Algorithmic Pricing with Restricted Consumer Characteristics

Abstract

With the growing adoption of algorithmic pricing, its potential bias against certain groups (e.g., along the dimensions of age, gender or race) has become a rising concern. This paper investigates two approaches aimed at ensuring fairness in algorithmic pricing and examines their implications for firm profits and consumer welfare. Specifically, the ``Ban" approach prohibits the use of restricted characteristics, like age, race or gender, in pricing algorithms. However, biases may persist, as demographic data often correlates with other factors. In contrast, the ``Parity" approach allows the use of all information but requires that prices be equal across demographic groups. By comparing Ban and Parity, this paper connects two previously separate strands of the literature on privacy protection and group fairness.

We employ a stylized model in which a firm uses algorithmic pricing to target consumer groups with different willingness to pay. The accuracy of targeting may depend on the availability of demographic data as well as other information (e.g., purchase history). Our findings suggest that Parity can lead to a Pareto improvement relative to Ban, benefiting both the firm and consumers. This occurs when demographic information complements other targeting data, especially when the restricted demographic group is small (e.g., an ethnic minority) but more vulnerable to surplus extraction due to targeted pricing.  Thus, Parity could be a favored regulatory approach to achieve fair algorithmic pricing, especially for minority groups at risk of price discrimination. We demonstrate that these results remain robust when the firm endogenously invests in improving targeting accuracy. Furthermore, in scenarios where consumers make decision on whether or not firms can contact them (necessary for personalized pricing), Parity may further enhance welfare by promoting consumer participation.

Keywords: Algorithmic Pricing, Restricted characteristics, Privacy, Parity