Ask two actuaries to estimate the same reserve and you may get two defensible answers — often because one leaned on the chain-ladder method and the other on Bornhuetter-Ferguson. The chain-ladder vs Bornhuetter-Ferguson question is one of the most practical decisions in loss reserving, and the right answer depends less on mathematical elegance than on how much you trust your data. This article explains how each method works, where each one breaks down, and how experienced reserving actuaries decide between them.
What the chain-ladder method actually assumes
The chain-ladder method is the workhorse of claims reserving. You arrange paid or incurred claims in a development triangle — accident years down the side, development periods across the top — and compute age-to-age factors that describe how claims have historically grown from one evaluation point to the next. Multiply the latest known position of each accident year by the appropriate cumulative development factor, and you have an estimate of ultimate losses. The reserve is simply ultimate losses minus what has already been paid.
The method's core assumption is easy to state and easy to forget: the future will develop the way the past did, in proportion to what has already emerged. Every chain-ladder projection is anchored to claims reported or paid to date. That anchoring is a strength when the data is mature and stable, because the projection responds directly to actual experience. It is a serious weakness when the anchor itself is unreliable.
Consider a recent accident year on a long-tailed casualty line. Twelve months in, perhaps 10% of ultimate losses have been reported. The chain-ladder multiplies that thin sliver of experience by a large development factor — often 8, 10, or more. A single large claim reported early, or an unusually quiet first year, gets amplified into a wildly distorted ultimate. The method is leveraging noise.
How Bornhuetter-Ferguson blends in an outside view
The Bornhuetter-Ferguson (BF) method, introduced by Ronald Bornhuetter and Ronald Ferguson in their 1972 paper "The Actuary and IBNR," addresses exactly this problem. Instead of projecting from reported claims alone, BF splits the ultimate into two pieces: the part that has already emerged, and the part still to come. The emerged portion is taken at face value from the data. The unemerged portion — the IBNR — is estimated from an *a priori* expected loss ratio, typically derived from pricing assumptions, business plans, or industry benchmarks, rather than from the immature claims themselves.
Formally, the BF ultimate equals reported losses plus the expected losses multiplied by the proportion still unreported (one minus the reciprocal of the cumulative development factor). The elegance is in what this does to sensitivity: early in an accident year's life, when little has emerged, the estimate leans almost entirely on the prior expectation and barely reacts to random claim noise. As the year matures and the percentage reported grows, the estimate gradually hands weight back to actual experience. In the limit, at full maturity, BF and chain-ladder converge.
In this sense, BF is a credibility-weighted compromise between two extremes: the pure chain-ladder (full credibility to observed experience) and the expected loss ratio method (zero credibility to observed experience). That is also its Achilles heel — a BF estimate is only as good as its a priori. Feed it a stale or optimistic expected loss ratio and the method will confidently, smoothly produce the wrong answer while the emerging data quietly disagrees.
Where each method earns its keep
In practice, the choice tends to follow a few reliable patterns.
Chain-ladder is usually preferred for mature accident years and stable, short-tailed lines. Property, motor physical damage, and other fast-developing business generate credible data quickly. Once a meaningful share of ultimate losses has emerged — many practitioners use rough thresholds like 50–70% reported — actual experience deserves the weight, and the a priori becomes an unnecessary intermediary.
Bornhuetter-Ferguson shines on immature years, long-tailed lines, and low-volume portfolios. Excess casualty, professional liability, and reinsurance business can take a decade or more to develop. For the most recent accident years on these lines, a chain-ladder projection is little more than a random number generator with a spreadsheet interface. BF's stability is worth the dependence on a prior.
BF is also the safer choice after structural change. New products, shifting attachment points, a reworked claims department, or an unusual event year (a pandemic-distorted accident year is the obvious recent example) all undermine the chain-ladder's assumption that past development patterns will repeat. A carefully constructed a priori can incorporate what you know about the change; a development triangle cannot.
