Your actuarial report likely shows a Bornhuetter-Ferguson (BF) estimate alongside the chain ladder for your most recent accident years. For the oldest years, the two methods probably produce similar numbers. For the newest years, they may diverge materially, and BF is often the one the actuary selected. If you do not understand what BF is doing differently from the chain ladder, you cannot evaluate whether that selection was justified.
This article explains BF in plain language: the core idea, the arithmetic, the assumptions it makes, the situations where it is the right tool, and the situations where it fails. It is written for the risk manager, CFO, or captive board member who receives a reserve estimate and needs to evaluate the method behind it.
What BF does, and why it exists
The chain ladder makes one bet: the past development pattern predicts the future. It multiplies current reported (or paid) losses by a cumulative development factor (CDF) to project ultimate claims. That bet works well for mature accident years with stable data. But for the most recent accident years, where only a small fraction of ultimate claims have been reported, the chain ladder is highly leveraged. A single large claim or a single late report can swing the projected ultimate by millions.
The expected claims method makes the opposite bet: an externally derived claim estimate (based on pricing, benchmarks, or the insured’s own mature-year experience) is more reliable than the data in the triangle. That works when there is no meaningful data, but it ignores actual experience entirely.
BF splits the difference. Friedland describes it directly: “The Bornhuetter-Ferguson technique is essentially a blend of the development and expected claims techniques” (Friedland, p. 152). For immature accident years, BF leans heavily on the expected claim ratio. For mature years, it leans heavily on actual development. The weight shifts automatically as the accident year ages and more data arrives.
Actuaries use BF nearly as often as they use the chain ladder. It is the second most common reserving method in practice and the natural default for the one or two most recent accident years in most self-insured reserve reviews.
The core assumption
The chain ladder assumes that future development will follow the historical pattern applied to actual reported (or paid) claims. If reported claims are unusually high or low at a given maturity, the chain ladder amplifies that signal.
BF makes a different assumption: unreported (or unpaid) claims will emerge based on expected claims, not based on claims reported to date. The claims that have already been reported stand as they are. The claims that have not yet been reported are estimated using an a priori expectation, scaled by how much development remains.
This distinction is subtle but consequential. If an accident year has unusually high reported claims at 12 months (perhaps because a single large claim was reported early), the chain ladder multiplies those high reported claims by the full CDF, producing an inflated ultimate. BF does not. It takes the actual reported claims at face value and adds an expected amount for the unreported portion, which is not influenced by the reported amount. The early large claim does not cascade through the projection.
The arithmetic, without a spreadsheet
The BF formula has three inputs:
- Actual reported claims for the accident year at the current evaluation date.
- Expected ultimate claims, an a priori estimate derived outside the triangle.
- Percent unreported, calculated from the cumulative development factor as (1 minus 1/CDF). This comes from the same development pattern the chain ladder uses.
BF Ultimate = actual reported + (expected ultimate x percent unreported)
A worked example makes the mechanics concrete.
Suppose your workers compensation program’s 2025 accident year has $300,000 in reported losses at 12 months of development. The actuary has selected a CDF from 12 months to ultimate of 4.00, which means historically only 25% of ultimate claims are reported at this maturity. The percent unreported is therefore 75% (1 minus 1/4.00 = 0.75).
The actuary has also selected an expected ultimate of $1,000,000 for the 2025 accident year, based on trended mature-year experience and current exposure.
BF Ultimate = $300,000 + ($1,000,000 x 0.75) = $300,000 + $750,000 = $1,050,000
Compare this to the chain ladder, which would project $300,000 x 4.00 = $1,200,000. The difference ($150,000) arises because BF does not amplify the $300,000 base through the full development factor. Instead, it anchors the unreported portion to the expected claims estimate.
