Measurement of Cost Efficiency in the Presence of Unobserved Heterogeneity
This project addresses the problems involved in the parametric estimation of cost-efficiency in the presence of unobserved heterogeneity. Our focus is on the cost-efficiency of companies operating in sectors such as network industries and health care, where the production function is characterized by a high degree of heterogeneity across production units. The data sets used in this project consist of two panel samples respectively on local bus transportation companies and nursing homes in Switzerland. We also may use other data from Swiss hospitals and/or railway companies. The main objective is to evaluate the performance of the recently developed models such as true random effects model, in estimating inefficiency and the cost function parameters using panel data. These two aspects are considered separately as we contend that an accurate estimation of cost function parameters does not necessarily imply a good performance regarding inefficiency estimates. In a panel data framework, consistent estimation of cost function parameters usually requires an appropriate control for the potential correlations between firm-specific effects and explanatory variables. In conventional panel data models this issue leads to the fixed effects estimators. However, most of the fixed effect models used in frontier analysis have a poor performance in efficiency estimates. This study attempts to develop econometric specifications that satisfy both purposes, namely models that can overcome the consistency problem (heterogeneity bias) without affecting the inefficiency estimates. Our focus is on recent panel data frontier models with time-invariant heterogeneity and time-variant efficiency. We would also use alternative approaches based on conventional panel models with time-variant efficiency. Finally, another important issue in panel data frontier models is that most specification tests for frontier models are based on the skewness of the OLS residuals. As the residuals contain the presumably symmetric effect of heterogeneity, these tests could over-reject the skewness hypothesis especially if the inefficiencies are dominated by firm-specific heterogeneity. In the panel data framework, the time-varying part of the GLS residuals could be used instead to evaluate the skewness. These tests will be applied to several specifications using both data sets and the alternative tests will be compared. To our knowledge, this project will be the first comprehensive study of the validity and effectiveness of these models. This project is not limited to the existing models but plans to develop a few improvements in several aspects of the econometric specification.