Causality and Exogeneity in Econometric Models

Abstract

The following pages are intended to show how statistics enters into a relatively young discipline as econometrics. The role of statistics in the classical inference procedures, namely estimation and hypotheses testing, will not be dealt with. This should not be taken to suggest that within econometrics such procedures should not be the object of attention. On the contrary, several authors have stressed the need for extensive use of statistical tests as tools for mis-specification analysis and diagnostic checking. Here we will be pursuing a different task, namely to point out that statistics also plays an explicit role in the earlier phase of model specification. In particular, we want to show how the form of the model — and hence what kind of inferences can be drawn on its basis — depends on the crucial concepts of non-causality and exogeneity defined in terms of conditional distributions of observable variables.