Australasian Microeconomic Theory Seminar - Jonathan Libgober (U Southern California)

Title: Machine Learning for Strategic Inference (joint with In-Koo Cho)

Abstract: We consider a simple buyer-seller game, with a buyer whose strategy is determined via access to data and some statistical algorithm. Our model builds off Rubinstein (1993), who showed, for this environment, that the seller can exploit the limited ability of simple classifiers to implement the ex-post optimal decision rule. Taking either the set of baseline classifiers as given or dropping the assumption that the seller is profit maximizing, we argue that no statistical algorithm is capable of approximating the rational benchmark. However, allowing for algorithms to “combine” classifiers and using the seller’s incentive to maximize expected profit, we show the existence of an algorithm which induces (approximately) rational behavior from the buyer. Our construction uses boosting, a common technique from machine learning. This algorithm shows that it is unnecessary for the buyer to be able to fit sophisticated classifiers, provided they can combine rudimentary classifiers in a particular way.