Self-Insured & Captive

Workers Compensation IBNR for Self-Insured Employers

How actuaries estimate unpaid workers compensation claims for self-insured programs, what drives the number, and what to require in the report.

By Sam · April 2026 · 13 min read

If your organization is self-insured for workers compensation, you carry a liability on your balance sheet for claims that have already happened but have not yet been fully paid. Part of that liability is IBNR: incurred but not reported, which in actuarial practice means both the claims nobody has told you about 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 they can represent half or more of a self-insured employer’s total workers compensation reserve for recent accident years.

This article explains how actuaries estimate that number for self-insured workers compensation programs, what makes the line different from other casualty lines, where the standard methods break down, and what you should require in documentation. It is written for the risk manager, CFO, or benefits director who receives a reserve study and needs to evaluate it, not for the actuary who produces it.

Why workers compensation is its own reserving problem

Workers compensation shares the basic reserving machinery with general liability and auto liability: 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. A workers compensation claim can take more than fifteen years to settle. Permanent disability claims, lifetime medical awards, and Medicare set-asides keep files open long after the injury occurred. That tail length means the actuary must select development factors and tail factors for maturities where the triangle has little or no data, and those selections carry a disproportionate share of the judgment in the review.

The exposure base is payroll. Friedland identifies payroll as the standard exposure base for workers compensation (Friedland, p. 132). That means the expected-claims anchor in a Bornhuetter-Ferguson or expected-claims analysis is built from payroll dollars times a pure premium rate (or an expected claim ratio applied to standard premium). If your payroll has shifted materially between accident years because of headcount changes, acquisitions, or reclassifications between class codes, the expected-claims anchor has to be adjusted accordingly. A flat expected-claims ratio applied to a growing payroll base without normalizing for mix is one of the most common silent errors in a self-insured reserve study.

Trend has three components. Unlike a short-tail property line where trend is mostly inflation, workers compensation trend decomposes into indemnity trend (driven by wage inflation and benefit-level changes), medical trend (driven by medical CPI, fee schedule changes, and utilization), and allocated loss adjustment expense trend (driven by defense and cost-containment costs). Each component can move independently, and a report that selects a single blended trend without disclosing the components is hiding the assumption that matters most for your most recent accident years.

Operational changes are frequent. Self-insured employers change TPAs, implement return-to-work programs, restructure medical management networks, shift opioid protocols, and respond to legislative expansions (PTSD presumption laws, for instance) more often than a typical insurer changes its claims operation. Each of those changes can distort the historical development pattern that the chain ladder rides forward.

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 workers compensation, the actuary builds two triangles (sometimes more): one for paid losses and one for reported losses (paid plus case outstanding). Each triangle organizes your claims by accident year on the rows and evaluation date on the columns.

Suppose your 2024 accident year shows $3.5 million in reported losses as of December 2025 (24 months of development). The actuary’s selected development pattern says that, historically, reported losses at 24 months for your program are about 65% of their ultimate value. The cumulative development factor (CDF, also called an LDF) is the reciprocal: 1.538.

Projected ultimate = $3.5 million x 1.538 = $5.38 million

The broad IBNR for this year is the gap: $5.38 million minus $3.5 million = $1.88 million.

Now look at the 2025 accident year, which has only 12 months of development and $1.8 million in reported losses. At 12 months, the CDF might be 2.50 (meaning 12-month losses are historically about 40% of ultimate). The projected ultimate is $1.8 million x 2.50 = $4.5 million, and the IBNR is $2.7 million, which is 60% of the projected ultimate.

The pattern is the same one the chain ladder article describes: the more immature the accident year, the more the CDF does the work, and the more sensitive the projection is to the factor the actuary selected. For workers compensation, the 12-month CDF is typically larger than for a shorter-tail line like auto physical damage, which means more of your most-recent-year reserve is an actuarial estimate rather than an observed number.

Method selection by accident-year maturity

A competent workers compensation reserve study does not use one method across all accident years. It matches the method to the maturity and data quality of each year. The standard approach, drawn from Friedland and confirmed by practice, looks like this:

Oldest accident years (seven or more years of development). Chain ladder on the reported triangle is the workhorse. At this maturity, most claims are closed or have well-established case reserves, and the historical development pattern is a reliable guide. The actuary may still run paid chain ladder as a cross-check, but reported is generally preferred because the case outstanding for mature years is a better signal than the timing of residual payments.

