A Stochastic Frontier Model with Unspecified Time-Varying Firm Effects

Abstract

We propose a new way of estimating a stochastic frontier model with time-varying firm effects. By means of nonparametric smoothing of categorical data we allow for completely unspecified time-varying firm effects. Furthermore it is well established that smoothing of cell probabilities of sparse contingency tables improves the finite sample performance compared to the frequency estimator. We are taking advantage of this in a panel data context with time-varying firm effects. By Monte Carlo simulations it is shown that smoothing of the time-varying firm effects works very well compared to the parametric panel data method proposed by Cornwell, Schmidt & Sickles (1990). The proposed method is applied on Indonesian rice farmer data. The most pronounced difference in our analysis relative to previous studies that analyzed this data is considerable larger average estimated efficiency levels. Thus the farmers appear to be less inefficient than in previous studies.