A0368
Title: Combining forecasts based on the evidence against equal weights
Authors: Lukas Bauer - University of Freiburg, Statistics and Econometrics (Germany) [presenting]
Abstract: A novel class of performance-based forecast combination schemes is proposed. The schemes determine the combining weights using the standardized loss difference of each model relative to the average model, and thus account for the statistical magnitude of the differences in predictive ability. The risk of two such combining schemes is characterized relative to the best individual model. These schemes build an intuitive bridge between model selection and forecast combination while showing robust performance in finite samples. An empirical application is performed, combining forecasts of financial risk, i.e., realized volatility and value-at-risk. The data are large-cap stocks from the NYSE trade and quote database, for which the novel scheme's performance is found to be competitive.
