Casualty lines posted $15.8 billion in adverse prior-year development in 2024, the highest level on record and equal to 5.9% of prior reserves, as nuclear verdicts jumped 52% in count to 135 cases and 116% in dollar terms to $31.3 billion (Milliman, 2026; Marathon Strategies, May 2025). That single accident-year shock is now sitting inside every carrier's loss triangle, distorting the development factors actuaries use to project 2024, 2025, and 2026 accident years forward.

Reviewing reserve actuarial opinions for carriers with excess casualty exposure across three consecutive accident years, the most consistent gap is not a wrong number. It is actuaries citing loss development factor instability in the 36-to-48 month window without quantifying what that instability does to the ultimate estimate, and without disclosing what methodological adjustment, if any, was made to compensate. That gap matters more in 2026 than it did in 2022, because the rate cushion that used to absorb a reserving miss is shrinking at the same time the triangles are getting harder to read.

How One Bad Diagonal Poisons a Triangle

A loss development triangle organizes historical claims by accident year down the rows and maturity age across the columns. Age-to-age (link) factors are typically selected as a weighted or simple average of the ratios observed across several historical accident years at each maturity, on the assumption that the process generating those ratios is stable over time. Nuclear verdict activity breaks that assumption in a specific, mechanical way: it does not raise severity evenly across the triangle, it concentrates in the diagonal representing the current calendar period, because that is when the verdicts, settlements struck in their shadow, and case reserve strengthenings actually get booked.

When a carrier's excess casualty or commercial auto book experiences an unusually severe run of verdicts in calendar year 2024, that shows up as an outsized age-to-age factor on the 2024 diagonal, wherever each accident year happens to sit in its own development. An accident year 2021 claim maturing from 36 to 48 months in calendar 2024 and an accident year 2023 claim maturing from 12 to 24 months in the same calendar period both pick up the same verdict-driven severity shock, even though they sit in different rows and columns of the triangle. If the actuary then averages historical link ratios across accident years to select the 36-to-48 month factor for a fresh accident year, that single contaminated diagonal pulls the selected factor upward for every accident year evaluated afterward, including ones where the actual verdict exposure has not yet emerged in the data because those claims have not reached that maturity age yet.

The CAS's Jim Lynch and Dave Moore worked through the mechanics of this bias for the Bornhuetter-Ferguson method specifically: when link ratios accelerate from one accident year to the next, multiyear-average cumulative development factors come out understated, and because industry data shows roughly a three-year lag between loss cost increases and the premium increases that follow them in commercial auto liability, the a priori expected loss ratio used in BF is also biased low (CAS Actuarial Review, March 2023). Both inputs to the method move in the same wrong direction at once, which is why a BF reserve built on unadjusted historical selections tends to miss low precisely in the years verdict severity is accelerating, not merely drift.

Milliman's year-end 2024 statutory data shows where that bias has already landed. Other liability occurrence carried $10.0 billion of adverse development against $150.8 billion of prior reserves, a 6.6% miss; commercial auto added $3.8 billion against $66.7 billion, 5.7%; non-proportional reinsurance liability contributed $1.7 billion against $37.1 billion, 4.6% (Milliman, 2026). The largest percentage misses by accident year sit in the 2015-2019 soft-market cohort, with accident year 2017 alone running 12% adverse relative to its prior carried reserves, evidence that today's contamination problem has been compounding since well before the 2024 verdict spike made it visible.

The Outlier Treatment Dilemma

The most direct response to a triangle skewed by a handful of extreme verdicts is to treat those claims as outliers: cap them at a threshold, exclude them from the link-ratio calculation, and reserve for them separately through a scenario load or excess-layer analysis. The appeal is real. A capped triangle produces smoother, more credible-looking development factors, and it isolates a genuinely different loss-generating process, litigation-funded, venue-selected, nuclear-verdict severity, from the attritional claims the chain-ladder method was designed to project.

The problem is that capping removes exactly the risk the reserve exists to cover. Thermonuclear verdicts, those exceeding $100 million, hit a record 49 cases in 2024, up from 27 in 2023, and five verdicts topped $1 billion, more than double the two seen in 2023; the median verdict itself climbed to $51 million from $44 million a year earlier (Marathon Strategies, May 2025). An actuary who excludes every claim above, say, a $10 million cap is not removing noise from the triangle. Given that 135 verdicts exceeded $10 million in 2024 alone across 34 states and 77 courts, that actuary is removing the specific tail event that the excess and umbrella layers were priced and reserved to absorb, and pushing the estimation problem entirely into a separate load that carries its own, often less rigorous, selection process.

