Financial Forecasting Accuracy: Exploring the M3-Competition

Resumen

Forecasting availability has widely increased, suggesting the need of analyzing the adequacy of different alternative methods. One of the main empirical researches in this field is the M-Competition, whose last edition (M3) is referred to year 2000 and includes 3003 time series (308 of them related to finance) and 24 forecasting techniques. Since the statistical theory of information provides a suitable framework for the evaluation of econometric models, including their goodness of fit and their forecasting ability, we propose the use of two information-based accuracy measures: the U Theil Index and the Quadratic Information Measure. The obtained results are compared with those provided by the five accuracy measuresincluded in the M-competition: the symmetric mean absolute percentage error, the average ranking, the percentage better, the median symmetricabsolute percentage error, and the median relative absolute error). A sensitivity analysis is presented.