New algorithms for automatic modelling and forecasting of decision support systems

Decision support systems often rely on time series forecasting, making the accuracy of such systems of paramount importance for their efficiency. Since most systems nowadays require the processing of massive amount of data, automatic identification of time series models has become inevitable. This automatism is inherent in artificial intelligence methods, but it often goes unnoticed that forecasting 'classical' methods have also been developing their own automatic methods for a long time. The radical novelty of this paper is the development of a brand new algorithm for identification of structural Unobserved Components models from which decision support systems may benefit. A second point is that combination of forecasts is more fruitful than competition or method selection in some cases. Both points are illustrated in two examples that show the effectiveness of the identification procedure and the forecasting gains when fairly different methods are combined.

Automated summary

This paper discusses the importance of time series forecasting in decision support systems, introduces a new algorithm for identifying structural Unobserved Components models, and highlights the effectiveness of combining forecasts over competition or method selection in certain cases.