A1707
Title: A Bayesian approach to discrimination-free insurance pricing with variational inference
Authors: Lydia Gabric - Arizona State University (United States)
Shuang Zhou - Arizona State University (United States) [presenting]
Kenneth Zhou - Arizona State University (United States)
Abstract: In recent years, many jurisdictions have implemented anti-discrimination regulations that require protected information to be excluded from insurance pricing calculations. To adapt to anti-discrimination regulations while maintaining accurate pricing outcomes, recent studies have sought to develop discrimination-free pricing methods through probabilistic inference. However, most of the existing estimation processes require individual-level discriminatory data, which are often prohibited due to regulatory constraints. We propose a novel Bayesian discrimination-free pricing method that no longer requires individual discriminatory information. To achieve this, we consider a Bayesian finite mixture model that treats the discriminatory variables as unknown latent variables with a hierarchical prior distribution to represent the assumed discriminatory relations. Through such model, the indirect discrimination can be inferred via the posterior distribution. Based on the hierarchical model, we propose a novel implementation of the variational inference to construct the discrimination-free pricing family with a mean-field family. Additional techniques such as the importance sampling are also used to obtain accurate prices. Supported by a simulation study and an empirical analysis based on real insurance data, our Bayesian approach is capable of inferring the hidden relationship between variables and consequently producing unbiased discrimination-free pricing results.
