Self-Insured & Captive

Public Entity General Liability IBNR: Municipalities, Schools, and Special Districts

How actuaries estimate unpaid general liability claims for public entities, why latent exposures and revival windows make the tail longer than it looks, and what to require in the report.

By Sam · April 2026 · 13 min read

If your municipality, school district, transit authority, or special district self-insures its general liability, your balance sheet carries a liability for claims that have already happened but have not yet been fully paid. The most uncertain piece of that liability is IBNR: incurred but not reported, which in actuarial practice means both the claims nobody has filed yet (pure IBNR) and the future development on claims your adjuster already knows about (IBNER). Together, those two components are what actuaries call broad IBNR, and for public entity general liability they present a problem that no other self-insured line quite matches: latent claims from decades-old conduct can surface years or even decades after the events that caused them, and state legislatures can reopen closed statute-of-limitations windows, turning a triangle that appeared complete into one that is suddenly developing again.

This article explains how actuaries estimate unpaid general liability claims for self-insured public entities, what makes the line different from workers compensation or commercial auto, where the standard methods break down, and what you should require in documentation. It is written for the finance director, risk manager, or pool administrator who receives a reserve study and needs to evaluate it, not for the actuary who produces it.

Why public entity GL is its own reserving problem

Public entity general liability shares the basic reserving machinery with other long-tail casualty lines: loss development triangles, cumulative development factors, Bornhuetter-Ferguson anchoring for recent years. But the line has characteristics that change how those tools behave and where they fail.

The tail is long. Bodily injury and property damage claims arising from government operations can take five to ten years to resolve through litigation, and that is the conventional tail. The unconventional tail is longer. Civil rights claims, sexual abuse allegations, police misconduct cases, and environmental contamination claims can emerge ten, twenty, or thirty years after the underlying conduct occurred. A school district that operated a facility in the 1990s may face abuse claims today under a revival-window statute enacted last year. That kind of latent exposure does not appear in any historical development pattern because the pattern was built from claims that followed the conventional reporting cycle.

The exposure base is unusual. Friedland identifies population or operating expenditures as the standard exposure base for public entity general liability (Friedland, p. 132). Neither is as clean as payroll (used for workers compensation) or vehicle count (used for commercial auto). Population changes slowly and does not capture the intensity of government operations: a city of 200,000 that operates a transit system, a jail, and a public hospital has a fundamentally different liability profile than a city of 200,000 with none of those operations. Operating expenditures are a better proxy for activity level but can shift dramatically with budget cycles, capital projects, and grant funding that may not correlate with claims-generating activity. The actuary building the expected-claims anchor must normalize for whichever exposure measure is used, and the buyer should understand which one was chosen and why.

Severity trend is high. Friedland uses 7.5% annual trend in a public entity GL example (p. 133), which is materially higher than general inflation and higher than the trend typically selected for workers compensation indemnity or commercial auto PD. That trend reflects the litigation environment surrounding government liability: rising jury awards in police misconduct cases, expanding theories of municipal liability, and plaintiff attorneys who specialize in public-entity claims. A trend selection that understates severity growth compounds across every future accident year in the expected-claims calculation, and the error is largest for the most recent years where Bornhuetter-Ferguson and expected-claims methods rely most heavily on the anchor.

The arithmetic in plain language

The core equation is the same one that applies to every casualty line:

Ultimate claims = paid to date + case outstanding + IBNR

If you are not familiar with that equation, the IBNR explainer walks through it in detail.

For public entity GL, the actuary typically builds triangles on a net basis (within the entity’s self-insured retention or pool retention) and may build separate triangles by claim type if the data supports it (for instance, separating law enforcement liability from premises liability or employment practices).

Consider a mid-sized city that self-insures its GL up to $1 million per occurrence. The 2024 accident year shows $4.2 million in reported losses (paid plus case outstanding) as of December 2025, at 24 months of development. The actuary’s selected development pattern says that reported GL losses at 24 months are historically about 50% of ultimate. The cumulative development factor (CDF, also called an LDF) is 2.00.

Projected ultimate = $4.2 million x 2.00 = $8.4 million

The broad IBNR for this year is $8.4 million minus $4.2 million = $4.2 million, fully half of the projected ultimate. At 12 months, the CDF might be 3.00 or higher, meaning two-thirds or more of the projected ultimate for the latest accident year is an actuarial estimate rather than an observed number.

Now compare this to a more mature accident year, say 2019, which has 84 months of development and $7.1 million in reported losses. At that maturity, the CDF might be 1.08, producing a projected ultimate of $7.67 million and an IBNR of only $570,000. The mature year is mostly settled; the recent year is mostly estimated. That asymmetry is exactly what the chain ladder article describes, and it is more pronounced for public entity GL than for shorter-tail lines because the CDFs at early maturities are larger.

