to-surplus leverage — e.g., ceding 40% of a book frees roughly 40% of the surplus the gross premium would have tied up — but it does not solve the large-line problem: on a \$10M risk, a 40% quota share still leaves the cedent net on \$6M. Surplus share does the reverse emphasis: it leaves small risks fully retained (no surplus relief on those) but caps the net line on a large risk at the retention — e.g., on a \$10M risk over a \$1M retention, the cedent keeps only \$1M net.

Exercise 14 (†)

Tower: retention \$1M; L1 = \$4M xs \$1M (band \$1M–\$5M); L2 = \$5M xs \$5M (band \$5M–\$10M); L3 = \$10M xs \$10M (band \$10M–\$20M). (a) \$700,000** — below the attachment; cedent pays all **\$700,000, no recovery. (b) \$3,000,000** — cedent \$1M; L1 pays \$2M (the band \$1M–\$3M). Cedent \$1M + L1 \$2M. (c) \$7,500,000** — cedent \$1M; L1 \$4M (fully used, band \$1M–\$5M); L2 \$2.5M (band \$5M–\$7.5M). L3 untouched. (d) \$22,000,000** — cedent \$1M; L1 \$4M; L2 \$5M; L3 \$10M (all three layers fully exhausted, to \$20M); the remaining \$2M above \$20M is net (or needs cover above the program). Cedent's total out-of-pocket on this loss = \$1M + \$2M above program = \$3M** net.

Exercise 15 (odd)

Reinstatement: an XOL layer provides its limit a limited number of times per year; after a loss exhausts (or partly exhausts) the layer, the cedent usually pays a reinstatement premium to restore the cover for the rest of the term, and a layer may permit only one (or a few) reinstatements. On a catastrophe XOL in an active hurricane season this is acute: if the first storm exhausts the cat layer and the reinstatements are used up, a second or third storm arrives with the cat cover spent — the company is then fully net on a further catastrophe. That is why cat-program design weighs not just the height of the tower but how many times each layer can pay.

Exercise 17 (†)

A sample design (answers will vary): cedent retention \$2M**; **L1 = \$8M xs \$2M** (band \$2M–\$10M); L2 = \$10M xs \$10M (band \$10M–\$20M); L3 = \$10M xs \$20M (band \$20M–\$30M). Total program = \$30M. A **\$18M loss is paid: cedent \$2M + L1 \$8M (exhausted) + L2 \$8M (band \$10M–\$18M; partly used). L3 untouched.

Exercise 18 (†)

\$15M limit; \$5M net retention; rest ceded via surplus share. (a) A \$15M total loss: company pays **\$15M gross and recovers \$10M from the surplus-share reinsurer, so it pays \$5M net (its retention). (b) A 30% ceding commission on the ceded premium is income to the cedent that offsets its acquisition/overhead costs, so it lowers the company's net expense ratio** (the cede comes back as a credit against expenses on the retained book).

Exercise 19 (odd)

The statement means your real exposure — and therefore your true line size, appetite, and price — is set by what your company keeps, not by what it writes. You can offer a \$20M limit only because reinsurance lets you cede most of it; the company's actual bet is the net retention behind that gross limit. So when you near the edge of the treaty, you must arrange facultative cover to bring net back to the retention; and when you price, you must charge for the net risk plus the cost of the reinsurance that makes the gross line possible.

Exercise 20 (†)

Two reasons the gross-profitable coastal account can be unprofitable net: (1) it consumes a slice of the company's catastrophe-treaty capacity, whose cost (the allocated cat-treaty premium) must be charged against it; (2) it adds net volatility to a book that already carries coastal accumulation, which has a real capital and reinsurance cost (and may require facultative on any over-treaty slice). The price must be adequate against the net risk — what the company retains after paying for the reinsurance the account forces it to buy — not against the gross expected loss.

Exercise 21 (odd)

In a catastrophe year, reinsurance recoveries pull the net loss ratio sharply down (the cover working), so the net combined ratio is far better than gross. In a quiet year, the ceded premium with no offsetting recovery pushes the net combined ratio up relative to gross (you paid for protection you didn't need that year). A well-bought program should therefore lower the volatility of the combined ratio across the cycle — trimming the catastrophic peaks — even if it slightly raises the average combined ratio; you judge the program across the cycle, not in any single year.

Exercise 22 (†)

\$25M line offered; treaty capacity \$20M (retention \$5M + 3 lines). (a) No — the treaty holds only \$20M, so you cannot bind the full \$25M on the treaty alone. (b) The \$5M excess above the treaty must be placed facultatively (or declined / co-insured); otherwise that \$5M sits net. (c) Protect the company by confirming the facultative cover before firming the full \$25M quote — never "fac to follow." (d) The Chapter 13 term: attach the facultative placement as a subjectivity (a condition precedent to binding the full limit).

Exercise 23 (odd)

Risk B (in the heavy-accumulation named-windstorm zone) may be the worse net risk even at the same gross price because it adds to a correlated accumulation: a single storm hits B and the company's existing coastal book at once, consuming cat-treaty capacity and adding net volatility that the inland Risk A does not. The cost that makes the difference is the account's allocated share of the catastrophe reinsurance (and the capital for the retained event loss). Same gross rate, very different net economics — which is why concentration (Chapter 29) and cat cost belong in the price.

Exercise 24 (†)

Net economics of the catastrophe-exposed account. Net premium ≈ gross premium − cat-treaty cost − facultative premium − surplus-share cession + ceding commission = \$60,000 − \$14,000 − \$9,000 − \$10,000 + \$3,000 = **\$30,000 net premium kept. (b) Against a net expected loss of \$22,000**, the account keeps \$30,000 of net premium — leaving only \$8,000 to cover all internal expenses, profit, and the cost of the capital the retained risk ties up. That is thin: it may be inadequate once net expenses and the capital charge (Chapter 28) are loaded. The lesson: an account can look healthy gross and be marginal-to-inadequate net once the reinsurance it consumes is charged against it.

Exercise 25 (odd)

Moving from a 20% to a 40% quota share: the cedent's retained premium falls (it keeps 60% instead of 80%), its retained expected loss falls proportionally, and its surplus strain falls (more leverage relieved by the larger cession). The ceding commission (33% either way) offsets more expense at the larger cession. The larger cession is the right call when the cedent needs the surplus relief — to fund growth, to support a new or volatile line it does not fully trust its own pricing on, or to bring leverage within capital limits — and is willing to give away more expected profit for that relief. If capital is ample and the book is profitable, the smaller cession keeps more margin.

Exercise 26 (†)

Red flag: a deeply discounted cat cover from an unrated, offshore, uncollateralized reinsurer priced well below the market. The concept at stake is collectability / credit for reinsurance (§27.6): the cedent stays fully liable to its policyholders (privity), so an uncollectible reinsurer leaves the company exposed after the loss. Before relying on the cover, require either an acceptable financial-strength rating (AM Best, Chapter 3) and authorized status, or, if unauthorized, collateral (a trust or letter of credit) sufficient to take credit for the reinsurance under the NAIC framework. A cheap cession from a shaky reinsurer is a deferred loss, not a bargain.

Exercise 27 (odd)

The problem is adverse selection (Chapter 1, §1.4) operating against the reinsurer: the cedent's underwriters are keeping the best risks net and ceding the worst. Fac-oblig treaties are especially exposed because the cedent chooses what to cede while the reinsurer must accept whatever is sent — so the cedent can systematically pass along its weaker business. Reinsurers grant fac-oblig sparingly and monitor the ceded book's loss experience precisely to guard against this.

Exercise 29 (odd)

(Sample memo, ~200 words.) To: Underwriting Manager Re: Pricing the \$20M coastal account above gross expected loss This account looks profitable on a gross basis, but our economics are net, and net it needs a higher price. The \$20M property line is shared — we keep our **\$5M net retention and cede the surplus above it — so our per-risk bet is modest. The real cost is catastrophe: the plant sits in a named-windstorm zone, and a single storm correlates it with our whole coastal book. That exposure is ceded through our cat XOL treaty, and this account consumes a slice of that tower's capacity. The cat-treaty cost allocated to the account, plus any facultative premium on an over-treaty slice (net of ceding commission on the surplus share), must be charged into the rate. Priced only against gross expected loss, the account would be adequate gross and a loser net — the classic coastal trap. Recommendation:** load the rate for the net cost of the cat capacity consumed and confirm any facultative before binding the full limit; if the broker won't accept the net- adequate price, tighten terms or decline rather than grow unprofitable coastal premium.

Exercise 30 (†)

(Sample reply.) "Not quite — we don't get to forget about the reinsurer. Because of privity, our company stays fully liable to the policyholder whether or not the reinsurer pays, so if our reinsurer is insolvent when the loss comes, we owe the claim with no recovery — that exposure is called collectability. It is also why regulation only lets us take balance-sheet credit for reinsurance when the reinsurer is authorized or posts collateral. So the reinsurer's financial strength is part of our risk; a cheap cession from a weak reinsurer is a deferred loss."

Exercise 31 (odd)

(Ethics — competing considerations and a defensible position.) On one side: relying on the cheaper, lower-rated reinsurer lets the company write more coastal premium, hit growth targets, and earn bonus — and the reinsurer is probably fine. On the other: in a genuine 1-in-100 event — exactly when the cover must perform — the lower-rated reinsurer's ability to pay is less certain, and the cedent remains liable to its policyholders regardless (privity/collectability). A defensible recommendation: do not trade catastrophe security for growth and compensation. The cat cover is the thing standing between one storm and the policyholders' claims and the company's solvency; its reliability is not the place to economize. Themes in tension: the combined ratio / pricing-discipline and growth pulling toward the cheaper cover, against insurance serves a social function (the policyholders' protection) and underwriting is judgment (refusing the seductive but fragile economy). Buy the secure cover, or write less coastal business.

Exercise 33 (†)

Harbor Steel \$20M building limit; \$5M net retention; surplus-share treaty above. (a) Gross vs. net of a \$20M total fire loss:** company pays **\$20M gross, recovers \$15M from the surplus-share treaty, pays \$5M net (its retention). (b) Facultative needed? Only if the surplus-share treaty's per-risk capacity tops out below \$20M. If the treaty (retention + lines) reaches \$20M, the line is fully within the treaty and no facultative is required; if it caps lower, the over-treaty slice is placed facultatively. (c) The subjectivity (Chapter 13) to attach if facultative were required: "facultative reinsurance confirmed and bound for the over-treaty slice" as a condition precedent to binding the full \$20M limit (never "fac to follow").

Exercise 35 (odd)

The three questions this chapter explicitly hands forward: - Chapter 28 (Capital): how much capital / cat-charge the account consumes, and whether it earns its cost of capital. - Chapter 29 (Portfolio): whether the coastal-property concentration in the Port Hadley zone (and the industry/broker concentration) still has room — the portfolio fit. - Chapter 30 (Catastrophe modeling): the account's PML/AAL contribution, the named-storm return-period loss, and the Port Hadley accumulation against the zone aggregate.

Chapter 28

Worked solutions to the daggered (†) and odd-numbered exercises. All dollar figures are illustrative constructed teaching examples; the methods are what matter.

Exercise 1

Policyholder surplus is an insurer's assets minus its liabilities — the net-worth cushion that absorbs losses worse than expected. It is named for policyholders (rather than shareholders) because its first function is to protect them: it is the buffer that stands between a bad year and a broken promise, ensuring claims are paid even when losses exceed what the company expected and reserved for. (§28.1)

Exercise 3

Premium-to-surplus ratio = net written premium ÷ policyholder surplus. A "2-to-1" ratio means the company is supporting two dollars of net written premium on every one dollar of surplus — it is writing twice as much business as it holds capital, which is more leveraged (and so more exposed to an underwriting miss) than a 1-to-1 carrier. (§28.2)

Exercise 4

The four major RBC risk categories (P&C labels): - Asset risk — invested assets may lose value or a bond may default; risky assets draw bigger charges. - Credit risk — money owed to the insurer may not be collected, including reinsurance recoverables — so uncollectible reinsurance is penalized. - Reserve risk (underwriting) — loss reserves may prove inadequate; long-tail/volatile lines draw more. - Premium risk (underwriting) — this year's written business may run worse than priced; volatile, catastrophe-exposed lines draw more. (§28.3)

Exercise 5

RBC action levels, in order: ≥200% — no action; 150–200% — Company Action Level (file a plan to restore capital); 100–150% — Regulatory Action Level (regulator examines, issues corrective orders); 70–100% — Authorized Control Level (regulator may seize the company); <70% — Mandatory Control Level (regulator must seize). Intervention escalates before the company is actually insolvent. (§28.3)

Exercise 7

Enterprise risk management (ERM) is the discipline of identifying, measuring, aggregating, and managing all of an organization's material risks together, as a single portfolio governed at the top of the company. It differs from traditional siloed risk management — in which each function (underwriting, investments, reinsurance) manages only its own risk — by owning the question no silo answers: what happens when several risks move together. (§28.5)

Exercise 9

The cost of capital is the return the providers of an insurer's surplus require on it — at least what they could earn on a comparable-risk investment elsewhere. It makes some combined-ratio-profitable accounts value-destroying because every risk ties up a slice of finite surplus that must earn that required return: an account can beat its combined-ratio target yet still earn less than the cost of the capital it consumes, in which case the company would have been better off deploying that capital elsewhere. (§28.6)

Exercise 11

Surplus = \$600M; net written premium = \$900M. - (a) Premium-to-surplus ratio = \$900M ÷ \$600M = 1.5 to 1. - (b) A 6-point loss-ratio miss applies to \$900M of premium ≈ **\$54M of unexpected loss, which is \$54M ÷ \$600M ≈ 9% of surplus. - (c) Same \$600M surplus but \$1,800M premium → a 6-point miss ≈ \$108M**, which is \$108M ÷ \$600M ≈ 18% of surplus — double the bite from the identical underwriting error. What changed is leverage: doubling the premium on the same surplus doubles the consequence of any miss. (§28.2)

Exercise 13

Premium leverage (net written premium ÷ surplus) measures the risk you are taking on with newly-written business. Reserve leverage (loss reserves ÷ surplus) measures the risk you have already taken and may have mis-measured — the chance that reserves for past business prove inadequate. Reserve leverage matters most for a workers'-comp book because WC is long-tail: claims (e.g., serious injuries with lifetime medical and indemnity) settle over many years, so the company holds large reserves whose ultimate cost is genuinely uncertain, and an inadequacy comes straight out of surplus years after the business was written. (§28.2)

Exercise 14

Three things a 1.1-to-1 premium-to-surplus ratio cannot see: 1. Catastrophe concentration — the book could be heavily exposed to one peril/zone (one hurricane could take a quarter of surplus); the ratio counts volume, not correlated tail risk. 2. Reserve inadequacy — deficient reserves overstate the surplus in the denominator, so true leverage is worse than reported (the surplus is partly illusory). 3. Underpricing / rate inadequacy — business may be written below an adequate rate, so the premium volume that looks "supported" is actually a delayed-action charge against surplus as losses develop. (Also acceptable: asset risk in a risky investment portfolio.) (§28.2, §28.3)

Exercise 15

RBC takes the square root of the sum of squared charges rather than a simple sum because the risk categories are assumed largely independent — your bond portfolio crashing, your reserves deteriorating, and a hurricane hitting are unlikely to all happen in the same year. Holding enough capital for every risk to go wrong at once (the naive sum) would be absurdly conservative; the covariance adjustment credits diversification across kinds of risk, just as the law of large numbers credits diversification across policies. It relies on the real-world fact that these risks are normally weakly correlated. (§28.3)

Exercise 16

Charges (\$M): asset 50, credit 30, reserve 80, premium 60. - (a) Naive sum = 50 + 30 + 80 + 60 = \$220M. - (b) Covariance-adjusted = √(50² + 30² + 80² + 60²) = √(2,500 + 900 + 6,400 + 3,600) = √13,400 ≈ \$115.8M. - (c) The adjusted number is far lower because the square-root-of-sum-of-squares assumes the risks are largely independent and so won't all peak together. It is dangerous to rely on whenever the risks become correlated — e.g., a financial crisis (2008) that simultaneously depresses asset values and impairs reinsurers' ability to pay, or a single catastrophe the formula treats as one independent risk but that hits a whole coastal book at once. In those cases true required capital exceeds the adjusted figure. (§28.3)

Exercise 17

An RBC ratio of 165% places the carrier in the Company Action Level zone (150–200%). It is required to file a plan with its regulator describing how it will restore its capital. "165%" sounds like a healthy multiple, but the denominator is the Authorized Control Level — the point at which the regulator may seize the company — so 165% means the carrier holds only 1.65× the seizure level and has already fallen below the 200% "no action" floor. Healthy carriers run far higher. (§28.3)

Exercise 18

The RBC credit-risk charge penalizes a carrier for reinsurance recoverables it may not collect (the charge rises with the reinsurer's weakness/uncollectibility). This makes an underwriter's choice of reinsurer a capital decision: ceding to a weak or unrated reinsurer not only risks non-payment when a loss hits (Chapter 27's warning) but increases the primary carrier's own required capital through a higher credit charge — so a cheap reinsurance deal with a shaky counterparty can cost more in required capital than it saves in ceded premium. (§28.3; Ch. 27)

Exercise 19

Build any three of these differences, with the pricing point: - Approach: RBC is a factor-based formula by risk category; Solvency II is market-consistent and calibrated to a confidence level. - Calibration: RBC has no single explicit confidence level; Solvency II's SCR targets 99.5% over one year (a 1-in-200-year loss). - Internal models: RBC does not use them for the requirement; Solvency II permits approved internal models. - Valuation: RBC uses statutory book values; Solvency II uses market-consistent values. - Disclosure: RBC's detail is limited/confidential; Solvency II adds Pillar 3 public reporting.

The difference that matters most for a catastrophe-exposed account is the explicit 1-in-200-year calibration: it forces capital to be held against the monstrous tail event the account contributes to (not the average season), which is exactly the cost that must be loaded into a coastal account's price. (§28.4)

Exercise 21 (odd; build a comparison — see Exercise 19 for the differences)

See Exercise 19 for three concrete differences and the pricing point. In short: the SCR's explicit 1-in-200, market-consistent calibration is the feature most relevant to pricing catastrophe risk, because it sizes capital to the severe tail loss — the very thing a coastal account's premium must pay for. (§28.4)

Exercise 23

An aggregate risk no single specialist would see: the firm's total hurricane exposure assembled across all functions at once — property and business-interruption claims, commercial-auto and marine losses, and the reinsurance recoverables a big storm puts at risk simultaneously — plus the investment-portfolio hit if the same event disrupts markets. Each desk's piece looks manageable alone; only an enterprise view reveals that one event strikes them together. (Other valid answers: a recession hitting credit/surety + investments + new-business volume; a pandemic hitting life/health + business interruption + workforce.) (§28.5)

Exercise 25

• (a) Economic profit = underwriting profit − capital cost.
• Account X: 8% − 5% = +3% of premium (creates value).
• Account Y: 2% − 1% = +1% of premium (creates value, but less).
• (b) The company prefers Account X (+3% vs. +1% of economic profit). The lesson: judging on combined ratio alone is incomplete — both accounts are "profitable," and X has the worse combined ratio (92% loss side vs. 98%), yet X creates more value because its capital cost is much lower relative to its underwriting profit. Economic profit, not the combined ratio, ranks them correctly. (§28.6)

Exercise 27

RAROC (risk-adjusted return on capital) measures an account's, line's, or book's profit against the capital it consumes, rather than as a raw margin. By dividing the (risk-adjusted) profit by the capital tied up, it puts a capital-light inland casualty line and a capital-heavy coastal property line on the same footing: each is expressed as a return per dollar of surplus consumed, so the company can compare them directly and allocate scarce capital to whichever earns the best risk-adjusted return — even though their combined ratios and capital appetites differ wildly. (§28.6)

Exercise 29

For most carriers the rating-agency capital model governs catastrophe-writing capacity (not the regulatory RBC floor) because a strong financial-strength rating is effectively a license to do business: brokers won't place serious accounts with a weakly-rated carrier, reinsurers price by the cedent's rating, and many contracts require a minimum rating. Agencies demand more capital than the RBC minimum to support a rating and model catastrophe exposure explicitly and severely. So a carrier hits the rating-agency cat limit long before it approaches an RBC action level — the rating, set by the market, binds tighter than the law. (§28.7)

Exercise 30

The capital "stack," from lowest to highest: the RBC Mandatory Control Level (legal seizure floor) → the RBC "no-action" floor (200% of ACL) → the Solvency II SCR (if applicable) → the company's own ERM/risk-appetite buffer → the rating-agency requirement to hold the target rating (usually the highest). The regulator sets the floor; the carrier manages to the rating-agency bar, because that is what the market demands. A downgrade acts like a slow death spiral: it shrinks the flow of business (brokers and clients leave a lower-rated carrier), which shrinks the profit that replenishes capital, which can threaten the rating further — a self-reinforcing decline, milder but structurally like the adverse-selection death spiral of Chapter 1. (§28.7; Ch. 3)

Exercise 31

The CUO means the company lacks the capital appetite for the account, not that it is unprofitable. Even at a 94% combined ratio, a large coastal-condo book ties up heavy surplus against its catastrophe contribution, and the binding constraint is almost certainly the rating-agency capital model (and the ERM cat-aggregate), not the loss ratio: writing it would force the company to hold more capital to keep its rating or put the rating at risk. The five calm years are a trap — the capital charge is for the monstrous storm, not the average season (§28.4). The question to ask: "What does this add to our modeled severe- event, net-of-reinsurance loss and our zone aggregate, and do we have rating-agency capital room for it?" To write any of it, you would need either room under the cat-aggregate/rating-agency limit, reinsurance that genuinely transfers the tail (from a collectable counterparty), or a price rich enough to earn the cost of the capital consumed — most likely all three, and probably only a portion of the book. (§28.6, §28.7)

Exercise 33

Red flags the headline numbers hide: - Catastrophe concentration in one hurricane-exposed state — the 1.4-to-1 premium-to-surplus ratio and the 210% RBC ratio cannot see correlated tail risk; one storm could take a large share of surplus in a day (the §28.3 RBC blind spot; the §28.5 aggregate the ORSA stress test is for). - 210% RBC is only just above the "no-action" floor — for a healthy carrier that is thin, especially with a concentrated book; there is little margin for adverse development. - Record growth in a soft market — fast growth often means underpriced business (Case Study 2's pattern), a delayed charge against surplus as losses develop. The single event that turns it catastrophic: one severe hurricane in that state, which hits the concentrated book all at once, blows through the thin capital margin, and (if reinsurance is inadequate or uncollectable) drops the RBC ratio through the action levels. (§28.2, §28.3, §28.5)

Exercise 35 (memo — model answer)

To: Underwriting Committee — Re: Harbor Steel pricing margin. I am recommending we bind Harbor Steel at the indicated, debit-rated price, which carries a margin richer than its loss experience alone would justify. The reason is capital, not caution. Harbor Steel's named-windstorm property and its long-tail WC and GL reserves tie up materially more of our surplus than an inland account of the same size, and that surplus has a cost. Priced to a "merely adequate" combined ratio, the account would clear our loss-ratio hurdle and still fail to earn the cost of the capital it consumes — destroying value every calm year, before any hurricane. The richer margin is what makes the account capital-adequate: it earns a fair return on the surplus we must hold against its catastrophe contribution. What I am protecting is that surplus — the capacity that backs every promise on our book — and our rating-agency capital position, which the cat exposure draws against most heavily. (§28.6, The Underwriting File)

Exercise 37 (ethics / governance — discussion)

How confidentiality protects policyholders: publishing a near-action-level carrier's RBC results, or using them to rank insurers, could trigger a run — brokers and policyholders fleeing, which collapses the business that generates the profit needed to recover, turning a recoverable weakness into a self-fulfilling failure (an adverse-selection death spiral). Confidentiality lets the regulator work the recovery plan quietly while the company still has a chance. The cost: a public — and even sophisticated buyers — cannot easily distinguish a strong insurer from a weak one using RBC, so the market leans on the rating agencies as the public signal of strength, concentrating enormous power in private firms whose methods are their own. The honest answer holds both: confidentiality is defensible as a stability measure, but it shifts the transparency burden onto ratings and onto the regulator's competence, and it asks the public to trust a system it cannot directly inspect. (§28.3)

Exercise 38 (Underwriting File extension)

Chapter 27 ceded Harbor Steel's catastrophe exposure to the cat XOL treaty over a ~\$5M net retention. This reduces the capital this chapter assigns to the account: most of the catastrophe tail — the heaviest capital charge — is transferred to the reinsurer, so the net premium-risk/cat charge the company holds against Harbor Steel is far smaller than the gross exposure would imply. But it creates a new, smaller charge: a reinsurance recoverable sits on the balance sheet, drawing an RBC credit-risk charge for the possibility the reinsurer cannot pay (§28.3) — so the capital saved on cat is partly given back as counterparty capital, and the quality of the reinsurer matters to the carrier's own capital. Running disposition reached: the account is capital-adequate at the indicated price — it earns a fair return on the (now largely net-of-reinsurance) surplus it consumes because it is priced to a richer margin. The two questions explicitly left open: whether the Port Hadley zone has aggregate room for it (Chapter 29) and the modeled PML/AAL contribution that sizes the cat charge precisely (Chapter 30). (The Underwriting File; Ch. 27)

Exercise 39 (Underwriting File extension)

Binding at a "merely adequate" combined ratio would be a mistake even with no hurricane because of economic profit = underwriting profit − cost of capital (§28.6). Harbor Steel ties up heavy surplus (the cat charge plus long-tail reserves), and that surplus has a cost. A price that merely clears the loss-ratio hurdle produces an underwriting profit too small to cover the cost of the capital the account locks up — so every calm year the account earns less than the capital deserves and destroys company value, quietly, regardless of whether a storm ever comes. The hurricane is the dramatic risk; the value destruction is the certain one at an inadequate price. The cat load is what converts the account from value-destroying to value-creating. (§28.6, The Underwriting File)

Chapter 29 — Worked Solutions to Selected Exercises

Solutions are provided for the daggered (†) exercises and the odd-numbered exercises. All figures are illustrative teaching numbers; the point is the method and the portfolio reasoning, not the arithmetic.