Now consider what happens if a single large claim pushes reported losses to $500,000 at 12 months:
- Chain ladder: $500,000 x 4.00 = $2,000,000 (an increase of $800,000)
- BF: $500,000 + ($1,000,000 x 0.75) = $1,250,000 (an increase of $200,000)
The chain ladder’s ultimate moved by $800,000 because it multiplied the entire $200,000 increase by 4.00. BF’s ultimate moved by only $200,000 (the dollar-for-dollar impact of the new reported claim) because the unreported portion did not change. This is the stability that makes BF valuable for immature accident years with thin data.
How the credibility weighting works
BF can be understood as a credibility-weighted average between the chain ladder ultimate and the expected claims ultimate. The credibility assigned to the chain ladder signal is Z = 1/CDF.
At 12 months, with a CDF of 4.00, Z = 0.25. That means BF gives 25% weight to the actual development signal and 75% weight to the a priori expectation. At 36 months, with a CDF of 1.333, Z = 0.75: the method now gives 75% weight to the data and 25% to the expectation. By the time the CDF reaches 1.05 (development nearly complete), Z = 0.95, and BF is effectively the same as the chain ladder.
This automatic rebalancing is BF’s defining feature. You do not need the actuary to manually choose how much to trust the data versus the expectation. The development pattern itself determines the weighting. When data is sparse, the expectation dominates. When data is mature, the data dominates.
When it works
BF is the natural choice when two conditions hold:
1. The most recent accident years have volatile or thin data. Self-insured programs generate far fewer claims than a carrier book, and the most recent year may have only a handful of claims reported. The chain ladder amplifies that thin base through a large CDF. BF stabilizes the projection by anchoring the unreported portion to an external expectation.
2. A defensible expected claim ratio is available. The a priori estimate must come from somewhere credible: trended mature-year experience from the same program, pricing data, industry benchmarks adjusted for the insured’s specific exposure profile, or a combination of sources. If the expected ratio is poorly supported, BF inherits that weakness.
In practice, most self-insured reserve reviews use BF (or a close variant) for the one or two most recent accident years and shift to the chain ladder for older years where the data is credible on its own. The five core methods overview shows this pattern in the decision matrix.
When it breaks
BF has three principal failure modes.
1. The expected claim ratio is wrong. For immature accident years, BF places most of the weight on the expected ultimate. If that number is too high, BF overstates the reserve. If it is too low, BF understates it. Unlike the chain ladder, BF does not self-correct through emerging data until the accident year is mature enough for the credibility weight to shift toward the data. A bad a priori estimate can persist through multiple quarterly evaluations before the data accumulates enough to override it.
This is the most common BF failure mode in self-insured programs. Expected ratios that are “carried forward from last year” without adjustment for changes in exposure, workforce composition, or claims management deserve scrutiny. The expected ratio is an input, not an output, and the buyer should understand its derivation.
2. The CDF falls below 1.00. This happens in lines where salvage and subrogation recoveries (S&S) reduce cumulative incurred claims at later maturities, producing negative development. When the CDF is below 1.00, the credibility Z = 1/CDF exceeds 1.00, and the percent unreported becomes negative. The formula breaks down: BF would subtract from the expected ultimate instead of adding to it, which inverts the logic of the method.
Practitioners address this by flooring the CDF at 1.00 for purposes of the BF calculation (Friedland notes this practice at p. 156). If your lines of business include physical damage, property, or any line with material S&S, ask whether any CDFs were floored and how the actuary handled the negative development.
3. The development pattern is wrong. The percent unreported comes from the same CDF the chain ladder uses. If the development pattern is distorted (by the same operational changes that break the chain ladder, such as case reserve strengthening or settlement speed shifts), the credibility weights are miscalibrated. BF partially insulates the estimate because it does not multiply the base by the full CDF, but the weighting between data and expectation will be off.
The self-insured wrinkle
BF is the workhorse method for self-insured programs, and for good reason.
Self-insured programs have thin triangles. A workers compensation program with 200 claims per accident year produces noisier development factors than a carrier book with 20,000 claims. The most recent accident year’s chain ladder projection is typically the most leveraged and the least reliable number in the entire analysis. BF provides a stabilizing anchor.
