Methods

Berquist-Sherman in Plain English

When your claims operation changes mid-stream, the historical triangle lies. Berquist-Sherman is the standard correction, and you should know what it does, what it assumes, and when to demand it.

By Sam · April 2026 · 14 min read

Your actuarial report probably references the chain ladder as the primary reserving method. The chain ladder works by projecting historical development patterns forward. But that projection rests on a premise: the pattern that produced last year’s development will produce next year’s development. When something in your claims operation changes (a new TPA, a case reserve audit, a faster settlement initiative), the historical pattern stops being a reliable predictor. The triangle is still there, and the chain ladder will still produce a number. But the number is wrong, because the data it was built on no longer means what it used to mean.

Berquist-Sherman is the standard actuarial technique for correcting that problem. It restates the historical triangle so that the data reflects a consistent operational environment, even when the actual environment changed. If your reserve report mentions a Berquist-Sherman adjustment, this article explains what was done, what it assumed, and whether the result deserves your confidence. If the report does not mention Berquist-Sherman and your claims operation changed during the experience period, this article explains what should have been done.

What Berquist-Sherman corrects

The core problem is simple. Every development method uses a loss development triangle to measure how claims grow from one maturity to the next. The age-to-age factors derived from the triangle are the building blocks of the projection. If those factors were distorted by an operational change, the projection inherits the distortion.

Two types of operational changes produce the most common distortions:

Changes in case reserve adequacy. When adjusters start setting higher (or lower) initial reserves, the reported triangle is affected. The incurred amounts at early maturities are inflated (or deflated) relative to prior years, and the development factors that measure the jump from one maturity to the next are distorted. An unadjusted reported chain ladder will overstate the ultimate if adequacy increased, or understate it if adequacy decreased.

Changes in settlement speed. When claims start closing faster (or slower) than the historical norm, the paid triangle is affected. More dollars are paid at earlier maturities than the historical pattern would predict, inflating the paid development factors. An unadjusted paid chain ladder will overstate the ultimate if settlement accelerated, or understate it if settlement slowed.

Berquist-Sherman addresses each distortion with a separate adjustment. The two adjustments are conceptually independent: one fixes the reported triangle, the other fixes the paid triangle. An actuary might apply one, both, or neither, depending on which triangle was affected and how severe the distortion is.

For a broader framework on how these operational changes map to specific reserve errors, see the diagnostic review guide.

The case adjustment: restating to consistent adequacy

The case adjustment corrects for changes in how adjusters set case reserves (also called case outstanding). The idea is to restate the entire case outstanding triangle as if adjusters had always reserved at the same adequacy level, specifically the most recent level.

Here is the intuition. Suppose your claims adjusters historically set an initial case reserve of $30,000 for a typical bodily injury claim. After a case adequacy audit in 2023, the standard initial reserve moved to $45,000. The claims themselves did not change. The same type of claim that used to show $30,000 in the triangle at 12 months now shows $45,000. If you compute the age-to-age factor from 12 to 24 months using the old diagonal (where the starting point was $30,000 and the claim developed to $42,000), you get a factor of 1.40. For the new diagonal (where the starting point is $45,000 and the claim develops to roughly the same $55,000 ultimate at 24 months), you get a factor of 1.22. The decline in the factor is not because claims are developing less. It is because the starting point moved up.

The chain ladder does not know this. It sees the 1.22 alongside the historical 1.40 and, depending on how the actuary selects factors, may average them. That average understates future development for the older years and overstates it for the newer years. The result is a blended distortion that is hard to see in the output but easy to see in the triangle if you know to look for it.

The Berquist-Sherman case adjustment fixes this by restating the case outstanding at each historical evaluation point to what it would have been under the current adequacy standard. The actuary selects a severity trend (typically annual claim severity inflation, often between 3% and 8% depending on the line of business) and uses it to deflate the latest diagonal back through the triangle. The restated triangle then reflects a consistent adequacy level across all diagonals, and the development factors derived from it measure genuine development rather than an artifact of changing reserve practices.

The severity trend selection is critical. A higher trend produces a larger restatement of older diagonals, which increases the restated development factors. A lower trend produces less restatement. The ultimate claims estimate is directly sensitive to this single parameter.

