Title: Optimal combination of forecasts under mean absolute deviation
Authors: Laurent Pauwels - University of Sydney (Australia) [presenting]
Felix Chan - Curtin University (Australia)
Abstract: Theoretical motivations are presented for combining forecasts optimally under the Mean Absolute Deviation (MAD) criterion. While the literature has covered extensively the optimal combination of forecasts under Mean Squared Errors (MSE) theoretically and empirically, it is sparse with respect to MAD. Under normality, the optimization problem for MAD is demonstrated to yield the same solutions as the optimization problem for MSE. The forecast errors are normally distributed as the number of forecasts combined and/or the forecasting sample are increasingly large. Furthermore, a set of conditions for which the simple average is the optimal weight is provided. Simulation studies support the theoretical results and show their relevance when combining a large number of forecasts and/or for long time-series.