January 22, 2021, 15:30–17:00
Job Market Seminar
In this paper, I show that dynamic discrete choice (DDC) models with continuous permanent unobserved heterogeneity are identified. The existing DDC literature controls for permanent unobserved heterogeneity through finite mixtures — that is, by assuming there is a finite number of agent ‘types’. In contrast, I show that DDC models with infinitely many agent types are identified. Relative to the existing literature, I exploit commonly imposed assumptions to show identification under low-level conditions. My results apply to both finite- and infinitehorizon DDC models, do not require a full support assumption, nor a large panel, and place no parametric restriction on the distribution of unobserved heterogeneity. The results provide a number of advantages for applied work. First, commonly used structural models can be estimated with more flexible heterogeneity. Second, my results do not require that the number of types be known a priori. Although there is rarely a theoretical reason for the number of types to be known, it is a common assumption in applied and theoretical work. Finally, the proposed seminonparametric estimator can be implemented using familiar parametric methods. I illustrate these advantages by applying my results to the labor force participation model of Altu˘g and Miller (1998). In this model, permanent unobserved heterogeneity may be interpreted as individual-specific labor productivity, and my results imply that the distribution of labor productivity can be estimated from the participation model.