Forecasting peak system load using hybrid models

Realizing that Box-Jenkins time series analysis lacks any basis in economic theory but at the same time wanting to take advantage of the excellent forecasting accuracy of the approach, the first part of this paper takes the basic Box-Jenkins structure and integrates it into an econometric system. Relating Box-Jenkins parameters to economic and weather variables allows one to forecast these parameters. The forecast parameters are in turn used in a Box-Jenkins model to provide a forecast of the series of interest. The result, for a specific example, proves to be clearly superior to a pure Box-Jenkins model. An alternate approach is suggested whereby an econometric model is estimated, the residual errors analyzed and forecast using a Box-Jenkins time series model, and then the results of both these models combined. The result, for the same example, proves to be an improvement over a simple Box-Jenkins model or the econometric model. When the two hybrid models are compared, the first model is the best, judged on forecasting accuracy. An important feature to consider when using the approaches, however, is the ease of use. On this basis, the alternate approach is best.