Recommender Systems Under Privacy Protection

We examine how a profit-driven platform designs its recommender policies under varying privacy protection regimes. We demonstrate a functional equivalence between opt-out protection and unprotected privacy. Notably, granting consumers the autonomy to misrepresent personal data can compel the platform to distort recommendations, inadvertently harming both parties. When recommendations are purely informative, such privacy controls can deter personalization entirely. However, when recommendations also serve an allocative role—controlling product access—algorithmic discrimination may emerge, restricting the disadvantaged minority from valuable opportunities. These findings remain robust even with endogenous profit margins.

This paper was accepted by Omar Besbes, revenue management and market analytics.

Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2024.05007.