An Evaluation of Econometric Models of Adaptive Learning (original) (raw)
Related papers
A Predictive Evaluation of Econometric Models of Adaptive Learning
2020
Game-theoretic models of learning are hard to study even in the laboratory setting due to econometric and practical concerns (such as limited length of the experimental session). In particular, as (Salmon, 2001) simulations shows, in a “blind testing” of several models data generated by those models does not correspond correctly to the estimated parameters. Thus, even when the real data generation process is knownwe cannot distinguish correct and incorrect models by looking at the estimates. However, we demonstrate that under the same conditions models are clearly distinguishable if instead of the parameters of the models we compare predictions that the models make. We also provide a rationale for why predictive quality is a particularly relevant metric for a learning model.
Evidence for Learning to Learn Behavior in Normal Form Games
Theory and Decision, 2004
Evidence presented in Salmon (2001) indicates that typical tests to identify learning behavior in experiments involving normal form games possess little power to reject incorrect models. This paper begins by presenting results from an experiment designed to gather alternative data to overcome this problem. The results from these experiments indicate support for a "learning to learn" or rule learning hypothesis in which subjects change their decision rule over time. These results are then used to construct an adaptive learning model which is intended to mimic more accurately the behavior observed. The Þnal section of the paper presents results from a simple simulation based analysis comparing the performance of this adaptive learning model with that of several standard decision rules in reproducing the choice patterns observed in the experiment.
Monetary Payoff and Utility Function in Adaptive Learning Models
2019
Non-technical Summary When players repeatedly face an identical or similar game (e.g., coordination game, technology adoption game, or product choice game), they may learn through experience to perform better in the future. This learning behaviour has important economic implications. It determines which economic outcome a game will reach and how fast it will get there. Given the importance of players’ learning behaviours, economists have proposed various adaptive models to study them. These models are usually estimated and tested using experimental data. Moreover, economists usually assume that individuals’ preference—their utility—is equal to the monetary reward they obtain. However, such an assumption can be wrong since players are not necessarily risk neutral. They could be risk averse or risk loving. I study the consequences of this false assumption and propose a method to deal with it. I then apply the method to an existing experimental dataset. The estimation results show that...
The American economic review, 1998
We examine learning in all experiments we could locate involving 100 periods or more of games with a unique equilibn'um in mixed strategies, and in a new experiment. We study both the ex post ( "best fit") akscrtptive power of learning models, and their ex ante predictive power, by simulating each experiment using parameters estimated from the other experiments. Even a one-parameter reinforcement learning model robustly outpelfonns the equilibrium predictions. Predictive power is improved by adding ' yorgemng" and "experimentation, " or by allowing greater rationality as in probabilistic jictitious play. Implications for developing a low-rationality, cognitive game theory are discussed (JEL C72. C92 ) * Erev: Faculty of industrial Engineering and Management. Technion. also contributed to the design and programming of the new experiment. We are indebted to Barry O'Neill, Jack Ochs, and Amnon Rapoport for access to unpublished parts of their data. The present version reflects numerous comments by three anonymous referees on several earlier drafts. This work was completed while Roth was at the University of Pittsburgh.
Adaptive vs. eductive learning: Theory and evidence
Adaptive learning and eductive learning are two widely used ways of modeling learning behavior in macroeconomics. Both approaches yield restrictions on model parameters under which agents are able to learn a rational expectation equilibrium (REE) but these restrictions do not always overlap with one another. In this paper we report on an experiment where we exploit such differences in stability conditions under adaptive and eductive learning to investigate which learning approach provides a better description of the learning behavior of human subjects. Our results suggest that adaptive learning is a better predictor of whether a system converges to REE, while the path by which the system converges appears to be a mixture of both adaptive and eductive learning model predictions.
A simple test of learning theory
2006
Abstract We report experiments designed to test the theoretical possibility, first discovered by Shapley (1964), that in some games learning fails to converge to any equilibrium, either in terms of marginal frequencies or of average play. Subjects played repeatedly in fixed pairings one of two 3�� 3 games, each having a unique Nash equilibrium in mixed strategies. The equilibrium of one game is predicted to be stable under learning, the other unstable, provided payoffs are sufficiently high.
ARTICLE NO. GA970540 Learning and Decision Costs in Experimental
1994
Subjects played strategically similar 4 � 4 and 6 � 6 constant sum games under varying payoff scales. Substantial divergences from equilibrium predictions were exhibited. The dynamic pattern of play is best explained by a stimulus learning model whereby players allocate weight to different actions according to their relative Ž time average. payoff experience in past plays. The results do not provide much support for the hypothesis that players select best responses to beliefs about opponent play based on observed choice frequencies in past plays, modified by random errors or preference shocks. Journal of Economic Literature Classification
Adaptive versus Eductive Learning: Theory and Evidence
Adaptive and eductive learning are two widely used ways of modeling the process by which agents learn a rational expectation equilibrium (REE). In this paper we report on an experiment where we exploit differences in the conditions under which adaptive and eductive learning converge to REE so as to investigate which approach provides the better description of the learning behavior of human subjects. Our results suggest that the path by which the system converges appears to be a mixture of both adaptive and eductive learning model predictions.