Plaintiff win probability in US liability cases rose 20 to 30 percent from 2009 to 2024, and verdict awards more than doubled since 2020, according to a May 2026 peer-reviewed study of insurance outcomes (Fung, Ma, Peng, and Yang, arxiv 2605.27265). A single aggregate trend factor applied to historical loss triangles cannot isolate those components, and the CAS 2026 Reserves Call Paper's $15,000 prize program is directly funding methods to fill that gap.

Tracking casualty reserve development disclosures across 10-K filings since 2019 reveals a consistent pattern: carriers that separated social inflation from underlying trend in their stated reserve assumptions posted fewer negative development surprises in subsequent years, with the gap widest in commercial auto and umbrella. The arxiv study, authored by Tsz Chai Fung, Lie Ma, Liang Peng, and Fang Yang, is the first statistically rigorous cross-portfolio measurement of social inflation with actuarially usable output variables, and its publication timing lands precisely as the CAS Reserves Working Group closes its June 15, 2026 submission window for ML-based reserving methodology papers. The two developments converge on a single reserving question: what does the profession now owe a casualty reserve analysis that has been absorbing social inflation through a single undifferentiated trend factor for 15 years?

What the Chain-Ladder Method Cannot See

The headline number from the study is a 20 to 30 percent relative increase in plaintiff win probability from 2009 to 2024. The methodological value is in what the authors measured and how: rolling-window logistic regression applied to case outcomes across the full observation period, combined with quantile regression for severity and random-weighted bootstrap methods for uncertainty quantification. That approach breaks the aggregate signal into components a development factor cannot. Settlement probability declined more than 10 percent over the same period (Fung et al., May 2026), meaning the mix of cases that proceed to verdict has itself shifted adversely. The cases staying in litigation are not a stable cross-section of the portfolio; they skew toward the profiles that produce larger verdicts.

The study also found social inflation more pronounced in corporate-defendant cases and in states lacking tort caps or third-party litigation funding regulations. That jurisdictional heterogeneity is exactly what a single chain-ladder development factor erases. Standard development methodology applies a historical pattern estimated from the full portfolio aggregate to every accident-year tail, without attribution to the geographic or defendant-profile composition of that year's open case inventory. A carrier whose commercial auto and umbrella book is concentrated in three high-plaintiff-win-rate jurisdictions is not the carrier whose reserve adequacy is served by an industry-average trend factor, even if that factor accurately reflects the national picture.

The verdict severity result sharpens the reserve implication. Settlement amounts showed limited and often statistically insignificant inflation across the 2009 to 2024 period (Fung et al., May 2026), but verdict awards more than doubled since 2020. Paid loss triangles mix settlements and verdicts in proportions that vary by accident year and development lag. A triangle built on aggregate paid losses will understate social inflation's contribution to development velocity because settlements, the numerically dominant case resolution category, are not the primary carrier of the signal. The reserve actuary applying a single LDF to a paid triangle is, in effect, averaging together two populations with very different development profiles: the settled cases, which carry limited social inflation exposure, and the litigated-to-verdict cases, where the structural shift in plaintiff win probability and verdict magnitude concentrates.

Decomposing the Signal: Three Components Instead of One

A defensible social inflation decomposition in a casualty reserve analysis works across three dimensions. The first is litigation frequency on a given accident-year cohort, the proportion of claims that enter suit versus resolving without litigation. The second is the plaintiff win probability among litigated cases, the figure the arxiv study measured and found to have shifted materially. The third is severity conditional on a plaintiff verdict, the channel through which nuclear verdict inflation compounds the first two effects.

Separating these components requires data beyond the loss triangle. Lex Machina's court docket analytics, which tracked 17,654 insurance cases filed in federal district courts in 2023 (Lex Machina, 2024 Insurance Litigation Report), provide jurisdiction-level plaintiff win rate data that makes the second component operational. In business liability cases that proceeded to trial, claimants prevailed more than 55 percent of the time in Lex Machina's 2024 data. When all case outcomes including dismissals are counted together, defendants prevail three times more often, but that aggregate figure buries the trial-specific plaintiff win rate that drives reserve exposure on cases that reach verdict. The distinction between "all outcomes" win rates and "trial-only" win rates is exactly the segmentation a paid triangle cannot produce.

