Simulated-Based Estimation of Dynamic Discrete Games with an Application to Banks' Spatial Competition

Trustees Of Boston University, Boston

Investigators

Abstract

The research program develops new techniques for the estimation of structural parameters in dynamic discrete games and applies these methods to study spatial competition in the US banking industry. Empirical discrete games are useful tools in the analysis of economic and social phenomena where strategic interactions play an important role in explaining behavior. The estimated models allow one to infer how different factors affect behavior, and one can evaluate the effects of changes in economic policies. This research project has three parts. The first part develops a class of estimators that deals with two important issues in the estimation of dynamic discrete games: the indeterminacy problem because of large numbers of possible solution to a game, and the large dimensions in the solution and estimation of these models. The method combines nonparametric techniques, to deal with the indeterminacy problem, and simulation techniques to avoid the large dimensions problem. This part of the project provides a class of estimators that extends the range of estimable dynamic discrete games. The second part of the project is to study the finite sample properties of the proposed class of estimators. When the dimension of the state space is large relative to the sample size, initial nonparametric estimates of players' choice probabilities can be very imprecise, which induces significant finite sample biases in the estimates of structural parameters. The proposed estimator is consistent only when the number of simulations to approximate players' best responses goes to infinity. The project proposes several techniques to reduce these sources of bias and will provide researchers an automated procedure to select the estimator in this class that minimizes mean square error. The third part of the project applies these techniques to study competition in the banking industry using a unique panel data set from the Federal Deposit Insurance Corporation. This data set includes annual information on the location and deposits of all banks' branches in US between 1994 and 2001. The estimated model is then used to answer several questions including market power, mergers and acquisitions, and the effects of the Reigle-Neal Act on competition in the Banking industry.