Method. We'll review some of the statistical background, especially hypothesis testing, needed to understand the issues and see how it applies to reserve modeling with aggregate loss triangles. We'll make use of the oncept of statistical power, associated with Type II error, which has been previously absent from reserve modeling discussions. This concept can be used to question the reliability of modeling results and certain common modeling recommendations. A few simplified reserving models and results of simulations that help illuminate the issues are described and reported.
Results. We'll see that significance tests, and testing more generally, might have little power and recommendations based on these tests can be unwise. We'll also see the benefits of a deeper understanding of the claims process and the dangers of relying on statistical methods without that understanding.
Conclusions. This particular argument for stochastic modeling in reserving with aggregate triangles is almost certainly unsound. If this were the only reason to resort to modeling, there are more productive uses of an actuary’s time. With or without modeling, better approaches probably rely on simpler methods, hard work, a skeptical and inquisitive attitude, and a deeper knowledge and understanding of the claims generation, reserving and settlement processes.
Keywords Reserving methods; reserve variability; statistical models and methods