The defensible version of outlier treatment keeps both pieces visible rather than discarding one. Cap or Winsorize the triangle for link-ratio selection purposes, document the threshold and the rationale (a percentile of historical severity, a multiple of attachment point, or a fixed dollar level tied to layer structure), and then reserve the excluded severity through an explicit large-loss or scenario-based load calibrated to the actual frequency and severity distribution of nuclear verdicts in the relevant venue and line, not to a single averaged multiplier. The Statement of Actuarial Opinion should show the capped ultimate, the large-loss load, and the sum, so a reviewer can see what was excluded from the triangle and where it came back in. An opinion that shows only the final blended ultimate, without that decomposition, is the pattern that recurs across the excess casualty files reviewed here: the adjustment happened, but nothing in the workpapers lets a second actuary reconstruct how much of the ultimate came from the triangle and how much came from judgment layered on top.

Methods That Reduce Reliance on a Contaminated Triangle

Three method families give an actuary more control over how much weight a contaminated triangle carries in the final ultimate, and each fits a different casualty line differently.

Method Mechanism Best fit
Bornhuetter-Ferguson Blends an a priori expected loss ratio with reported development; limits how far a single contaminated diagonal can move the ultimate for immature years Commercial general liability, where premium and exposure data support a credible a priori pick
Cape Cod Derives the expected loss ratio from the triangle's own reported experience weighted by used-up exposure, rather than an external pricing assumption Commercial auto liability, where recent rate actions make an external a priori assumption unreliable but exposure data is credible
A-priori-first (frequency × severity) Builds the ultimate from separately trended frequency and severity assumptions rather than aggregate link ratios, isolating the verdict-severity shock as its own trend component Excess casualty and umbrella, where a handful of large claims dominate the triangle and frequency is too thin for chain-ladder credibility

For commercial general liability, Bornhuetter-Ferguson remains the workhorse specifically because it caps the influence of a contaminated triangle on immature accident years, provided the a priori loss ratio is not itself derived from the same contaminated history. The corrective the CAS analysis recommends is direct: select link ratios from the most recent development year or an extrapolated trend rather than a multiyear average, and make an explicit upward adjustment to the a priori loss ratio for expected social inflation rather than trusting an average of trend-lagged historical loss ratios (CAS Actuarial Review, March 2023).

For commercial auto liability, Cape Cod has gained ground precisely because BF's weakness, an a priori loss ratio anchored to a pricing assumption, becomes a bigger liability when rate actions have moved fast and recently. Cape Cod instead derives its expected loss ratio from the triangle's own used-up-exposure-weighted experience, which sidesteps the question of whether last year's filed rate indication was itself adequate. It does not, however, escape triangle contamination on its own; the same capped-and-loaded treatment of extreme verdicts described above still needs to happen before Cape Cod's exposure weighting is applied, or the method simply launders the same contaminated severity through a different weighting scheme.

For excess casualty and umbrella books, where claim counts are too thin to support a stable chain-ladder pattern at all, an a-priori-first approach that separates frequency and severity trend assumptions gives the actuary a place to isolate the verdict-driven shock explicitly, rather than letting it hide inside a blended aggregate link ratio. Milliman's Brown and Julga, writing for Insurance Thought Leadership from full-year 2024 data, found that commercial-lines-heavy carriers with limited workers' compensation exposure posted 2.6% adverse development for accident years 2022 and prior, against 0.3% for carriers diversified across commercial, workers' comp, and personal lines, and roughly 3% favorable development for personal-lines-and-workers'-comp-heavy cohorts (Insurance Thought Leadership, January 2026). Their recommendation tracks the same logic: segregate lines granularly enough to separate litigated from non-litigated claims, and reevaluate expected loss ratios against current claim activity rather than historical adverse development alone.

Why the Q3 2025 Cross-Tabulation Matters

The same Milliman analysis found that within its "adverse" cohort, 49% of companies actually reported favorable development, and within its "favorable" cohort, 41% reported adverse development. That cross-contamination inside supposedly clean groupings is itself a warning against relying on peer-cohort benchmarking as a substitute for company-specific triangle diagnosis. A carrier cannot assume its reserve position tracks its line-of-business mix; the dispersion within each cohort is nearly as wide as the dispersion between them.

Quantifying and Communicating Reserve Uncertainty

When historical development patterns are structurally unreliable, a defensible range of ultimate estimates has to be built from the specific sources of that unreliability rather than a generic confidence interval around the point estimate. In practice that means running the ultimate under at least three explicit scenarios: the unadjusted chain-ladder projection (a reasonable upper bound on how much weight the raw triangle would assign to recent verdict severity if extrapolated forward), the capped-triangle-plus-large-loss-load estimate described above, and a BF or Cape Cod estimate with an a priori loss ratio adjusted for the CAS's recommended social inflation loading. The spread across those three is the actuary's actual measure of parameter uncertainty introduced by verdict contamination, and it should be disclosed as a range, not averaged into a single point that hides how much the selection judgment is driving the number.

