Achieving Individually Fair Personalized Pricing

Abstract

The market is an indispensable aspect of our lives, enabling us to exchange valuable goods and services. Allocation and pricing are two significant challenges in determining the equilibrium dynamics of a market. With the advent of AI and technology, the way markets function has evolved greatly. With data analytics tools, sellers can engage in personalized pricing to maximize revenue. However, it is shown that personalized pricing is prone to unfairness among consumers. This thesis precisely studies the fairness challenges in personalized pricing. We aim to examine the concept of fairness in the context of monopoly markets that implement feature-based pricing. Our proposal introduces a new form of individual fairness, referred to as 𝛼-fairness, which ensures that individuals with similar characteristics are subjected to comparable pricing. To evaluate the loss in revenue to the seller in pursuit of fairness, we additionally introduce the notion of Cost of Fairness (CoF) – the ratio of the expected revenue in optimal feature-based pricing to the expected revenue in the given fair feature-based pricing. First, we investigate the discrete valuation space and present an analytical solution for the most suitable fair feature-based pricing strategy. Our findings indicate that CoF, can be arbitrarily high for any fair pricing strategy. We observe that such valuation spaces are uncommon, so we focus on continuous valuation spaces that are well-studied in economics. With the standard assumption of the revenue function being continuous and concave with respect to the prices, we demonstrate that CoF is strictly less than two, regardless of the model parameters. Finally, we present a polynomial time algorithm that computes a fair feature-based pricing approach, successfully achieving CoF less than two.