The Bornhuetter-Ferguson method is an actuarial technique for estimating the incurred but not reported (IBNR) portion of insurance claims. It combines two sources of information: the claims actually reported so far, and a prior expectation of total losses based on premium and an expected loss ratio. The unreported share of the expected losses is added to what has already emerged to give an estimate of the ultimate loss.
The method was published in 1972 by Ronald Bornhuetter and Ronald Ferguson in a paper titled "The Actuary and IBNR" in the Proceedings of the Casualty Actuarial Society. It sits between the chain ladder method, which relies entirely on reported data, and the expected loss ratio method, which ignores it. It is a standard tool of claims reserving in non-life insurance and reinsurance.
Formula and worked example
The Bornhuetter-Ferguson IBNR estimate for an origin year is:
- is the earned premium for the origin year.
- is the a priori expected loss ratio, the losses expected per unit of premium before looking at the year's own claims.
- is the cumulative development factor from the current age to ultimate, taken from the claims triangle; is the proportion of ultimate losses expected to be reported by now.
The ultimate loss is then the reported amount plus this IBNR:
Suppose an accident year has earned premium of 1,000, an expected loss ratio of 70 per cent, reported losses of 300 and a cumulative development factor of 2.5. Expected losses are . The expected reported share is , so the unreported share is 0.60. The IBNR is and the ultimate is . For comparison, the chain ladder ultimate would be and the expected loss ratio ultimate would be 700.
Why the method is used
The appeal of the method is stability. In a young accident year only a small share of claims has been reported, and the chain ladder multiplies that small figure by a large development factor. One unusually large early claim, or an unusually quiet start, can move the chain ladder estimate a long way. Bornhuetter-Ferguson does not scale up reported losses: a large reported claim raises the ultimate only by its own amount, and the unreported portion stays anchored to the prior expectation.
As an origin year matures, the factor shrinks towards zero and the estimate converges on reported losses, so the method moves naturally from relying on the prior to relying on experience. Many reserving teams therefore use Bornhuetter-Ferguson for the most recent two or three years, especially in long-tail classes such as liability and workers' compensation, and switch to the chain ladder for older years. The method is also common for new lines of business, where the company's own triangle is too thin to produce reliable factors and development patterns are borrowed from similar classes or market data.
Choosing the expected loss ratio
The quality of the result depends heavily on the a priori loss ratio. It is typically drawn from pricing assumptions, business plans, the loss ratios of earlier mature years adjusted for rate changes and claims inflation, or industry benchmarks. A loss ratio that is too optimistic leads to under-reserving that may not become visible for years, which is one source of later reserve strengthening.
The development factor also matters. It comes from the same triangle used for the chain ladder, including any tail factor, so errors in the selected pattern flow into both methods.
Variants and limitations
Several refinements exist:
- Benktander method. A credibility-weighted blend in which the Bornhuetter-Ferguson ultimate is used as the prior for a second iteration, giving more weight to reported data.
- Cape Cod or Stanard-Bühlmann method. Derives the expected loss ratio from the portfolio's own data across years rather than setting it externally.
- Paid and incurred versions. The method can be applied to either paid or incurred triangles, with the corresponding development factors.
The main criticism is that the method responds slowly to genuine change. If a year is truly deteriorating, the estimate moves only as fast as claims are reported, while the unreported part still reflects the original, perhaps outdated, expectation. It is also a deterministic point estimate; stochastic versions are used when a range is needed for capital modelling or for the risk adjustment for non-financial risk. Actuaries usually present Bornhuetter-Ferguson results alongside other methods and explain differences when setting the booked claims reserve and judging reserve adequacy.