Prediction in Two­Equation Linear Regression Models

Abstract

In regression models using time series data, the errors are generally correlated. The sample residuals contain useful information for predicting post­sample observations. This information, which is generally ignored, has been exploited here in deriving the best linear unbiased predictors in a 2­equation linear regression model. The gain in efficiency of the proposed predictors over the usual generalized least ­ squares predictors has been obtained and the particular case when error terms in the two equations follow AR(l) process has also been disscussed.