Annuity Table Validation in Pension Funds: A Comparative Power Analysis of Goodness-of-Fit Tests

Authors:

Abstract
Assessing the goodness-of-fit of an annuity table to the observed mortality in a pension fund is a fundamental step in actuarial management, with direct impact on the actuarial valuation of benefits, provisioning, and risk model validation. In this paper, we compare the efficiency of different hypothesis tests, both those commonly used by the industry-such as the chi-squared test and the so-called Kolmogorov-Smirnov (KS) test-and the alternatives we propose in this article. Our proposals assume that death counts by age are modelled through a Poisson generalized linear model with offset and a deflator parameter that is either constant or linear in age, under both frequentist and Bayesian approaches. A comprehensive simulation study was carried out in three modalities: (i) simulation with age-constant deflators; (ii) simulation with age-linear deflators; and (iii) direct simulation from pairs of real annuity tables, in which the data-generating process corresponds to a known table and the null hypothesis is evaluated against a second also known table. The results show that some of our proposals exhibit type I error rates close to the nominal level while minimising type II error compared to the tests currently in common use in the industry. The results indicate that, unlike the tests we propose, the traditional chi-squared and KS tests inflate type I error and exhibit low power (high type II error), especially for populations that are not large and when the annuity table has higher mortality than the fund's experience-precisely the scenario requiring the most care from a solvency standpoint for survival benefit plans. To the best of our knowledge, the tests we propose in this article are novel and have never been subject of prior research.