COMPOSITE FORECASTING METHODS: AN APPLICATION TO NIGERIA'S PALM PRODUCE PRICES

Abstract

This paper uses combined time series and regression analysis rather than either method alone to produce better quality forecast of palm kernel prices.

A regression model was selected for the price data while a time series model was constructed for the regression residual series.

The result of the regression analysis lends credence to the claim that statistical elegance of an econometric model is not directly related to its ability to forecast well. Values of the root-mean-square (RMS) error, RMS forecast error and the Theil coefficient (U₂) of the linear model were 43.19, 52.76 and 1.05 respectively whilst those of the double log models (having better fit) were 43.84, 61.08 and 1.21 respectively.

The combined regression-time series model outperformed the regression model. Values of the RMS simulation error, RMS forecast and U₂ of the combined model were 85.27, 44.22 and 0.878 respectively, whilst those of the regression model were 93.73, 52.76 and 1.05 respectively.