Bias-Variance Games

Firms engaged in electronic commerce increasingly rely on machine learning algorithms to drive a wide array of managerial decisions. The goal of this paper is to understand how competition between firms affects their strategic choice of such algorithms. We model the interaction of two firms choosing learning algorithms as a game, and analyze its equilibria in terms of the resolution of the bias-variance tradeoffs faced by the players. We show that competition can lead to strange phenomena---for example, reducing the error incurred by a firm's algorithm can be harmful to that firm---and provide conditions under which such phenomena do not occur. We also show that players prefer to incur error due to variance than due to bias. Much of our analysis is theoretical, but we also show that our insights persist empirically in several publicly-available data sets.