Forecasting with quantitative methods the impact of special events in time series

Abstract

Quantitative methods are very successful for producing baseline forecasts of time series; however these models fail to forecast neither the timing nor the impact of special events such as promotions or strikes. In most of the cases the timing of such events is not known so they are usually referred as shocks (economics) or special events (forecasting). Sometimes the timing of such events is known a priori (i.e. a future promotion); but even then the impact of the forthcoming event is hard to estimate. Forecasters prefer to use their own judgment for adjusting for forthcoming special events, but humans’ efficiency in such tasks has been found to be deficient. This study after examining the relative performance of Artificial Neural Networks, Multiple Linear Regression and Nearest Neighbor approaches proposes an expert method which combines the strengths of regression and artificial intelligence.