Research

Access to consumer data enables firms to segment markets and charge different prices, but the use of such data raises privacy concerns. This paper studies a monopoly market in which information used for pricing must satisfy differential privacy, a restriction on the likelihood ratio of any disclosed message across consumer valuations. We characterize the attainable combinations of consumer and producer surplus. Relative to unrestricted disclosure, differential privacy can lower the maximum attainable consumer surplus by restricting information that supports lower prices that benefit some consumers, while the minimum attainable producer surplus remains equal to the uniform monopoly profit. Sufficiently strict privacy permits only uniform monopoly pricing, eliminating consumers’ gains from segmentation. Protecting consumer data therefore need not improve consumer welfare when we consider only its effects on market outcomes.

As firms increasingly adopt AI-powered pricing algorithms, a key and urgent policy concern is how to regulate the potential algorithmic collusion. This paper studies the regulation of algorithmic collusion through information design when firms delegate pricing to Q-learning algorithms under stochastic demand. A third-party intermediary commits to one of three disclosure rules: no disclosure, full disclosure, or upper censorship. Upper censorship truthfully reveals low-demand states while pooling high-demand states and yields higher profits than full disclosure, consistent with theoretical predictions. However, the profit ranking of no disclosure and full disclosure reverses the theoretical benchmark’s prediction: no disclosure yields higher profits at high discount factors, whereas full disclosure yields higher profits at low discount factors. These findings suggest that limiting information sharing may raise profits when pricing algorithms are sufficiently patient, complicating efforts to regulate algorithmic collusion.

How do people address self-confidence-concerned audiences? This paper analyzes this question under the framework of strategic communication games proposed by Crawford and Sobel (1982) by assuming the receiver cares about the expectation of his ability, i.e., his self-confidence. No fully revealing equilibrium exists even if the sender’s and the receiver’s preferences are aligned. Indeed, all equilibria in this game are interval partitional as in Crawford and Sobel (1982). In the uniform quadratic environment, the partition equilibria resemble those in standard cheap talk games.

We study how pricing algorithms learn to communicate when firms hold private information about demand. Our theoretical benchmark compares profit-maximizing pooling and separating communication in repeated price competition. In the numerical baseline, greater patience can shift communication from pooling to separation; with highly precise signals, separation can give way to pooling before returning. We are developing Q-learning experiments in which algorithms choose both messages and prices to test whether learned communication follows these predictions.