What is Fair? Proxy Discrimination vs. Demographic Disparities in Insurance Pricing

Indirect discrimination and fairness are major concerns in algorithmic models. This is particularly true in insurance, where protected policyholder attributes are not allowed to be used for insurance pricing. Simply disregarding protected policyholder attributes is not an appropriate solution, as this still allows for the possibility of inferring the protected attributes from non-protected covariates. This inference leads to so-called proxy or indirect discrimination. Though proxy discrimination is qualitatively different from the group fairness concepts in the machine learning literature, these group fairness concepts have been proposed to control the impact of protected attributes on the calculation of insurance prices. The purpose of this paper is to discuss the differences between direct and indirect discrimination in insurance and the most popular group fairness axioms. In particular, we show that one does not imply the other, as these concepts are materially different. Furthermore, we discuss input data pre-processing methods and model post-processing methods that achieve both discrimination-free insurance prices and demographic parity group fairness. The main tool of these methods is the theory of optimal transport