The principal focus of this project is panel data,i.e., data which contain repeated observations over time for each individual. Panel data are increasingly the data of choice for empirical economic research in labor and industrial organization because the data capture both changes over time and cross- sectional relationships at a given point of time. Existing statistical tools do not permit economists to utilize all of the rich informational detail provided by panel data. This project provides applied economists with statistical tools that will do as better job in exploiting the information in panel data. The research project also extends methods for estimating changes of industry technological inefficiency and it improves forecasting methods for rational expectations models. More specifically, the research on panel data has three components. First, the project considers the dynamic panel data model, in which one of the explanatory variables is the lagged value of the dependent variable. Existing estimators for this model are inefficient, because they fail to utilize all of the available moment conditions, and the efficiency gains from using all moment conditions are quantified. Second, various panel data models with autocorrelated errors are considered. Once again it is possible to find efficient estimators that make use of all available moment conditions, and the interesting question is to find equally efficient estimators that use less conditions; that is, to find out which conditions are superfluous. Third, the project investigates a model with interactive individual and time effects. A connection is made between generalized method of moments and likelihood-based approaches to this model, and the model is extended to the case of more than one interactive component. In addition, the project extends the literature on frontier production functions in ways that allow a flexible pattern of technical inefficiency over time. The project investigates the possible efficiency gains in estimation of rational expectations models from the use of observable forecast errors.
