Forecasting with mixed regression/ARIMA model

Abstract

The purpose of this study is to develope a technique to improve a forecasting accuracy of regression method by refining the residuals of regression with ARIMA process. The noble features of the method is that, by refining residuals, autocorrelations and/or cross-correlations inherent in the model can be removed, which is practically impossible in the classical regression method. The mixed regression/ARIMA model is set up for the cases of single equation and simultaneous equations, and these are applied to forecast monthly gasoline consumption in Korea. Major findings are as follows: First, the mixed regression/ARIMA model improves the forecasting accuracy significantly in terms of sum of square error (S.S.E.). Second, the forecasting accuracy of the mixed regression/ARIMA model is not seriously diminished as the lead time increases. Third, it is reavealed to be more accurate than any other individual model especially for unstable data.