Chain Ladder vs Bornhuetter-Ferguson: A Practical P&C Reserving Guide

The Chain Ladder method estimates age-to-age development from an observed cumulative paid or incurred triangle. Selected factors are chained from the latest observed maturity to ultimate and applied to each origin period. The method is driven by the triangle: if early reported or paid loss changes materially, the indication for an immature origin period can change materially with it.

Bornhuetter-Ferguson also needs a selected development pattern, but it uses that pattern to estimate the share of expected loss that remains unreported or unpaid. The remaining percentage is applied to an expected ultimate selected before considering the still-unobserved portion of the current origin period. The result then combines what has already emerged with an expectation for what has not.

For mature origin periods, the cumulative development factor approaches one and the unreported percentage approaches zero. Chain Ladder and Bornhuetter-Ferguson will usually place increasing weight on the observed value, although differences can remain when a tail is material or the selected pattern is unstable. A mature row is not automatically reliable: case strengthening, reopening, late claims, commutations, large losses, and operational changes can still distort it.

For immature origin periods, Chain Ladder can amplify a small amount of early experience through a large cumulative development factor. Bornhuetter-Ferguson is designed to reduce that sensitivity by applying the unobserved percentage to the expected ultimate instead. That stability is useful only when the expected ultimate is credible, appropriately trended and on-level, and sufficiently independent from the immature losses it is meant to complement.

A mechanical volume-weighted average is a starting candidate, not an automatic selection. Review individual link ratios, the number and volume of observations at each age, calendar-period effects, claim handling changes, large or catastrophe losses, changes in policy limits and deductibles, mix shifts, and whether paid and incurred patterns tell consistent stories. Exclusions and adjustments should be explicit and reproducible.

The tail deserves separate support. Long-tailed liability lines may have material development beyond the last observed age, while a short-tailed property segment may not. Compare reasonable curve fits, prior selections, external benchmarks, and the observed stability of late-age data. Show the impact of the selected tail on ultimate loss and unpaid claims instead of burying it inside one cumulative factor.

An expected ultimate may come from earned premium multiplied by an expected loss ratio, an exposure and frequency-severity view, a pricing model, a business plan, or another actuarially reviewed source. Record the source, valuation date, trend and rate-level treatment, exposure changes, mix, coverage, limits, and any management or underwriting assumptions. A single unexplained loss ratio entered into a workbook is not enough evidence.

Independence is a practical concern. If the prior is repeatedly revised to follow the same early emergence that makes Chain Ladder volatile, the BF indication can appear stable without providing genuinely separate information. Compare the prior with historical actual-versus-expected results and explain why it remains reasonable for the current portfolio and valuation date.

Method comparison should be performed by coherent segment and origin period, not only at the portfolio total. Show the latest observed value, maturity, selected cumulative factor, Chain Ladder ultimate, expected ultimate, BF ultimate, remaining claim amount, and the change from the prior valuation. Investigate where the methods diverge most and whether the cause is emergence, the prior, the tail, segmentation, or a data issue.

ASOP No. 43 requires the actuary to consider methods or models appropriate to the circumstances and relevant material factors. The final selection may use one method, a weighted combination, or another approach, but the decision belongs to the qualified actuary. The selected result should retain its source data, assumptions, calculations, scenarios, rationale, peer review, approval, and communication basis.

A reserving product earns its place after the calculation. It should connect approved claims, premium, and exposure data to reproducible triangles; expose factor and tail decisions; run several deterministic methods through versioned engines; preserve scenarios; record actuarial selections; explain movement; and render consistent reports from the approved result.

ACTUIT is Facio’s governed reserve-close product for that operating layer. The product itself runs in its dedicated environment; facio.io remains the indexable research and evaluation surface. This separation keeps search content public and fast while authentication, customer data, actuarial calculations, approvals, artifacts, and audit evidence stay inside the real product boundary.