Nuclear verdict frequency and magnitude round out the severity picture. In 2024, 135 cases produced nuclear verdicts totaling $31.3 billion in awards, a 52 percent increase over the prior year. The median nuclear verdict reached $44 million, up from $21 million in 2020 (Marsh). Against that backdrop, a liability book with meaningful exposure to corporate-defendant litigation in high-plaintiff-win-rate jurisdictions carries a right-tail severity distribution that is structurally heavier than the historical LDF pattern would suggest. The LDF captures what actually developed in prior accident years; it does not capture that the underlying probability distribution of verdict outcomes has shifted during the observation period.

Data Sources Enabling the New Methods

The analytical gap between what actuaries need and what loss triangles provide is closing through three categories of external data. Lex Machina court docket data provides case-outcome statistics at the jurisdiction, case type, and defendant-profile level, updated regularly enough to track trends within an accident year rather than waiting for development to emerge in the triangle. Litigation finance disclosure filings are a newer source of leading-indicator data. Third-party litigation funding is forecast to exceed $30 billion as a US industry (Insurance Information Institute), and in the growing number of jurisdictions with disclosure requirements, the presence of litigation funding in a case predicts trial persistence and verdict magnitude well enough to serve as a case-level risk flag before development occurs.

Attorney advertising spend data by jurisdiction, compiled by providers including Kantar Media, completes the set. States with high attorney advertising intensity systematically produce higher rates of represented claims and more claims that escalate to suit. That geographic signal is observable years before the litigation it generates shows up as loss triangle development, which is precisely the leading-indicator property a social inflation reserve margin needs to be defensible rather than backward-looking.

The CAS 2026 Reserves Call Paper: Where the Prize Money Points

The CAS 2026 Reserves Call Paper Program describes its scope in language that maps directly onto the social inflation decomposition problem. The specific topic that most squarely addresses this: case studies on how LASSO, Generalized Additive Models, or Bayesian Markov chain Monte Carlo methods can be used to evaluate and reflect the effect of changes in loss cost or inflation trends, and changes in company operations for underwriting or claims (CAS, 2026). That framing is a description of the social inflation signal-decomposition challenge, not an abstract modeling exercise.

Gradient boosting (XGBoost), neural networks via PyTorch and TensorFlow, Bayesian MCMC, LASSO regression, and GAMs are all named explicitly in the program's list of methods it wants applied to reserving. The $15,000 prize fund covers papers accepted for presentation at the Casualty Loss Reserve Seminar, September 14 to 16, 2026 (CAS, 2026). The evaluation criteria prioritize originality and practical contribution to reserving literature. The underlying methodology gap the program is addressing: development factors estimated from historical aggregate triangles carry an implicit assumption that the drivers of loss development velocity are stable, when the arxiv study's output shows a 20 to 30 percent shift in plaintiff win probability over 15 years is precisely not stable.

A Bayesian MCMC approach to development factor estimation allows the actuary to model the posterior distribution of loss development velocity as a function of observable covariates, including jurisdiction-level plaintiff win rate data, litigation finance proxy variables, and accident-year vintage, rather than treating development factors as fixed parameters estimated from historical averages. A GAM with jurisdiction and defendant-profile spline terms can capture the nonlinear relationship between legal environment characteristics and development velocity. The CLRS September 2026 sessions will be the first major public forum where practitioners can compare these approaches against real portfolio data and production-quality loss triangles.

Reserve Margin Implications: Calibrating the IBNR Load

The reserve margin question becomes concrete once the structural shift is quantified. The joint CAS and Insurance Information Institute study on social inflation and loss development estimated it added more than $20 billion to US commercial auto liability claims costs between 2010 and 2019 (CAS and Insurance Information Institute, 2022). That is a cumulative figure across the period before the 2020 verdict severity acceleration began. The post-2020 dynamic, with verdicts more than doubling and nuclear verdict frequency rising 52 percent in 2024 alone, suggests the rate of social inflation accumulation accelerated precisely when many reserve triangles were being disrupted by pandemic-related claim closure delays that complicated LDF selection on its own.

For a carrier with meaningful commercial auto, GL, or umbrella concentration in high-plaintiff-win-rate jurisdictions, the appropriate IBNR load is not the industry-average social inflation trend. The arxiv study's finding that the shift is more pronounced in corporate-defendant cases and in states without tort caps or litigation finance restrictions means a book dominated by personal auto in tort-reform states carries substantially lower social inflation exposure than a book of commercial trucking or construction GL in states like California, Florida, or New York. Applying an industry-aggregate trend to either book produces a wrong answer; the question is which direction the error runs for a given carrier's specific portfolio composition.

