Abstract
Quantitative methods are very successful in producing baseline forecasts of time series; however, these models 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 judgement for adjusting for forthcoming special events, but human efficiency in such tasks has been found to be deficient. This study after examining the relative performance of Artificial Neural Networks (ANNs), Multiple Linear Regression (MLR) and Nearest Neighbour (NN) approaches proposes an expert method, which combines the strengths of regression and artificial intelligence.
Acknowledgements
The Manchester Business School (MBS) (Research support fund 2006/07) funded this work as part of a 1-year interdisciplinary project on the development of an innovative forecasting methodology for TV ratings. The author would like to thank especially Mr Konstantinos Giannelos for providing the software for running the experiments with the ANNs, a software developed in July 2004 in the Forecasting Systems Unit in National Technical University of Athens for his Diploma dissertation entitled ‘Irregular Events Impact Estimation Methodology and Information System’, under the supervision of Dr K. Nikolopoulos, Assistant Professor in MBS and Professor Vassilis Assimakopoulos, Special Secretary for Digital Planning in the Ministry of Economics and Finance in Greece. The author would like to thank Prof. Paul Goodwin of the School of Management at the University of Bath for his suggestions in the first submission of this research as well as the Editor of AE and an anonymous referee for their suggestions in the revision of this article.