A1070
Title: Nonparametric panel data estimators of the nonparametric panel stochastic frontier analysis
Authors: Kai Sun - Shanghai University (China) [presenting]
Xin Geng - SAS Ins. (United States)
Abstract: Procedures are proposed for estimating individual effects panel stochastic frontier models where the frontier function is unknown via nonparametric panel data estimators. The unobserved individual heterogeneity is decomposed into (i) individual effects that affect output through the frontier function and (ii) time-invariant technical inefficiency that affects output through the inefficiency function. The individual effects can be either fixed or random: they can affect output either as non-stochastic individual characteristics or as an individual-specific random shock. When the individual effects are random, the associated idiosyncratic term in the context of stochastic frontier analysis (SFA) is shown to be heteroskedastic, and therefore, a modified nonparametric random effects estimator of the conditional variance of the idiosyncratic term is also proposed. To guide practitioners to deciding between fixed versus random effects, a Hausman-type test statistic with a bootstrap procedure that works under the SFA setting is provided. Simulations show that the nonparametric panel data estimators and testing procedures perform well in finite samples. Finally, the panel data framework is applied to estimating a panel stochastic frontier model where the base inefficiency follows a half-normal or exponential distribution.