There is a subtler consideration too: incentives and anchoring. Because BF estimates move slowly, they can mask genuine deterioration. If reported losses on a soft-market underwriting year keep coming in above expectations, a BF estimate with an unchanged a priori will under-recognize the bad news quarter after quarter. Reserving literature on the underwriting cycle has repeatedly flagged this pattern: methods that lean on plan loss ratios tend to under-reserve in soft markets, precisely when plans are most optimistic. The discipline that matters is revisiting the a priori — not defending it.
Using both, deliberately
Sophisticated reserving practice rarely picks a single winner. A common and defensible approach is a maturity-based blend: expected loss ratio or BF for the greenest accident years, transitioning to chain-ladder as years mature — either through judgmental selection or a formal credibility framework such as the Benktander method, which is itself an iterated blend of BF and chain-ladder.
Whichever route you take, document the rationale. Under regimes like Solvency II and rising auditor scrutiny of management judgment, "we always use chain-ladder" is not a methodology — it is an absence of one. The method selection, the derivation of the a priori, and the triggers for changing weights are exactly the judgments a reserve review will probe.
The takeaway
Chain-ladder trusts the data; Bornhuetter-Ferguson trusts your expectations, then gradually lets the data take over. Neither is inherently superior. Chain-ladder is the better tool when experience is mature, credible, and stable; BF is the better tool when experience is thin, volatile, or disrupted — provided the a priori is honest and regularly re-examined. The real skill in reserving is not running either method; it is knowing how much credibility today's triangle has earned, and being willing to change your answer when the data changes its story.
Related reading
See also Insurance Risk and Model Risk.
About the author
Jonas Osman Abdelghafour is an actuary and risk expert who advises insurers, reinsurers and pension funds on reserving, capital, underwriting governance, financial-crime exposure and enterprise risk management. His work sits at the intersection of quantitative actuarial practice and the governance, risk and compliance (GRC) frameworks that regulators now expect boards to evidence. See qualifications and expertise for background, or get in touch to discuss a consulting engagement.
Frequently asked questions
What the chain-ladder method actually assumes?
The chain-ladder method is the workhorse of claims reserving. You arrange paid or incurred claims in a development triangle — accident years down the side, development periods across the top — and compute age-to-age factors that describe how claims have historically grown from one evaluation point to the next. Multiply the latest known position of each accident year by the appropriate cumulative development factor, and you have an estimate of ultimate losses. The reserve is simply ultimate lo...
How Bornhuetter-Ferguson blends in an outside view?
The Bornhuetter-Ferguson (BF) method, introduced by Ronald Bornhuetter and Ronald Ferguson in their 1972 paper "The Actuary and IBNR," addresses exactly this problem. Instead of projecting from reported claims alone, BF splits the ultimate into two pieces: the part that has already emerged, and the part still to come. The emerged portion is taken at face value from the data. The unemerged portion — the IBNR — is estimated from an *a priori* expected loss ratio, typically derived from pricing...
Where each method earns its keep?
In practice, the choice tends to follow a few reliable patterns.
What should risk leaders know about using both, deliberately?
Sophisticated reserving practice rarely picks a single winner. A common and defensible approach is a maturity-based blend: expected loss ratio or BF for the greenest accident years, transitioning to chain-ladder as years mature — either through judgmental selection or a formal credibility framework such as the Benktander method, which is itself an iterated blend of BF and chain-ladder.
What should risk leaders know about the takeaway?
Chain-ladder trusts the data; Bornhuetter-Ferguson trusts your expectations, then gradually lets the data take over. Neither is inherently superior. Chain-ladder is the better tool when experience is mature, credible, and stable; BF is the better tool when experience is thin, volatile, or disrupted — provided the a priori is honest and regularly re-examined. The real skill in reserving is not running either method; it is knowing how much credibility today's triangle has earned, and being will...