Exercise 1 (†)

A book of business is the entire collection of policies an underwriter, unit, or company has on the risk at a given time, managed as one whole. Three properties a book has that no single account has: (1) a blended result — its loss/combined ratio is driven by the mix of all accounts, not the adequacy of any one price; (2) an aggregate volatility that depends on the correlation among its risks (diversified or concentrated); and (3) a trajectory — it is growing or shrinking, improving or deteriorating at the margins where new and renewal business come and go. The slogan: you can be right on every risk and wrong on the book.

Exercise 2

Diversification fills the book with risks whose losses do not move together, so the spread of the book's possible aggregate outcomes shrinks — the bad outcomes don't all arrive on the same day. But each account still carries the same probability and severity of loss it always had, so neither its expected loss nor the book's expected loss changes. Diversification buys lower volatility, not a lower mean — which is valuable precisely because, in insurance, one bad year can be existential and capital held against volatility is costly (Chapter 28).

Exercise 3

The four axes and the correlation each breaks: Geography — breaks the correlation of one hurricane, quake, freeze, or flood hitting everything at once. Industry — breaks one sector's downturn, one recalled component, or one mass tort that names a whole class. Size — breaks the dominance of a single large account whose bad year swamps the law of large numbers for the book (severity concentration). Line — breaks one line's cycle or systemic loss (e.g., a casualty tail) from sinking the whole book.

Exercise 4 (†)

Concentration risk is the danger that a large share of a book is exposed to the same loss event, so one occurrence damages many policies at once and the law of large numbers (which assumed independence) fails. Accumulation is the mechanism — the build-up of many separate policies' exposure to a single event, place, peril, or counterparty. So accumulation is the thing that builds; concentration is the state of danger it produces. A portfolio manager measures accumulation (by zone, class, counterparty) precisely to keep concentration within what capital and reinsurance can survive.

Exercise 5

The four flavors: Peril-zone — many property accounts in the same hurricane/quake/wildfire/flood footprint (the classic; Harbor Steel in Port Hadley). Industry/class — many accounts in one business class exposed to a shared recall, regulation, or mass tort. Counterparty — concentration in one reinsurer, one large insured across policies, or one broker controlling a large share of the book. Clash / correlation — a single event triggering different lines at once (a hurricane hitting property + BI + cargo; an explosion hitting property + GL + WC + auto on one insured).

Exercise 7

The retention ratio is the share of a book that renews term to term (premium or count retained ÷ that up for renewal). It must be read for quality because who stays and leaves is not random. In a soft market, competitors pick off your best accounts (they can underprice them and profit), while your worst accounts — the ones nobody else wants — renew gratefully. Retention that holds up only because the bad risks have nowhere to go is adverse selection through the back door of renewals — a strong number masking a deteriorating book.

Exercise 8 (†)

The new-business penalty is the well-established practitioner observation that newly written business runs a meaningfully worse loss ratio than seasoned renewal business in its first year or two. Why: a renewal you've seen for years is a known quantity — you hold its loss runs, you know its management, familiarity has largely solved the adverse-selection problem. A new account shows only what its submission reveals; the prior carrier's reasons for any non-renewal may be opaque; and the very fact that it is shopping means someone chose to let it go. The discipline is to budget growth knowing new business must earn its way into the seasoned book, and never to fund aggressive growth entirely with new business without pricing the penalty in.

Exercise 9

A serious plan answers: (1) premium volume by segment (growth broken down so "grow 10%" becomes segment-level direction); (2) target loss/combined ratio by segment (the line in the sand for the variance review); (3) new-business/renewal split and retention targets (with the new-business penalty priced in); (4) rate change by segment; and (5) expense and capital budget (including the catastrophe charge the planned book will consume). An undifferentiated "grow 10%" is dangerous because the easiest way to grow is to cut price or loosen standards — so growth-for-its-own-sake buys premium today at the cost of losses two or three years out, and an undifferentiated target invites exactly that.

Exercise 11 (†)

The book is not diversified in any way that matters. Strong axis: size — no account exceeds \$5M, so no single loss dominates. Fatally weak axis: geography — all 1,200 policies sit in one 60-mile coastal strip, so they are exposed to the same named-windstorm event. Industry spread (eight industries) is irrelevant to a wind peril that strikes by location, not by class. The single event that undoes the apparent spread is one hurricane through the strip: it hits all 1,200 at once. By count this is a pool of 1,200; by correlation it is a single bet wearing 1,200 costumes — the violation of the word independent.

Exercise 13

At equal price and quality the two accounts are not equally valuable to the book because a risk's diversifying value is part of its value. The Mountain West property balances an existing coastal concentration — it adds exposure to a peril uncorrelated with the rest of the book, so it lowers the book's volatility. The tenth Gulf-zone warehouse piles onto an existing accumulation — it adds correlated exposure, raising the book's probable maximum loss, its catastrophe capital charge, and its reinsurance need. A portfolio manager reflects this by appetite (relax on diversifying business, tighten or cap concentrating business) and sometimes by price, charging the concentrating risk for the accumulation it deepens.

Exercise 14 (†)

Two scenarios in which "uncorrelated" lines lose together: (1) a systemic shock that shares a hidden driver — a pandemic that simultaneously hits life, health, business-interruption, and event-cancellation lines; or a financial crisis that hits D&O, mortgage, and surety at once. (2) A single physical event that clashes across lines — a hurricane or industrial explosion triggering property, BI, marine, liability, WC, and auto on overlapping insureds. The model could not see it because it learned correlation from historical loss data, and these correlations were either absent from the history (a novel systemic event) or rare enough to read as zero. The human judgment the model cannot supply is the question "what could make these supposedly independent risks lose together — something not in the data?"

Exercise 15

"A book of a thousand homes in the same flood zone is a single bet wearing a thousand costumes" because the thousand policies are not independent — the flood that hits one hits essentially all, so the loss arrives as one giant correlated event, not a thousand small unrelated ones. The law of large numbers, which would stabilize a thousand independent risks, has nothing to work with. The management instrument that prevents this is an aggregate exposure cap by peril zone — a hard limit on how much the book may hold in any one hurricane/quake/flood footprint — enforced by a portfolio referral that routes any account which would breach the cap to a manager before it can be bound (§29.7).

Exercise 16 (†)

Counterparty accumulation in three forms: (1) Reinsurer concentration — if too much of the book's catastrophe protection sits with one reinsurer and that reinsurer fails or disputes, the "ceded" losses come home to the cedent (collectability risk, Chapter 27). (2) Single large insured across policies — one insured carrying property, GL, WC, and auto with you means a single event on that insured triggers a correlated multi-policy loss. (3) Broker concentration — if one broker controls a large share of the book, the exposure is to the relationship: lose it (or have it sour) and you lose the whole segment at once, and a soft-priced broker book adversely selects against you en masse. Each is a portfolio exposure because no individual policy looks risky — the risk lives in the correlation across the counterparty.

Exercise 17

Concentration is "expensive every year" because a concentrated coastal book has a larger probable maximum loss in its peril zone (Chapter 30), which forces the insurer to hold a larger catastrophe capital charge (Chapter 28) and to buy more reinsurance (Chapter 27) — both real, recurring costs paid regardless of whether a storm ever arrives. So the concentration raises the book's cost of capital and its reinsurance spend continuously. A disciplined portfolio manager should therefore have been charging for the concentration all along, treating the catastrophe load as a cost of the business and not a free option that only bites in the storm year.

Exercise 19 (†)

(a) Profit engine: light manufacturing (\$14.0M at 58% — the best loss ratio and the largest premium). Bleeding: habitational (\$6.0M at 112% — paying out far more than it earns). (b) Accumulation worry: coastal property (\$11.0M at 66%, growing 18%) — its growth is part of the problem because each new coastal account deepens the peril-zone accumulation, raising the book's PML, capital charge, and reinsurance need even though the loss ratio looks acceptable in a quiet year. (c) Actions — Habitational: re-underwrite and re-price hard, tighten terms/deductibles, non-renew the worst, or exit the class; stop or reverse its growth immediately. Coastal property: refer further growth against the zone aggregate cap, slow or cap new business in the zone, and confirm reinsurance/capital cover the rising accumulation (and watch whether the 18% growth signals you've gotten cheap).

Exercise 21

Tell the sales team, in plain language: "Fast growth in a segment where the loss ratio is also climbing usually means we're winning that business because we're the cheapest carrier in it — and we're cheap because our price is below the risk, which is why the losses are rising. We're not beating the market; the market is handing us the business it priced higher for a reason, and the losses will keep coming until we fix the price." (Two sentences: the growth is adverse selection at portfolio scale, and the rising loss ratio is the bill arriving.)

Exercise 23 (†)

The case for disciplined shrinkage: if you follow the market down 15% to hold your growth number, you keep your retention up today — but you've now re-priced your whole renewal book below the risk, and you've won new business only by being cheapest. The premium looks healthy and the early loss ratio looks fine, because losses from underpriced business take two or three years to develop. When they arrive, the soft-market business is on the books at inadequate rates and the carrier that grew fastest bleeds worst in the correction. The disciplined alternative is to hold rate and terms, accept that retention falls and the book shrinks (let the bad business go to competitors who'll regret it), and preserve the capacity to grow when the hard market returns. The cost of discipline (visible lost premium) is immediate; its benefit (a book that doesn't blow up in the correction) is deferred — which is exactly why it's the hardest discipline in insurance.

Exercise 25

The cycle's lag is the delay between writing underpriced soft-market business and the losses from it developing — typically two or three years. It traps experienced managers because soft-market business looks good on the day it's written: the price is competitive, the broker is happy, the premium counts toward plan, and the early loss ratio is fine because nothing has developed yet. So the carrier growing fastest at the bottom of the soft market is often celebrated as the best-run shop in the room — right up until the development triangles turn and the losses arrive en masse. The lag makes the warning signs (soft rate, surging win rate, terms drift) leading indicators that you must act on before the loss ratio confirms the problem, because by the time it does, the underpriced business is already bound.

Exercise 27

The red flags hiding in the "banner quarter": premium up 25% plus a doubled win rate is the classic signature of having gotten cheap — you're suddenly winning a lot because your price fell below the market, which means you're writing the business everyone else priced higher (adverse selection). All-time-high retention in this context likely means you're also holding renewals by under-pricing them. And the early loss ratio on plan proves nothing, because losses take years to develop (the lag). What's most likely actually happening: the unit is buying premium with rate, the book's true loss ratio is deteriorating invisibly, and the "good news" is the appearance every soft-priced book presents right before the losses arrive. The leading indicators (rate change, win rate, terms) are all flashing; only the lagging one (loss ratio) looks fine.

Exercise 28 (†)

The worst-case interpretation of "high retention in a soft market": you are retaining business only because you kept your price flat while the market fell, which means you are now the expensive carrier on your good accounts (they're staying out of inertia and will eventually leave) or — more dangerously — the accounts you're retaining are the ones competitors don't want, so you're holding the worst of a softening market while the best risks get picked off. The one cut of the data that would confirm or refute it: a quality-weighted retention analysis — segment retention by account quality/loss ratio and look at what is renewing versus lapsing. If you're keeping the good accounts, the stability is real; if you're keeping the worse accounts and losing the better ones, the high retention is adverse selection wearing a flattering number.

Exercise 29 (†)

Sample memo (≈200 words):

To: A. Line Underwriter From: Portfolio Manager Re: Declination — coastal account [broker: Meridian]

Thanks for the workup on this one — to be clear, this is a good risk and a clean submission, and the decline is ours, not yours. The account is adequately priced and well-termed; on its own merits I'd write it. The problem is the book: we are at our aggregate cap in this named-storm zone for the year. Adding this account would push our zone exposure past what our cat treaty and capital are sized to survive, which raises our PML and our reinsurance cost across the whole coastal book — a portfolio cost the single account's premium doesn't justify.

Please decline to Meridian on that basis explicitly: "This is a sound account and our constraint, not a reflection on the risk or on you — we're simply full in this zone this year and can't add coastal capacity." That keeps the relationship intact and tells a strong broker we want their other business. If we free up zone capacity (shedding or non-renewing weaker coastal accounts at their term), let's revisit this one first. Document the decline reason as portfolio/zone-aggregate, not risk quality.

Exercise 31

Three portfolio referral triggers (distinct from authority triggers): (1) "Adds to a peril zone within X% of its aggregate cap" — guards against peril-zone accumulation breaching what capital/reinsurance can survive. (2) "Grows a segment more than X% over its planned premium" — guards against mix shift and soft-pricing in a hot segment (you may be getting cheap). (3) "Concentrates more than X% of the book (or a segment) in a single broker or source" — guards against counterparty/distribution dependence. (Each is about the book's room, not the underwriter's dollar authority or the account's quality.)

Exercise 33 (†)

Both sides. Keep writing it: it's currently profitable, it makes the growth plan, the price is what the market bears today, and you're not deceiving anyone — you're simply competitive. Stop / re-price: "currently profitable" is the soft-market illusion — you're winning because you're cheapest, which is adverse selection, and the losses from under-priced business arrive two or three years out; making the growth plan with business you'll lose money on is buying premium at the cost of future solvency and, ultimately, the policyholders' promise. What a disciplined manager does: treat fast growth + thin/soft pricing as a warning, not a triumph — verify rate adequacy against loss trend, raise rate or tighten terms even at the cost of growth, and refuse to hit a growth number with knowingly under-priced business. The ethics and the economics point the same way: the combined ratio tells the truth eventually, and writing business you expect to lose money on to make a plan fails both the shareholder and the insured.

Exercise 35

This chapter cannot settle whether the Port Hadley zone has aggregate room because answering it requires measuring the accumulation — running the book's coastal exposure through a catastrophe model to quantify the peril-zone PML against the cap — and that machinery is Chapter 30 (catastrophe modeling and accumulation management). The ordering is correct because portfolio management (this chapter) establishes the concept and the gate — "does the book have room?" — while cat modeling supplies the measurement that answers it for a specific zone and account. You must know the question (portfolio fit) before the measurement means anything, and the measurement (Chapter 30) before the final bind (Chapter 40). Hence Harbor Steel's disposition here is deliberately conditional: fits appetite if the zone has room, with the "if" handed to the next chapter.

Chapter 30

Worked solutions to the daggered (†) and odd-numbered exercises. All dollar figures are illustrative teaching numbers, not real losses. (Even, non-daggered items are discussion prompts whose reasoning is developed in the chapter text.)

Exercise 1 (why catastrophe breaks the law of large numbers)

The law of large numbers stabilizes aggregate losses only when the individual losses are independent — one loss telling you nothing about whether another occurs (§1.2). A catastrophe is a single common cause (one hurricane, one earthquake, one fire) that strikes a whole region's exposures at the same time, so the losses are correlated, not independent. The pool you thought was diversified by count turns out to be a single bet on whether the event occurs, multiplied across every exposure in its footprint — diversification by count is an illusion when the count is concentrated in space.

Exercise 3 (AAL and PML; price vs. balance sheet)

The AAL (average annual loss) is the mean of the entire catastrophe loss distribution — the long-run expected annual cat loss. It is the catastrophe pure premium and belongs in the price (it answers "what do we charge?"). The PML (probable maximum loss) is a loss far out in the tail, at a chosen return period — it answers "what must we survive?" and belongs on the balance sheet (it sizes reinsurance and the capital the company holds).

Exercise 5 (return-period conversions)

A 0.4% annual exceedance probability is the 1-in-250-year loss (1 ÷ 0.004 = 250). A "1-in-50-year" loss is a 2% annual exceedance probability (1 ÷ 50 = 0.02). Both are probabilities per year, not schedules.

Exercise 7 (the protection gap)

The protection gap is the portion of total economic catastrophe losses that is not covered by insurance — borne instead by households, businesses, and governments. (It is widest exactly where catastrophe risk is highest and least affordable.)

Exercise 9 (reading the illustrative EP curve)

(a) The 1-in-100-year loss is about \$250M**. (b) The 1-in-250-year loss is about **\$300M. (c) A rating agency is more likely to use the 1-in-250-year figure as a capital standard, because holding capital only to the 1-in-100 PML leaves the company, by construction, undercapitalized for the storm worse than 1-in-100 — and such storms occur on schedule with their probabilities. The longer return period buys a margin beyond "the hundred-year event."

Exercise 10 (the chief underwriter's confusion)

The chief underwriter is confusing the AAL with the PML. Collecting more than the \$6M AAL in cat load means the average storm is paid for — that is an AAL (pricing) statement. The \$240M 1-in-100 PML is a tail (survival) figure: it is what the company could lose in a bad single event and therefore what reinsurance and capital must be sized to. Being "covered" on the AAL says nothing about whether the company can survive the \$240M event. Price to the AAL; capitalize to the PML — they are different questions.

Exercise 11 ("PML" is a misleading name)

It is misleading on two counts. (1) It is not the maximum possible loss — the EP curve extends further right, to losses worse than whatever return period you chose. (2) It is not "probable" in any everyday sense — at, say, the 1-in-250-year point it is a rare loss, not a likely one. The three things that must always be specified when a PML is quoted: the return period, whether it is per-event or annual (aggregate), and whether it is gross or net of reinsurance.

Exercise 12 (the calm-years argument)

Catastrophe losses are heavy-tailed: most years sit near zero and a rare year is enormous (§30.1). The AAL is an average over that whole distribution, including the catastrophic years, so a run of calm years is exactly what the distribution predicts — it is not evidence the load is too high. Reading recent calm as proof of over-pricing is the soft-market trap (§30.6): you would be lowering the cat load for a tail that has not gone anywhere, and the loss arrives on its own probabilistic schedule, not the broker's.

Exercise 13 (the "hundred-year flood" homeowner)

"1-in-100-year" does not mean once per century on a schedule; it means a 1% chance every single year, independent of last year — the dice have no memory. So her flooding last year does nothing to reduce next year's risk: next year still carries roughly the same ~1% chance as any year, and over her remaining decades the cumulative chance of another is substantial (≈26% over 30 years). The right way to think: treat every year as carrying the same annual probability, and insure/mitigate accordingly.

Exercise 14 (mortgage-horizon arithmetic)

Chance of at least one 1-in-100-year event over 30 years = $1-(0.99)^{30}$. $(0.99)^{30}\approx 0.7397$, so the probability $\approx 1-0.7397 = 0.2603$, about 26%. It matters because a lender holding a 30-year mortgage on coastal collateral faces a roughly one-in-four chance, over the life of the loan, of the very event borrowers imagine they will never see — which is why lenders require insurance and why the collateral's catastrophe exposure is a credit issue, not just an insurance one.

Exercise 15 (two "hundred-year" floods in a decade)

Not a fair conclusion. A 1-in-100-year loss is a 1% annual probability, and independent 1% annual events can occur close together — two in a decade is unlikely but entirely consistent with the probability, not evidence the model is broken. (It might prompt a check of whether the baseline has shifted — §30.6 — but clustering alone does not indict the model.)

Exercise 16 (regulator vs. rating agency on return period)

They have different mandates. The regulator's concern is largely affordability and availability of coverage for the public, so a prior-approval regulator may push for a lower cat load (a shorter effective return period) to keep premiums affordable (Ch. 4). The rating agency's concern is solvency — whether the insurer can survive a severe event and keep its promises — so it pushes for capital held against a longer return period (e.g., 1-in-250). The tension between affordability and solvency is real and is itself a driver of the availability crisis (§30.7).

Exercise 17 (underwrite the zone — the new Port Hadley account)

The zone limit is \$250M PML and the zone already sits at \$210M, leaving \$40M of headroom. The new account adds about \$22M** at the 1-in-100 level — which *fits* within the \$40M headroom, but consumes more than half of it on a single account, leaving the tightest zone with very little room. Decision: writable on cat grounds, but only with eyes open that it nearly exhausts the zone; document that it is now the marginal account and that further Port Hadley growth is effectively blocked until something rolls off. The account's own adequate price (its AAL) does not by itself justify the zone consumption — that is the trap of defending price against an accumulation objection (§30.3). Two ways to write it anyway if headroom were shorter: (1) buy facultative reinsurance (Ch. 27) on this specific risk to cede its marginal cat contribution out of the zone; (2) tighten its terms — a higher percentage named-storm deductible or a wind sublimit — so its modeled net loss, and thus its zone consumption, shrinks (§30.2). You could also decline or seek an offsetting non-renewal elsewhere in the zone.

Exercise 19 (find the red flag — the blank-field hotel submission)

Two distinct data problems, each biasing the model toward understatement. (1) Blank year-built and roof-type: the vulnerability module must substitute default construction assumptions, which are usually optimistic, so it returns a lower damage ratio at every wind speed than an old, poorly-roofed building would actually suffer — understating the loss. (2) ZIP-code-only location: coarse geocoding can place the property on the wrong side of a storm-surge or flood line; a property mislocated by a fraction of a mile can have its surge loss understated by an order of magnitude (§30.2). Fix: demand the actual construction details and a rooftop-level geocode before trusting the account's modeled catastrophe contribution.

Exercise 21 (price the cat load)

Work with the catastrophe portion only. The cat pure premium (AAL) is \$48,000; the load (expenses + profit) is ~35% of premium, so the pure premium is the remaining ~65% of the cat-load premium. Required cat-load premium = \$48,000 ÷ (1 − 0.35) = \$48,000 ÷ 0.65 ≈ \$73,800 (illustrative). Build-up: pure premium \$48,000 → divide by 0.65 to gross up for the 35% load → ≈ \$73,800 of premium must be allocated to the catastrophe component so that, after the load, the \$48,000 expected cat loss is fully funded. (All figures illustrative; the point is that the AAL is grossed up by the load exactly like any other pure premium, Ch. 11 — underprice this and the average storm is unfunded.)

Exercise 23 (the financial module and your terms)

The financial module converts physical damage into the net insured loss your company retains, by applying the policy structure (Ch. 12) and reinsurance (Ch. 27). A 5% named-windstorm deductible on a \$20M building absorbs the first \$1M of wind damage before the policy pays anything, and a wind sublimit caps the exposure — so both lower the account's modeled net loss, not just its premium. Because the zone PML is built from modeled net losses, tighter terms directly reduce the account's zone consumption. The terms you set are therefore an accumulation lever, not only an incentive or a price lever.

Exercise 25 (the suppressed-rate chain)

Climate-conditioned model raises the indicated coastal rate → a prior-approval regulator suppresses it below the indicated level to protect affordability → at the suppressed price the insurer cannot earn its cost of capital (Ch. 28) on the exposure → the private market reduces writings / non-renews the most exposed risks → coverage becomes scarce and risk migrates to the residual market / FAIR Plan, which takes it on (often itself underpriced and undercapitalized for the tail) → the uninsured/under-insured share grows, widening the protection gap and shifting catastrophe risk onto the public balance sheet (§30.6, §30.7).

Exercise 27 (the CUO referral memo — model answer)

To: Chief Underwriting Officer. Re: Coastal manufacturing account, \$40M TIV, Port Hadley zone. I recommend we decline as presented, or write only with cat mitigants. The account is well-constructed and adequately priced on its own AAL, but it adds an estimated \$28M to the Port Hadley 1-in-100 zone PML, moving the zone from \$210M to ~\$238M against our \$250M limit — leaving only ~\$12M of headroom for the entire zone and effectively closing the Gulf's tightest zone to further growth. Given the limited headroom and the climate trend (our historical PML is a floor, not a center), I am not comfortable consuming this much zone capacity on one account. Recommended conditions if we proceed: (1) facultative cession of the marginal cat slice to keep the net zone contribution under ~\$15M; (2) a 5% named-windstorm deductible and a wind sublimit to shrink modeled net loss; (3) verified roof and structural mitigation credited in the vulnerability inputs; (4) a documented note that Port Hadley is now closed to new wind-exposed business pending roll-off. Without the facultative cession I recommend decline. (≈190 words.)

Exercise 29 (deferring to the model is not anti-science)

Using a model correctly means deferring to it where it is strong and reserving judgment where it is weak — and those are different parts of the same model output. The model's central tail estimate is where the machine genuinely beats human intuition (no one can hold tens of thousands of correlated events in their head), so you defer to it there. But the uncertainty around that estimate — which vendor, what calibration, how much model error, what the model is known to miss (newer construction, the non-stationary climate trend, §30.6) — is not "the science"; it is a known limitation of this model. Loading for that uncertainty is the ordinary discipline of using any estimate with error bars; treating a single model's number as a fact with no error is the actual departure from science.

