A regularization approach to estimating inefficiency in stochastic frontier panel models

Abstract

The paper proposes a regularized mode estimator of unit inefficiency in a panel data context, allowing inefficiencies to vary across units and over time. This regularized estimator penalizes the likelihood function by constraining the sample average of the idiosyncratic error to zero.

Extensive simulations demonstrate that the regularized conditional mode estimator outperforms existing estimators, such as the unregularized mode estimator and the conditional mean estimator, particularly for the least efficient firms. We have also developed a user-written Stata package that implements the regularized conditional estimator for three sets of inefficiency distributions: (i) half-normal distribution with zero mean and conditional heteroskedastic variance, (ii) exponential distribution with conditional heteroskedastic scale parameter, and (iii) truncated normal distribution with conditional heteroskedastic mean and variance parameters.

Presenter

Professor Firmin Doko Tchatoka is Professor at the School of Economics and Public Policy, The University of Adelaide. His research interests are in Econometric (Theory and Applied), Statistics and Financial Econometrics. He is currently working on various topics, including identification problems in structural models, model selection and optimal inference in GMM setting, Bootstrap and Monte Carlos methods, treatment effect and network econometrics, Forecasting, identification of macroeconomic shocks, and the housing market.

Prof Firmin published a wide range of articles on these topics, and is the leading investigator of two Australian Research Council (ARC) grant projects; the first on ``Selection of mixed strength moment restrictions and optimal inference" (DP20; 2020-2022) with Prof. Prosper Dovonon (Concordia University, Canada) and the second on ``Identification power and instrument strength for causal effect in discrete outcome models'' (DP21; 2021-2023) with Profs Donald Poskitt (Monash University), Xueyan Zhao (Monash University), Eric Renault (University of Warwick, UK) and Franck Windmeijer (University of Oxford, UK)