763 © 2019 The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd THE JOURNAL OF INDUSTRIAL ECONOMICS 0022-1821 December 2018 No. 4 Volume LXVI TESTING ALTERNATIVE LEARNING THEORIES: EVIDENCE FROM SUBSCRIPTION CONTRACTS * ITAI ATER † V ARDIT LANDSMAN ‡ Analyzing panel data of 32,650 checking-account holders facing a menu of three-part tariff contracts, we document several findings that indicate that subscribers use simple heuristics to learn about the desirability of the contracts they have chosen. Our main findings are: subscribers change contracts in a direction that diminishes the probability of re–experiencing the trigger for switching; subscribers exhibit recency effects in switching; and after switching the majority of switchers systematically pay higher fees than they did before. We argue that directional learning theory best explains why consumers behave in a manner that yields suboptimal economic outcomes. I. INTRODUCTION THE CAPACITY TO PROCESS NEW INFORMATION AND LEARN has long been acknowl- edged as a fundamental component of decision making and behavior, in economics and in general (Simon [1959]; Tversky and Kahenman [1971], [1973]). For instance, theorists assume that information processing enables individuals to reach equilibrium behavior (Smith [1982]; Sobel [2000]). Labor economists study how firms use new information to learn about productivity and wage dynamics (Farber and Gibbons [1996]), and macro- economists study how individuals form expectations about future inflation (Woodford [2003]). Bayesian or belief-based learning, combined with expected utility theory, is by far the most common approach used in economics to model learning. † Authors’ affiliations: The Coller School of Management at Tel Aviv University, Tel Aviv, Israel. e-mail: Ater@post.tau.ac.il ‡ The Coller School of Management at Tel Aviv University, Tel Aviv, Israel, and Erasmus School of Economics, Erasmus University, Rotterdam, The Netherlands. e-mail: landsman@post.tau.ac.il * We are very grateful for comments made by the Editor, and two thoughtful anonymous referees. We also benefitted from suggestions by Pedro Bordalo, Yves Breitmoser, Kfir Eliaz, Ido Erev and seminar participants at Tel Aviv University, Hebrew University, Mannheim University, Humboldt University, the Applied IO CEPR conference in London, Bar Ilan University and the Interdisciplinary Center (Herzliya). This research was supported by the Henry Crown Institute of Business Research in Israel. We also thank Emanuel Marcu for excellent research assistance. Remaining errors are our own.