The actuarially defensible approach under a persistent 20 to 30 percent step-up in plaintiff win probability is to construct the IBNR margin on a component basis: an explicit load for litigation frequency trend, a separate load for plaintiff win probability trend calibrated to the book's jurisdictional footprint, and a severity tail load reflecting the changed distribution of verdict outcomes for cases that reach trial. This three-component structure creates an explicit sensitivity: if plaintiff win probability reverts toward pre-2020 levels, the corresponding margin component releases; if it holds or rises, the margin is justified by the same observable data used to set it. A single undifferentiated trend factor cannot offer that transparency, and reserve actuaries defending IBNR adequacy in an adverse development environment increasingly need to be able to explain the mechanism, not just the aggregate trend selection.

Cross-Line Velocity: Commercial Auto First, Umbrella Behind It

Commercial auto liability carries the highest social inflation velocity across casualty lines because litigation frequency relative to premium volume is high, defendant profiles are predominantly corporate, and the line does not benefit from the tort reform provisions that partially offset social inflation in some GL classes. The CAS/III 2022 study identified commercial auto as the steepest accumulator of social inflation among the major lines it examined, with similar directional trends in other liability occurrence and medical malpractice.

Umbrella accumulates social inflation through the attachment point mechanism: as nuclear verdicts push more cases above primary limits, umbrella attachment frequency climbs even when the primary line's own frequency is unchanged. A commercial auto primary limit that adequately covered virtually all verdicts in 2018 may now be pierced routinely in the jurisdictions where median nuclear verdicts have reached $44 million, triggering umbrella layers on cases that would historically have closed within the primary policy. E&O and D&O lines show a similar dynamic at higher severity thresholds, where corporate-defendant status and sophisticated plaintiff litigation teams produce the same plaintiff-win-probability shift the arxiv study measured across the broader liability portfolio.

The cross-line reserve implication is portfolio-level, not line-by-line. A carrier writing commercial auto, GL, umbrella, and E&O on the same corporate account concentrations that produce both high plaintiff win rates and significant verdict magnitude carries a correlated social inflation exposure that individual-line reserve analyses may understate. The jurisdictional concentration of that exposure, not the line mix alone, determines the aggregate IBNR adequacy question.

What Casualty Actuaries Should Do Now

The convergence of the arxiv study's statistical output and the CAS prize program's methodology scope creates a practical near-term agenda. Reserve actuaries managing commercial auto, GL, and umbrella through the second half of 2026 now have a peer-reviewed quantification of the plaintiff-win-probability shift to cite when defending an explicit social inflation margin above the historical trend. The CAS CLRS in September 2026 will produce the first published ML-based development methodologies applied to real reserve triangles, and those papers will serve as methodological authority for actuarial opinions that separate social inflation from underlying trend.

The near-term work is data infrastructure: identifying which components of the social inflation decomposition the existing data sources can support, which require external data acquisition (Lex Machina, litigation finance disclosures), and which require triangulation from industry benchmarks pending internal data development. Carriers that have not begun building jurisdiction-level plaintiff win rate profiles into their reserving frameworks are operating on an assumption of development stability that the arxiv study's 15-year measurement now formally contradicts. The ones that close that gap before the CLRS papers land in September will be in a defensible position; the ones waiting for a vendor product to package the methodology may find the examination cycle arrives first.

Further Reading

Sources

  1. Fung, Ma, Peng, and Yang: Quantifying Social Inflation in Liability Insurance with Advanced Statistical Methods (arxiv 2605.27265, May 2026)
  2. CAS: 2026 Reserves Call Paper Program on Improved Methodologies and Technologies for Reserving (Casualty Actuarial Society, 2026)
  3. CAS: Casualty Loss Reserve Seminar 2026 (September 14-16, 2026)
  4. CAS and Insurance Information Institute: Social Inflation and Loss Development (2022)
  5. Lex Machina: 2024 Insurance Litigation Report (LexisNexis, 2024)
  6. Marsh: Social Inflation and Nuclear Verdicts (median nuclear verdict data)
  7. NAIC: Insurance Topics, Social Inflation
  8. INSURICA: Social Inflation and Nuclear Verdicts, Definitions, Trends, Facts, and Figures (third-party litigation funding market size)