Middle accident years (three to six 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: the method anchors the unreported portion to an expected claim estimate rather than riding the leveraged development factor forward, which stabilizes the projection when the data is still volatile. For a self-insured program with fewer claims than a carrier book, this stabilization is more important, not less, because a single large permanent-disability claim hitting the paid triangle can swing the chain ladder projection by hundreds of thousands of dollars.

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, the paid triangle has almost no information (many claims are still in the medical-only and initial-indemnity phase), and the reported triangle is dominated by initial case reserves that may or may not reflect the year’s true severity. The actuary’s judgment about the expected claim ratio, built from payroll, class-code mix, and historical loss rates, is the primary driver of the number.

This is exactly the maturity-based method selection that the five core methods overview describes in general terms. Workers compensation just makes it more consequential because the tail is longer and the CDF for the most recent year is larger.

What breaks the estimate

Every reserving method makes assumptions, and when those assumptions fail, the method still produces a number. It is just the wrong number. Workers compensation programs are especially prone to four categories of operational change that invalidate the historical development pattern.

TPA transitions

When a self-insured employer changes its third-party administrator, the new TPA typically reviews every open file and resets case reserves to its own standards. If the new TPA reserves more aggressively than the old one, the reported triangle shows a sudden jump in case outstanding that looks like adverse development but is really a one-time adequacy shift. The unadjusted reported chain ladder interprets that jump as a permanent change in loss level and projects it forward, overstating the ultimate. The case reserve strengthening article walks through this distortion in detail.

The fix is either a Berquist-Sherman case adjustment (restating the historical triangle to the new TPA’s adequacy level) or, where the data permits, relying on the paid triangle for the transition period instead. Either way, the actuary needs to know the full timeline of TPA changes during the experience period, and that information has to come from you.

Medical management and return-to-work programs

A new medical management vendor, a pharmacy benefit carve-out, or a structured return-to-work initiative changes the payment pattern even if it does not change the ultimate cost. Claims close faster, medical payments shift earlier in the life of the claim, and the paid triangle looks more mature than it really is. The paid chain ladder underestimates ultimate because it reads faster payment as higher completion. Conversely, if a program is dismantled or loses effectiveness, claims slow down and the paid chain ladder overestimates ultimate because it reads slower payment as less completion.

The actuary should document any known medical management or return-to-work changes during the study period and explain how they were addressed.

Legislative and regulatory changes

Workers compensation benefits are set by statute, and legislatures change them. PTSD presumption expansions (now covering first responders in more than forty states), changes to medical fee schedules, increases in maximum weekly indemnity benefits, and opioid prescribing restrictions all change the development pattern from the point of enactment forward. The actuary needs to identify which accident years straddle a benefit-level change and adjust the expected-claims anchor (or the trend assumption, or both) accordingly.

Opioid utilization shifts

Opioid-related claims represent a specific category of medical cost that has moved dramatically over the past decade. Programs that implemented aggressive utilization review or pharmacy benefit management saw medical severity flatten or decline for certain claim types, while programs that did not saw continued escalation. The effect is visible in the medical trend component, and a report that selects a single flat medical trend across all accident years without acknowledging the shift is likely wrong for at least some of those years.

The self-insured wrinkle: thin data and leveraged factors

Everything in the previous section also applies to a traditional carrier writing workers compensation. What makes the self-insured program different is the combination of thin data and long tail.

A carrier might have ten thousand workers compensation claims per accident year. A mid-sized self-insured employer might have three hundred. A smaller one might have fifty. When you build a loss development triangle from fifty claims, a handful of large permanent-disability awards can dominate an entire accident year, and the age-to-age factors derived from that triangle become volatile.

This is the leveraged-factor problem that Friedland illustrates with a paid bodily-injury example where the latest-year CDF is 90.00 (Friedland, p. 134). At that kind of leverage, a single $50,000 payment added to the latest accident year produces a $4.5 million change in the projected ultimate. No risk manager wants to explain to the CFO that the reserve moved by millions because one claim got a settlement check three days before the valuation date.

The standard response is to lean on Bornhuetter-Ferguson for recent years, which caps the influence of the actual data at a credibility weight that increases with maturity. For the most recent year, credibility might be 20% or less, meaning 80% of the projected ultimate comes from the expected-claims anchor. That anchor has to be right, which circles back to the payroll-mix and trend questions raised earlier.

