Empirical Comparisons of Five Procedures for Combining or Selecting Forecasts

Abstract

The Makridakis et al. (J. Forecasting 1, 111–153, 1982 [1]) 111 time series were used to evaluate five procedures for combining individual forecasts produced by different techniques. The five combinations considered were a simple average forecast, the median forecast, two versions of focus forecasting (Smith and Wright. Focus Forecasting: Computer Techniques for Inventory Control, 1978 [2]) and a procedure developed by Bunn (Opn. Res. Q. 26, 325 329, 1975 [3]) and Bunn and Kappos (Eur. J. Opn. Res. 9, 173-18O, 1982 [4]). Results indicate that each of the five combinations can potentially result in improved forecast accuracy. The conditions under which each combination will he most likely to improve forecast accuracy are discussed.