The critical question is what happens beyond the oldest maturity in your triangle. If your triangle has fifteen years of history, the tail factor (the CDF from the oldest maturity to ultimate) picks up any remaining development. For public entity GL, that tail factor must account for latent claims that may not follow the conventional pattern at all, which is one reason actuaries often supplement triangle-based tail factors with industry benchmarks or explicit loadings for latent exposure.

Method selection by accident-year maturity

A competent public entity GL reserve study matches the method to the maturity of each accident year, following the same logic described in the five core methods overview.

Oldest accident years (eight or more years of development). Reported chain ladder is the primary method. At this maturity, most conventional claims are closed or have stable case reserves, and the historical development pattern is a reasonable guide. The actuary may cross-check with paid chain ladder, but reported is generally preferred because residual payments on mature GL claims (structured settlements, periodic defense invoices) can create noise in the paid triangle that does not reflect changes in ultimate value.

Middle accident years (three to seven years of development). The actuary typically weights chain ladder and Bornhuetter-Ferguson, with Bornhuetter-Ferguson receiving more weight for years closer to three. The Bornhuetter-Ferguson article explains the mechanics. For public entity GL, the BF method is especially valuable in this range because a single large police misconduct verdict or abuse settlement in one accident year can distort the chain ladder projection for that year without affecting the underlying loss rate across all years.

Most recent accident year (twelve months or less of development). Bornhuetter-Ferguson or expected claims, with the expected-claims anchor doing most of the work. At this maturity, paid data has almost no information (many GL claims are still in the investigation and litigation phase), and reported losses are dominated by initial case reserves that reflect early assessments, not final outcomes. The actuary’s judgment about the expected claim ratio, built from the selected exposure base, historical loss rates, and trend, is the primary driver of the number.

What breaks the estimate

Latent claims and revival windows

This is the failure mode that distinguishes public entity GL from nearly every other self-insured line. Latent claims (those arising from conduct that occurred years or decades before the claim is reported) do not follow the conventional reporting pattern embedded in the development triangle. A development triangle built from fifteen years of data assumes that claims are substantially reported within that window. When a state enacts a revival-window statute (temporarily lifting the statute of limitations for childhood sexual abuse claims, for example), a wave of new claims arrives for accident years the triangle treated as fully developed.

The practical effect is that completed accident years reopen. IBNR that was zero becomes materially positive. The actuary may need to add an explicit loading for latent exposure that sits outside the standard development-pattern projection, and that loading requires assumptions about claim frequency, severity, and the scope of the revival window that cannot be derived from the entity’s own triangle. Industry data, peer-entity experience, and legal analysis of the specific statute are the typical inputs.

If your state has enacted or is considering a revival-window statute, your actuary needs to know. If the report does not mention latent exposure, ask whether it was considered and, if excluded, what the rationale was.

Case reserve adequacy shifts

The same distortion described in the case reserve strengthening article applies to public entity GL. When a self-insured entity changes its claims administrator, brings claims handling in-house, or undergoes an adequacy audit, the reported triangle shows a calendar-year jump in case outstanding that the unadjusted chain ladder misreads as adverse development. The Berquist-Sherman case adjustment is the standard correction.

Litigation environment shifts

Changes in state tort-immunity statutes, qualified-immunity doctrine, damage caps, or venue rules can shift the severity distribution for public entity claims in ways that do not appear in the historical triangle. A state that narrows sovereign immunity or removes caps on non-economic damages changes the prospective severity profile without changing the development pattern for prior accident years. The actuary should identify any such changes that took effect during the experience period and explain how the trend selection accounts for them.

Claim mix changes

Public entities face a diverse range of claim types: premises liability, police professional liability (law enforcement), employment practices, vehicle liability (if not separated), and infrastructure failures. If the mix shifts (for instance, a city creates a new police oversight structure that changes the frequency or severity of law enforcement claims), the historical development pattern may no longer be representative. Where data volume permits, the actuary should segment the triangle by claim type rather than projecting a single blended triangle.

The self-insured wrinkle: pools, thin data, and shared retention

Most self-insured public entities participate in a risk pool, joint powers authority, or intergovernmental cooperative rather than standing alone. This pooling structure changes the reserving problem in several ways.

Data volume improves, but heterogeneity increases. A pool of fifty municipalities has more claims than any single member, which stabilizes the development pattern and reduces the leveraged-factor problem. But the fifty members may have very different operations (a city with a police department versus a water district with no law enforcement exposure), and a single pooled triangle blends those different severity profiles. The actuary needs to decide whether to project the pool as a whole (more data, more heterogeneity) or segment by member characteristics (less data, more homogeneity). Either choice involves tradeoffs, and the report should disclose which approach was used and why.

Member allocation requires a methodology. The pool-level reserve has to be allocated back to individual members for assessments, collateral requirements, and financial reporting. The allocation methodology (by exposure, by incurred losses, by a combination, or by actuarial projection for each member) affects each member’s financial statements. The group captive and RRG article covers the allocation problem in the captive context; the same tensions apply to public entity pools.

