Reserve obtained by feeding the Bornhuetter-Ferguson ultimate back in as the new a priori, giving a credibility estimator whose mean squared error is lower than either of its two parents.
Chain ladder trusts the triangle entirely, Bornhuetter-Ferguson trusts the a priori entirely for the undeveloped share. Benktander goes one step further: it takes the ultimate produced by Bornhuetter-Ferguson and reuses it as the a priori loss ratio in a second pass. Algebraically the result is a weighted average of chain ladder and Bornhuetter-Ferguson whose weight is the percent developed itself, so the more mature the year, the closer the method sits to chain ladder, which is exactly the behavior you want. Benktander and Hovinen showed independently that this estimator has a lower mean squared error than both chain ladder and Bornhuetter-Ferguson across most realistic settings, which makes it a reasonable default when you are torn between the two. It inherits the same blind spot, though: if the development pattern is wrong, all three methods are wrong together and in the same direction.
Accident year 2023 of a construction book as of December 31, 2025, percent developed 40%. Chain ladder returns an ultimate of 18.0M EUR, Bornhuetter-Ferguson 14.4M EUR on a 12M EUR a priori. Benktander returns 0.40 x 18.0 + 0.60 x 14.4 = 15.8M EUR. The reserve therefore moves from 7.2M EUR under Bornhuetter-Ferguson to 8.6M EUR, and the 1.4M EUR difference is precisely the credibility given back to the triangle.
Gunnar Benktander, méthode GB, Benktander-Hovinen, credibility chain ladder