The paid adjustment: restating to a consistent disposal rate

The paid adjustment corrects for changes in how quickly claims close. Where the case adjustment fixes the reported triangle, the paid adjustment fixes the paid triangle.

Here is the intuition. Suppose your TPA historically closed 40% of claims within 12 months. After a process improvement in 2023, that rate jumped to 55%. The claims themselves are not settling for different amounts. They are just settling sooner. But the paid triangle at 12 months now shows more dollars than the historical pattern would predict, because more claims have been resolved.

The chain ladder does not know this either. It sees higher paid amounts at early maturities for recent accident years and interprets that as higher development. The result is an overstated ultimate for those years.

The Berquist-Sherman paid adjustment fixes this by restating the paid triangle to a consistent disposal rate. The actuary builds a model (Friedland describes an exponential regression between the cumulative number of closed claims and cumulative paid losses at each maturity) that relates how much has been paid to how many claims have been closed. Using that relationship, the actuary can estimate what the paid amounts would have been at each historical evaluation point if claims had closed at the current rate.

The restated paid triangle then reflects a consistent closure pattern, and the development factors derived from it measure genuine claim growth rather than an artifact of changing settlement speed.

The hierarchy: reorganize data first, adjust second

A point that Friedland emphasizes and that practitioners sometimes overlook: Berquist-Sherman is not the first tool to reach for. When the operational change allows it, reorganizing the data is a simpler and more transparent correction than a quantitative adjustment (Friedland, p. 283).

Examples of reorganization:

If limits or retentions changed between policy years, analyze by policy year rather than accident year. This isolates each retention level in its own row of the triangle and avoids mixing development at different retention levels.
If exposure growth has been rapid and uneven, analyze by accident quarter rather than accident year. Shorter intervals give the chain ladder less opportunity to blend development from periods with very different exposure volumes.
If the claim mix shifted because of a new coverage line, segment the triangle by coverage or claim type and project each segment independently.

Reorganization works when the data structure can isolate the change. Berquist-Sherman is needed when the change cuts across the existing data structure and cannot be segmented away. Case reserve practice changes are the canonical example: every claim in every accident year is affected by the new standard, so there is no way to reorganize the triangle to avoid the distortion.

When Berquist-Sherman is the right tool

The adjustment is most appropriate when three conditions hold:

The change is identifiable. You know (or the actuary knows from claims management) that case adequacy shifted or settlement speed changed. Berquist-Sherman is not a general-purpose smoothing tool. It corrects for a specific, diagnosed operational change.

The change affects the triangle systematically. A one-time event (a single catastrophic claim, a one-quarter processing backlog) may not warrant a full Berquist-Sherman adjustment. The adjustment is designed for changes that persist across multiple evaluation dates and affect the triangle’s structure.

The change cannot be addressed by data reorganization. If you can segment the data or switch to a different triangle basis to isolate the change, that is usually preferable. Berquist-Sherman involves modeling assumptions (the severity trend for the case adjustment, the regression specification for the paid adjustment) that add uncertainty. Simpler corrections are better when available.

When Berquist-Sherman breaks

The adjustment has real limitations, and an actuary who presents a Berquist-Sherman result as definitive is overstating the precision.

Sensitivity to the severity trend. The case adjustment hinges on the selected severity trend. A reasonable range for workers’ compensation medical severity might be 4% to 8%. Over a ten-year triangle, the difference between 4% and 8% annual trend produces a large difference in the restated historical case reserves, which flows through to a large difference in the selected development factors and the ultimate. Joseph Thorne’s caution, as cited by Friedland (p. 287), applies directly: the adjustment is only as defensible as the trend assumption behind it.

Multiple simultaneous changes. Berquist-Sherman works best when one thing changed. If case adequacy increased while settlement speed also changed while claim mix shifted, the adjustments interact. The case adjustment assumes the restated severity trend reflects only the adequacy change, but if severity also shifted because of mix, the trend captures both and the restatement is imprecise.

Sparse data. The paid adjustment relies on a regression model fit to the relationship between closed claim counts and paid amounts. For self-insured programs with few claims per accident year, the regression has limited data points and the fit may not be reliable. An exponential curve through five data points is a model, not a fact.