Exercise 31 (Underwriting-File extension — the Harbor Steel cat screen)

(a) What it settled: Running Harbor Steel through the cat model confirmed two things — its AAL contribution shows the cat load already built into the indicated price is adequate (the account pays for its average hurricane), and its marginal PML contribution to the Port Hadley zone fits within the zone aggregate net of the cat XOL treaty (the 5% named-storm deductible and the cession keep the net contribution modest). The account survives the catastrophe screen. (b) What it did not settle: the things the model is known to miss — chiefly that the backward-looking model understates an intensifying climate, so the historical PML is a floor, not a center; it also did not re-open the price, the capital charge (Ch. 28), or the final bind. (c) The caveat for renewal: the Port Hadley zone fit is real but with limited headroom and a climate-trend caveat — both flagged for re-examination at the next renewal. (Does not pre-empt the Chapter 40 bind.)

Exercise 33 (what could erase the limited headroom)

Two examples, with pre-emptive actions. (1) A model upgrade or climate-conditioned recalibration raises the Port Hadley zone PML — better measurement of an intensifying peril pushes the zone over its limit even with no new accounts (§30.6). Action: treat the historical number as a floor now, hold a margin, and pre-plan which zone exposures would roll off if the PML rises. (2) New coastal accounts (ours or via the same broker) fill the remaining headroom, or a reinsurance change shrinks the net capacity — either consumes the slack (§30.5). Action: monitor zone consumption in real time, flag Port Hadley as near-limit so new submissions are referred, and confirm the cat XOL treaty terms at renewal so the net contribution assumption still holds.

Chapter 31

Worked solutions to the daggered (†) and odd-numbered exercises. Section references point back to the chapter. All figures are illustrative.

Exercise 31.1

Pre-fill (data enrichment) is the automatic population of a submission's fields from third-party data sources, so that an applicant or producer enters minimal information (often just a name and address) and the system fills in the rest. Its three failure modes: wrong match (the address resolves to the wrong parcel/structure, filling the file with confident facts about the wrong building); stale data (a real value that is out of date — an assessor record predating a renovation, an aerial image predating a new roof); and false precision (a value that looks measured — a square footage to the foot, a roof age to the year — but is estimated or derived, carrying hidden uncertainty). (§31.3)

Exercise 31.3

The underwriting workstation is the integrated software environment that pulls the submission, runs pre-fill, calls the data sources, displays the assembled risk picture, runs the scoring and pricing models, surfaces the guidelines and referral rules, and records the decision — all in real time, in one place. It differs from the fax-and-folder workflow in that the old workflow assembled the risk picture by hand from a dozen separate inputs over days, while the workstation assembles it automatically in seconds and presents it as a single screen. The work shifted from assembling the picture to reading and questioning it. (§31.4)

Exercise 31.4

The four families and a can/cannot for each: - Satellite/aerial imagery — can: roof condition, shape, footprint, yard hazards (pools, debris, overhang). Cannot: the interior/wiring, intent, or anything that changed after the image date. - IoT / telematics — can: actual behavior and condition over time (driving patterns, leaks, equipment condition). Cannot: behavior it doesn't observe, or the context of an anomaly. - Public records — can: building age, size, ownership, permits, liens, years in business. Cannot: current condition, or whether the record matches this structure vs. the parcel. - Third-party aggregators — can: a broad enriched feed assembled from many sources. Cannot: the provenance/certainty behind each field (you inherit the vendor's matching and currency). (§31.2)

Exercise 31.5

The six dimensions, each with a failure example: accuracy (the roof is really 10 years old but a mismatched record says 30); currency (the aerial image predates the new roof); completeness (a price-driving field silently left blank or class-defaulted); consistency (the assessor's square footage disagrees with the image's footprint); provenance (a vendor's derived estimate presented as a measured fact); relevance (a field pulled because the system could, not because it bears on the decision). (§31.7)

Exercise 31.7

The referral logic is the set of rules in a straight-through-processing system that decides which risks the machine keeps and binds and which it hands to a human underwriter. The chapter calls a good referral rule "an honest statement of where the data and the model stop being trustworthy" because each trigger to refer — missing or conflicting data, large limits, catastrophe exposure, thin/novel exposure, low model confidence, a guideline flag — marks a condition under which the automated picture or price can no longer be relied on. The referral grid is therefore a direct map of the limits of the data and the model. (§31.5)

Exercise 31.8

(a) The "poor / end-of-life" roof flag reasonably confirms an exterior roof in apparent poor condition, and corroborates an existing loss/inspection read if they agree. (b) The image cannot establish: the condition of the wiring/interior; whether the roof was already replaced after the image date; or the controls, management, and corrective actions behind any prior losses. (c) The single most valuable follow-up: a physical inspection (or insured documentation such as a roofing contract/invoice), which resolves the currency question — whether the flagged roof is the roof that exists today. (§31.2, §31.3)

Exercise 31.9

A near-perfect six-month telematics score is strong evidence because telematics observes actual driving behavior over time — mileage, braking, speed, time-of-day — rather than a proxy like credit or territory. It is not proof because it observes only what it measures: it says nothing about a trip taken in a borrowed (un-sensored) car, a recent change in circumstances, or behavior outside the observation window. In Chapter 6's terms, telematics observes the frequency-driving behavior well but can miss context and the severity tail (one bad un-observed trip). The work is in the word data — it is a rich observation, not the whole driver. (§31.2)

Exercise 31.10

A data feed is a claim, not a fact, so two aggregators disagreeing about the same building is a flag to chase, not a discrepancy to smooth over. The right move: identify which value is correct by going to a more authoritative or independent source (the assessor record, the aerial footprint, a physical measurement, or the insured), and resolve it on any field that drives the price. Averaging is the wrong instinct because it manufactures a third number that matches neither source and obscures the fact that one of your inputs is wrong — the disagreement is information (one source is unreliable for this risk), and averaging discards it. (§31.2, §31.7)

Exercise 31.11

Most important limitation by source: (a) imagery — exterior, point-in-time; it can be stale and cannot see inside; (b) public assessor records — staleness and parcel-vs-structure mismatch; (c) IoT/telematics — uneven coverage (not every risk is sensored) plus privacy, and it measures only what it senses; (d) aggregator feed — you inherit the vendor's matching and currency errors without seeing the underlying provenance. (§31.2)

Exercise 31.12

Pre-fill can make a submission more honest because it fills fields from independent third-party data rather than from the applicant's self-report. A producer can no longer "forget" a prior loss the clearinghouse already records, or shave years off a building's age, because those fields come from sources with no stake in the submission. This attacks adverse selection (Chapter 1) directly: adverse selection exploits the information gap between what the applicant knows and what the underwriter knows, and pre-fill narrows that gap by giving the underwriter independent knowledge the applicant cannot suppress. (§31.3; Chapter 1)

Exercise 31.13

Automation bias on a pre-filled field is trusting a value more because a machine supplied it than you would if a human had typed it. A false roof age slides through more easily when pre-filled because a hand-typed "roof age: 5 years" on an obviously old building carries an implicit "the applicant says" that invites skepticism, while the pre-filled value carries an implicit "the data shows" that feels authoritative — even when it came from a mismatched record and is less reliable than the human's guess. The costume of "verified fact" disarms the scrutiny the same wrong number would otherwise receive. (§31.3)

Exercise 31.15

On a commercial property risk, always verify price-driving pre-filled fields such as roof age/condition and construction type/occupancy (and prior-loss indicators), because an error there moves the price materially and feeds any automated rating. A reasonable low-stakes field to let stand on the pre-fill alone might be the entity's years in business or a non-rating descriptive field — useful context, but not a number the price turns on, so the cost of a small error is low. The principle: verify in proportion to how much the field moves the decision. (§31.3)

Exercise 31.16

The three habits: (1) read the file before you over-weight the score — or read it as if the score did not exist, then compare; (2) ask what the score could not see — relationship facts, in-flight corrective actions, broker intelligence, one-off context; (3) document the divergence — record why your judgment differs from the score. "Read the file as if the score did not exist, then compare" is more than a formality because the score arrives first, and arriving first colors every fact you read afterward; reading independently first prevents the score from anchoring your judgment, and turns the score into a check on your read rather than a substitute for it. (§31.4)

Exercise 31.17

• Ignoring the score throws away a genuinely powerful triage tool that has, fast and consistently, read more data than a human could — wasteful.
• Deferring to the score surrenders the selection-and-pricing judgment that is the underwriter's reason for existing, and treats a partial, data-only opinion as the final answer — abdication.
• Reading it as one voice in the room weighs the score for exactly what its inputs and track record earn: a confident, well-informed opinion that has never seen the broker's note or the in-flight fix, set against the things the underwriter can see. The chapter rejects the first two because each replaces judgment with a posture — blanket rejection or blanket deference — rather than weighing the evidence. (§31.4)

Exercise 31.19

Routing decisions (single deciding feature in brackets): - (a) personal-auto renewal, clean, no changes → BIND via STP [simple, standard, fully inside a well-modeled class]. - (b) \$20M coastal commercial property with a loss flag → REFER [large-limit catastrophe exposure — one wrong bind is existential]. - (c) small low-hazard office BOP, clean pre-fill → BIND via STP [simple, well-described, modest limits]. - (d) brand-new exposure class, no loss history → REFER [novel — no pool for the model to learn from; the law of large numbers fails for lack of data]. - (e) mid-size fleet with one conflicting MVR field → REFER [internally conflicting data the system can't resolve]. (§31.5, §31.6)

Exercise 31.21

As STP's frontier advances, the clean, simple risks increasingly never reach a human — they bind straight through — so the mix that lands on an underwriter's desk shifts toward the hard residue: data-poor, novel, conflicting, large-limit, low-confidence accounts. The chapter calls this a concentration, not a demotion because the human is now spending a larger share of time on exactly the judgment-heavy risks where they add the most value, and a smaller share rubber-stamping easy ones — the job gets harder per file, not more trivial. The implication for a new underwriter: build the judgment skills (reading context, loss-run stories, novel exposures, defending an override), because the easy-file volume that once filled a trainee's day is exactly what automation takes first. (§31.1, §31.5)

Exercise 31.23

Data-quality problems by dimension: the blank flood-zone code priced as "low hazard" is a completeness failure compounded by a silent default (missing field filled with a favorable guess); the two-year-old aerial image is a currency failure; the "0" prior-loss field contradicting the producer's phone mention of a fire is an accuracy/consistency failure. The most dangerous is the silent-defaulted flood zone, because it is invisible — the file looks complete and nothing flags, so a catastrophe-relevant field was guessed favorably and then priced and (here) bound. What should have stopped the bind: a rule that missingness is visible and never silently defaulted on a price-driving field, plus a confidence threshold that refers a risk with an unresolved hazard code or a source conflict to a human. (§31.7)

Exercise 31.25

"Outsourced the data but kept the risk" means a carrier that buys a data feed and wires it into automated pricing has handed off the gathering of the data but retains every consequence of its being wrong: the mispriced book, the losses, and the regulator are the carrier's, not the vendor's. So the data-driven carrier must run controls on the feed itself, not only on its decisions: audit the vendor's match rate and the currency of its data, monitor its default behavior (does it silently fill blanks?), require corroboration on price-driving fields, and treat a degradation in the feed as an underwriting problem — because a vendor's quietly worsening match rate can poison every downstream price at once. (§31.7)

Exercise 31.27

Model memo (illustrative, ~180 words).

To: Underwriting Manager Re: STP binding of [low-hazard contractor-services class] I recommend we stop straight-through-binding this class and route it to manual review, for two data-driven reasons. First, a referral-logic gap: these risks frequently carry a mobile/jobsite exposure our pre-fill does not capture, so the model is pricing a partial picture and binding it — a classic case of automating a risk the data describes poorly. Second, a data-quality concern: spot checks show the construction/occupancy field is being silently defaulted on roughly [illustrative] of these submissions when the geocode fails to resolve, meaning we are pricing some on a favorable guess. The combined-ratio risk is the familiar lag: our expense ratio on this class looks great, but if we are binding mispriced exposure, the loss ratio will surface it in two to three years and give back the savings. I propose: refer this class pending (a) a pre-fill field that captures the jobsite exposure and (b) a rule that makes missingness visible rather than defaulting it. Happy to walk through the sample. (§31.5, §31.6, §31.7)

Exercise 31.29

The mechanism: automation lowers the expense ratio immediately (fewer human hours per policy), and that shows up in year one with no lag. But if the machine bound risks badly — mispriced or mis-selected — the loss ratio rises, and loss development has a two-to-three-year tail, so it surfaces late. A carrier can therefore look successful (lower expenses) precisely while it is quietly underpricing, and the verdict arrives only when the losses develop. The single number against which the initiative must finally be judged is the combined ratio, which captures both the expense saving and the loss-ratio effect; "we automated and cut costs" is only good news if the combined ratio fell. (§31.6; Chapter 3)

Exercise 31.31

(a) The data corroborates rather than changes the disposition because the satellite roof flag, the loss runs, the inspection read, and the broker's note all agree — independent sources pointing the same way raise confidence in the existing risk picture rather than introducing a new fact that would move the decision. (b) The one hypothesis the quarter-old image cannot settle: whether the roof was already replaced after the image was taken — currency, resolved only by the inspection (and relevant because a warranted replacement is already a subjectivity). (c) The open thread handed to Chapter 32: the real-time model recommends decline (a 7), and the file does not — Chapter 32 opens the model and earns the documented override to a 6. (The Underwriting File)

Exercise 31.33

"The enriched feed tells you the breach happened; only judgment tells you whether the company actually fixed it" means that data sources can reliably flag a past event (Tindall Stores suffered a ransomware breach) but cannot evaluate the remediation — whether controls genuinely improved or the company merely bought a policy to check a box. The evidence that would move that judgment is qualitative and verified: a security assessment, evidence of specific control improvements (multi-factor authentication, backups, segmentation, incident response), and the underwriter's read of management's seriousness — gathered through underwriting questions, a security questionnaire, or a specialist review. A data feed cannot supply it because remediation quality is a judgment about people and controls over time, not a field that can be looked up — the same limit the chapter draws around every alternative data source. (The Underwriting File; Chapter 24)

Chapter 32

Worked solutions to the daggered (†) and odd-numbered exercises. Figures are illustrative teaching examples.

Exercise 1

A generalized linear model (GLM) is a statistical model that relates a set of predictor variables to an outcome through a link function and an assumed error distribution, estimating all the predictors' effects simultaneously. For insurance pricing it typically uses two component models: a frequency model (a Poisson distribution with a log link) predicting the number of claims per exposure, and a severity model (a gamma distribution with a log link) predicting the size of a claim given one occurred. Their product is the modeled pure premium.

Exercise 3

A gradient boosting machine (GBM) builds its prediction by combining many small decision trees, each one trained to correct the errors left by the trees before it; the ensemble of hundreds or thousands of these weak trees becomes highly accurate. Because trees ask questions in sequence ("if coastal and old roof and prior losses…"), a GBM discovers interactions and nonlinearities automatically — the very things a GLM must be told about by the modeler adding them by hand.

Exercise 4

Feature engineering is the work of constructing, transforming, and selecting the input variables (the features) a model learns from. It is the underwriter's natural seat at the modeling table because it is the translation of domain knowledge into model inputs: a data scientist knows how to fit the model, but an underwriter knows that "two fires in five years" is a frequency signal, that a thirty-year-old roof in a windstorm zone is an interaction, and that hot-work is the hazard a class code only hints at. Encoding those insights as features is what makes a model underwrite rather than merely correlate — and it is work the algorithm cannot do for itself.

Exercise 5

Lift measures how well a model separates good risks from bad: sort the book by predicted loss cost, cut it into deciles, and compare the actual loss experience of the best decile to the worst. The Gini coefficient compresses that whole separation into a single number from ~0 (sorts no better than random) toward 1 (near-perfect separation). What both fail to prove is price adequacy — they measure whether the model ranks risk correctly, not whether the overall price level is high enough. A model can sort risks perfectly and still be priced too low across the board.

Exercise 7

The three corners of the modeling triangle: - The data scientist owns the model build — the algorithm, the features (with the underwriter's help), and the validation. - The actuary owns the rate level and the filing — whether the price, in aggregate, is adequate and defensible to the regulator. - The underwriter owns the risk decision — whether to accept this risk, on what terms and at what price, informed by the model but not dictated by it; and the override and its documentation. The model is best when all three contribute; it fails when one corner dominates.

Exercise 9

(a) Multiply the relativities: $1.00 \times 1.85 \times 1.30 \times 1.20 \times 0.92 \approx 2.45$. The risk is modeled to have about 2.45 times the claim frequency of the reference driver. (b) The 1.85 relativity means: the effect of being a 22-year-old driver, for two drivers who are identical in every other modeled respect — same territory, same vehicle, same prior-coverage status — multiplies expected claim frequency by 1.85. The crucial phrase is "holding everything else constant": because the GLM estimated all factors simultaneously, the youth factor is not contaminated by the cars young drivers tend to own or the places they tend to live (which is exactly the flaw of the one-way method, §32.1).

Exercise 11

Predicting the log of expected loss makes the individual factor effects add up on the log scale. By the algebra of logarithms, adding on the log scale is the same as multiplying on the normal scale (log(a) + log(b) = log(a×b)). So when you exponentiate the fitted coefficients back to the normal scale, each factor becomes a multiplier — a relativity — and the model reproduces exactly the multiplicative structure of a classical rate manual: a base rate times a relativity for each factor. This is convenient for an underwriter because the output then looks like a rate table you already know how to read, and each relativity is an interpretable price signal you can sanity-check, argue, and defend.

Exercise 13

A GLM assumes the factors multiply — that each variable's effect is the same regardless of the others' values. When the effect of one variable genuinely depends on another (the danger of a sports car depends on whether the driver is 22 or 52), that is an interaction, and a plain GLM will not discover it on its own. The modeler must add the interaction by hand — specifying an interaction term so the model can estimate, e.g., a separate sports-car effect for young drivers. (This is exactly the weakness a GBM, §32.3, fixes automatically — at the cost of interpretability.)

Exercise 14

• (a) Setting the filed rate → the GLM (Gini 0.31). A filed rate must be explained and defended to the regulator, and a GLM's relativities can be justified factor by factor; the GBM's 0.06 of extra Gini is not worth a price you cannot explain or file. (§32.3)
• (b) Triaging which submissions a human reviews first → the GBM (Gini 0.37). Triage does not face the same explainability bar, so the GBM's superior ranking power is pure upside; let it sort the queue and flag the worst risks for human attention. (§32.3) The general rule: GLM where you must explain the price; GBM where you must rank the risk — and many carriers run both.

Exercise 15

Overfitting is building a model so flexible that it memorizes the noise in the historical data rather than the signal, so it performs beautifully on the past and badly on the future. The losses from an overfit pricing model resemble soft-market underpricing because in both cases the bad outcome is invisible at the moment of decision and arrives two or three years later: the model loved certain risks for reasons that turn out to have been accidents of the sample, accepted or under-priced them, and the losses surface on the usual delay — long after everyone congratulated themselves on the model's "98% historical accuracy."

Exercise 17 (find the red flag)

Three reasons the image model could be confidently wrong about "metal / excellent / high confidence": 1. Stale or mismatched imagery — the satellite tile may be months or years out of date, or stitched from multiple passes, and may not reflect the current roof (or even the right building). 2. Visual confusion — shadows, sun angle, or a reflective tarp/coating can be misread as a roof material; ponding or patches can be invisible from a particular angle. 3. Training-data bias — if the model was trained mostly on suburban single-family roofs, it may be unreliable on an industrial structure, and the score will not warn you of this. The posture toward any high-confidence image score: treat it as a strong lead requiring confirmation, never a finding of fact, and insist that high-consequence decisions (declines, large surcharges) rest on something a human can verify — here, the inspection and the documents (including the signed roof contract the image can never see).

Exercise 19

• (a) year built + today's date → building age (and age relative to typical roof/equipment life): captures wear-driven loss exposure the raw dates don't.
• (b) list of prior claim dates → claim frequency (count per exposure) and years since last claim: captures the frequency dimension of risk (Ch.6) and recency.
• (c) street address → distance to coast (cat exposure) and distance to fire station (the fire protection idea, Ch.9): turns a meaningless string into geographic risk signal.
• (d) welding-shop class code → a hot-work hazard flag and a products-liability exposure flag: surfaces the specific hazards a broad class code only hints at. Each feature encodes an underwriting pattern the model would otherwise miss.

Exercise 21 (ethics dilemma)

The argument to keep the ZIP-proxy variable out: the variable is largely a proxy for a protected class (§32.5; Ch.35), so pricing on it would reproduce — and possibly amplify — discrimination the law forbids (Ch.4's unfair discrimination), even though the model never names the protected class. A measurable Gini gain does not change this, because accuracy and fairness are different questions and a proxy's predictive power often traces to historical bias in the data (redlining, under-investment), not to legitimate risk. "The algorithm chose it for predictive power" is not a defense (§32.1 Compliance Corner). For a correlated variable to be defensible, you must be able to show its predictive power has a legitimate, risk-related source independent of the protected characteristic — and, increasingly, pass a documented disparate-impact test before deployment.

Exercise 22

"Garbage in, garbage out" is the classic warning that a model fed low-quality or erroneous data produces worthless output. "Bias in, bias out" is the sharper modern version: a model trained on data that encodes historical discrimination will reproduce that discrimination, no matter how well-built or accurate the model is — because the bias is not a data-quality error the model can detect; the model predicts the biased data correctly. Feature selection determines fairness more than algorithm choice because the bias enters through which variables the model is allowed to see: a proxy admitted as a feature will carry the bias regardless of whether the algorithm is a GLM or a GBM, and a proxy excluded cannot.

Exercise 23

(a) Yes, this model has good lift. The tell is the monotone climb from the best decile (~45% loss ratio) to the worst (~190%): the model predicted which risks would be worst, and they were worst, by a wide and steadily increasing margin. A model with no lift would show every decile near the same loss ratio (a flat chart). (b) No — this chart does not prove the prices are adequate. Lift proves the model ranks risk correctly, not that the overall price level is right. Notice the deciles span from 45% to 190% and several exceed 100%: the model sorts beautifully, but whether the book as a whole is priced to a profitable combined ratio is a separate question, answered by checking the overall rate against your loss ratios and target (Ch.11 rate adequacy), not by the shape of the lift chart.

Exercise 25 (price this risk — model triage)

The lift chart says the model genuinely sorts risk, so act on the ranking: - Best three deciles (~55% loss ratio, below your ~62% target): these are profitable to your target — compete hard for them: accept, fast-track, and where the market allows, price keenly to win and retain them before a better-modeled competitor does (the Case Study 1 dynamic). - Worst two deciles (~170%): these are deeply unprofitable — decline, surcharge heavily, or impose terms that change the risk; do not write them at standard price. What the chart does not tell you: whether your overall rate is high enough. The whole book could be mispriced upward or downward by a constant and the lift chart would look identical. Price-level adequacy is a separate check against your loss ratios and target combined ratio (Ch.3, Ch.11).

Exercise 26

The four questions to ask of any model, and the failure each catches: 1. Out-of-sample? — catches overfitting (a model that flatters the data it was trained on and fails on new risks). 2. Does the lift hold across segments? — catches a model with great overall lift but flat or reverse lift in an important sub-book (a state, industry, size band) where it will quietly lose money. 3. Is it stable? — catches a brittle model whose lift swings when refit on a different year (signal vs. accident). 4. What's the price level? — catches the gap between ranking and adequacy: a model can sort risk perfectly and still be priced too low overall (Ch.11).

Exercise 27

There is no universal "good" Gini because the achievable separation depends on the line, the data, and the incumbent benchmark. Some lines have inherently more predictable, separable risk (lots of stable signal in the data) and support high Ginis; others are dominated by genuinely random loss and cap out lower, so a Gini that is mediocre in the first line could be excellent in the second. The Gini is therefore most useful as a relative measure — "is model B better than the model A it replaces?" — not as an absolute pass/fail threshold.

Exercise 28 (underwrite this submission)

Override the decline-leaning 8. Two of the three §32.7 justifications apply. First, the model is missing a material fact: it scored the risk before the broker supplied the post-loss sprinkler upgrade and the new safety-management contract — corrective controls that exist but never reached the model's inputs, so the model priced the risk as it was, not as it will be under the conditions I attach. Second, the model is out of its domain: the risk is in an industry the model has very few historical examples of, so its score is an extrapolation made with false confidence. I would write the risk with the corrective controls as subjectivities, at terms reflecting the improved-but-watched profile — and document that the override rests on named facts the model lacked, not on a preference. (§32.7)

Exercise 29

The three situations that justify an override: (1) the model is missing a material fact; (2) the model is out of its domain (novel risk, thin data, extrapolating); (3) the model is demonstrably wrong on this case (a data error, a misread image, a miscoded class). What all three share: the underwriter has information or context the model did not have. What is not on the list: "I disagree" or "it feels wrong" — a bare preference with no nameable fact, which simply reintroduces the inconsistency and bias the model was built to remove.

