FORECASTING WITH SERIALLY CORRELATED REGRESSION MODELS.pdfVIP

Journal of Statistical Computation and Simulation Vol. 74, No. 9, September 2004, pp. 625–649 FORECASTING WITH SERIALLY CORRELATED REGRESSION MODELS YUE FANGa and SERGIO G. KOREISHAa,b,* a University of Oregon, Eugene, OR 97403, USA and b University of Queensland, Brisbane, Qld 4072, Australia (Revised 5 November 2002; In final form 26 August 2003) In this article we investigate the asymptotic and finite-sample properties of predictors of regression models with autocorrelated errors. We prove new theorems associated with the predictive efficiency of generalized least squares (GLS) and incorrectly structured GLS predictors. We also establish the form associated with their predictive mean squared errors as well as the magnitude of these errors relative to each other and to those generated from the ordinary least squares (OLS) predictor. A large simulation study is used to evaluate the finite-sample performance of forecasts generated from models using different corrections for the serial correlation. Keywords : Asymptotic mean squared errors; Autoregressive disturbances; Generalized least squares; Incorrect generalized least squares; Predictive mean squared efficiency; Simulation 1 INTRODUCTION Since Cochrane and Orcutt (1949) and Durbin and Watson (1950) developed an approximate transformation to deal with and test for autoregressive disturbances of order 1, we have witnessed a plethora of studies dealing with the problems of serial correlation in regression models. Chaudhury et al. (1999), among others, have documented the evolution of approximate=exact transformations to deal with more complex serial correlation structures, as well as the development of n