Computational modeling of 1994 A.M. Best life/health insurer ratings

The insurance industry is the major intermediator of long term credit in the U.S., and insurance ratings agencies have become powerful de facto regulators of insurance company investment policy and hence of capital allocation for the economy as a whole. If the rating process can be shown to be dishonest, arbitrary, or even simply lacking in rigor, the implications for the proper role of insurance ratings in our capital markets is highly signiicant. This article examines the relationship between the A.M. Best Company's assigned Life/Health insurer ratings, 1994 edition, and the hundreds of quantitative and qualitative factors that A.M. Best discusses in its explanation of its rating system. The results indicate that company ratings are largely determined by company size and cash ow. Using tree modeling with only a few quantitative factors, with no additional qualitative input, it was possible to predict 49% of insurer's A.M. Best ratings exactly, and 93% if a miss by one rating in either direction is allowed. A simpler linear model was able to predict ratings with corresponding accuracies of 38% and 82%.