Exercise 30 (write the memo — sample answer)

Override note — [Risk name], [date]. The risk-selection model scored this submission 8/10 (decline-leaning). I am overriding to a writable grade and recommending a quote with conditions. The model's score was computed on incomplete information: it did not have (1) the post-loss sprinkler upgrade completed [date] or (2) the signed safety-management services contract the broker supplied after the score ran — both corrective controls that materially change the forward risk. The model also has few historical examples in this industry, so its score is an extrapolation outside its reliable domain. These facts move the risk from a decline to an acceptable-with-conditions profile. Subjectivities: proof of the completed sprinkler upgrade, the executed safety contract, and a confirmatory inspection within 60 days. Revised grade: 6/10, quote-with-conditions. Override justifications: missing material fact; out-of-domain. Logged for model feedback (feature gap: post-loss corrective controls).

Exercise 31

A model "no one overrides is a model no one is watching" — and an unwatched model drifts, encodes yesterday's world, and accumulates errors that surface as losses later. Logged overrides feed the pricing-model lifecycle: when underwriters consistently override the model in one direction on a particular kind of risk, that pattern, captured in the override log, tells the actuaries and data scientists where the model is blind — which feature is missing, which segment it misjudges — and that intelligence feeds the next version (e.g., adding a "post-loss corrective controls" feature). So the underwriter who overrides well — rarely, for nameable reasons, with documentation — is not the model's adversary but its complement: the one role that supplies the context the model lacks and keeps the build-validate-monitor loop honest.

Exercise 33 (Underwriting-File extension)

(a) What the model could see (drove the 7): two fire losses in five years (frequency signal); the ~\$1.2M 2023 fire (severity signal); the thirty-year-old built-up roof, confirmed by the image model; the named-windstorm zone; fire protection class 4; the welding/hot-work occupancy class; the pending products-liability claim. (b) What the model could not see (justifies the 6): the signed roof-replacement contract; the new hot-work permit program being implemented; the management context behind the loss history (the 2021 fire electrical, the 2023 fire hot-work, both predating corrective action); the fact that the risk is being priced as it will be under the attached subjectivities, not as it was. (c) Override justification relied on: #1, "the model is missing a material fact." The model's reading of the history is not wrong; it is uninformed about the corrective controls that exist but never reached its inputs. Adding those facts changes the grade from a decline-leaning 7 to a writable-with-conditions 6 — a modest, defensible adjustment a reasonable reviewer would accept.

Exercise 35

A disciplined override is a modest adjustment grounded in named facts because the model's score is a strong, well-validated prior — the distilled experience of a million files — and the only legitimate reason to move it is specific information the model lacked. The Harbor Steel facts (a signed roof contract, a hot-work program, the management story) justify moving a 7 to a 6: they meaningfully improve the forward risk under conditions, but they do not erase the aging sprinklers, the cat tail, or the pending claim. An override that swung the grade dramatically — 7 to a 2 — on the same evidence would itself be a red flag, because the evidence does not support that magnitude; such a swing would signal that the underwriter is substituting preference for analysis, which is exactly the caprice the documentation discipline exists to prevent.

Chapter 33

Worked solutions to the daggered (†) and odd-numbered exercises. All figures and scenarios are illustrative teaching examples. Throughout, the chapter's bright line holds: a red flag is a prompt to investigate, never a finding of fraud.

Exercise 1

Insurance fraud is the deliberate deception of an insurer for financial gain — knowingly making a false or misleading statement, or concealing a material fact, in connection with an application or a claim, to obtain coverage, a lower premium, or a payment one is not entitled to. The three elements are deliberate (separating fraud from honest error — the applicant who genuinely forgot a small claim is not a fraudster), material (separating the lie that matters from the one that doesn't), and financial gain (the motive that distinguishes fraud from an embarrassed omission). All three must be present: an honest mistake lacks intent; a trivial lie lacks materiality; a deception with no payoff lacks the motive that makes it fraud.

Exercise 3

A material misrepresentation is a false statement of fact, made by an applicant to obtain insurance, that is material and on which the insurer relied. Its three elements: (1) a false statement of fact — not an opinion or expectation, but an untrue assertion about something knowable and stated; (2) materiality — the falsehood would have affected the underwriting decision; (3) reliance — the insurer used the false statement in deciding. The classic test for materiality is the prudent-underwriter test: would a reasonable, prudent underwriter, knowing the truth, have declined the risk, charged more, or written it on different terms? If the truth would not have changed the decision, the misstatement is not material, however dishonest.

Exercise 5

Rescission is the legal undoing of an insurance contract from its inception — the premium is returned, coverage is treated as void from day one, and any claim is denied — on the ground that the policy was procured by a material misrepresentation or concealment. It differs from non-renewal (which ends coverage forward, at the term's end, leaving the past coverage intact) and from a coverage denial (which disputes one claim under a policy that remains valid). Rescission is the heaviest of the three: it unwinds the entire contract backward, saying the policy was poisoned at the root by the lie that procured it.

Exercise 7

The special investigation unit (SIU) is the specialized team within (or hired by) an insurer that investigates suspected fraud — gathering evidence, conducting interviews and surveillance where warranted, coordinating with industry databases (the NICB) and law enforcement, and supporting the legal actions a fraud finding can lead to. The division of labor: the underwriter is the front line of detection — the first professional to read the submission and notice the flag that starts everything — and spots and refers; the SIU is the specialist of investigation, with tools, authority, and legal backing the underwriter lacks, and investigates and substantiates; legal and claims then act on what the SIU finds. The underwriter's job ends at a clean, documented referral.

Exercise 9

Placements on the fraud spectrum: (a) forgetting a \$700 claim from five years ago — honest error, no intent, almost certainly immaterial. (b) rounding payroll from \$11.3M to "about \$11M" — gray zone / borderline soft, usually sloppiness rather than fraud, and small in materiality; verify rather than accuse. (c) adding a never-owned laptop to a real burglary list — soft fraud, because it starts from a genuine loss (the burglary) and pads it. (d) a ring staging collisions across dozens of policies — hard, organized fraud at the far right, premeditated and criminal. The gradient across these four is intent growing from none to premeditated and materiality growing with it.

Exercise 10

Soft fraud is harder to detect because it hides inside a legitimate transaction — the claim or application is genuinely real and merely stretched, so there is no manufactured event to expose, no structure to trace. A back injury exaggerated by three weeks leaves no fingerprint. Hard fraud, by contrast, must build something (a staged crash, a phantom employee), and that structure leaves traces (links, anomalies). Much of soft fraud's cost is therefore not caught but absorbed into the loss ratios that set the class rate — it pads severities a little across thousands of claims. The consequence is uncomfortable: the honest insureds in the class are effectively pre-paying for the soft fraud the class commits, because the "adequate" class rate is built on fraud-inflated losses. The lever the underwriter controls is terms (deductibles, documentation) that make padding less worthwhile and easier to expose.

Exercise 11

Seeing fraud as a binary "yes/no light" produces a calibration failure. Example: a new underwriter, having learned that fraud is bad and detection is virtuous, reads every rounded payroll figure, forgotten claim, and sympathetic story as a lie — and either accuses honest applicants (souring relationships, exposing the carrier to bad-faith and defamation claims) or, exhausted by false alarms, swings the other way and rubber-stamps everything. The cost runs both directions: treating an honest error as fraud defames a good customer and burns a broker; treating soft fraud as honest error lets it through. The disciplined posture holds two ideas at once — fraud is real and must be caught, and the overwhelming majority of applicants are honest — and reserves the legally weighty word "fraud" for cases where intent and materiality actually warrant it.

Exercise 12

Materiality judgments and actions: (a) misstating the year of construction by one year — immaterial; a one-year difference does not change the risk or the decision. Action: clear it (likely a transcription error), no further step. (b) not disclosing a \$300,000 fire two years ago — plainly material; a recent fire of that size would change the terms or lead to decline. Action: document, ask directly and in writing, refer to SIU, and do not bind until the cause and the disclosure are resolved. (c) listing the operation as "light assembly" when it includes daily welding — material; the misdescription hides the hot-work hazard that drives the rate and the controls. Action: investigate the operation, re-rate to the true class, and treat the misdescription's intent as the open question (a misunderstanding of "light assembly" vs. a deliberate dodge of a hazard class).

Exercise 13

"I expect to hire ten more employees next year," which turns out false, is generally not a misrepresentation because it is a statement of expectation, not a statement of fact: it describes a future intention that was honestly held when made, and a misrepresentation must be about a fact — something knowable and stated — that is untrue. A statement on the same application that would be a misrepresentation: "We currently employ 40 people" when the true figure is 95, or "We have had no fire losses in the past five years" when there was a \$300,000 fire two years ago. The difference is present, knowable fact vs. future expectation.

Exercise 14

A vague or ambiguous application question can defeat a later misrepresentation claim because the innocent-misrepresentation doctrine (§33.4) protects an applicant who answered reasonably given how the question was phrased. If a question is genuinely ambiguous, or invites a checkbox where the truth needs a paragraph, an applicant's "wrong" answer may have been the reasonable reading — and courts will not let an insurer that drafted sloppily then cry misrepresentation. Utmost good faith runs both ways: the insurer owed truthful answers is the insurer that asked clear, specific, material questions. The implication for drafting is direct: write questions that are unambiguous, specific, material, and answerable, because the cleanest rescission case is built on a precise question the applicant answered with a precise lie — and the sloppiest question is the one that produces an unrescindable gap.

Exercise 15

Establishing that an application answer was false tells you nothing about whether it was fraud because the falsehood alone does not establish intent — that the applicant knew the statement was false and intended to deceive. A loss-run order proves the discrepancy (the answer was false), but a false answer can arise from honest oversight, a misread question, a broker's transcription error, or a genuine dispute over whose claim it was. The only ways to establish intent are investigation (what the SIU does) and the applicant's own account (which is why you ask a neutral, documented question first). Data finds the gap; only investigation and the applicant's explanation fill in the state of mind that turns a clarifiable gap into a rescindable fraud.

Exercise 16

Four limits on rescission, each able to defeat it: (1) Incontestability — after a statutory period (notably ~2 years in life insurance), the insurer generally cannot contest the policy even for misrepresentation, so a late-discovered misstatement is barred. (2) Innocent misrepresentation — a misstatement made in good faith, especially to an ambiguous question, may not support rescission even if material. (3) Waiver/estoppel — if the insurer knew or should have known of the misrepresentation and issued or renewed anyway, or accepted premium after learning the truth, it may have waived the right to rescind. (4) Varying state standards / burden of proof — some states require intent to deceive or that the misrepresentation contributed to the loss, and in all states the insurer must prove its case in court; a rescission it cannot prove fails. (A fifth, the practical one: rescission is litigation-prone, and a failed rescission adds bad-faith exposure to the original loss.)

Exercise 17

Post-claim underwriting is doing at claim time the verification the insurer should have done at underwriting time, and using a trivial or immaterial discrepancy discovered only after a loss as a pretext to rescind and avoid a valid claim. Courts and regulators disfavor it because it lets an insurer collect premium, skip the underwriting, and then escape coverage only when a loss actually arrives — a heads-I-win-tails-you-lose proposition that guts the policy's value. The disciplined alternative is to do the verification up front — order the loss runs, run the databases, inspect the risk before binding — so that the few rescissions pursued rest on genuinely material, genuinely fraudulent misrepresentations and can survive litigation. The tools come from Chapter 8 (information gathering: loss runs, CLUE, MVRs, inspections, financials).

Exercise 18

Recommendation: do not rescind. Walk the analysis. Materiality: an 8% overstatement of square footage, unrelated to the fire, is not material — if anything it would have led the insurer to charge more, not to decline, so the truth would not have changed the decision in the insurer's favor. Post-claim- underwriting risk: the only reason the file is being combed is the size of the claim; the discrepancy was not a genuine fraud signal at underwriting, which is the textbook signature of post-claim underwriting that courts and regulators condemn. Litigation/bad-faith exposure: rescinding a covered, accidental fire on an immaterial size error converts a payable claim into a bad-faith lawsuit, exposing the carrier to consequential and possibly punitive damages well beyond the original claim. The recommendation is to pay the covered claim, note the square-footage correction for the record and future rating, and reserve rescission for material, knowing misrepresentations supported by up-front verification.

Exercise 19

A rescission that cannot be proven is "worse than no rescission" because it does not merely fail to avoid the claim — it creates a new, larger exposure. When the insurer rescinds and denies the claim, the insured sues; if the insurer then cannot prove the misrepresentation, its materiality, and (where required) the intent, it loses the rescission and faces a bad-faith judgment for the wrongful denial of a covered loss, which in many states carries consequential and punitive damages that dwarf the original claim. So a weak rescission turns a covered claim the insurer might simply have paid into a multiple of that claim plus regulatory scrutiny — the asymmetry that makes thin, pretextual rescissions one of the most expensive unforced errors in insurance.

Exercise 20

Red flags in the Crescent Auto submission, by family: Timing/urgency — wants to bind tomorrow. Mismatch/over-insurance — building values well above the modest leased-shop operation. History gaps — "none" on loss history with loss runs "to follow." Identity/opacity — entity 4 months old, two addresses in 18 months, and a request to hold off the inspection until after binding. Financial distress — business "slow" and behind on the lease. Innocent explanations (each plausible): a real deal on a deadline; a conservative owner or lender requirement; honest price-shopping with runs genuinely pending; a new/mobile business and scheduling friction; ordinary hard times. But the overall picture is a REFERRAL, not a question, because there is a cluster across five different families converging on one submission — and the combination of over-insurance + financial distress + urgency + identity opacity + declined inspection is exactly the shape that warrants specialist attention. Action: do not bind, order the loss runs and an inspection as conditions precedent, and refer to the SIU with a documented file.

Exercise 21

Change one fact to flip the picture: remove the over-insurance (make the requested building values consistent with the modest leased-shop operation). With values that match the operation, the remaining items collapse into an ordinary new account: a new business (hence the recent formation and address changes), price- shopping on a deadline (hence the urgency and pending runs), having a hard month (the lease comment), and a scheduling conflict on the inspection. None of those, without the over-insurance, suggests a motive to manufacture a loss — and the single routine question ("please provide the loss runs and let's schedule the inspection before we can bind") resolves it. The over-insurance was the keystone of the cluster because it supplies the motive (a property worth more burned than run) that turns ordinary new-business friction into a fraud concern; remove the motive and the cluster is just a new business in a hurry.

Exercise 22

The neutral, documented question: "Thank you for the submission. Our loss-history search returned a claim of approximately \$25,000 that is not reflected in the 'no prior losses' answer on the application. Could you help us reconcile this — confirm whether this loss relates to this insured/location and provide any detail? We'll need this resolved, along with the outstanding loss runs, before we can finalize terms." It states the specific fact, gives the applicant a fair chance to explain, asks in writing, and ties the answer to the binding decision — without accusing. Why a second false explanation is more damning: the first omission is consistent with honest oversight (the explanation column of §33.5), so by itself it supports only a question. But if the applicant, confronted with the specific claim, gives a new false account of it (denies it exists, mischaracterizes it), that second falsehood — made with the fact squarely in front of them — is strong evidence of knowledge and intent, the very element (§33.3) that a single omission lacks. The documented exchange is how a clarifiable gap reveals itself as fraud.

Exercise 23

The innocent-explanation column is a genuine part of the method, not mere courtesy, because it encodes the base rate: for every red-flag family, the innocent explanation is usually the true one (most distressed businesses never commit fraud; most urgency is a real deadline; most rounded numbers are sloppiness). An underwriter who ignores it commits a base-rate error — treating a weak, common indicator as if it were strong evidence of a rare event. Statistically, because honest applicants vastly outnumber fraudsters, most accounts showing any single flag are honest, so acting on one flag as if it were proof generates far more false accusations than true catches. The column keeps detection calibrated: it forces the underwriter to ask "what's the innocent explanation, and is this one flag or a cluster?" before escalating — which is exactly the discipline that separates vigilance from paranoia.

Exercise 24

This is a clarification, not an SIU referral — and the reasoned defense earns full credit. The Harbor Steel application shaded the cause of the 2023 fire (hot-work, described as general/accidental). It is material (cause bears on hazard and controls), so it warrants a response. But: (1) the fire's existence and size were honestly disclosed — only the cause was shaded; (2) there is no cluster — no over-insurance, urgency, distress, identity opacity, or shifting story; (3) a strong broker (Meridian) has voluntarily attached corrective controls; and (4) there are innocent explanations at least as likely as deception (a coarse application "cause" field; the broker's summarization; an honest characterization of a welding fire as "accidental"). Per the calibrated-referral discipline of §33.6, lighting up the SIU for a single coarse-form cause description on an otherwise honest, controls-attached submission would flood the unit and bury real fraud. The right action: a neutral, documented written question to Meridian ("the loss detail indicates the 2023 fire involved hot-work/welding; please confirm the cause so we can finalize the hot-work-permit subjectivity"), documented in the file. (If instead drafting the referral note: it would state the specific fact, attach the application and loss runs, note the explanation received, record that you did not accuse/bind/deny, and flag any time pressure — but here the facts do not warrant it.)

Exercise 25

"Most referrals come back cleared" is a feature because a referral is a question for specialists, not a verdict — and a healthy fraud program clears the innocent as diligently as it pursues the guilty, which is how it protects honest customers and avoids the bad-faith exposure that wrongful fraud findings create. For the underwriter reluctant to refer "because I don't want to accuse anyone," the correction is that referring is not accusing: handing a documented concern to the people equipped to investigate is the responsible act, and declining to refer a genuine cluster — to spare an accusation you were never making — lets real fraud through and leaves the honest pool to pay for it. (The opposite error, over-referring, is also real: it floods the unit and buries the serious cases. The skill is the calibrated referral.)

Exercise 27

Three legal duties/constraints on fraud investigation, with the disciplined practice for each: (1) Unfair- claims-practices timelines — you may not deny or delay a legitimate claim because an investigation is convenient; an investigation used as a pretext for delay is bad faith. Practice: investigate promptly, within statutory timelines, and only on a genuine basis. (2) Defamation exposure — stating as fact that someone committed fraud, before it is established, is a real legal exposure. Practice: use the operative words "red flag" and "referral," and reserve "fraud" for what is substantiated. (3) FCRA / privacy limits — the consumer data you pull to investigate is regulated. Practice: pull only permitted data, for a permissible purpose, and honor adverse-action obligations. (A fourth: good-faith reporting to a state fraud bureau is typically immunized when done in good faith — so report honestly, never recklessly or maliciously.)

Exercise 28

Match technique to strength and characteristic false positive: Anomaly detection — best at finding the unusual case (over-insured, inconsistent, statistically strange); characteristic false positive: the unusual-but-honest (a novel operation, a size outlier) flagged as suspect, since unusual ≠ fraudulent. Link analysis — best at finding hidden connections across claims (shared people, clinics, shops, phones), lethal to rings; characteristic false positive: a coincidental link (neighbors share a body shop; one attorney legitimately represents many claimants). Predictive fraud models — best at scoring fraud likelihood to triage the flood toward the SIU; characteristic false positive/limit: a score, not a finding, that encodes past bias and produces a probability requiring human substantiation. The common thread: every technique yields a lead, never a verdict — analytics decide what to look at; people decide what is true.

Exercise 29

Using the Chapter 32 framework: a fraud score is a strong, well-validated prior — a distilled opinion from many cases — but it produces a likelihood, not a fact, and the consequences of acting on it (a denial, a rescission, a referral to law enforcement) are far heavier than a price, so the bar for action is higher than the bar for attention. A high score therefore decides where to look, routing scarce investigative attention to the highest-odds cases; it cannot, alone, deny a claim in good faith or rescind a policy, because those require a person to establish the underlying facts with defensible evidence. The fraud-side analogue of Chapter 32's documented override is the substantiated SIU finding: just as the underwriter must name the specific fact the model lacked to override a pricing model, the SIU must establish the specific evidence of intent and materiality to convert a fraud score into an adverse action — and a low score is not a guarantee of honesty, only a reason this case wasn't chosen for scarce attention.

Exercise 30

Response: The colleague is missing that a model trained on past investigations learns who got investigated, not who committed fraud — and if historical investigations skewed toward certain ZIP codes (through past human bias, over-policing of some communities, or correlated socioeconomic factors), the model will reproduce that skew under a veneer of math, flagging those claimants more not because they offend more but because they were looked at more. "The model just found where the fraud is" mistakes the training signal (investigations) for the target (fraud). This is proxy discrimination / algorithmic bias, which Chapter 35 owns and treats in full. Before trusting the model to triage, the carrier should: test for disparate impact across protected and proxy groups, examine whether ZIP code is acting as a proxy for a protected class, validate against substantiated fraud (not mere investigation) where possible, and ensure a person establishes the facts before any adverse action — because an algorithm may decide where to look, never who is guilty.

Exercise 31

A ring is vulnerable to link analysis specifically because its strategy depends on each policy looking innocent in isolation — and to operate at scale a ring must reuse people, clinics, body shops, attorneys, phones, addresses, and bank accounts. Every reuse is a connection, and link analysis is precisely the tool that refuses to view the claims in isolation: run across the whole book (and across carriers via shared databases), the recurring elements that are invisible to any single adjuster light up as a cluster. The very thing that makes a ring profitable (industrialized repetition) is the thing that exposes it. An honest coincidence: people who live in the same neighborhood genuinely use the same body shop, and a single personal-injury attorney legitimately represents many unrelated claimants — so a shared shop or attorney is a lead, and the SIU must still establish that the connection is collusion, not coincidence, before it becomes a finding.

Exercise 32

Sample decision memo (for the file):

Submission: [account]. During pre-bind review, the ordered loss runs revealed a prior loss (a ~\$40K water-damage claim, 18 months ago) not listed on the application's loss-history answer. I sent a neutral, written inquiry to the broker on [date] asking the applicant to reconcile the discrepancy. The applicant responded on [date] that the claim was at a prior leased location they no longer occupy and that they misread the question as applying only to the current premises. Assessment: the omission is material (it shifts the loss picture and bears on the rate), so it required a response; but the applicant's explanation is plausible and consistent with how the question was phrased, there is no cluster of other red flags, and there is no evidence of intent to deceive. Disposition: I am treating this as a documented clarification, not a fraud referral and not a rescission matter, and I have re-rated to reflect the disclosed loss. I did not refer to SIU because a single, explained, material discrepancy with no cluster and no indication of intent does not meet our referral threshold, and over-referring degrades the unit's ability to catch real fraud. The discrepancy and the explanation are documented above.

This hits every required element: the flag, the action, the explanation, the materiality/intent assessment, the disposition, and the reasoned decision not to refer or rescind.

Exercise 33

Charging both classes the same "adequate" rate is a problem because Class A's rate is built on fraud- inflated losses: its reported loss ratio looks like Class B's, but the losses underneath it are padded by soft fraud, so the "adequate" rate is adequate only in the sense that it covers the fraud too. Who overpays: the honest insureds in Class A, whose premiums absorb the soft fraud their class commits — they are subsidizing the padders. (Class B's honest insureds pay a clean rate; Class A's honest insureds pay a fraud-loaded one.) Terms, not just price, to push back: higher or per-claim deductibles (which make small padding less worthwhile and give the insured skin in the game — the moral-hazard logic of Chapter 1), documentation and proof-of-loss requirements (which make exaggeration harder and easier to expose), coinsurance and scheduled limits on the items most often padded, and targeted inspection/verification at claim time for the patterns soft fraud follows. The pricing fix alone just spreads the fraud cost more "accurately"; the terms fix attacks the fraud itself.

Exercise 34

Decision process for the over-insured, non-renewed, runs-pending, 48-hour-bind account: First, do not bind on the timeline — the urgency is itself a red flag, and binding before verification is exactly how fraud and bad risk get in. Require before binding (conditions precedent): the 5-year loss runs, an inspection, financials sufficient to test the over-insurance, and an explanation of the prior carrier's non-renewal. Red-flag families present: timing/urgency (48 hours), over-insurance (values exceed the operation), history gaps (runs not provided), and possibly distress (why is the prior carrier leaving?). What moves it from "writable with verification" to "decline or refer": if the loss runs reveal concealed material losses (not just unprovided ones), if the values cannot be justified by the assets (motive for a manufactured loss), if the non-renewal was for fraud or serious loss history, or if a cluster hardens (over-insurance + distress + concealment + urgency) — then refer to SIU and do not bind. Avoiding both errors: rubber-stamping is avoided by making verification a condition precedent (no runs, no bind); false accusation is avoided by asking neutral, documented questions and letting the verification — not a hunch — drive the decision. Most such accounts, once the runs and inspection arrive clean, are simply new business in a hurry; the discipline is to verify before committing coverage, not to presume guilt.

Exercise 35

Counterfactual: the application concealed the 2023 fire's existence entirely. This changes the file fundamentally. On the fraud spectrum (§33.2): concealing a \$1.2M fire's existence moves from a left-of- center disclosure gap to a serious matter well right of the gray zone — a material concealment, and if knowing, application fraud. Is it now a referral? Yes. Unlike the real file (one shaded cause, existence disclosed, no cluster, controls attached), a concealed major loss is precisely the kind of material omission that warrants a documented SIU referral, because the question of intent (did the applicant knowingly hide a \$1.2M fire?) now genuinely matters and exceeds what a clarifying question resolves. Is rescission on the table (§33.4)? Potentially yes — a knowing concealment of a material prior fire is the archetype of a rescindable misrepresentation (cf. Case Study 2, Scenario A), though the standard still depends on materiality (clearly met), intent (to be established), the policy language, and the state. What you do before binding: do not bind; refer to SIU with the loss runs, the application, and a documented note; ask the applicant directly and in writing to explain the omission; and let the investigation establish intent before any coverage attaches. What changed from the real file: the real file had the fire's existence and size honestly disclosed with only the cause shaded, no cluster, and voluntary controls — making it a clarification; concealing the existence removes the honest disclosure, supplies a strong intent question, and converts the matter into a referral with a live rescission analysis.

