A generalized approach to the treatment of autocorrelation

Abstract

One of the earliest problems considered and analysed by econometricians was that posed by the presence of autocorrelation in regression models In the now classic article by Cochrane and Orcutt a definitive treatment of the standard Markov alternative was presented, and a simple transformation was proposed enabling users to circumvent the computational difficulties found in attempts to apply the (theoretically correct) generalised least squares technique. Such was the influence of this article that only small modifications were later made to the procedure: the seeming lethargy having a two-fold cause (a) The emphasis upon the use of yearly data in econometric models made the Markov process a realistic one and (b) the rigours of computation demanded a simple transformation device. In recent years there appears to have been a reconsideration of this dual justification, stemming primarily from a strong desire to free econometric modelling from traditional restrictions¿thus, the development of transformations for fourth and second order correlation