Chapter 36 Quiz

Twenty questions to check your grasp of continuous underwriting, AI as co-pilot, climate-driven repricing, insurability under stress, and the new product frontier. Fifteen multiple-choice, five short-answer. The answer key is in the collapsed block at the bottom — try the whole set before you open it.

Multiple choice

  1. Continuous underwriting primarily changes which aspect of the underwriting decision? a) The legal nature of the policy contract b) When the risk is assessed — from a one-time snapshot to ongoing, real-time monitoring c) The rate-regulation regime that governs the price d) Whether the risk needs a broker

  2. According to the chapter, the greatest value of continuous underwriting is usually: a) Re-pricing the policy in the middle of the term b) Replacing the underwriter with a sensor c) Preventing the loss by catching deterioration while it is still cheap to fix d) Eliminating the need for a renewal

  3. "AI as co-pilot, not autopilot" means the artificial intelligence: a) Makes and owns the final underwriting decision b) Handles the high-volume, mechanical work while the human keeps the judgment and accountability c) Is used only for marketing, never for underwriting d) Replaces the actuary but not the underwriter

  4. Which work is most appropriately automated, per the chapter's division of labor? a) The first cyber policy for a novel exposure b) The Harbor Steel package with its catastrophe exposure c) High-volume, well-understood, data-rich risks like standard personal auto d) An account where the model says decline and the file says write

  5. A large language model's tendency to produce confident, plausible, but false output is called: a) Basis risk b) Adverse selection c) Hallucination d) Non-stationarity

  6. The chapter's preferred reformulation of "AI will replace underwriters" is: a) "AI will never affect underwriting" b) "Underwriters who use AI will replace underwriters who don't" c) "Actuaries will replace underwriters" d) "Underwriting will disappear by 2035"

  7. To say the catastrophe baseline is non-stationary means: a) The model never changes b) The underlying frequency/severity of perils is shifting over time, so history is a biased guide c) Losses are perfectly predictable d) The risk is uninsurable by definition

  8. A "1-in-100-year" flood defined on twentieth-century data that now occurs far more often is an example of: a) Moral hazard b) A non-stationary baseline making the historical record understate today's risk c) Basis risk d) Schedule rating

  9. Per the chapter, a risk most often becomes uninsurable when: a) The peril cannot be modeled at all b) The risk-adequate price outruns the price the market will (or by regulation can) pay c) The insured has any prior loss d) The broker stops submitting it

  10. When large carriers pull back from writing new homeowners business in the most exposed parts of states like California and Florida, the chapter characterizes this primarily as: a) A measurement failure — the risk can't be priced b) An availability and affordability failure — the adequate price and the payable/approved price no longer meet c) Pure profiteering unrelated to risk d) A temporary data-quality glitch

  11. A FAIR plan is best described as: a) A federal flood program b) A reinsurance treaty c) A state insurer-of-last-resort providing basic property coverage the standard market won't write d) An AI fairness-testing standard

  12. Parametric insurance pays out based on: a) The proven, adjusted actual loss b) A measured trigger being met (e.g., a storm of a given intensity within a given distance) c) The insured's credit-based insurance score d) The underwriter's discretion after the event

  13. Basis risk in a parametric policy is: a) The risk the insurer becomes insolvent b) The gap between the parametric payout and the policyholder's actual loss c) The risk of a data breach d) The interest-rate risk on reserves

  14. On-demand insurance (switch coverage on and off as needed) is especially vulnerable to which classic enemy from Chapter 1? a) Moral hazard only b) Adverse selection — people switch coverage on precisely when they perceive the risk as highest c) The protection gap d) Coinsurance penalties

  15. The chapter argues that parametric, embedded, and on-demand products: a) Eliminate underwriting judgment entirely b) Relocate underwriting judgment upstream into the product and algorithm design c) Require a human to underwrite every individual policy d) Are illegal in the United States

Short answer

  1. In one or two sentences, explain why continuous underwriting generally cannot be used to raise a policy's rate in the middle of the term, and what the live data is therefore most useful for instead.

  2. State the rule the chapter gives for an LLM co-pilot ("the co-pilot drafts, the underwriter ___"), and explain in one sentence why "the AI wrote it" is not a usable defense.

  3. Explain the difference between a trend and an event in climate-driven pricing, and say which one should move an underwriter's rate.

  4. Give the chapter's definition of insurability in your own words, emphasizing the clause about price that distinguishes "modelable" from "writable."

  5. Name three of the six skills the chapter says will matter most in 2035, and for one of them explain why AI makes it more valuable rather than less.

Answer key (try the questions first) **Multiple choice** 1. **b** — Continuous underwriting changes *when* the risk is assessed (ongoing real-time monitoring vs. a one-time snapshot); it does not change the legal nature of the contract or the rate-regulation regime. 2. **c** — Its greatest value is loss *prevention* (catching deterioration early), not mid-term re-pricing, which the policy contract generally forbids anyway. 3. **b** — The AI handles the mechanical, high-volume work; the human retains judgment, override, and accountability. 4. **c** — High-volume, well-understood, data-rich risks (standard personal auto, small BOP) are where the machine is already better; the novel/complex/contested cases belong to human judgment. 5. **c** — Hallucination: confidently wrong output. (Basis risk and adverse selection are different concepts; non-stationarity is a climate term.) 6. **b** — "Underwriters who use AI will replace underwriters who don't." 7. **b** — A non-stationary baseline means the peril distribution itself is moving over time, so history-based estimates are biased. 8. **b** — A classic non-stationary-baseline effect: the historical record systematically understates today's (higher) frequency. 9. **b** — Uninsurability is usually a price problem: the adequate price exceeds the payable/permitted price, not that the peril cannot be modeled. 10. **b** — An availability/affordability failure: the adequate price and the payable/approved price no longer meet. The risk remains modelable. 11. **c** — A FAIR plan is a state insurer of last resort for property the standard market won't write. 12. **b** — Parametric pays on a measured trigger, not on proven loss. 13. **b** — Basis risk is the gap between the parametric payout and the actual loss. 14. **b** — Adverse selection: on-demand invites switching coverage on when risk is perceived highest. 15. **b** — They relocate judgment upstream into product/algorithm design rather than eliminating it. **Short answer** 16. A bound policy is a fixed-term promise (Chapter 5), and rate regulation (Chapter 4) governs pricing, so you generally can't raise the rate mid-term. The live data is most useful for *preventing the loss* (loss-control intervention now), informing the *next* term's price, and deciding whether to renew. 17. "The co-pilot drafts, the underwriter **verifies and owns**." "The AI wrote it" is not a defense because accountability for the decision rests with the human who signs it — to a regulator, a committee, or a court (the override/documentation discipline of Chapters 32 and 13). 18. An *event* is a single hurricane/fire/flood — weather drawn from a distribution. A *trend* is the slow shift in the distribution itself. The **trend** should move the rate; over-reacting to events (up after a bad year, down after quiet ones) just adds a weather amplifier to the underwriting cycle. 19. Insurability is a risk's capacity to be insured — possessing the Chapter 1 characteristics well enough — *at a price someone will both charge and pay*. A risk can be perfectly **modelable** (we can compute the price) yet not **writable**, if that adequate price exceeds what the market or the regulator will bear. 20. Any three of: judgment under uncertainty; model literacy (reading/questioning/overriding the model); climate and catastrophe fluency; data judgment; communication/negotiation/trust; ethical reasoning. For example, *judgment under uncertainty* becomes more valuable because as the machine prices the routine, the human's worth concentrates in exactly the novel, contested, climate-stressed cases the model cannot handle — so demand for that scarce judgment rises.