Abstract

Fairness-aware statistical learning is crucial for mitigating discrimination against protected attributes, such as gender, race, and ethnicity, in data-driven decision-making. This is particularly important in high-stakes applications like insurance underwriting and annuity pricing. Factor models and principal component analysis (PCA) are widely used for risk assessment and pricing in these domains, but their predictive outputs may inadvertently introduce bias. In this paper, we introduce a novel fairness notion, decision error parity, which ensures that expected errors in decision making are equal across demographic groups. To promote fairness in data-driven decision making with factor models and PCA, we propose a fair decision model that mitigates outcome disparities through fairness regularization, which is different from existing prediction-focused fair models. Applying  our framework to annuity pricing, we demonstrate its effectiveness in reducing gender disparities and achieving decision error parity across genders. We validate our model’s performance through both simulation studies and empirical data analysis, underscoring its potential for fairer decision-making in financial and .insurance applications. (Joint work with Junhao Shen, Yanrong Yang, and Ran Zhao)

Speaker

Fei Huang (UNSW Sydney)