Robust estimation of panel data : with an application to investment equations

Abstract

Testing economic theory on the basis of available data sets through econometric modelling is only legitimate if each condition of the entire collection of underlying assumptions is satisfied. This, at first glance obvious, statement has more implications than many of the practitioners of econometrics in the field are aware of. Even on a theoretical level, econometricians seem to show little agreement about the recommended modelling strategies or estimation procedures. This is imputed to the highly non-experimental nature of economic data and the fact that in many cases the theoretical variables do not coincide one-to-one with the available observed data. Disciples of both Fisher’s experimental design paradigm or Gauss’s theory of errors have to assume that the mechanism which generates the data is a nearly isolated system. See Spanos (1993) for an exposition of these different methodologies. This means that the systematic explanatory part of the model is somehow fixed or predetermined and the error terms can be viewed as non-systematic effects of omitted influencing factors or mismeasurement. In sharp conflict with this assumption, economic data is held to be generated by a fickle process where explanatory variables are often endogenous and influential factors show up and disappear. Usually, a specific economic theory only considers a sub set of the systematic explanatory components and makes predictions under the assumption that the influential factors which have been put aside are Constant. These ceteris paribus clauses often do not hold when analyzing real data sets.