For self-insured programs specifically, the actuary may also cap individual claims at a per-occurrence retention (say, $500,000) and project the capped triangle separately from the excess layer. This reduces the leverage problem within the working layer and isolates the large-claim volatility into a layer that can be priced or reserved using frequency-severity methods or industry benchmarks. If your program has a per-occurrence retention and the report does not show a limited-loss analysis, ask why.

Trend: the assumption that drives recent years

For accident years older than five or six, 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 estimate because it feeds directly into the expected-claims anchor that Bornhuetter- Ferguson and expected claims rely on.

Workers compensation trend decomposes into three parts:

Indemnity trend reflects changes in wage levels and statutory benefit rates. It tends to track general wage inflation with periodic jumps when a state raises its maximum weekly benefit.

Medical trend reflects changes in medical costs per claim. Medical CPI is the starting point, but fee schedule changes, utilization review effectiveness, and pharmacy costs all modify it. Medical trend for workers compensation has historically exceeded general inflation, though the gap has narrowed in states with aggressive fee schedules.

ALAE trend (allocated loss adjustment expense, which covers defense costs and cost-containment expenses; the NAIC’s current terminology is DCC for defense and cost containment and A&O for adjusting and other) is typically selected separately, driven by legal costs and the complexity of litigated claims.

A good report discloses each component, the source of the selection (the program’s own data, a benchmark, or a blend), and the combined annual trend applied to the expected-claims anchor. If the report shows a combined 5% trend without breaking it into components, you cannot evaluate whether the medical piece is realistic given your program’s actual medical management results.

What a buyer should ask their actuary

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

Which method drove the selected ultimate for each accident year, and why? A competent report discloses the method selection by year. Workers compensation should show chain ladder for older years and Bornhuetter-Ferguson or expected claims for recent years. If chain ladder is selected for the latest year without explanation, ask what justifies trusting the leveraged factor.
What expected claim ratio are you using for the most recent years, and how was it derived? The expected-claims anchor for a self-insured program should come from the program’s own historical loss rates adjusted for trend and payroll mix, not from an industry benchmark applied without normalization. If the ratio is benchmark-based, ask what the program’s own data shows and why it was not used.
What are the indemnity, medical, and ALAE trend selections, and what is the source for each? Trend selections should be documented from the program’s own experience where credible. Where the program’s data is too thin, the actuary should explain what benchmark was used and why it is appropriate for your state, class-code mix, and benefit level.
Were there any TPA changes, medical management program changes, or benefit-level changes during the experience period, and how were they addressed? The actuary should have asked you this before starting the analysis. If the report does not mention operational changes, either there were none (confirm this), or the actuary did not ask (which is a red flag).
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 workers compensation specifically, the report should include:

Paid and reported triangles at annual or semiannual evaluation intervals, with enough history to see the full development pattern for older accident years.
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 its basis.
Trend selections decomposed into indemnity, medical, and ALAE components, with source documentation (program experience, benchmark, or blend).
Exposure base reconciliation confirming that payroll by accident year ties to the employer’s records and that class-code mix changes are reflected in the expected-claims calculation.
Operational change disclosure documenting every TPA change, medical management program change, return-to-work initiative, legislative change, or benefit-level change that occurred during the experience period, with a description of how each was addressed.
Sensitivity analysis showing how the reserve changes under alternative trend, CDF, or expected-claims assumptions for the most recent two or three 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, you are not getting what the professional standards require. The right response is not to find a new actuary; it is to send the list back and ask for the missing sections. Most actuaries will provide them.

Interim monitoring: catching problems between annual reviews

A full reserve study is typically an annual exercise. Between studies, the risk manager should be comparing actual development against what the actuary’s analysis predicted. If the 2024 accident year was projected to reach $4.0 million in paid losses by June 2026 and the actual paid is $4.8 million, something has changed: either a large claim settled earlier than expected, the payment pattern accelerated, or the ultimate is higher than projected. Any of those is worth a conversation with the actuary before the next annual study.

Friedland frames this directly as comparing actual quarterly development against the predictions from the full annual analysis (Friedland, p. 345). That comparison is the core of interim monitoring, and it is the mechanism that justifies a standing actuarial relationship rather than a point-in- time opinion purchased once a year. The five leading indicators article covers the claim-level metrics that supplement triangle-level monitoring.