A0853
Title: Predictive density combination using a tree-based synthesis function
Authors: James Mitchell - Federal Reserve Bank of Cleveland (United States) [presenting]
Tony Chernis - Bank of Canada (Canada)
Niko Hauzenberger - University of Strathclyde (United Kingdom)
Florian Huber - University of Salzburg (Austria)
Gary Koop - University of Strathclyde (United Kingdom)
Abstract: Bayesian predictive synthesis (BPS) combines multiple predictive distributions based on agent opinion analysis theory and encompasses a range of existing forecast pooling methods. The key ingredient in BPS is a synthesis function. This is typically chosen from particular parametric forms (e.g., linear or following a random walk). A nonparametric treatment of the synthesis function is developed using regression trees. The advantages of the tree-based approach are shown in two inflation forecasting applications. The first uses density forecasts from the Euro Area's survey of professional forecasters. The second combines the density forecast of US inflation produced by many regression models involving different predictors.