Chapter 34

Worked solutions to the daggered (†) and odd-numbered exercises. (Even, non-daggered items are discussion or decision prompts whose reasoning is developed in the chapter text.)

Exercise 1 (define InsurTech; the four player types)

InsurTech is the wave of technology-driven companies — and the technologies themselves — that aim to change how insurance is distributed, underwritten, priced, serviced, and paid. The four types, sorted by what they disrupt in the value chain: (1) full-stack carriers (be the insurer; own the paper, the capital, and the loss); (2) digital MGAs/MGUs (underwrite on a backing carrier's paper under delegated authority for a fee/commission; distribution + underwriting, but not the risk); (3) enablers/SaaS (sell software — intake, claims, data, pricing — to incumbents; the incumbent keeps the risk); (4) distributors/embedded platforms (the front door — a digital agency, comparison site, or embedded API, with a carrier or MGA behind them).

Exercise 3 (full-stack carrier vs. digital MGA; why it matters most)

A full-stack InsurTech carrier holds its own license and policyholder surplus; the losses land on its balance sheet and it lives or dies by its combined ratio. A digital MGA underwrites on a backing carrier's paper under delegated authority, collects a fee/commission, and passes the loss to the carrier and its reinsurers. The distinction is the chapter's most important because who eats the loss ratio predicts almost everything: the MGA is capital-light and runs a distribution business with software margins, while the full-stack carrier must hold capital against catastrophe and survive its own loss ratio through a hard market. That is why the MGAs and embedded platforms aged better than the full-stack carriers.

Exercise 5 (peer-to-peer insurance; where the residual risk goes)

Peer-to-peer insurance is a model in which a group of policyholders pools premiums to cover each other's claims, often with a fixed fee or a "giveback" of the unused pool. The residual risk — the chance claims exceed the pool — does not vanish; it is ceded to or backed by a traditional carrier or reinsurer. The P2P framing is largely a distribution and brand layer on top of conventional risk-bearing.

Exercise 7 (basis risk; cutting both ways)

Basis risk is the gap between a parametric policy's payout (which is tied to a measured trigger) and the policyholder's actual loss. It cuts both ways for the insurer: the objective trigger removes most fraud and dispute about the amount of loss and slashes claims-handling expense (no adjuster), but it also removes the adjuster's role in confirming that an insurable loss occurred at all — so the insurer can pay out when little or no real loss happened (and the insured can suffer a real loss and collect nothing when the trigger just misses).

Exercise 8 (the MGA incentive misalignment and the defenses)

The MGA earns its money on volume — fees and commissions scale with premium written — while the carrier bears the losses. That is a textbook misalignment of incentive, a structural moral hazard (Chapter 1) sitting at the heart of the relationship. Three defenses a disciplined backing carrier uses: (1) tight delegated-authority limits with a clear appetite; (2) a referral grid that forces anything outside appetite to a human at the carrier; (3) a hard audit of the MGA's underwriting (Chapter 38); and (4) a contract clause to claw back the pen the moment the developed loss ratio drifts (any three). The carrier should judge the relationship on developed, ultimate loss ratios, not on this year's premium growth.

Exercise 9 (capital efficiency: genius and danger)

The same fact — that the MGA does not put up the capital and does not own the loss in the existential way a carrier does — is both the genius and the danger. It is the genius because it lets a startup be a capital-light technology business: it needs only enough money to build software and survive to breakeven on commissions, not hundreds of millions in policyholder surplus. It is the danger to the backing carrier because the MGA, paid on volume while the carrier eats the losses, has every incentive to write more business than is prudent; a wide pen granted with a thin audit is letting someone else write business on the carrier's balance sheet and hoping they do it well. The capital efficiency that attracts founders is exactly what requires the carrier's tightest supervision.

Exercise 10 (embedded as class underwriting; the new failure mode)

Embedded cover usually involves no individual selection: every qualifying customer gets the same offer, the same terms, the same price. So the underwriting becomes class underwriting (Chapter 20) — you are pricing the class of "everyone who buys this product through this platform," not the individual — negotiated once between carrier and platform. The new way it can go wrong: because the decision is made once and applied automatically to thousands, a single error (misjudging the embedded population's risk, or handing the platform terms that invite adverse selection) becomes the same bad decision made automatically, thousands of times, before anyone reads the loss runs.

Exercise 11 (year-one premium growth is not evidence of a good book)

Two reasons (from the chapter): (1) Growth attracts adverse selection (Chapter 1) — a book that grows fast by saying yes cheaply disproportionately attracts the eager buyers, who skew toward the worse risks; the volume itself can be a warning. (2) The losses lag the premium — the MGA's revenue arrives now with the premium, while the claims emerge and develop over the next several years, so a fast-growing delegated book looks benign for two or three years and only reveals its true loss ratio as the claims mature and the growth slows. Year-one premium growth measures appetite, not quality.

Exercise 12 (parametric basis risk: two scenarios)

(Figures illustrative.) Policy pays \$250,000 if recorded sustained wind at the location crosses 110 mph. (a) Hurts the insured: a storm brings sustained winds of 108 mph and tears off part of the plant's roof — a real, large loss — but because the trigger (110 mph) was not crossed, the policy pays nothing. The insured has a genuine loss and no recovery: basis risk against the insured. (b) Hurts the insurer: a brief gust pushes the recorded reading just past 110 mph while causing little or no actual damage; the policy pays the full \$250,000 against a trivial loss. The objectivity of the trigger that protects the insurer from amount-fraud also obliges it to pay when there is no real loss: basis risk against the insurer.

Exercise 13 (parametric and insurable interest)

A parametric policy pays on a measured parameter regardless of the insured's actual loss, which makes it look uncomfortably like a wager on the weather (or the earthquake, or the flight delay). The doctrine of insurable interest (Chapter 4) — the requirement that the insured stand to suffer a genuine financial loss from the insured event — is what keeps insurance from being gambling, so a parametric product must be designed so that the trigger and payout track a real exposure the insured actually has. If it is not, the structure risks being characterized as a wager (gambling) rather than insurance, which is both legally void as insurance and a regulatory problem.

Exercise 15 (parametric: model's job vs. judgment's job)

The model's job (well-handled by the catastrophe models of Chapter 30): estimating the probability the trigger is crossed — that is exactly what an exceedance-probability curve produces — and therefore the pure premium for the parametric payout. The underwriter's judgment (which the model cannot supply): the design of the product — where to set the trigger, how to structure the payout so it tracks real loss closely enough to be useful and to satisfy insurable interest, and which measuring station to trust. The model tells you the odds the trigger fires; it cannot tell you whether a trigger that fires when the insured has no loss is a product you should sell.

Exercise 16 (low expense ratio, still an underwriting loss)

Combined ratio = loss ratio + expense ratio (Chapter 3). A full-stack InsurTech can drive its expense ratio well below average with automation and direct distribution — but for most lines the loss ratio is by far the larger term, and software does nothing to make a risk less likely to have a claim. If the company's selection and pricing are loose (often because it is growing fast and attracting adverse selection), the loss ratio runs high enough that loss + expense still exceeds 100%, i.e., an underwriting loss, no matter how low the expense ratio. What would have saved it: underwriting — better risk selection and adequate pricing (Chapters 7–13) — because that is the only thing that moves the loss ratio.

Exercise 17 (the master lesson; "our customers love us")

The master lesson, in one sentence: growth is easy in insurance and underwriting is hard, and the loss ratio cannot be outrun. "Our customers love us" is not a rebuttal because customer love is not a loss-ratio signal: several InsurTechs were simultaneously beloved by customers, admired by the press, richly valued, and losing money on nearly every policy. An insurer must pay less in claims and expenses than it collects in premium over time to survive; affection does not change that arithmetic. The combined ratio is the only scoreboard.

Exercise 19 (the pivot is the market teaching the lesson)

Several InsurTechs that began as full-stack carriers ended up as digital MGAs, enablers, or heavy reinsurance buyers. The chapter reads this not as failure but as the market correctly relocating the insurance risk to balance sheets built to hold it — and the technology to the parts of the chain (distribution, intake, claims, data) where it adds durable value. From an underwriting standpoint this is the right move because the question was never "can technology replace the carrier?" but "where does technology create value, and where does it merely move the loss ratio around?" Relocating the risk to a reinsured, capital-strong carrier while keeping the software is answering that question correctly.

Exercise 21 (find the red flag: the delegated MGA dashboard)

Three things that should worry underwriting management: (1) 80% premium growth — a book growing that fast is presumptively attracting adverse selection and may be winning on price; growth is not quality. (2) The current-year reported loss ratio looks "fine" — but it is immature; the claims on a fast-growing book have not developed, so a low current-year figure is exactly what a mispriced book looks like in year one. (3) Strong commission revenue with no audit since the pen was granted — the carrier is booking fee income while flying blind on the actual underwriting, with the MGA's volume incentive unchecked. What to demand instead of the current-year loss ratio: the developed / ultimate loss ratio (loss-development-adjusted), ideally by accident year/cohort, plus an immediate underwriting audit of a sample of bound risks.

Exercise 23 (memo: partner with a digital MGA?)

A model answer takes a clear position and structures it around the three questions. Example (≈170 words): "Recommendation: proceed to a pilot with [MGA], but on tightly supervised terms, not an open pen. The strategic case is sound — they reach a small-commercial class we cannot acquire cost-effectively, and their platform speeds intake and binds the clean risks straight through. But the economics that make this attractive to them (capital-light, paid on volume) put the loss ratio on our balance sheet, so the deal must be built around our three questions. Whose loss ratio: ours — therefore tight delegated-authority limits, a referral grid for anything outside appetite, and a hard quarterly audit. Growth vs. profitability: we judge them on developed, ultimate loss ratios by cohort, never on premium growth, with a contractual repricing/claw-back trigger if the developed loss ratio breaches plan. What is automated: intake and the binding of simple risks — fine; but selection on anything unusual must refer to our underwriters. Net: a real opportunity, written with the discipline that the InsurTech stumbles proved is non-negotiable."

Exercise 25 (usage-based: true and a fairness question)

It is genuinely true that usage-based auto insurance is "fairer" in one sense: it attacks adverse selection by replacing a proxy for driving risk (age, zip code) with a measurement of how the person actually drives, which is harder to game and lets a safe-but-risky-looking driver prove it — pricing follows real risk (theme 4). It also raises a fairness question deferred to Chapter 35: continuous measurement is a data relationship with privacy and consent implications, the scoring can embed or correlate with protected characteristics, and "price optimization" and proxy effects can creep in. Both are true at once because actuarial fairness (price reflects measured risk) and social fairness (access, privacy, non-discrimination) are different axes — the very tension Chapter 35 exists to weigh.

Exercise 27 (Underwriting-File extension: could an MGA platform bind Harbor Steel?)

Drop the submission into a digital-MGA quote-in-seconds flow and trace it. Where the intake helps: the address pre-fills the cat zone and pulls satellite roof imagery (Chapter 31); the rating engine pulls the class code — the intake genuinely is faster. Where the referral rule should trip: every feature that makes this account this account — two fire losses in five years including a \$1.2M hot-work fire, an original 1994 roof and original sprinklers at end of life, a named-windstorm exposure the prior carrier non-renewed over, a pending products-liability claim, and a multi-line program (property/GL/WC/auto/\$10M umbrella). A sound referral rule (§34.4) should catch any one of these and stop the auto-bind, kicking the account to a human. Correct outcome: the platform may quote/intake faster, but it should refuse to bind automatically and refer to an underwriter. Harbor Steel belongs on the exception path of even a good small-commercial flow — it is the risk the referral rule exists to catch — not on the happy path of quote-and-bind, which is for the small clean machine shop down the road (Chapter 20).

Exercise 29 (why a non-renewed, multi-line, cat-exposed, loss-active account stays human)

The InsurTech machine handles simple, high-volume, well-understood, homogeneous risks — the renters policy, the phone plan, the clean small BOP — where the underwriting can be encoded once and run at scale consistently. Harbor Steel is the paradigm of the opposite: it is complex (multi-line), novel in its combination (cat-exposed coastal fabrication with a specific loss story), low-frequency/high-severity on its worst exposures, and relationship-dependent (it arrived through a human broker because the prior carrier non-renewed it). Every one of those features is exactly where data and models run out and judgment is irreplaceable — reading the loss runs for the management story (Chapter 9), structuring terms to make it writable (Chapter 12), pricing a risk no model has clean training data for, and defending the decision. It is a textbook case of the work that stays human in an InsurTech world (theme 5).

Chapter 35

Worked solutions to the daggered (†) and odd-numbered exercises. (Discussion-only items whose answers are developed at length in the chapter text are summarized briefly.)

Exercise 1 (fair vs. unfair discrimination)

Fair discrimination distinguishes among risks on the basis of their expected loss (charging a reckless driver more than a careful one); insurance cannot function without it because a flat price across unequal risks triggers the adverse-selection death spiral (Chapter 1) and destroys the pool. Unfair discrimination distinguishes on a prohibited basis — a protected class, or a factor unmoored from cost — and is forbidden by every state's unfair-trade-practices law. The first is the mechanism of insurance; the second is its betrayal.

Exercise 3 (disparate impact, defined)

Disparate impact is a discriminatory effect on a protected group produced by a facially neutral practice, regardless of intent. It differs from intentional discrimination in that no one need mean to discriminate; the practice's outcome, measured by comparing groups, is what makes it disparate. A model with no protected variable and no discriminatory intent can still produce a disparate impact.

Exercise 4 (the four tests)

(1) Protected-class test — is the factor a prohibited characteristic (race, religion, national origin; variably sex, etc.)? Categorical: no actuarial exception. (2) Actuarial-justification test — does the permitted factor have a genuine relationship to expected loss? (3) Disparate-impact test — does the factor fall much more heavily on a protected group in effect? (4) Causation-vs-correlation test — is the factor closer to a cause of loss or merely a correlate of a protected trait? The weaker the causal story, the more suspect the factor.

Exercise 5 (redlining)

Redlining is the historical practice of denying or pricing up insurance and lending for entire neighborhoods based on racial/ethnic composition. The name comes from the 1930s federal Home Owners' Loan Corporation "residential security" maps, which colored predominantly Black and immigrant neighborhoods red as "hazardous." Its legacy makes geography contested because territory (a legitimate factor — coastal ZIPs really do have more hurricane exposure) is, in a segregated society, also a proxy (§35.3) for the populations those historical lines enclosed.

Exercise 8 (the GINA gap)

The Genetic Information Nondiscrimination Act (GINA) prohibits the use of genetic information in health insurance and in employment. It does not reach life, disability, or long-term-care insurance — the "GINA gap." So a life insurer may, in most states, ask about and use genetic-test results, meaning a person who responsibly got tested could be rated up or declined on a predisposition they did not choose and cannot change. The statute is real; the gap is real; the ethics are contested.

Exercise 9 (the flat-price advocate)

A market that charged one flat price regardless of risk would adverse-select: good risks overpay and leave, bad risks get a bargain and pile in, losses exceed the price, the price rises, the next-best risks leave — the pool spirals and coverage becomes unavailable or unaffordable for everyone, including those the policy meant to help. The strongest surviving point on the advocate's side: a "predictive" factor that merely encodes a protected group's accumulated disadvantage may be punishing history rather than measuring risk, and society may legitimately decide some accuracy is worth sacrificing to refuse that — which is the social-fairness claim of §35.7, not a reason to abolish risk pricing wholesale.

Exercise 10 (classify the factors)

(a) At-fault accident in last three years — (i) clearly permissible: direct causal loss relationship. (b) Religion — (ii) clearly prohibited: a categorical protected class, no actuarial exception. (c) Credit-based insurance score — (iii) contested proxy: predictive and facially neutral but with a disparate impact correlated with race/income; permitted in most states, restricted in some. (d) Fire-protection class — (i) clearly permissible: direct causal relationship to fire loss (Chapter 9). (e) ZIP code in a segregated metro — (iii) contested proxy: legitimate where it captures a real peril, but suspect where it carries racial information without a causal loss story (§35.3, §35.5).

Exercise 11 (why deleting race doesn't work)

A capable model reconstructs a hidden variable from its correlates: remove race, and a gradient boosting machine (Chapter 32) will learn race from the interaction of ZIP code, name, shopping behavior, and dozens of other features — because race carries predictive signal and the model is built to extract every drop of signal. "Colorblindness" therefore disables your ability to see the proxy effect, not the model's ability to produce it. The fix is to measure protected-group impact (the §35.4 audit), not to hide the variable.

Exercise 13 (three mechanisms of algorithmic bias)

(1) Biased training data — the model faithfully learns a discriminatory pattern already in the history (harsher claim adjusting, historical steering) and treats it as ground truth. (2) Proxy variables — the model reconstructs a prohibited characteristic from its correlates even when the characteristic itself is absent. (3) Feedback loops — the model's own pricing decisions distort the data it later retrains on, compounding a small bias into a large one over successive retrainings.

Exercise 15 (why metric conflict is a values problem)

Demographic parity (equal average prices/approval across groups), equalized odds (equal error rates), and calibration (a score means the same expected loss for everyone) are mutually incompatible when base loss rates genuinely differ — this is a mathematical theorem, not an engineering shortfall. Because you cannot have all three, you must choose which to honor, and that choice is not derivable from the data: it is a judgment about what "fair" should mean. For the governance committee this means there is no "objectively fair" model to build — only a model whose fairness trade-off has been chosen deliberately and documented, rather than chosen by accident and hidden.

Exercise 17 (underwrite the "prior insurance lapses" factor)

Run the four tests. Protected-class: "number of prior lapses" is not itself a protected class — passes. Actuarial justification: it is predictive of loss — passes. Disparate impact: it correlates with income, and income correlates with race in your markets — strains/fails; this is the danger zone. Causation-vs-correlation: weak — lapses may proxy for financial instability rather than cause claims, so the causal story is thin. Required analysis before adoption: a formal disparate-impact audit across protected groups; an attempt to establish a causal (not merely correlational) loss mechanism; a test of whether the lift survives once you control for legitimate risk; consideration of whether a narrower, more-causal variable captures the same signal without the proxy effect. Provisional recommendation: do not adopt on predictive power alone; escalate to model governance; if the lift turns out to be largely a relabeling of income/race with no causal driving story, the factor is unfairly discriminatory by proxy and should be rejected. (Theme 1: judgment; theme 6: social function.)

Exercise 19 (find the red flag — the "willingness to pay" vendor)

The red flag is "personalizes every premium to exactly what each customer is willing to pay." The practice is price optimization (§35.7). It violates the §35.2 principle that price differences between insureds must reflect cost (risk) differences — here the price reflects the customer's price sensitivity, not their expected loss. Two customers with identical risk would be charged differently based on who will tolerate more. Widely deemed unfairly discriminatory and banned/restricted in personal lines in many states. Decline the pitch.

Exercise 21 (find the red flag — genetics in accelerated life UW)

The red flag is that an input is derived partly from genetic-test data in a life-underwriting program. The statute implicated is GINA. The line: GINA prohibits genetic-information use in health insurance and employment, but not in life, disability, or long-term-care insurance — so the use may be legal for a life product in most states, yet it raises the §35.6/§35.7 ethical problem (the GINA gap) of rating people on predispositions they did not choose. Flag for legal review and for the ethics/fairness governance discussion; legality is the floor, not the answer.

Exercise 22 (calibrated yet disparate)

A model can be calibrated — a 7 means the same expected loss for every group — and still decline Group A at twice the rate of Group B if Group A simply has more high-scoring (genuinely higher-risk) members. Calibration is about whether a score means the same thing across groups; the decline rate is a demographic-parity question about average outcomes. When base loss rates differ, these two notions diverge — exactly the §35.4 incompatibility. It is not a bug to be "fixed": you cannot make the model both calibrated and demographically equal when the underlying risk differs. What it demands is a documented values decision about which fairness to honor and what, if anything, to do about the disparate decline rate.

Exercise 23 (sketch the disparate-impact audit)

Group the book by the protected attribute (used only as an audit variable — never as a pricing input; that distinction is the whole legal point). Compare outcomes across groups: average score/price, decline rate (share scored above the action threshold), and actual loss per unit of score (the calibration check). The one inviolable rule: the protected attribute audits outcomes but never enters the model's pricing features. Read the result like an underwriter — if decline rates diverge sharply but loss-per-score is roughly equal, the model is calibrated yet disparate, and the decision passes to governance.

Exercise 25 (ethics dilemma — is calibration enough?)

Strongest actuarial position: the model is calibrated, so a given score means the same risk for everyone; charging by accurate expected cost is the definition of actuarial fairness, and forcing equal average prices across groups with different real risk would make the good risks in the lower-risk group subsidize the bad risks in the higher-risk group — itself unfair, and an invitation to adverse selection. Strongest social position: accurate pricing of a disadvantaged group can still deepen inequality and price the vulnerable out of essential coverage; if the higher loss rate itself reflects accumulated disadvantage (worse roads, older housing, historical disinvestment), "accurate" pricing punishes circumstance, not choice. Where to come down: calibration is necessary but not sufficient for fairness — it answers the actuarial question and is silent on the social one. A defensible stance: keep calibrated pricing as the baseline, run the disparate-impact audit, and where the disparity is large and traces to circumstance rather than risk-relevant behavior, escalate to the social-fairness mechanisms (§35.7) that exist precisely because pricing alone cannot resolve the tension. Don't pretend "it's calibrated" ends the conversation.

Exercise 27 (the two traps)

The technocrat's trap: "I just price the risk; fairness is the regulator's problem, not mine" — false, because pricing decisions are where actuarial and social fairness collide, and refusing to see the social dimension does not make the harm disappear. The advocate's trap: "all risk-based pricing is unjust; insurance should just be fair" — false, because pricing that ignores risk destroys the pool through adverse selection and leaves everyone worse off. Both are failures of the same skill: the capacity to hold two real, conflicting values at once without collapsing into either pole. The disciplined underwriter prices by risk and examines for proxy/disparate effects and supports the social mechanisms that handle what pricing cannot.

Exercise 29 (Harbor Steel: actuarial vs. social)

The social-fairness problem is the protection gap (Chapter 30): the same coastal catastrophe exposure that makes Harbor Steel's rate defensibly high is pricing out or non-renewing homes and small businesses across Port Hadley — exactly as Harbor Steel's own prior carrier did. No single underwriter created it (it is the aggregate of many individually actuarially-fair decisions) and none can solve it at the desk; the remedies (residual markets, public programs, rate policy) sit above the desk (§35.7). What belongs in the file: a note that the Harbor Steel price is actuarially fair and risk-based, and an honest acknowledgment that it sits inside a protection-gap question that is real but not the underwriter's to resolve in this quote. What belongs above the desk: the social-fairness policy response. Writing the social answer into the rate filing would be the §35.7 technocrat's-inverse error.

Chapter 36

Worked solutions to the daggered (†) and odd-numbered exercises. Figures and scenarios are illustrative teaching examples; the forward look on Harbor Steel does not pre-empt the capstone's binding decision (Chapter 40).

Exercise 1

Continuous underwriting is the practice of monitoring an insured risk in real time, with live data (sensors, satellites, telematics), throughout the policy period, so the risk picture is updated continuously rather than captured once at the point of sale. The assessment it produces differs from the traditional annual snapshot in timing and form: the snapshot model assesses the risk once at bind, then looks away for twelve months until renewal — a photograph of the risk frozen at a point in time. Continuous underwriting makes the assessment a video: the risk is observed across its whole life, so deterioration that used to hide between bind and renewal (and surface only in the claim file) is now visible as it happens.

Exercise 3

Insurability is the condition of a risk being capable of being insured — possessing the Chapter 1 characteristics well enough (a pool of similar exposures, definite and fortuitous loss, a calculable chance of loss, losses not catastrophic to the insurer, an economically feasible premium) — at a price someone will both charge and pay. A risk can become uninsurable while remaining perfectly modelable because the limit of insurability is a price limit, not a measurement limit: we may be able to compute the risk-adequate premium precisely, yet that price may exceed what customers can afford or what the regulator will approve. The clause that does the work is "at a price someone will both charge and pay." When the adequate price outruns the payable/permitted price, insurers face a bad choice — write at an inadequate rate and bleed, or stop writing — and the risk becomes uninsurable in practice even though the math never failed.

Exercise 5

The three frontier products and the indemnity weakness each addresses: - Parametric — addresses indemnity's slowness (and adjustment cost): it pays on a measured trigger, so money arrives in days, not months, with no claim to adjust. - Embedded — addresses indemnity's distribution friction: it sells coverage inside another transaction at the moment of need, at near-zero acquisition cost, reaching the customer exactly when the risk is salient. - On-demand — addresses indemnity's all-or-nothing duration: it lets the insured switch coverage on and off for the period actually needed, pricing usage in real time rather than over a fixed annual term.

