Combining Forecasts Using Ensemble Learning

The problem of combining individual forecasters to produce a forecaster with improved performance is considered. The connections between probability elicitation and classification are used to pose the combining forecaster problem as that of ensemble learning. With this connection in place, a number of theoretically sound ensemble learning methods such as Bagging and Boosting are adapted for combining forecasters. It is shown that the simple yet effective method of averaging the forecasts is equivalent to Bagging. This provides theoretical insight into why the well established averaging of forecasts method works so well. Also, a nonlinear combination of forecasters can be attained through Boosting which is shown to theoretically produce combined forecasters that are both calibrated and highly refined. Finally, the proposed methods of combining forecasters are applied to the Good Judgment Project data set and are shown to outperform the individual forecasters.