Per-occurrence retentions create layered reserving. Most pools carry a per-occurrence retention (say, $500,000 or $1 million) above which excess or reinsurance coverage applies. The actuary should project the limited (within-retention) triangle separately from the excess layer. The limited triangle is less volatile because large claims are truncated, and the excess layer can be reserved using expected-claims methods or industry benchmarks for public entity GL severity. If the report does not show a limited-loss analysis, ask why.

For entities that self-insure outside a pool, the thin-data problem is severe. A single municipality generating twenty GL claims per year cannot build a credible development triangle from its own experience. The actuary will rely heavily on Bornhuetter-Ferguson with an industry-based or peer-based expected-claims anchor, and the quality of that anchor is the single largest driver of the reserve estimate for recent accident years.

Trend: the assumption that drives recent years

For accident years older than seven or eight, the CDF is small enough that the trend assumption does not matter much. For the two most recent years, trend is often the largest single driver of the reserve because it feeds directly into the expected-claims anchor that Bornhuetter-Ferguson and expected claims rely on.

Public entity GL severity trend reflects the litigation environment surrounding government liability. Friedland’s example uses 7.5% annual trend for a public entity GL line, which incorporates rising jury awards, expanding theories of government liability, and increasing attorney involvement in public-entity claims.

Trend for public entity GL should be decomposed into at least two components:

Loss trend (indemnity and medical combined, reflecting changes in verdict size and settlement values) driven by jury behavior, statutory changes, and the plaintiff bar’s focus on government defendants.

ALAE trend (allocated loss adjustment expense, which covers defense costs and cost containment; the NAIC’s current terminology is DCC for defense and cost containment and A&O for adjusting and other) driven by defense counsel rates, litigation complexity, and the volume of claims requiring outside counsel.

A report that selects a single combined trend without disclosing the components is hiding the assumptions that matter most. If the combined trend is 6% but the loss component is 8% and the ALAE component is 3%, those components will not move together in the future, and a change in the litigation environment may require updating one without the other.

What a buyer should ask their actuary

These are the concrete questions that separate a buyer who understands the number from one who simply accepts it.

What exposure base are you using, and how is it normalized across accident years? Population and operating expenditures both have limitations. The actuary should explain which was selected, why, and how changes in the entity’s operations (new facilities, expanded services, closed programs) are reflected in the expected-claims calculation.
Is latent exposure included in the reserve, and if so, how is it quantified? If your entity has exposure to abuse claims, civil rights claims, or environmental liability, the reserve should either include an explicit loading for latent exposure or the report should explain why it was excluded. If your state has a revival-window statute, the loading should reflect the specific scope of that statute.
What severity trend is selected, and what is the basis? A 7.5% trend is materially different from a 4% trend over five or ten accident years of projection. The selection should be documented from program experience, pool experience, or an appropriate industry benchmark, with a clear rationale.
Are the triangles segmented by claim type, and if not, why? If the entity has material law enforcement liability, a blended triangle that combines police professional claims with premises slip-and-fall claims is mixing two fundamentally different severity distributions. Segmentation improves the projection where data volume supports it.
Is the estimate a central estimate, a management best estimate, or something else? ASOP 43 defines the actuarial central estimate as an expected value over reasonably possible outcomes. A management best estimate may be different. Know which one you are getting.

What to require in documentation

The documentation checklist consolidates the professional-standards requirements into a single reference. For public entity GL specifically, the report should include:

Paid and reported triangles at annual or semiannual evaluation intervals, with enough history to observe the full development pattern. If the entity participates in a pool, the report should clarify whether the triangles represent the pool aggregate, the member’s share, or both.
Method selection by accident year with a stated rationale for each selection and a comparison of chain ladder, Bornhuetter-Ferguson, and expected-claims results for every year.
Development factor selections showing the historical age-to-age factors, the averages considered, and the selected factor at each maturity, including the tail factor and the basis for the tail (program experience, industry benchmark, or a blend).
Trend selections decomposed into loss and ALAE components, with source documentation (program experience, pool data, industry benchmark, or blend).
Exposure base reconciliation confirming the selected exposure measure (population, operating expenditures, or other) by accident year, with adjustments for changes in the entity’s operations.
Latent exposure disclosure stating whether the reserve includes a loading for latent claims (abuse, civil rights, environmental), the basis for the loading, and any revival-window statutes that affect the entity’s exposure.
Operational change disclosure documenting any changes in claims administration, risk pool membership, coverage structure, retention levels, or litigation environment during the experience period.
Sensitivity analysis showing how the reserve changes under alternative trend, CDF, or expected-claims assumptions for the two or three most recent accident years, where most of the uncertainty is concentrated.
Diagnostic commentary explaining whether actual development since the prior study was consistent with what the prior analysis predicted, and if not, what changed and what the actuary did about it.

If the report does not include these elements, send the list back and ask for the missing sections. A report that projects all claim types together, selects a single trend without decomposition, and omits any discussion of latent exposure is not giving you the information you need to evaluate the number on your balance sheet.