Exercise 7

Four of the six 2035 skills, with the direction AI pushes their value: - Judgment under uncertainty — more valuable (the human's worth concentrates in the cases the model can't handle). - Model literacy (read/question/override) — more valuable (you must use the model without being captured by it). - Climate and catastrophe fluency — more valuable (the moving baseline makes forward-looking judgment harder and scarcer). - Communication, negotiation, and trust — more valuable as a share of the job (as the routine automates, the human-to-human work the machine can't do becomes a larger fraction of what's left). (The remaining two — data judgment and ethical reasoning — are likewise made more valuable, not less.)

Exercise 8

Continuous underwriting generally cannot be used to raise a policy's rate mid-term because a bound policy is a fixed-term promise (Chapter 5): the insurer agreed to provide the coverage for the policy period at the agreed price, and rate regulation (Chapter 4) governs when and how a rate can change — you cannot simply re-price in month four because a sensor flagged a problem. So the live data is most valuable for things other than mid-term re-pricing: (1) loss prevention — intervening with loss control now, while the deterioration is still cheap to fix, so the loss never happens; (2) informing the next term's price at renewal; and (3) deciding whether to renew or non-renew a deteriorating risk rather than discovering its decline in a claim. The greatest of these is loss prevention — the best claim is the one that never happens.

Exercise 9

Continuous underwriting "collapses the gap between underwriting and loss control" because the same live signal serves both functions that used to belong to two separate departments. In the snapshot world, underwriting (Chapter 7) decides whether to write the risk and loss control (Chapter 9) visits once to recommend improvements. In the continuous world they fuse. Single signal doing both jobs: a sensor reporting that the sprinkler riser's water pressure has dropped. As underwriting, it says the risk just got worse (the property is now less protected than when you priced it). As loss control, it says someone should fix the sprinkler system today, before a fire finds the gap. One data point; two jobs.

Exercise 11

The two flaws in "we can just monitor it continuously and react": 1. Monitoring is not mitigation. A sensor that tells you the building is burning does not put the fire out. Many of the worst losses — a catastrophe, a sudden structural failure, a fraud — give no useful warning window even with perfect instrumentation, so the ability to watch the risk fail in real time does not prevent the failure. 2. The data stops when the insured unplugs the box — and the insured most likely to unplug it is exactly the one whose risk just got worse. This is adverse selection (Chapter 1) in a new costume: the worst risks self-select out of the monitoring precisely when the monitoring would matter most. So continuous underwriting sharpens a good decision; it cannot rescue a bad one. Write the risk only if you would write it anyway, and let the monitoring make it better.

Exercise 13

Preferred reformulation: "Underwriters who use AI will replace underwriters who don't." The kind of underwriter being automated away is the mechanical one — whose value was looking up the rate, re-keying the application, and stamping the file. That was never the craft; the machine does it faster, cheaper, and more consistently, and it should. The kind being amplified is the judgment underwriter — whose value is reading the risk, structuring the deal, overriding the model with documented reasons (Chapter 32), and defending the decision. AI hands that underwriter an assistant that does the drudgery so the judgment has room to work. This is the book's fifth theme exactly: technology augments underwriters; it does not replace them — for the simple risks the algorithm writes; for the complex ones, human judgment is irreplaceable; and the future belongs to those who can do both.

Exercise 15

A trend is the slow shift in the underlying loss distribution itself (the climate baseline rising); an event is a single hurricane season drawn from that distribution. You should not spike the coastal rate after one bad season and cut it after two quiet ones because that is reacting to events — adding a weather amplifier to the ordinary underwriting cycle (Chapter 3), which whipsaws the rate up and down around the truth and tends to underprice exactly when capacity is short. The disciplined alternative is to hold the rate to the forward-looking expected loss — the climate-adjusted average annual loss (AAL, Chapter 30) — through both the loud years and the quiet ones, letting only the trend (a genuine shift in the distribution) move the rate. That is rate adequacy (Chapter 11) applied to a moving target: the hardest version of the book's fourth theme.

Exercise 17

The trap: a clean five-year coastal loss run looks like a good risk by every traditional measure (the loss run is the history that predicts the future, Chapter 8). But if the underlying frequency of the destructive storm is rising (a non-stationary baseline, §36.3) and the five quiet years were simply luck, the clean record describes a baseline that no longer exists. The history is not merely noisy; it is biased downward — it systematically understates tomorrow's risk. This is the climate version of the Chapter 1 catastrophe error: pricing for the average year while the tail gets fatter. In Chapter 1 the error was ignoring correlation (a thousand homes in one flood zone is one bet, not a pool); here it is ignoring drift (the distribution itself is moving). The disciplined move is to price off the climate-conditioned model (Chapter 30), treating a clean coastal loss run as necessary but nowhere near sufficient.

Exercise 18

(a) The risk is measurable — the catastrophe model can simulate the named-storm losses and compute the climate-conditioned AAL; this is not a measurement failure. (b) What has failed is the meeting of the adequate price and the payable/approved price: the risk-adequate premium now sits far above both what the owner can afford and what the filed rate allows — an availability and affordability failure (§36.4). (c) The company-level decision: non-renew or restrict new business in the zone unless the rate can reach adequacy or the catastrophe exposure can be ceded/shared (reinsurance, Chapter 27). Writing it at the inadequate allowed rate would violate rate adequacy (Chapter 11) and the combined ratio (Chapter 3). (d) The system-level mechanism meant to catch such a risk is the public-private response: residual markets and FAIR plans (the insurer of last resort) and public catastrophe programs such as the NFIP for flood — the machinery that exists precisely to provide basic coverage the standard market won't write. Note: a good, careful, loss-free insured can become uninsurable purely because the priced risk of their location outran the price the system will bear.

Exercise 19

The single lever that lowers the adequate price (not merely shifts who pays it) is mitigation / resilience — physically changing the risk: hardening the roof, creating wildfire defensible space, installing flood barriers, improving the building to code. The credit/debit logic of Chapter 11 (schedule rating) rewards it directly: verified mitigation earns credits that reduce the indicated premium, because it genuinely reduces the expected loss. This is the only durable answer because every other move — rate suppression by regulation, a subsidy, pushing the risk into a residual market — only changes who pays the same (or a hidden, growing) cost; it suppresses the payable price while leaving the adequate price untouched, which accumulates losses someone eventually bears. Mitigation alone lowers the true risk, and therefore the true price.

Exercise 21

The largest threat to an on-demand drone policy's loss ratio (Chapter 3) is adverse selection (Chapter 1): hobbyists will tend to switch the coverage on precisely when they perceive the risk to be highest — flying in marginal weather, over a crowd, in an unfamiliar area — and leave it off for the easy flights. It is the oldest enemy in the book in a new app: the people who most expect a loss buy the coverage for exactly the moments they expect it. Two design rules that push back: (1) minimum activation units / waiting periods — require coverage to be bought for a block (a full day, a full month) or with a short delay before it takes effect, so it can't be switched on the instant before a risky flight; and (2) usage- and behavior-based pricing with telematics (the drone's own flight data, §36.1) — price each activation by the actual conditions and history, so the high-risk moments cost more and the selection is priced rather than absorbed. Either way, the anti-selection judgment is built into the product rules up front.

Exercise 23

For an API-bound embedded shipping-protection product with no human in the loop at the point of sale, the three design decisions that now carry the underwriting judgment a desk used to carry are: 1. The class definition / eligibility rules — which shipments, values, destinations, and item categories the product will automatically accept, and which it excludes. This is the risk-selection decision, made once for all policies (the straight-through-processing logic of Chapter 20, industrialized). 2. The rating algorithm — how price is set from the available data (value, route, carrier, item type), since no human will re-rate any individual policy. 3. The guardrails / circuit-breakers — the limits, fraud checks, and anti-selection rules (caps, exclusions, anomaly flags) that protect the book when the algorithm meets a case its designers didn't anticipate. The judgment is relocated upstream into design; it is not removed.

Exercise 24

Three red flags a senior underwriter would raise about a fully-automated, sub-minute coastal-homeowners product: - Climate / insurability (§36.4): coastal homeowners is the line where insurability is failing first under climate stress. Binding new coastal business fast does nothing about the core problem — whether the rate can reach adequacy against a rising, non-stationary catastrophe baseline and a hardening reinsurance market. Speed is not the constraint; adequacy is. A product that races to write the very risks the market is fleeing should be presumed to be writing them at an inadequate rate until proven otherwise. - The AI (§36.2): an LLM that reads the application and imagery and binds in under a minute is making an unverified decision a human never checks — and LLMs hallucinate (state confident, plausible falsehoods). Who verifies the coverage position and owns the decision to a regulator? "The AI did it" is no defense. - The continuous-monitoring claim (§36.1): "continuous sensor monitoring after bind" is doing reassurance work it can't support — monitoring is not mitigation, the data stops when the insured unplugs the box (adverse selection), and a sensor can't prevent the catastrophe that is the dominant coastal peril anyway. The monitoring cannot rescue a risk that shouldn't have been written at that rate.

Exercise 25

It is not genuinely forward-looking because its only "forward" input is the carrier's own past losses in the zone, trended at a flat historical rate — i.e., it assumes the future distribution is the past distribution scaled by a constant. That is precisely the assumption climate change breaks: a non-stationary baseline (§36.3) means the shape and level of the peril distribution are shifting, not just scaling at the old rate. What's missing is a climate-conditioned catastrophe-model view (Chapter 30) — an explicit, physically-grounded forward projection of how named-storm/wildfire/flood frequency and severity are changing for that location — rather than an extrapolation of a record that systematically understates tomorrow's risk.

Exercise 26

(Model memo, ~200 words.) "Team, I recommend we adopt an LLM co-pilot to summarize our large commercial submissions. Benefit: it can read a hundred-page submission and a five-year loss run and produce a draft risk summary, a first-cut assessment, and a list of broker questions in seconds — giving underwriters back the hours that judgment actually needs, and improving consistency on intake. Guardrail 1 (verification): the co-pilot drafts; the underwriter verifies and owns. Every summary is checked against the source before it informs a decision, because LLMs can be confidently wrong (hallucination) — they will quietly drop the one loss-run claim that mattered or state a coverage position that sounds right and isn't. Guardrail 2 (accountability): the underwriter signs the quote, not the model; 'the system said so' is not a defense to a broker, a regulator, or our committee. Compliance risk to manage (Chapter 35): an LLM can encode and explain away bias — producing principled-sounding rationales while a protected-class proxy drives the result. We will not let the co-pilot touch the pricing/selection decision without the fairness testing and documentation the NAIC guidance and laws like Colorado SB21-169 increasingly require. Used this way, the co-pilot amplifies our underwriters; it does not decide for them."

Exercise 27

(Spoken explanation to the broker, ~5 sentences.) "I know it's frustrating — they've never had a claim, and the number still moved, and that feels wrong. But the price isn't about their history; it's about the location's risk, and the catastrophe math on that coast has changed: warmer water and rising seas mean the storm we're pricing for is genuinely more likely and more severe than it was a decade ago, and the reinsurance behind the policy costs us more every renewal. A clean loss run tells me they've been lucky and careful — it doesn't tell me the storm isn't coming, and if I priced off the quiet years I'd just be hiding a cost that shows up later in a claim. So this is me pricing the risk honestly, not penalizing a good client. Here's what we can do together: let's get them mitigation credits for any hardening they'll do — that's the one lever that actually lowers the price rather than just moving it — and let's talk about a parametric supplement for fast cash after a storm." (Honest, risk-based, relationship-preserving — Chapter 39.)

Exercise 28

"It's just the cat model — it's actuarially fair, end of discussion" is the wrong posture for two reasons. First, actuarial fairness is necessary but not the whole question (Chapter 35): a risk-based price can be actuarially sound and still raise a genuine social-fairness problem when it makes whole communities — often those least able to move or absorb the loss — uninsurable. The book's sixth theme says the underwriter must hold both fairnesses honestly, not resolve the tension by declaring one side the only side. Second, "it's just the cat model" should not be taken on faith; what I would actually check is whether the higher prices are truly driven by location-based catastrophe risk or whether a proxy (Chapter 35) is doing hidden work — i.e., whether the geographic variable is standing in for a protected characteristic (redlining's legacy, Chapter 35). I'd want the model's drivers tested and documented for disparate impact (the fairness testing Chapter 35 and laws like Colorado SB21-169 call for). The genuine, unresolved tension: even if the price is purely and verifiably risk-based, a true price that prices a vulnerable community out of coverage is a real harm the math does not answer — that is the actuarial-vs-social-fairness collision at its sharpest, and it has no clean underwriting solution, only honest pricing plus support for mitigation and public backstops.

Exercise 29

(Both sides.) Access/convenience case for embedded insurance: selling coverage frictionlessly at the moment of need reaches people who would otherwise be uninsured, at near-zero acquisition cost, exactly when the risk is salient (the shipment, the trip, the rental) — genuine value delivered where traditional distribution never reaches. Suitability/value case against: coverage bought with almost no thought is coverage bought without comparing price or reading terms; the customer may be paying a poor price for a thin product they didn't need, with consent that is technically given but barely considered. The regulatory question now being asked: whether embedded products meet suitability and value standards — i.e., whether the frictionless sale is delivering real protection at a fair price, or exploiting the absence of deliberation at the point of sale. The honest answer is that embedded insurance is neither inherently good nor bad; it depends on the value of the specific product and the clarity of what's being sold.

Exercise 31

(Harbor Steel 2035 addendum, 3–4 sentences.) "2035 addendum (forward look — not a decision). Under continuous/IoT underwriting, the electrical panels and hot-work areas that caused the 2021 and 2023 fires would be sensored, so the deterioration behind those losses would likely be caught at the smoldering stage — the two fires that drove this whole file might never have happened. An AI co-pilot would pre-fill and pre-score the submission and draft the assessment and subjectivity list in seconds, but the judgment that defines the account — reading the fires as a story of management being fixed, weighing the prior carrier's non-renewal, and the model override of Chapter 32 — would still be the underwriter's to make and defend. Honestly, a climate-conditioned 2035 view raises the Port Hadley named-storm catastrophe load above today's figure, and the cat reinsurance behind the property line costs more, narrowing the margin between the adequate and payable price. This is a forward look only — it does not change the disposition; the account remains the quote-with-conditions the building chapters set, and the capstone (Chapter 40) states and defends the binding decision."

Exercise 33

A higher climate-conditioned Port Hadley named-storm AAL makes Harbor Steel's fate depend less on its own quality and more on whether the zone has room because of how catastrophe risk is managed at the portfolio level (Chapter 29), not the account level. Catastrophe exposure is governed by accumulation management (Chapter 30): the carrier limits its total loss to any one event by peril zone, and judges each new account on its marginal contribution to the zone's PML rather than on its standalone merit. If the zone's modeled loss has risen and the cat aggregate for the Port Hadley zone is near its limit, then even a well-run, well-priced, well-mitigated Harbor Steel can be turned away — not because the account is bad, but because the portfolio cannot absorb one more dollar of correlated coastal exposure (the failure of "independent" from Chapter 1, managed at scale). The account's own quality is necessary; the zone's headroom is the binding constraint.

Chapter 37

Worked solutions to the daggered (†) and odd-numbered exercises. (Items not reproduced here are discussion or planning prompts whose answers are developed in the chapter text or are individual to the reader.)

Exercise 1 (the trainee program)

The trainee program is a structured first job in which a carrier hires a person with little or no insurance experience and, over one to two years, develops them into a working underwriter trusted with a small grant of authority. Its first year is for building judgment that does not yet exist — above all a mental model of the whole value chain — not for proving knowledge the trainee already has.

Exercise 3 (the two axes)

The horizontal axis is complexity of risk: personal lines → small commercial → middle-market → specialty/E&S; a move along it (e.g., personal auto → middle-market commercial) builds range and judgment. The vertical axis is scope of responsibility: underwriter → senior underwriter → manager → chief underwriting officer; a move along it (e.g., underwriter → underwriting manager) adds people, appetite, and accountability for a book — a different job, not a bigger one.

Exercise 4 (the four designations)

AINS = Associate in General Insurance — a broad multi-line foundation; best for trainees and career-changers. AU = Associate in Commercial Underwriting — the commercial-craft credential; best for working commercial underwriters. CPCU = Chartered Property Casualty Underwriter — the rigorous P&C capstone (underwriting, law, finance, ethics); best for the career-committed aiming at senior/management roles. ARM = Associate in Risk Management — risk identification and treatment; best for those on the risk-management or broker/buyer side.

Exercise 5 (the analytic path)

The analytic path runs from underwriting into the quantitative functions that increasingly drive it, in the order: underwriter → underwriting analyst (book/loss-ratio analysis) → pricing/actuarial analyst (rate indications) → data scientist (builds and validates predictive models) → product/strategy (owns a line's economics — price, appetite, growth).

Exercise 7 (professional brand)

A professional brand is the reputation you build and the relationships you maintain, which over time determine which opportunities, submissions, and roles come to you rather than ones you chase. Its foundation, per the chapter, is being the underwriter brokers want to call — responsive, consistent, technically credible, and honest about what you can and cannot do.

Exercise 8 (the decision journal)

A decision journal records, for every risk you touch, three lines: what you saw, what you decided, and what you were unsure about. Months later you pull the losses on the accounts you bound and read your own notes against the outcomes. The mechanism: judgment is calibration, and calibration improves fastest when past reasoning is confronted with realized results. Ordinary experience blurs which judgments were sound because outcomes arrive on a delay and detached from the reasoning; the journal reconnects them, so the trainee learns from their own record rather than from vague impression.

Exercise 9 (the claims rotation)

Two parts. (1) Claims is where the promise comes due and the losses you accepted actually appear — rotating through it lets the trainee see, viscerally, what a bad risk costs when it returns, grounding risk selection in consequences rather than abstraction. (2) It builds the value-chain mental model (Chapter 1): underwriting is one gear in distribution → underwriting → pricing → issuance → claims → reserving → reinsurance, and an underwriter who has stood in the claims station reads loss runs and structures terms with a sharper sense of how losses actually develop and settle.

Exercise 11 (the over-specialization trap)

The warning: over-specializing too early in a line that is being automated, and mistaking deep skill in shrinking work for durable expertise — the loss arrives on a delay, felt only once the market has moved past the skill. The diagnostic question to answer by year three or four: Is my line deepening (more judgment required over time, like specialty) or thinning (more automation, less human judgment)? If deepening, going deeper is rational; if thinning, steer toward complexity, analytics, or management before the skill's value erodes.

Exercise 13 (declining a management promotion)

Because the two career paths — management and senior specialist/analytic individual contribution — reconverge at the top of the compensation arc (§37.5). A senior specialty underwriter (cyber, D&O, marine) can out-earn a generalist manager, since the judgment is rare and hard to automate. So an excellent underwriter who does not want to develop others, set appetite, and own a book they no longer underwrite can rationally decline management and earn well as an individual contributor — the only mistake is making that choice by accident.

Exercise 14 (two candidates, one seat)

Candidate B is the stronger candidate for this middle-market commercial seat: B has built the specific judgment the role needs — multi-line reasoning, terms structuring, broker negotiation — plus the AU and ~40 package accounts under referral. Candidate A has impressive depth but in personal auto, the wrong dimension for a multi-line commercial role; A built depth where B built range, and this seat rewards range. This does not make A the weaker underwriter (A may be excellent and better paid today). A's next move: either deliberately broaden (take the lateral B took years ago) or go deep-analytic/managerial within personal lines. The principle: underwrite the candidate like a risk — not "is this person good?" but "is this the right experience for the seat?"

Exercise 15 (ex-underwriter vs. pure modeler)

An ex-underwriter who learns to model brings a felt sense of what the variables mean on the ground — why a thirty-year-old roof in a named-storm zone differs from one inland, why a loss run is a story about management and not just a frequency count — and so knows which features are signal and which are noise before the data confirms it, and recognizes when a model is confidently wrong because it cannot see context (the Harbor Steel 7→6 override, Chapter 32). A stronger pure modeler with no insurance background can fit the same features but cannot supply that domain interpretation, which is precisely what keeps a pricing model honest and legally clean (e.g., spotting a proxy for a forbidden factor, Chapter 35).

Exercise 16 (AU or CPCU next)

Earn the AU first if committed to commercial underwriting — it is the directly relevant, craft-certifying credential, and it sharpens the actual work while signaling competence to the market. Start the CPCU (chipping away one or two courses a year) if aiming at senior individual-contributor, management, or executive roles, because it is the broad, rigorous capstone that keeps the most doors open. The deciding condition is where you are trying to go: stay-in-the-craft favors AU first; aim-at-leadership favors getting the CPCU underway early rather than waiting for a perfect window.

Exercise 17 (AINS high early, modest later)

High early: for a newcomer or career-changer, the AINS teaches the shared vocabulary fast and signals to an employer that you are serious enough to invest your own time — both valuable when you have little track record. Modest later: by the time you are a senior underwriter, the AINS on your signature line does little — it is a foundation credential whose signaling job is done, superseded by the AU/CPCU and, above all, by your demonstrated judgment.

Exercise 18 (book analysis — loss ratio by segment)

Loss ratio = incurred loss / earned premium. Light manufacturing = 2,310,000 / 4,200,000 = 0.55; metal fabrication = 1,530,000 / 1,800,000 = 0.85; warehousing = 1,240,000 / 3,100,000 = 0.40; contractors = 2,000,000 / 2,500,000 = 0.80. Metal fabrication (0.85) most needs attention, with contractors (0.80) close behind; warehousing (0.40) is the healthy segment. The domain knowledge to bring before changing appetite or rate: you know why metal fabrication runs hot (hot work and the products-liability tail — exactly the Harbor Steel hazards) and why contractors run loss-heavy (mobile operations, subbed work, auto exposure). That interpretation tells you whether the high loss ratio is a pricing problem (raise rate), a selection problem (tighten appetite or controls), or noise in a small segment — a judgment the raw numbers cannot make on their own.

Exercise 19 (designation ↔ path)

Mapping: AINS → any entry/trainee path (foundation for all); AU → the commercial-lines path (§37.2, the middle-market/specialty craft); CPCU → the management/leadership path (§37.2 vertical axis, the broad capstone for senior roles); ARM → the risk-management/large-account/broker path (the buyer-side bend). Justification (CPCU→management): the capstone's breadth across underwriting, law, finance, and ethics matches what a manager or CUO must integrate — they own appetite, governance, and a book's economics, not a single line's craft, so the broad credential fits the broad job.

Exercise 20 (steepest in the middle)

The largest pay jumps come in the middle years because that is where scarce judgment and scarce responsibility concentrate. Entry (trainee → underwriter) is a modest step because the work is still bounded and supervised. The very top (into the CUO/executive band) is a smaller population and a slower climb. The middle — underwriter → senior/specialty, and underwriter → manager — is where you cross from doing bounded work to owning judgment (specialty) or owning a book and a team (management); both are scarce and hard to automate, so the market pays steeply for the transition.

Exercise 21 (manager pay tied to combined ratio)

It is "exactly right" because it aligns the underwriter-manager's incentive with the one number that tells the truth about underwriting (Chapter 3): below 100% the book made money on underwriting, above 100% it lost money before investment income. Tying pay to the combined ratio of the book the manager owns rewards profitable underwriting rather than mere premium growth — directly countering the temptation to chase volume at inadequate rates (the rate-adequacy discipline of Chapter 11).

Exercise 23 (the golden-handcuffs plateau)

The trap is the golden-handcuffs plateau: a comfortable senior salary in a stable, automating line, with a small steady raise that makes leaving feel costly, and no real development. Red flags: top production but in shrinking work; security that feels high precisely as the ground (automation) shifts; no new skills in three years; a raise that lifts the point without changing the slope. The defense: the §37.2 diagnostic read every couple of years (is the line deepening or thinning?), and the willingness to make a lateral or developmental move that costs a little now to avoid the larger loss later — the career version of pricing for the loss that arrives on a delay.

Exercise 25 (soft skills become more important)

As routine underwriting automates (Chapters 20, 32), the algorithm absorbs the clean, high-volume, pure- analysis work, leaving the human precisely the judgment-heavy residue: the referrals, the complex multi-line accounts, the model-says-decline-but-context-says-otherwise cases. That residual work is communication, negotiation, and the confidence to override and decide — so the soft skills become the job, not a soft part of it. Connection to promotion: the underwriters who rise in an automated era are those strongest in the judgment-and-relationship skills the machine cannot own, because that is where the remaining human value concentrates.

Exercise 26 (decline-letter memo — model answer)

A strong reply: thanks the broker for the submission and the relationship; states plainly that you cannot reach an adequate price for the exposure at terms that would be competitive, naming the specific driver (e.g., the loss history or the catastrophe accumulation) honestly; offers, where possible, a constructive path ("if the insured adds [control] / accepts [higher deductible], bring it back and I'll re-look"); and closes by inviting the next submission. The two soft skills deliberately exercised: communication (explaining the why clearly so a good decision is not mistaken for an arbitrary one) and negotiation / relationship management (declining this risk while keeping the broker bringing the next one — Chapter 13's "say no and keep the broker"). The point is that the decline preserves trust precisely because it is reasoned and respectful.

Exercise 27 (confidence to say no ↔ themes)

(1) Underwriting is judgment — the confidence to decline is judgment made personal: the underwriter, not the algorithm or the sales target, decides. (2) Adverse selection is the enemy — the broker always brings the risks that most want to be written, so declining the wrong ones is the front line of managing adverse selection. (3) The combined ratio tells the truth — saying yes to inadequately priced risk drives the combined ratio above 100%, on a delay. (4) Pricing follows risk — the confidence to decline is what rate adequacy (Chapter 11) feels like from the inside. (Optionally (6) the social function — an underwriter who cannot decline a bad risk eventually cannot keep the promises made to the good ones.)