Communicating that range to management and the board carries its own risk. A range that is too wide invites second-guessing of the actuarial function's competence; a range presented as a single confident point when the underlying triangle is known to be unreliable creates a disclosure exposure if the eventual outcome falls outside it. The workable middle ground is to present the point estimate as the primary carried reserve while separately quantifying, in the actuarial report rather than the boardroom summary, the dollar distance between the chain-ladder-implied ultimate and the adjusted-method ultimate. That distance is a direct, defensible measure of how much of the carried reserve depends on the actuary's judgment about verdict contamination rather than on the triangle's raw arithmetic, and it is the number a regulator or auditor will ask for first if development continues adverse.

The Soft-Cycle Amplification

Reserve under-estimation is more dangerous in 2026 than it would have been during the 2020-2023 hard market, because rate adequacy is declining at the same time the reserving problem persists. Analysts covering the sector project a 96% to 97% combined ratio for 2026, reflecting a normalizing catastrophe year and materially lower favorable reserve development than the outsized releases that flattered 2025 results (Reinsurance News, 2026, citing Fitch and Aon analysis). Aon's own 2026 outlook data shows general liability rates rising 5.6% in the fourth quarter of 2025 with a forecast climb to 9% in the first quarter of 2026, and auto liability rates up 9.2% in the fourth quarter of 2025 with 7% to 15% forecast for the following quarter (Aon, 2026 P&C Outlook), rate increases that are themselves a lagging response to the verdict severity already booked, not a forward-looking buffer against verdicts still to come.

The mechanism is compounding, not additive. A reserving miss on accident years 2021-2023 forces a correction that flows through current-year loss picks; a softening current-year rate environment, meanwhile, means the margin available to absorb that correction without a combined-ratio hit is thinner than it was in 2022. Aon's data shows the average lead umbrella limit carriers deploy has fallen to $10 million from $20 million in 2019, even as the median verdict for top U.S. casualty cases climbed to $98 million in 2024 from $49.7 million in 2019 (Aon, 2026 P&C Outlook). Primary and lower excess layers are absorbing a larger share of severity that used to sit higher in the tower, which means the loss development factor problem described above is migrating down into layers with thinner historical loss experience and less credible triangle data of their own.

How Reinsurers Are Responding at July 2026 Renewals

Casualty reinsurers are pricing this dispersion directly rather than averaging across it. Howden Re's mid-year renewal report found ceding commissions held flat overall at the July 1, 2026 casualty and financial lines renewals, but with a 43-percentage-point calendar-year loss ratio gap between the top and bottom quintile of carriers writing other liability business between 2021 and 2025, 59% for the top quintile against 102% for the bottom (Howden Re, June 2026). Reinsurers rewarded cedants that could demonstrate stronger underlying portfolio performance and applied pricing pressure where loss trends had visibly deteriorated, which functions as a real-time market test of exactly the reserve-quality question actuaries are wrestling with internally: cedants whose triangles show credible, well-documented adjustment for verdict contamination are getting priced on their own merits, while cedants with opaque or unadjusted reserve positions are being priced to the market's assumption of the worst case.

That dynamic is showing up inside individual carrier results as well as at the treaty level. W.R. Berkley's first-quarter 2026 filing reported $8 million of adverse prior-year development in its Insurance segment driven by umbrella and excess liability claims concentrated in accident years 2019 through 2023, with the company attributing part of the pressure to social inflation in auto-related claims and a rising frequency of litigated claims driving both indemnity and loss adjustment expense beyond expectations (W.R. Berkley Corp, 10-Q, Q1 2026). RLI Corp's casualty segment moved the other direction in the same quarter, releasing $14 million of favorable prior-year reserves concentrated in accident years 2018, 2019, 2021, 2022, and 2025, led by executive products, professional services, and commercial transportation, even as its personal umbrella book posted its own adverse development (RLI Corp, 10-Q, Q1 2026). Two carriers, the same quarter, opposite signs, both concentrated in the same 2018-2023 accident-year band: that dispersion is the market-level evidence that segment mix, and the rigor of each carrier's method choice within its own triangles, now separates winners from losers more than any industry-wide trend assumption could.

Why This Matters

The actuarial response to a contaminated triangle is not a single formula swap. It is a documented chain of decisions: which claims get capped out of the link-ratio calculation and why, how the excluded severity is loaded back in, which method's a priori assumption is least exposed to the same contamination the triangle shows, and how the resulting range of estimates gets communicated without either overstating false precision or inviting a solvency-adequacy challenge. Carriers entering 2026 with rate increases that lag the verdict severity already on their books, and with reinsurers now pricing cedant-specific reserve transparency directly into treaty terms, do not have the cushion to treat that chain of decisions as a footnote. The gap this analysis keeps finding, actuaries who name the LDF instability without quantifying its dollar impact or disclosing the adjustment made, is the gap regulators and reinsurers are both now positioned to price against.

Further Reading

Sources