No substitute for judgment. Berquist-Sherman is a mechanical adjustment. It does not tell the actuary whether the adjustment was needed, only what the triangle looks like after the adjustment is applied. The diagnostic step (identifying the operational change, quantifying its effect, deciding whether an adjustment is warranted) still requires judgment and operational context. A Berquist-Sherman number without that diagnostic foundation is a black box with a different shape.

The self-insured amplification

Everything about Berquist-Sherman matters more for self-insured programs than for carriers, for the same reason that every reserving issue matters more for thin data: the margin for error is smaller.

A carrier with thousands of claims per accident year has enough volume that a case adequacy shift, while real, produces a moderate and predictable change in the aggregate development factors. The Berquist-Sherman adjustment operates on stable averages and the sensitivity to the trend assumption is bounded.

A self-insured program with 150 claims per accident year has no such cushion. A case reserve practice change affecting twenty claims can shift the aggregate development factor for an entire accident year. The Berquist-Sherman adjustment for that year depends heavily on whether the severity trend captures the true adequacy change or whether it is distorted by a few large claims that happened to coincide with the practice change.

This means two things for self-insured buyers. First, when Berquist-Sherman is applied, you should ask about the sensitivity of the result to the trend assumption (not just the point estimate, but what the ultimate would be under a higher or lower trend). Second, when Berquist-Sherman is not applied despite a known operational change, you should ask why the actuary believes the unadjusted triangle is reliable.

For more on why thin data amplifies every reserving judgment call, see the chain ladder article’s discussion of leveraged development factors. For the broader set of operational changes that produce triangle distortions, see the diagnostic review framework.

What a buyer should ask their actuary

These five questions test whether a Berquist-Sherman adjustment (or the decision not to adjust) was well-supported.

1. Did case reserve adequacy change during the experience period? If the actuary does not know, or has not asked, the diagnostic foundation is missing. The actuary should be able to describe the direction and approximate timing of any adequacy shift, supported by metrics like average initial case reserves, paid-to-incurred ratios, or closed-without-payment rates.

2. Did settlement speed change during the experience period? Same logic. The actuary should be able to point to disposal rates (claims closed as a percentage of claims open at each maturity) and identify whether the pattern shifted.

3. If you applied a Berquist-Sherman adjustment, what severity trend did you select, and how did you derive it? The trend should be supported by data: claim severity trends from the insured’s own experience, industry benchmarks, or medical/indemnity inflation indices. A trend selected without documentation is an unsupported assumption driving the entire adjustment.

4. How sensitive is the adjusted ultimate to the trend selection? The actuary should be able to show the ultimate under at least two alternative trend assumptions (one higher, one lower). If the range is wide, the adjustment adds uncertainty rather than resolving it, and the actuary should explain why the point selection is preferable.

5. Did you consider data reorganization before applying the quantitative adjustment? Reorganization (by policy year, accident quarter, or claim segment) is often simpler and more transparent. If the actuary went directly to Berquist-Sherman without considering reorganization, the report should explain why the data structure did not allow it.

What to require in documentation

A reserve report that includes a Berquist-Sherman adjustment should document:

The operational change that motivated the adjustment: what changed, when, and how it was identified (from claims management, from the triangle, or both).
Which adjustment was applied: case, paid, or both.
The severity trend selected for the case adjustment, with the data or benchmarks that support it.
The regression specification for the paid adjustment, including the fit quality and any data points that were excluded or weighted differently.
The adjusted triangle alongside the unadjusted triangle, so you can see what the adjustment actually did to the development factors.
A sensitivity analysis showing the ultimate under at least two alternative trend assumptions.
The ultimate with and without the Berquist-Sherman adjustment, so you can see the magnitude of the correction.

A report that applies Berquist-Sherman without showing the unadjusted triangle is hiding the baseline. A report that shows the adjustment without documenting the trend assumption is presenting a result without its most important input.

If no Berquist-Sherman adjustment was applied, and the claims operation changed during the experience period, the report should document why the unadjusted triangle was considered reliable. Absence of an adjustment is a judgment call, and it should be documented like any other judgment call in the analysis.