Case Study 2: When the Math Is Skipped — The Small-Sample Book That Blew Up

DataField.Dev

Case Study 2: When the Math Is Skipped — The Small-Sample Book That Blew Up

What this case is. A study of the chapter's failure mode: what happens when underwriters price off raw, immature, low-credibility loss data and let a small sample — or a falsely flattering loss ratio — drive a book of business. The specific account below is a clearly labeled composite, assembled from a pattern that has recurred across real soft markets and real long-tail lines in the U.S. property-casualty industry. The pattern is thoroughly documented in the public record — the long-tail reserve deficiencies that surfaced after the soft markets of the late 1990s and early 2000s, the way immature loss ratios flatter a growing book, the recurring "underwriting cycle" in which discipline erodes and losses arrive years later (Chapters 3 and 11 own the cycle). No specific company, statistic, combined ratio, or reserve figure is asserted as a real fact; the numbers are constructed to make the mechanics legible.

Background: a hungry book in a soft market

Picture a mid-size carrier that decides to grow its commercial general liability book for contractors and small manufacturers — a long-tail line (claims take years to settle and develop, §10.4) — during a soft market, when capacity is plentiful, competitors are cutting price, and brokers shop every renewal. The growth target is aggressive. The underwriting team is rewarded on premium volume and loss ratio as reported today. New submissions pour in, many of them accounts being non-renewed or re-shopped by other carriers precisely because they are marginal — a quiet adverse-selection pressure (Chapter 1) that nobody flags because the loss runs "look fine."

And they do look fine, which is the trap. Most of the incoming accounts are small to mid-size, with only a year or two of usable experience each. On their own raw, recent loss runs, the loss ratios look low — because, as §10.2 and §10.4 warned, the premium is freshly written and not yet fully earned, the claims are paid to date and not yet incurred to ultimate, and a long-tail line's losses develop upward for years. The book reports a gorgeous loss ratio in its first eighteen months. Management celebrates. The appetite widens. The book doubles, then doubles again.

The insurance / underwriting issue: every discipline in the chapter, skipped

Walk the failure against this chapter section by section, because the book broke each rule in turn:

Loss ratio computed on the wrong basis (§10.2). The reported loss ratio leaned on paid losses over written premium — the most flattering and least honest combination (the §10.2 table). On an incurred-over-earned basis the picture would already have looked worse; the carrier was, in effect, reading the "30% darling" column of a book that was really heading for the "83% problem" column.

THE SAME BOOK, TWO READINGS — illustrative           [constructed composite]

   as reported (paid / written, immature)  ......  loss ratio looks ~40%   → "grow it!"
   as it truly was (incurred / earned, developed)  loss ratio headed ~95%+ → "stop!"

   The gap was not fraud. It was the difference between the number that flatters a young,
   growing, long-tail book and the number that tells the truth (§10.2, §10.4).

No trend, no development (§10.4). Recent accident years were priced at face value. Nobody multiplied the immature losses by development factors to estimate ultimates, and nobody trended severities forward for the inflation and rising verdicts that were quietly running through liability costs. So the priced loss cost was systematically too low — pricing the past, and an under-reported version of the past at that.
Credibility ignored — in the wrong direction (§10.5–10.6). Here is the cruel twist. The classic credibility error is to over-react to a small sample. This book made the opposite error in a way that was just as fatal: it took each account's thin, flattering recent experience as good news worth pricing on — effectively assigning the accounts' own (low-credibility, immature) experience high weight when it pointed down, while ignoring what the class experience and the broader market were screaming (that contractor GL was deteriorating). Credibility discipline cuts both ways: a small sample that looks good deserves exactly as little weight as one that looks bad. The book trusted favorable noise.
Adverse selection unmanaged (Chapter 1). Because the price was soft and the screening loose, the accounts that found the carrier most attractive were disproportionately the ones other carriers had pushed away. The pool filled with the very risks the low price was least adequate for.

What it shows

This is the chapter's argument in the negative. Trend, development, and credibility are not bureaucratic steps; they are the only things standing between a flattering number and an honest one. Skip them and a young long-tail book will always look better than it is, for a predictable window of one to three years — long enough for the appetite to open, the volume to balloon, and the underwriters to be praised — and then the losses develop and trend right through the inadequate premium, on schedule. The blow-up is not bad luck; it is arithmetic deferred.

It also shows the asymmetry that makes the trap so seductive: doing the math always makes your number higher than the competitor who skips it. The disciplined underwriter who trends, develops, and credibility-weights produces a less competitive quote and loses business in the soft market — and looks worse on volume in the short run, precisely while the undisciplined book is winning awards. The discipline costs you in exactly the years it is hardest to defend, and pays off only later, when the undisciplined book is hemorrhaging and yours is not. That is why rate adequacy (Chapter 11) is called the hardest discipline in insurance.

Outcome

In the documented real-world versions of this pattern, the ending is consistent even though the details vary: two or three years in, the reserves on the long-tail book develop adversely — actuaries strengthen them as claims mature and late ones report — the reported loss ratio lurches upward toward the truth, the combined ratio crosses well above 100% (Chapter 3), and the carrier reacts hard: it raises rates steeply, tightens appetite, non-renews swaths of the book, and sometimes exits the line. The underwriters who were rewarded for the growth are often gone or reassigned by the time the losses surface — one of the structural reasons the cycle repeats, because the incentives reward the early volume and punish the later losses on a different watch. The market as a whole hardens in response, prices rise across the industry, and the next soft market sets the stage to do it all again.

Lesson

The lesson pairs exactly with Case Study 1 and completes the chapter. Case Study 1 (the X-mod) showed credibility done right, hard-wired into a filed plan that refuses to over-trust a small sample. This case shows credibility skipped — and skipped, fatally, in the flattering direction, with the trend and development adjustments abandoned alongside it. A loss ratio is only as honest as the premium and losses you put in it; a pure premium is only as honest as the trend and development behind it; and a price is only as honest as the credibility weighting that decides how much the data is allowed to say. The underwriter who internalizes this chapter will distrust any loss ratio that looks too good on a slow-paying line, will demand trended and developed numbers before quoting, and will weight a thin, favorable sample exactly as skeptically as a thin, unfavorable one. That skepticism is not pessimism. It is the discipline that keeps you solvent when the soft market is rewarding everyone around you for abandoning it.

Discussion questions

The book's reported loss ratio used paid losses over written premium. Reconstruct, using the §10.2 table, why that combination flatters a young, growing book, and state the combination that would have told the truth.
This case calls the credibility error "ignored in the wrong direction." Explain how trusting a thin, favorable sample is the same statistical mistake as over-reacting to a thin, unfavorable one — and why the favorable version is more dangerous in practice. (§10.5–10.6)
Doing trend, development, and credibility "always makes your number higher" than a competitor who skips them. In a soft market, how does this asymmetry punish the disciplined underwriter in the short run and reward them later? Tie your answer to rate adequacy (Chapter 11) and the combined ratio (Chapter 3).
The underwriters who grew the book were often gone before the losses surfaced. What does this say about how incentives — not ignorance — drive the underwriting cycle, and what could a carrier change to align them with long-run results? (§10.4; the cycle, Chapter 3)
Contrast this case directly with Case Study 1. The X-mod is a filed formula that enforces credibility discipline automatically; this book relied on individual underwriters' judgment and discipline failed. What does the contrast suggest about where credibility should be embedded in a rating plan versus left to judgment? (§10.5–10.7)