A robust version of the FGLS estimator for panel data

Abstract

Panel or longitudinal data sets are frequent in financial and economic studies. This type of data combines cross-sectional with time-series data, providing extra information and allowing to evaluate and measure the statistical effects that would otherwise keep unknown. Different degree of restrictions upon the structure of the data leads to different approaches with least squares methodology. This results in estimators that can be highly affected by a violation of those assumptions. The Feasible Generalized Least Squares estimator (FGLS) is an estimator that preserves good properties without requiring strong distribution requisites. In spite of this, it is highly affected by the presence of observations too much different from all the rest. These are known as atypical observations or outliers. Economical and financial real data often present this type of data and the FGLS estimator may be seriously affected by those observations. This might be avoided if a robust option is chosen. Although robustness is the main concern in recent econometric modelling, there is still much to do in this field. Recent studies in those fields point to the advantage of using robust estimators. With this work, we want to contribute to the use of robust methodologies in the estimation of panel data models and present a robust version of FGLS, the RFGLS (Robust Feasible Generalized Least Squares). In this paper, the performance of the proposed estimator is compared with the FGLS using real data previously analysed by some authors.