A Sampling Study of the Merits of Auto-Regressive and Reduced Form Transformations in Regression Analysis

Abstract

This paper is concerned with some aspects of regression analysis when the error terms are autocorrelated and there exists more than one relationship between the variables. In particular, we investigate the merits of autoregressive transformations and the reduced form transformation in dealing with these complications. An important result is that, unless it is possible to specify something about the intercorrelation of the error terms in a set of relations and to choose approximately the correct autoregressive transformation, a certain amount of scepticism is justified concerning the possibility of estimating structural parameters from aggregative time series of only twenty observations.