Identification and Estimation of Industry Dynamic Models with Persistent and Hidden State Variables

National Bureau Of Economic Research Inc, Cambridge MA

Investigators

Abstract

This research project focuses on the formulation and estimation of structural models of producer entry, growth, and exit. The data source is new; a file of alcohol tax returns that contains the date of birth, date of exit, and a complete monthly history of the dollar value of alcohol sales for all licensed restaurants and bars in Texas. In the model, a firm's sales is proportional to its profits, but transitory cost shocks make it an imperfect indicator of profitability's persistent component. Because of this imperfect observation, the model's state variable is both persistent and hidden. The state variable's persistence distinguishes this model from the many estimable models of dynamic discrete choice that incorporate Rust's (1987) conditional independence assumption. To disentangle the persistent and transitory components of profitability, this project uses the fact that producers' exit decisions depend only on the persistent component. In the model with normally distributed shocks, the information in producers' exit decisions identifies the parameters describing both the persistent and transitory components of profitability as well as the producer's optimal exit threshold. An important component of this research project is the extension of this identification proof to semi-parametric and non parametric environments. The project's initial empirical research focuses on distinguishing Gaussian models of entrepreneurial learning similar to Jovanovic's (1982) from models with perfect entrepreneurial information, such as Hopenhayn's (1992). Additional information in the data set describing each restaurant or bar's location and the characteristics of its parent firm suggest further generalizations of the model and estimation technique.