Exercise 29 (designations and the fairness line)

A formally grounded grasp of the fair-vs-unfair-discrimination line (Chapters 4 and 35) becomes more important as you rise because your decisions carry regulatory and reputational weight for the whole company, not just one file: you set appetite, approve overrides, and shape rating and selection at scale, where a proxy-discrimination problem (Chapter 35) or a misuse of a restricted factor becomes a systemic and legal exposure rather than a single mistake. The designations' curricula and codes of conduct embed that legal and ethical frame formally, which is one of the cleaner ways to build the understanding the senior role demands.

Exercise 31 (everyday networking)

Three concrete actions: (a) return brokers' calls promptly and handle their hard accounts fairly — the broker remembers and brings you the best submissions first (Chapter 39); (b) mentor a junior underwriter — they become a trusted colleague and, over a career, a peer or leader who vouches for you; (c) take the adjacent functions seriously (return the claims adjuster's call, engage the actuary's analysis) — the cross-functional trust earns you referrals, heads-ups about openings, and the benefit of the doubt in hard calls. Each compounds because insurance is a small world with long memories: everyday reliability accumulates, over decades, into a brand that brings opportunity to you.

Exercise 32 (Underwriting File — who writes it)

Harbor Steel is a middle-market commercial account (multi-line, individually underwritten, ~\$45M revenue, hard story), so it is written by a middle-market commercial underwriter — several years in, full authority within commercial lines, the craft for a multi-line account. It exceeds that underwriter's individual authority because the catastrophe exposure, the \$20M property line, the loss history, and the predictive model's decline recommendation push it past what a line underwriter binds alone — so it is referred to a senior underwriter or manager (Chapter 13's referral logic). Per the brief, no binding decision or new terms are stated here — only the role, the level, and the referral the account triggers.

Exercise 33 (Underwriting File — the profile that binds it)

A realistic profile of the underwriter trusted to bind a coastal, debit-rated, model-overridden middle-market account like Harbor Steel: roughly eight-plus years of experience (past trainee and the high-volume early roles); prior segments spanning small and middle-market commercial across multiple lines (property, GL, WC, auto) so multi-line reasoning is second nature; the AU earned and likely the CPCU underway or complete (for the breadth and the credibility); and a senior grade of authority (or a manager) whose limits cover the property line size, the catastrophe exposure, and a documented model override. Tie-ins: the property/cat size demands authority and cat judgment (Chapters 19, 30); the multi-line nature demands range (§37.2); the model override demands the analytic literacy to interrogate the score (§37.3, Chapter 32); the debit rating and loss story demand the pricing discipline and the confidence to set adequate terms (§37.6, Chapter 11).

Chapter 38

Worked solutions to the daggered (†) and odd-numbered exercises. (Even, non-daggered items are discussion or memo prompts whose reasoning is developed in the chapter text and not reproduced here.)

Exercise 1 (the four levers)

The four levers of the underwriting leader: Appetite — what we will and won't write, and how much (captured in the risk-appetite statement → referral grid); Authority — who may say yes, to what, and up to what limit (the letter of authority → escalation path); Audit — whether the team is actually following appetite and authority and underwriting well (the underwriting audit, a leading indicator); and Profit & Team — whether the book makes money (the manager's combined ratio) and whether the team can sustain it (hiring, training, culture). The levers form a loop: audit feeds back into appetite and team. (§38.1)

Exercise 3 (letter of authority)

A letter of authority is the formal written grant specifying what underwriting decisions a named individual or role may make on the carrier's behalf. At minimum it specifies: (1) the lines/products the holder may underwrite (and may not); (2) limit and premium thresholds — the maximum policy limit and account premium they may bind alone; (3) class restrictions — which appetite tiers (target/accept/restrict) they may write without referral; (4) pricing latitude — how far from technical or filed rate they may move; (5) binding authority — whether they may bind or only recommend; (6) referral triggers — conditions that require escalation regardless of the above; (7) term and review — that the grant is reviewed at least annually and is revocable. (§38.3)

Exercise 5 (CUO vs. underwriting manager)

The chief underwriting officer (CUO) is accountable for the quality and profitability of all of the carrier's underwriting — they own the appetite, the authority framework, the audit function, and the enterprise underwriting result, and defend the underwriting strategy to the CEO and board. An underwriting manager is accountable for a team's book — applying the appetite and authority the CUO sets, running the audit and the combined ratio for their segment, and developing their underwriters. The difference is scope and ownership: the manager runs the levers for a team; the CUO owns the levers for the enterprise and answers for them within governance. (§38.7)

Exercise 7 (three lines of defense)

The three lines of defense in underwriting governance: the first line is the underwriters and their managers, who own the risk and make the decisions; the second line is independent risk management and compliance, who set the framework and challenge the first line; the third line is internal audit, who independently assure that the first two are working (and that the underwriting audit itself is real). The structure exists so the people writing the business are not the only ones judging whether it is sound. (§38.7)

Exercise 8 (rewrite the appetite line — usable form)

The board line — "We have a moderate appetite for coastal commercial property" — is true but unusable: a line underwriter cannot apply it. A usable rewrite pushes it down to concrete boundaries, e.g.: "We write coastal commercial property up to \$25M total insured value per location, outside the mapped storm-surge zone, with a roof no older than 20 years (or an ACV roof endorsement until replacement), subject to a 5% named-windstorm deductible; coastal capacity is capped at \$X of modeled PML per peril zone; accounts that breach any of these route to a senior-underwriter referral, not an automatic decline." The test of a good rewrite: a new underwriter could look at a live submission (e.g., Harbor Steel) and know in thirty seconds whether it is in, out, or a referral. (§38.2)

Exercise 9 (the "light manufacturing" drift)

The failure is appetite drift / unstated expansion: a class grows without any single account tripping an alarm, because the path of least resistance (bind the account in front of you) accumulates into a book nobody designed. Mechanism: each individual heavy-fab account looked in-appetite or close enough to write; no threshold flagged the cumulative shift from light to heavy. The metric the leader should have watched besides the loss ratio is the shape / mix of the book — premium distribution by class, zone, and size — because the shape changes before the loss ratio does and is therefore the early warning. (§38.2, §38.5)

Exercise 10 (appetite vs. adverse selection)

An individual underwriter fights adverse selection one file at a time, with classification and pricing. A leader fights it by shaping the submission flow: appetite tells the market which risks to bring and which not to bother, and prices the restricted classes so only the genuinely good ones clear the bar. Set appetite too wide and price too soft and you become the market's dumping ground — every risk another carrier declined finds you, because you are the one who says yes. So appetite is a filter on the flow itself, the first and cheapest line of defense against adverse selection, operating before any single file is even classified. (§38.2; adverse selection from Ch.1, §1.4)

Exercise 11 (classify into tiers)

For a regional middle-market carrier (illustrative): (a) a clean office building — target or accept; standard, low-hazard, the kind of business you want. (b) Harbor Steel — restrict; heavy metal fabrication with hot-work fire history and coastal PML — writable, but not by a junior alone; requires referral and conditions. (c) a fireworks manufacturer — decline; a hazard class outside the carrier's appetite entirely. (d) a well-run machine shop with one small loss — accept; light manufacturing, good management, a minor loss that doesn't change the grade. The point: most real accounts live in accept and restrict — the "yes, but" middle. (§38.2)

Exercise 12 (build the referral grid; route Harbor Steel)

A four-row grid (illustrative thresholds):

  characteristic           line UW        senior UW       manager         CUO
  property limit           ≤ $5M          ≤ $25M          ≤ $50M          > $50M
  appetite tier            target/accept  + restricted    + by-exception  any (within treaty)
  loss history (5 yr)      ≤ 1 large loss ≤ 2 large       referral        referral
  pricing vs. technical    at/above       −10% band       −20% band       below −20%

Routing Harbor Steel: it is a restricted class (heavy fab — beyond the line UW's tier), it has two large losses in five years (the 2021 and 2023 fires — at the senior UW's limit), it carries a debit-rated price the broker is contesting (pricing pressure), and (adding the catastrophe row) a material coastal PML. It trips the line-UW boundary on at least the class and loss-history cells, so it routes to a senior underwriter, and — because two triggers are quality-related — gets a peer review. The coastal/zone-aggregate appetite call escalates to the CUO. Rule: route to the lowest column satisfying all triggers. (§38.3)

Exercise 13 (the $30M account on a $5M letter)

Two distinct problems, both real even though the account never has a loss. About this account: the carrier is now on a \$30M risk it never agreed to take — the exposure, the catastrophe contribution, and the reinsurance treatment were never evaluated or approved at the right level; "it looks fine" is luck, not process, and a \$30M limit is six times the vetting the account received. About the system: an underwriter has demonstrated they will exceed their authority, which means the authority framework is not being respected — and if it happened once undetected, it is happening elsewhere. "No loss" doesn't make it okay because authority is about controlling the risk the carrier accepts, not about whether a given gamble happened to pay; the next \$30M account bound outside authority is the one with the loss. (§38.3)

Exercise 15 ("everything gets referred")

First, diagnose which problem it is. If the referrals are routinely approved without change — the manager is rubber-stamping — it is an under-delegation (authority-calibration) problem: authority is set too low and the referrals add delay without adding judgment; the fix is to raise the line underwriters' letters to the level they can actually handle, freeing the seniors for the genuinely hard accounts and speeding broker response (Ch.39). If instead the referrals are frequently being corrected or declined on review — the underwriters are genuinely sending up accounts they couldn't handle well — it is a competence/development problem; the fix is coaching, calibration, and incremental authority as judgment is demonstrated, not blanket delegation. Tell them apart by the referral override rate: high approvals-without-change = over-referral; high corrections = capability gap. (§38.3, Ch.39)

Exercise 16 (the audit sampling plan)

Where 5% of accounts hold 50% of the premium, pure random sampling misleads because a random 2% sample is dominated by small accounts and will likely miss the few large ones that actually drive the result — you'd estimate a book-wide error rate that says nothing about your biggest exposures. Design a stratified plan: (1) a random component across the whole book to estimate the true error rate and catch systemic problems; plus (2) a targeted component covering the large accounts (review all or most above a premium threshold), the exceptions (overrides, below-technical pricing, restricted-class binds), the new underwriters, and classes running hot. Weight the sample toward where the premium and the risk are concentrated. The targeted sample protects the book; the random sample measures it. (§38.4)

Exercise 17 (finding vs. bad account)

A finding records that the file did not properly defend its own decision — e.g., an undocumented price basis or an unexplained model override — whereas a bad account is a poor risk or a mispriced one. They are different: the audited account may be perfectly well-selected and well-structured and still fail on documentation, because the audit also asks "could an auditor, manager, reinsurer, or court two years from now see the reasoning?" What the audit is protecting is the defensibility and teachability of the decision — the file documentation discipline of Chapter 13 — precisely because the loss runs can't check it and because an undocumented decision can't be reviewed, taught, or defended (and becomes unreadable when the underwriter leaves). (§38.4, Ch.13)

Exercise 18 (find the red flag — the 94% pass rate)

The red flag is not the 94% headline pass rate; it is the pattern hiding under it: a 22% failure on the rate adequacy dimension, clustered in two underwriters and one class. That clustering is the tell — it is not random error but a systematic softness in pricing, concentrated where it will do damage. It most likely means those two underwriters are consistently shaving rate (perhaps under producer pressure) in a class that is therefore being written underpriced — a leading indicator of a future loss-ratio problem in that segment. Next three actions: (1) pull and review the specific underpriced files to confirm the pattern and quantify the rate gap; (2) have the targeted coaching/calibration conversation with the two underwriters and check whether it's a capability gap or a pressure/incentive problem; (3) examine the class itself — is the technical rate right, is the appetite too loose — and tighten pricing/appetite and re-audit to confirm the fix took. (§38.4, §38.5)

Exercise 19 (audit-as-punishment)

When findings are used primarily to discipline, the data the audit produces degrades: underwriters learn to hide the messy accounts, document defensively rather than honestly, and treat the auditor as an adversary — so the audit stops measuring reality and starts measuring what people are willing to show. The posture a leader should take instead: treat the audit as a coaching and calibration tool first, an accountability tool second — most findings produce a development conversation, not a write-up. The goal is to find the small errors quickly so they can be corrected before they compound (the epigraph's argument). An organization that shoots the messenger soon has no messengers, only loss runs. (§38.4)

Exercise 21 (the 30%-growth, 92% combined ratio book)

This is dangerous because of, not despite, the good number. The reported 92% is a lagging figure — it is the loss ratio on business written two and three years ago, when rates were higher. Meanwhile every leading indicator points the other way: rate is down 6 points (the premium is becoming inadequate), the hit ratio is up 15 (you're winning more, which often means you've gone soft on price or terms), new business is 35% of the book (new business runs worse than renewals), and it's concentrated in two restricted classes (mix drift toward hot business). Four indicators moving together is a pattern, not noise: the book you are writing now is softer than the book the 92% describes. Four leading indicators to watch: rate change, hit/quote ratio, new-business loss ratio, and mix/shape (any of retention or audit-findings rate also valid). Don't celebrate the 92%; expect it to deteriorate as you fix the book, because the fix arrives on a lag too. (§38.5)

Exercise 23 (four leading indicators and why they lead)

(1) Rate change — average premium change at constant exposure; it leads because falling rate makes the premium inadequate now, but the resulting losses don't surface in the combined ratio until those policies mature (Ch.11). (2) Hit/quote ratio — share of quotes that bind; a surging hit ratio leads because it often means you've softened price or terms, and the consequences of that softening are still in the pipeline. (3) New-business loss ratio — loss ratio on first-year accounts; it leads because new business runs worse than renewals and is the freshest read on selection quality before the whole cohort matures. (4) Mix / shape — premium distribution by class, zone, size; it leads because drift into hot or restricted classes changes the expected loss ratio before the realized one moves. (Retention and audit-findings rate are also acceptable.) (§38.5)

Exercise 25 (hire A or B)

Hire Candidate B — thinner product knowledge but a record of honest, well-documented declines. The reasoning is the chapter's core hiring principle: product knowledge can be taught; the underlying aptitudes cannot. You can teach a smart, honest, commercially-minded person the entire property rate plan in a year. You cannot easily teach a "make-the-deal-work" underwriter the intellectual honesty to document the messy truth or the temperament to say no under pressure — and those are exactly the traits that protect a book through a soft market. Candidate A's instinct to find a way to write everything is the adverse-selection risk wearing a competent costume. Hire for judgment, honesty, commercial sense, and the temperament to decline; train the knowledge. (§38.6)

Exercise 27 (the CUO and the governance that says no)

The line means the CUO's seniority is proven not by unlimited authority to approve but by the humility and discipline to build, and submit to, the structures that can overrule them — the board-approved appetite, the independent second line, internal audit. Why a CUO who resents the second and third lines is a warning sign: an underwriting function in which the people writing the business are also the only ones judging it will, sooner or later, talk itself into a book it shouldn't have written, because everyone inside it shares the same growth incentive and the same optimism. The independent challenge exists precisely to catch that. A CUO who wants the challenge removed is removing the structural protection against the most expensive failures in insurance history — the ones where a profitable-looking, fast-growing line ran with no one empowered to say stop. (§38.7)

Exercise 29 (ethics dilemma — the high-producing, thinly-documenting underwriter)

This is the chapter's central tension in a person. The producers love them and the loss ratio "looks fine," but the loss ratio is only two years old (lagging), the files are thinly documented (undefendable), and several accounts were bound below technical price with no rationale (rate-adequacy erosion — the §38.5 leading indicator). What to do: do not simply celebrate the production, and do not wait for the loss ratio to prove the problem — by then years of underpriced, undocumented business are bound. Act on the leading indicators now: (1) require the documentation and the price rationale going forward — this is non-negotiable, because an undefendable decision is a liability regardless of outcome; (2) audit a deeper sample of their book to size the rate gap and the concentration; (3) coach, and if the below-technical pricing continues, tighten their authority/pricing latitude. Weigh present production against the unseen risk by remembering that the production is certain and visible while the risk is real but lagged — and that the leader exists precisely to act on the risk before the numbers force everyone's hand. Reward the disciplined behavior you want, not just the premium. (§38.4, §38.5, §38.6)

Exercise 31 (Underwriting-File extension — route Harbor Steel; why the appetite call is the CUO's)

Routing Harbor Steel through the §38.3 grid, it trips four cells at once: restricted class (heavy fabrication with hot-work fire history), two large losses in five years (2021 + 2023 fires), material catastrophe PML in the Port Hadley zone, and debit-rated pricing the broker is contesting. Any one might sit within a line underwriter's letter; all four together do not, so it routes to a senior underwriter and — because two triggers are quality-related — a peer review. Why the appetite call belongs to the CUO and not the technical analysis: the technical work (is the risk sound, is the price adequate, are the terms right) can be done and approved by a senior underwriter. But the question "do we want one more coastal-fabrication account in this zone, given the Port Hadley aggregate (Ch.30) and the broker concentration with Meridian (Ch.29)?" is a portfolio and capital question that affects the whole book and the cat treaty — it is the appetite the CUO owns and the board approves. The individual account can be excellent and still be the wrong marginal addition to the portfolio; only the owner of the appetite can make that call. (§38.3, §38.7, The Underwriting File)

Exercise 33 (CUO appetite decision when the zone is at its cap)

Illustrative two-sentence decision: "With the Port Hadley zone aggregate already at its catastrophe cap (Ch.30) and Meridian sourcing a third of our coastal book (Ch.29), I will not add Harbor Steel as net additional coastal exposure at the current capacity — the concentration, not the account's quality, is the binding constraint. If we still want it, we make room rather than make an exception to the cap." The one lever to pull to make room: reinsurance (Ch.27) — buy or arrange additional facultative or treaty catastrophe capacity for the zone so the account is ceded rather than retained net, or free aggregate by shedding weaker coastal risks at renewal (an appetite/portfolio action). The disciplined move is to expand capacity deliberately, not to quietly breach the cap one good-looking account at a time — which is exactly the drift §38.2 warns against. (§38.2, §38.7, The Underwriting File)

Chapter 39

Worked solutions to the daggered (†) and odd-numbered exercises. Section references point back to index.md. All figures are illustrative/constructed.

Exercise 39.1

Submission quality is the completeness, accuracy, organization, and honesty of the information a broker provides about a risk — the quality of the presentation, not the risk. The two are independent. Great risk, poor submission: a clean, well-run manufacturer with five years of excellent losses whose broker sends one unsigned application and "loss runs to follow" — a fine risk you cannot yet see. Bad risk, beautiful submission: a distressed, heavily-loss-prone account documented immaculately, with every form, SOV, supplemental, and a candid narrative — the presentation is excellent; the risk is still a decline. Submission quality tells you what you can trust and how hard you must dig; it does not, by itself, grade the risk. (§39.3)

Exercise 39.3

An agent legally represents the insurer (agency agreement, often binding authority, a duty to the carrier); a broker represents the insured (a duty to find the client the best coverage and price across the market). It matters at the table because the broker's professional and fiduciary obligation is to extract the best deal for the other side — not from hostility, but as the job. Knowing this keeps you from the two rookie errors: treating the broker as an adversary to be beaten, or as a teammate who shares your loss ratio. They are neither — they are your most important counterparty. (§39.1)

Exercise 39.4

A retail broker deals directly with the insured: they sit across from the business owner, learn the account, and shop it to carriers they are appointed with — for vanilla risk, straight to an admitted carrier. A wholesale broker is the specialist who connects the retail broker to the excess-and-surplus (E&S) market, because most retailers do not hold direct appointments with surplus-lines carriers and Lloyd's syndicates. The wholesaler's role: take a hard risk the admitted market won't write, broker it into the non-admitted market (with its freedom of rate and form), and bring back terms — handling the diligent-search and surplus-lines-tax mechanics that route the risk legally. An MGA is a wholesaler who also holds delegated binding authority. (§39.2)

Exercise 39.5

The three pillars, behaviorally: Trust — be exactly as good as your word; honor your quotes and binders, never surprise the broker after they've sold your number, and protect their relationship with their client. Responsiveness — answer fast; "received it, terms Thursday" within the hour, or a fast specific no the same day, beats a slow maybe (the chief broker complaint about underwriters is slowness and silence, not strictness). Consistency — be predictable in appetite, pricing, and standards, so a broker can pre-qualify an account and steer the right risks to you before sending them. Each is something you do, repeatedly, not something you are. (§39.4)

Exercise 39.7

A normally-complete commercial submission includes: signed ACORD applications for all lines; five years of currently-valued loss runs; a detailed statement of values; supplemental applications for the account's specific hazards (hot-work, products, drivers, cyber controls); a narrative on operations, management, and controls; the broker's own assessment and target price; honest up-front disclosure of the bad facts; the reason the account is in the market / the incumbent's status; and a realistic timeline. (§39.3)

Exercise 39.9

Three things change in a surplus-lines (non-admitted) placement versus admitted. For the insured: (1) no state guaranty-fund backstop if the carrier becomes insolvent; (2) rates and forms are not filed/standardized, so coverage can be tailored (good) but is less predictable and must be read carefully; (3) a surplus-lines tax applies and a required disclosure must be given that the coverage is non-admitted. For the underwriter: (1) you have freedom of rate and form — you can price the cat or distressed exposure as you see fit and manuscript the terms; (2) you rely on the diligent-search having been done (the admitted market genuinely declined first); (3) you must know you are quoting in the non-admitted market, because it changes the forms you use and the backstop the insured has. (§39.2; Ch.4)

Exercise 39.11

When a carrier delegates underwriting authority to an MGA, your job shifts from underwriting individual risks to underwriting the MGA itself: you audit their book, their risk selection, their pricing, and their adherence to the agreed guidelines and authority limits. You are, in effect, underwriting a portfolio and a process rather than a single account — verifying that the delegated authority is being used as intended and that the book the MGA produces fits your appetite and runs to an acceptable loss ratio. This is closer to the leadership/oversight discipline of Chapter 38 than to risk-by-risk underwriting. (§39.2; Ch.38)

Exercise 39.12

The warning signs and what each signals: one unsigned ACORD — incomplete and not yet a firm application; "loss runs to follow" — the single most important document is missing, and the phrase often means it never quite arrives; a lump building value with no SOV — you cannot value the property or assess concentration; no supplemental for the known welding hazard — the real driver of the risk is unaddressed; "need it bound tomorrow" — deadline pressure on a hazard-heavy class, designed (consciously or not) to rush you past the gaps. Taken together they signal a risk you cannot see and possibly one being shopped. What you do: do not quote. Send a specific, professional information request — the full five-year currently-valued loss runs, a detailed SOV, the hot-work supplemental, and the loss-control history — explain that the welding hazard cannot be responsibly priced without them, and note the pattern on the broker scorecard. (§39.3)

Exercise 39.13

You treat the two submissions differently because Broker A has given you a risk you can grade and trust — the bad fact (the dust-collection fire) and the fix (the housekeeping program) are on the table, so you can price the hazard and condition the account — while Broker B has hidden the very hazard that defines the risk, so anything you quote is a guess and B may be shopping a problem. What submission quality does tell you: how complete and candid the presentation is, and therefore how much you can trust it and how hard you must dig. What it does not tell you: whether the underlying risk is actually good or bad — A's account could still be a decline on the dust hazard, and B's could be a clean risk buried in a sloppy file. Quality governs trust, not the grade. (§39.3)

Exercise 39.15

"You are always teaching your market what you will accept" because brokers learn, from how you respond, what gets them a quote. Accepting thin submissions teaches a broker that they can get terms from you without doing the work — so they stop gathering the loss runs, the SOV, and the supplementals, and your incoming submission quality falls (and your adverse selection rises). Requiring quality — a specific information request every time, and a refusal to quote blind — teaches the broker exactly what you need, so their next submission arrives more complete; you have raised your own input quality by training the source. Every response is a lesson; make sure it is the one you want repeated. (§39.3)

Exercise 39.17

Consistency is the most underrated because it is what lets the broker pre-qualify and steer. If your appetite and pricing are predictable, a broker can look at an account and know, before sending it, whether you will want it and roughly what you will charge — so they send you what fits and spare you both the misses. That is the whole prize: the right risks flowing to you because the broker can route them confidently. If your appetite swings with your mood, the month's production number, or a new appetite statement, the broker cannot pre-qualify anything; every submission becomes a coin flip, and they route their best, most durable accounts to carriers who are predictable. Consistency turns a market into a steered market. (§39.4)

Exercise 39.19

A fast, specific decline is more valuable than a slow maybe because it saves the broker time on a deadline-driven placement, tells them exactly what you don't want (so they can pre-qualify next time), and treats them as a professional — and because the chief broker complaint about underwriters is slowness and silence, not strictness. Rewrite: "Declined." → "This isn't one for us this year — the cat aggregate in that zone is full, and the roof at end-of-life with no replacement plan is past my appetite. Bring me the same risk with a signed roof-replacement contract and I'll look hard at it." The second version ends the deal but keeps the broker, and points them at the account you would write. (§39.4, §39.5)