But the anchor is only as good as the expected claim ratio behind it. For a self-insured program, the expected ratio is often derived from the same program’s mature-year experience, trended forward for medical or indemnity inflation. If the program has only three or four years of history (common after a TPA change or a restructuring), the mature-year base is thin, and the expected ratio carries more uncertainty than it appears to.
Buyers should ask two things: where did the expected claim ratio come from, and how sensitive is the BF ultimate to a change in that ratio? If a 10% increase in the expected ratio moves the BF ultimate for the most recent year by a material amount, the estimate is not as anchored as it looks.
For captive-specific considerations on how BF interacts with fronting arrangements and gross-to-net bridges, see IBNR for Single-Parent Captives and Fronting, Reinsurance, and Net vs. Gross.
What a buyer should ask their actuary
These questions test whether the BF projection in your report is supported and transparent.
1. What expected claim ratio are you using, and how was it derived? The expected ratio is the most influential input for immature accident years. A good answer identifies the source (mature-year experience, pricing, industry data), the trend applied, and any adjustments for changes in the insured’s exposure profile. An answer that says “same as last year” without further support is insufficient.
2. Is any CDF below 1.00, and if so, how is it handled? A CDF below 1.00 collapses the BF credibility logic. The actuary should either floor the CDF at 1.00 for the BF calculation or use an alternative method for those years. If this was not addressed, the BF number for those years may be mechanically unsound.
3. How do BF and chain ladder compare for each accident year, and where do they diverge? Agreement between the two methods is a strong signal that the estimate is well-supported. Divergence is a diagnostic signal. For mature years, the two should be close. For immature years, BF should be more stable. If the chain ladder is lower than BF for a recent year, it may mean reported claims are running below expectation. If the chain ladder is materially higher, a single large claim or case reserve change may be inflating the development signal.
4. How sensitive is the BF ultimate to the expected claim ratio? Ask the actuary to show what happens to the most recent accident year’s BF ultimate if the expected ratio moves up or down by 10% or 15%. This sensitivity test reveals how much of the estimate is driven by the a priori assumption versus the actual data.
5. At what maturity does the actuary switch from BF to chain ladder? The transition point reveals the actuary’s assessment of when the data becomes credible enough to stand on its own. If the switch happens at 24 months for one line but 48 months for another, ask what drives the difference.
What to require in documentation
For any accident year where BF is the selected method, your report should include:
- The expected claim ratio, with its derivation, source data, and trend applied.
- The cumulative development factors used, with the averaging method for the underlying age-to-age factors.
- The BF ultimate for each accident year, shown alongside the chain ladder indication so divergence is visible.
- An explanation of any CDF below 1.00 and the treatment applied.
- A sensitivity exhibit for the most recent accident year showing the effect of reasonable variation in the expected claim ratio on the BF ultimate.
- A narrative explanation of why BF was preferred over the chain ladder for the selected accident years, tied to specific data characteristics (thin data, volatile factors, operational change, or leveraged CDFs).
If the report does not contain these items, the BF number is unsupported. It may be reasonable, but you have no way to evaluate whether the expected claim ratio, the development pattern, and the credibility weighting are defensible. For why a single number without context is the least useful output, see Point Estimate vs. Range.
Further reading
For the foundational concept of IBNR that BF estimates, see IBNR, Explained Without the Jargon. For a walkthrough of the triangle that provides the development pattern BF relies on, see How to Read a Loss Development Triangle. For the chain ladder method that BF blends with expected claims, see Chain Ladder: How Actuaries Project Claims From Historical Patterns. For the broader method landscape, see How Actuaries Estimate Your Unpaid Claims. For diagnostic signals that the data underlying your reserve estimate has shifted, see What’s Actually Driving Your IBNR Higher? and Five Leading Indicators of Adverse Reserve Development.