Exercise 39.20

The three asks, worked: (a) Lower the named-windstorm deductible — HOLD. This is the firm line: the wind deductible is what makes the catastrophe exposure writable at all, and it is what the cat XOL treaty (Ch.27) and the Port Hadley zone aggregate (Ch.29–30) assume. Explain why so Meridian can sell it as protection rather than a grab. (b) Restore replacement cost on the roof now — FLEX, via the roof itself. The ACV-roof endorsement is not a penalty; it is a temporary, self-curing condition that converts automatically to replacement cost once the warranted new roof is installed. The path back to RC is the roof contract — give the broker that roadmap. (c) Loosen the subjectivities — FLEX the timing, not the substance. Roof replacement and the hot-work program are conditions of the risk and non-negotiable in substance, but their timing can flex (ACV until the roof is done; a reasonable window to stand up the hot-work program). The single firm line is the named-windstorm deductible, non-negotiable because giving it away leaves the cat exposure uncovered by the protection the reinsurance and portfolio aggregate are built on — you could not defend the file to your manager or the treaty without it. (§39.5; The Underwriting File)

Exercise 39.21

Four non-price levers from Chapter 12: (1) raise the deductible — lowers the premium and improves the risk by keeping the insured's skin in the game (Ch.1), so it does double duty; (2) add or tighten a sublimit on the specific exposure you are worried about, reducing your downside; (3) adjust limits — right-size the per-occurrence or aggregate limit to the real exposure; (4) endorsements — a coverage restriction (e.g., excluding a peril the insured doesn't need covered) that safely reduces price, or a manuscript endorsement that fits the coverage to the operation. The deductible lever is the one that improves the risk while it lowers the price — the insured now has a direct financial incentive to prevent and minimize loss. (§39.5)

Exercise 39.22

(a) Cutting straight from \$60,000 to \$54,000 with no change to the risk consumes the \$3,000 profit/contingency load first and then eats into the expense load, because the \$42,000 expected-loss component is fixed by the risk — you cannot wish it away. The result is a rate at or below the expected loss plus expenses, i.e., inadequate: there is no margin for adverse development or error, and the account is underpriced from the day it binds (Ch.11). (b) A legitimate path to a lower defensible price changes the risk, not just the number: e.g., raise the deductible so the insured retains more of the working losses, which genuinely lowers the expected-loss component flowing to you — now a lower premium can be built that still carries its expense and profit loads. (c) The one thing you must not do: cut the price by assumption — dropping the rate while pretending the expected loss is lower than the risk supports. That is not competing; it is underpricing, and the loss it underwrites surfaces two or three years later. (§39.5, §39.6)

Exercise 39.23

Disciplined response to the relationship-plus-deadline rate-cut push: Hold the adequate price. Offer instead a non-price solution (a higher deductible that legitimately lowers expected loss, a sublimit, or a coverage difference) or, if the gap can't be bridged honestly, let the account go. Explain the number so the broker can carry the rationale to the client — "here's what's driving the rate, and here's the structure that could lower it." Over a career, holding the line (explained) earns the good brokers' respect, keeps all your other prices credible, and protects the combined ratio; folding teaches the broker your prices are negotiable — so none can be trusted — and converts the relationship into a reverse auction. A broker who only values you when you're cheapest is not a relationship. (§39.5)

Exercise 39.24

Two ways to try to win without matching the \$8,000-cheaper competitor: compete on coverage — offer a better-fitting form, a sublimit the competitor omitted, or a manuscript endorsement matched to the insured's real exposure, so the broker can tell the client "B's policy actually covers your business-income exposure the way you need"; and compete on service — responsiveness, a clean binder on time, claims advocacy, ease of doing business, the durable moat that earns a service premium. The move you should not make: match the price absent a genuine expense/data/scale advantage. Winning purely on price is quiet adverse selection because the cheapest carrier wins the account the disciplined carriers didn't want at that price — you've selected the risks others priced more accurately and walked away from, and the soft-market loss surfaces when the cycle turns (Ch.1, Ch.3). (§39.6)

Exercise 39.25

Competing on price trades on the premium number — visible, comparable, and a race to the bottom absent a real cost edge. Competing on coverage trades on fit, breadth, and structure — offering the policy that responds when the insured has a claim, at an adequate price. Coverage competition rewards line knowledge because only an underwriter who deeply understands the insured's exposures can identify the broadening worth offering and price it correctly — exactly what Parts III–IV build. A winning one-sentence pitch the broker carries back: "Carrier A is \$2,000 cheaper, but Carrier B's form covers your business-income exposure the way you actually need and won't leave you with a coinsurance penalty — for two grand, that's the policy that pays when you have a loss." (§39.6)

Exercise 39.27

The analytics and the judgment combine like this. The analytics give the scorecard: a broker you enjoy working with whose book runs at 75% is a problem the friendship is hiding, and a less-pleasant broker at 48% is, on the numbers, the more profitable relationship — the portfolio report (Ch.29) and the model (Ch.32) surface this unsentimentally and tell you where to look. But before acting, judgment must decide what the 75% means: a one-off catastrophe year on otherwise sound business is very different from a pattern of adverse selection or thin submissions, and a strong long-term relationship may be worth investing in despite a weak recent number. So: let the data point you at the 75% book, diagnose why with judgment, and then have the human conversation the numbers imply — tighten submission requirements, re-price, or re-set expectations — rather than reflexively cutting the broker on a single ratio. (§39.7; Ch.29, 32)

Exercise 39.29

The ethics dilemma surfaces three lines and a calculation. The rebating/inducement line (Ch.4): the expensive dinner edges toward, and the quid-pro-quo hint crosses into, improper inducement — relationships are built on professionalism, not bought with entertainment or favors, and many states' anti-rebating and unfair-trade rules (and every carrier's code) govern this. The pricing-discipline line (Ch.11): "always finds a way to shave a few points" describes underpricing to buy business — a rate cut to win is inadequate the moment it binds. The relationship-vs-capture line (§39.4): the goal is better, honest submission flow, not capture; a broker worth having wants a solvent market that will be there at renewal, not a pushover. The long-game calculation: caving to the threat trains this broker (and word travels) that your prices bend to pressure — destroying the credibility of all your prices. Response: decline the implied deal politely and unmistakably — "I'll always give you fast, fair terms and creative structure, and I'll compete hard on coverage and service, but my pricing reflects the risk and I don't shave points to win business; that's what keeps me a market you can rely on at renewal." Keep entertainment within the code; put nothing in writing that hints at a quid pro quo. This protects your integrity and, over the long run, your book — and the genuinely good brokers respect it. (§39.4, §39.5; Ch.4, 11)

Exercise 39.30

File status note (illustrative): "Negotiation with Meridian Risk Partners completed [date]. Named- windstorm deductible held at 5% (cat-protection line; assumed by cat XOL treaty and Port Hadley zone aggregate); rationale communicated to broker. ACV-roof endorsement framed as temporary, self-curing — converts to replacement cost upon installation of warranted roof. Subjectivity timing flexed, substance held." Subjectivity tracker: roof-replacement contract — DELIVERED (signed; replacement within 12 mo); hot-work permit program — DELIVERED (insured committed); sprinkler certification — IN PROGRESS; infrared electrical scan — IN PROGRESS; telematics installation — IN PROGRESS. What this chapter settles vs. leaves: This chapter settles the deal — a meeting of minds with the broker on price, terms, and the roadmap to satisfy the subjectivities, with the roof contract and hot-work program now delivered. It deliberately leaves the final assembly and bind — the complete file, the coverage recommendation memo, the reinsurance/portfolio sign-off, and the formal binding with its stated subjectivities and residual risks — to the capstone (Ch.40); do not declare the account bound here. (The Underwriting File; Ch.13, 40)

Exercise 39.31

Had a different broker brought Harbor Steel as a thin submission — two years of loss runs, no SOV, the 2023 fire's cause omitted, "see what you can do" — the handling would have changed completely. You could not have graded the cat exposure (no SOV), could not have read the loss story (two years hides the 2021 electrical fire and obscures the 2023 welding fire), and the omitted cause of the 2023 fire is exactly the kind of material non-disclosure that triggers the red-flag review the file meets in Chapter 33. You would not have reached terms; you would have sent a specific information request and, until the loss runs and the fire's true cause were on the table, treated the account as un-underwriteable — and possibly as a disclosure-integrity concern. The contrast underscores the chapter: it was Meridian's high-quality, candid submission — and the relationship behind it that produced the roof contract and hot-work program — that made the account placeable at all. Submission quality was the precondition for everything downstream. (§39.3; The Underwriting File; Ch.33)

Chapter 40

Worked solutions to the daggered (†) and odd-numbered exercises. Numbers in worked examples are illustrative teaching figures, consistent with the constructed Harbor Steel file. Section refs point back to index.md.

Exercise 1 (†)

The complete underwriting file is the assembled, ordered, self-documenting record of an underwriting decision — its inputs, its analyses, and the reasoned decision and terms — built so a reader who never met the underwriter can reconstruct what was decided, why, and on what conditions. It is built in the order it is read — inputs → analysis → decision — because a decision is only as good as the information behind it, each analytical layer builds on the one before, and the decision should fall out of the reasoning rather than precede it. A file that opens with the conclusion and then justifies it reverses the logic of a defensible decision, and an auditor reads that order as "decided first, reasoned after."

Exercise 3

A quote is an offer the insured may take or leave; the insurer is not on risk. Bound coverage is the point at which the insurer's promise legally attaches and the risk transfers, subject to the policy's terms and any conditions precedent (subjectivities). In short: a quote puts no one on risk; binding puts the insurer on risk. The difference between them is the binding sequence (clearance → quote accepted → subjectivities stated → binder issued → policy issued → subjectivity tracking).

Exercise 4 (†)

The thirteen sections, in order: (1) Submission & clearance — who/what/limits, broker, conflict check; (2) Information gathered — application, 5-yr loss runs, inspection, financials, MVRs, SOV; (3) Risk-assessment summary — COPE, hazards and controls, the grade; (4) The math — frequency/severity, credibility of the loss history; (5) Pricing rationale — manual → experience + schedule → indicated premium; (6) Terms & conditions — deductibles, limits, sublimits, endorsements, by line; (7) Reinsurance treatment — cession of the cat exposure, net vs. gross; (8) Portfolio & capital — fit, concentration, cost of capital, cat aggregate; (9) Data & model view — pre-fill/enrichment, the score and the override; (10) Disclosure / SIU check — the application-accuracy review; (11) The recommendation memo — decision, price, terms, subjectivities, residual risks; (12) Authority & sign-off — referral, peer review, audit trail; (13) Binder & subjectivities — the conditions precedent and the bound record. The first four are inputs, the next six are analysis, the last three are the decision.

Exercise 5

A subjectivity (owned by Ch.13) is a condition the insured must satisfy before, or as a condition of, coverage attaching — a condition precedent. "Condition precedent" means the stated event (e.g., the infrared scan completed, the hot-work program in place) must occur for the coverage grant to be effective; if a subjectivity is not met, the coverage or terms are revisited. For Harbor Steel the subjectivities are the roof replacement within 12 months (ACV until then), the hot-work permit program, the sprinkler certification, the infrared electrical scan, and telematics installation.

Exercise 7

Synthesis adds the consistency and the single judgment that individual analyses cannot produce on their own. Example: the assessment grade ("average-to-below-average but controllable") and the cat-model result ("fits within the Port Hadley zone aggregate, with the 5% wind deductible shrinking the modeled net loss") are two separate truths. Synthesis is the act of confirming they agree — the same wind deductible appears in the terms and in the cat model's financial module — and of producing one recommendation that reflects both. Absent synthesis, you have two correct pages that might quietly contradict each other (e.g., a 2% deductible in the binder and a 5% in the model), which is exactly the defect the capstone exists to catch.

Exercise 8 (†)

Opening with "Recommendation: bind" and then presenting the analysis reverses the order in which a defensible decision is made. An auditor infers that the conclusion came first and the analysis was assembled to justify it — the hallmark of a decision driven by something other than the evidence (a desire to win the account, a relationship pressure). The correct opening logic presents the inputs and analysis first so the recommendation falls out of them: submission and information → assessment → math → pricing → terms → reinsurance/portfolio → model and disclosure checks → then the recommendation memo. The memo may summarize the decision at the top of the file as a reader's aid, but the file's reasoning must run inputs → analysis → decision, and the memo itself must show the decision resting on the documented analysis.

Exercise 9

Four Harbor Steel judgment calls and the sentence each requires in the file: (1) the net schedule debit — "a net debit for the end-of-life roof and the hot-work loss history, partly offset by control credits for the hot-work program, contracted roof replacement, and infrared scan"; (2) the model override — "model scored 7/10 (decline); overridden to 6 because the model could not see the signed roof contract, the new management, and the loss history read as a problem being fixed"; (3) the 5% (not 10%) named-windstorm deductible — "5% retention is adequate given the agreed-value structure, the ACV roof carve-out, and the cat-model net-loss result, and is the term the broker can place"; (4) the products-claim watch rather than an exclusion — "products-completed ops retained with a documented watch because the single pending bracket claim is not yet adverse and an exclusion would gut the coverage the insured most needs." Each is a discretionary call, so each needs its why on the record.

Exercise 11 (†)

A defensible one-paragraph assessment, grade-first: "Harbor Steel grades as an average-to-below-average but controllable commercial-property risk. The drivers are a metal-fabrication occupancy with hot-work and electrical hazards, a fire protection class 4 with original wet-pipe sprinklers and ~600 ft to the nearest hydrant (not HPR), a named-windstorm coastal exposure, an original 1994 built-up roof at the end of its life, and a loss history of two fires in five years (electrical, then hot-work) plus several workers'-comp claims and one pending products claim. The controls that change the grade are the contracted roof replacement (moving the roof from a decline-driver to a time-limited ACV subjectivity), a hot-work permit program, sprinkler certification, an infrared electrical scan, and — read carefully — a loss history that is a management story now being addressed. The grade is conditional on those controls taking hold, which is why they become subjectivities at binding."

Exercise 13

"Summarize the verdict, not the data" because the assessment section's job is to deliver a read a reviewer can absorb in two minutes and defend, not to re-list every loss (the loss runs themselves are already in the information section). If the summary is just a list of every loss, three things go wrong: the reviewer cannot tell which losses matter (the two fires) from which are noise (minor auto claims); the story (a management problem being fixed) is lost in the tally; and the grade — the actual output — is buried. A list is data; an assessment is a graded, reasoned verdict with the controls that change it.

Exercise 14 (†)

Illustrative figures from the chapter's build. The indicated property rate is the sum of the build-up: \$4.50 (manual/class) + \$0.45 (experience adjustment) + \$0.80 (net schedule debit) = **\$5.75 per \$1,000 of building value (debit-rated). Applied to the \$20M building: \$5.75 × (\$20,000,000 / \$1,000) = \$5.75 × 20,000 = \$115,000 indicated building premium (illustrative; equipment, BI, and the other lines are priced separately and summed to the program premium). Every figure here is a constructed teaching number, not a filed rate — the lesson is the traceable shape of the build, not the digits.

Exercise 15 (†)

The two-sentence justification the file should contain: "The net schedule debit reflects two real, above-class features — the original 1994 built-up roof at the end of its life in a named-windstorm zone, and a hot-work/electrical loss history (the 2021 ~\$180K and 2023 ~\$1.2M fires). It is held to a net figure because genuine control credits offset part of it: the contracted roof replacement, the hot-work permit program, and the infrared electrical scan, each of which addresses the very hazards driving the debits." Every debit and credit names a feature, which is what makes the schedule modification defensible rather than arbitrary.

Exercise 17

Two things wrong with using the model's 7/10 score as the price: (1) a score is a ranking, not a rate — it tells you the risk is relatively adverse, but it does not build from pure premium through expense and profit loads to an adequate premium (Ch.11); a flat "high-risk surcharge" is untraceable and unfilable. (2) the score is built on inputs the underwriter has already reasoned past — it reflects the fires, roof, and hot-work class as they were, not the contracted roof replacement, new management, and controls; pricing off the raw score double-counts the very features the override and the control credits already addressed, and ignores the documented reasons the score was overridden to a 6. The price must follow the underwriter's assessment, with the score recorded only as a cross-check.

Exercise 18 (†)

Each term's financial and behavioral jobs: - (a) 5% named-windstorm deductible — financial: the insured retains the first ~\$1M of a wind loss on the building, reducing the insurer's payout and the modeled net loss. Behavioral: keeps the insured invested in storm preparation and roof maintenance, since the first slice of any wind loss is theirs. - (b) ACV roof endorsement — financial: caps the roof recovery at actual cash value (not replacement cost) so the insurer does not pay new-roof money for a worn-out roof. Behavioral: removes the moral-hazard incentive to let the roof ride and converts the roof into a 12-month replacement deadline. - (c) Return-to-work credit — financial: lowers the workers'-comp premium for an account that returns injured workers to modified duty, reflecting lower claim cost. Behavioral: rewards (and thereby encourages) the safety and return-to-work program that reduces loss. - (d) Mandatory telematics — financial: enables better pricing and faster detection of deteriorating fleet risk. Behavioral: changes driver behavior (drivers drive better when monitored) and gives the file an early-warning system on the auto exposure.

Exercise 19

The single term appearing in both the property terms (§40.4) and the cat-model sign-off (§40.6) is the 5% named-windstorm deductible. In the terms, it shares the catastrophe pain and keeps the insured invested in storm preparation. In the cat model (Ch.30), it is fed into the financial module, where it absorbs the first slice (~\$1M on the building) of every modeled wind loss before the policy pays — which measurably shrinks the account's modeled net loss and therefore its zone consumption. Showing that the same number does consistent work in both places is exactly the internal-consistency check the capstone requires.

Exercise 21 (†)

The three red flags and their fixes: (1) a 12% schedule credit with no stated reason — an undocumented discretionary decision; to an auditor it looks arbitrary or potentially discriminatory. Fix: document the specific risk features that justify the credit, or remove it. (2) a recommendation memo with no residual-risk line — the memo is a sales document, not an underwriting memo; it hides what could go wrong. Fix: add the residual-risk line and the re-pricing/non-renewal triggers. (3) subjectivities listed only in an email, not the binder — the conditions precedent are not part of the binding record, so coverage may attach without them being enforceable. Fix: state every subjectivity in the binder itself as a condition precedent, and track each to completion.

Exercise 23

The file is not yet bindable because, for a \$20M coastal property, the individual-risk analysis (pricing, terms, grade) is only one of three nested decisions that must all be "yes" (Case Study 1). The missing check is the portfolio/reinsurance sign-off (§40.6): how the catastrophe exposure is ceded (so one storm cannot sink the carrier) and whether the account's marginal PML contribution fits the peril-zone aggregate (so it does not push the zone past the point where one storm threatens the whole book). A perfectly-priced coastal account written without that sign-off is how carriers fail in aggregate even when each risk was individually sound.

Exercise 24 (†)

A model one-page memo (compressed):

Harbor Steel & Fabrication, Inc. — Coverage Recommendation Memo 1. The ask: Full commercial program (property/GL/WC/auto/umbrella + cyber + IM), new business via Meridian Risk Partners; expiring carrier non-renewing for cat exposure + loss history. 2. Recommendation: Quote-and-bind, with conditions, at the indicated debit-rated terms. 3. Why: Average-to-below-average but controllable risk; the loss history is a management story now being addressed; the price is adequate for the risk and the capital; the terms align incentives; the cat exposure is ceded and within the Port Hadley zone aggregate; the account earns its cost of capital and fits appetite. 4. Price: Indicated, debit-rated whole-program premium (property rate built §40.3; traceable line by line). 5. Terms: Agreed value; 5% named-windstorm deductible; ACV roof (12 mo); 12-mo BI; products watch; debit X-mod + RTW credit; telematics + one driver removed; \$10M umbrella; cyber; IM. 6. Subjectivities (conditions precedent): roof replacement in 12 mo (ACV until then); hot-work permit program; sprinkler certification; infrared electrical scan; telematics installation. 7. Model note: Scored 7/10 (decline-leaning); overridden to 6 — documented reasons the model could not see (signed roof contract, new management, controls). 8. Residual risk: aging sprinklers; pending products claim; cat-and-roof tail until replacement verified. Triggers for re-pricing/non-renewal listed. 9. Authority: Exceeds line-UW authority → referral, peer review, CUO appetite sign-off (Ch.38); audit trail attached.

Exercise 25

A committee-defense script: "The model scored Harbor Steel a 7, decline-leaning, and it was right on its inputs — it saw the two fires, the \$1.2M severity, the end-of-life roof, the wind zone, the FPC 4, the hot-work class, and the pending claim. What it could not see is the signed roof-replacement contract, the new plant management, the hot-work permit program, and the fact that the loss history is a problem being fixed, not a frequency to fear. I bound it at a 6, and the terms make the override prudent — the ACV roof endorsement, the 5% wind deductible, the subjectivities, and the cession to the cat treaty. And it's cross-checked: the satellite imagery and data enrichment corroborated my manual read rather than the score. The override is in the file with all four points, so it can be backtested."

Exercise 27 (†)

A non-renewal-trigger memo (four events tied to residual risks): (1) the roof is not replaced within the 12-month subjectivity window → re-price or non-renew; the cat-and-roof tail does not close and the ACV interim was a bridge, not a permanent posture. (2) a new fire or serious workers'-comp loss reveals the controls did not take → re-underwrite; the "controllable grade" assumption has failed. (3) the pending products-liability claim develops adversely (or a second products claim appears) → re-price the GL and re-evaluate the products exposure. (4) the Port Hadley zone aggregate fills (the carrier writes more coastal exposure) → the account may no longer fit appetite regardless of its own merits (§40.6). Each trigger maps to a residual risk the file already named, which is what makes the renewal decision a documented one rather than a surprise.

Exercise 29

The "social function" reading: the underwriter's craft accomplished what a pure decline would not — a real business, abandoned by its prior carrier, received the catastrophe and liability protection it needed, at a fair, risk-based price, because an underwriter read the file well enough to make a hard risk writable (with terms, subjectivities, and reinsurance). That is insurance doing its job: pooling and transferring a risk one party could not bear. The limit of the argument: the social function never justifies writing a risk that genuinely cannot be made adequate. If Harbor Steel had no roof contract, no controls, an adverse and worsening loss trend, and a zone with no aggregate room, the correct — and equally professional — decision is to decline. The craft is making writable risks writable; it is not writing unwritable ones to be kind, which would simply transfer the loss onto every other policyholder in the pool.

Exercise 31 (†)

A one-page file index (thirteen sections, the headline fact each records): (1) Submission — full commercial program via Meridian; prior carrier non-renewing. (2) Information — 5-yr loss runs, inspection, financials, MVRs, SOV ordered; gaps flagged. (3) Assessment — average-to-below-average but controllable. (4) Math — two fires = low-credibility frequency, real severity/hazard signal. (5) Pricing — indicated, debit-rated, traceable (property ~\$5.75/\$1,000 illustrative). (6) Terms — agreed value, 5% wind deductible, ACV roof (12 mo), 12-mo BI, products watch, debit X-mod + RTW, telematics + one driver out, \$10M umbrella, cyber, IM. (7) Reinsurance — cat exposure ceded to cat XOL treaty; net within retention. (8) Portfolio & capital — earns cost of capital; fits Port Hadley zone aggregate. (9) Data & model — enrichment/satellite corroborate the read; model 7 → override to 6, documented. (10) Disclosure — 2023 fire cause clarified via SIU; no rescission issue. (11) Memo — quote-and-bind, with conditions. (12) Authority — referral, peer review, CUO sign-off; audit trail. (13) Binder & subjectivities — bound; roof/hot-work/sprinkler-cert/IR-scan/telematics as conditions precedent. Disposition: bound, with conditions, at adequate terms.

Exercise 33

A renewal assessment one year in (compressed two paragraphs): "We keep Harbor Steel. The roof has been replaced — the single biggest driver is resolved, so the ACV endorsement comes off and the building returns to full replacement cost, agreed value — and the property line has run clean, which validates last year's 'controllable' grade and the override. The cat exposure and zone fit are re-checked and, assuming the zone aggregate still has room, the property terms can ease modestly to reflect the new roof. But two items move the other way and must be re-priced or re-conditioned: the pending products claim has developed adversely, so the GL is re-rated for the worse products experience and the watch becomes a hard look at whether the products exposure needs a sublimit or tighter terms; and the new welder injury revives the workers'-comp frequency question, so we re-examine the X-mod, reinforce the return-to-work credit's conditions, and may require a documented safety-program review.

To Meridian I'd say: the account earned its renewal by fixing the roof, and we reward that with the endorsement coming off and slightly better property terms — but the products development and the new welder injury are real and we're re-pricing GL and re-looking WC accordingly. Here's the revised quote, here's exactly why each line moved, and here's what we need from the insured on the safety review. That's the deal that keeps it profitable for us and renewable for them."

Exercise 34 (†)

The file "defends itself without you in the room" because the file, not the underwriter's memory, is the authoritative record — and a claim, an audit, or a successor will read it years later when you are unavailable. Three features make that possible: (1) a documented reason beside every judgment call (every credit, debit, override, and subjectivity says why), so no decision looks arbitrary; (2) the inputs → analysis → decision order, so a reader can trace the conclusion back through the reasoning to the information; and (3) the residual risks and the model override disclosed on the face of the recommendation memo, so the hard calls are surfaced and owned rather than buried. A file with those three features answers the auditor's, the claim adjuster's, and the successor's questions before they are asked — which is the whole point of writing it.