CMStatistics 2016: Start Registration
View Submission - CFE
Title: Adaptive combination schemes for point and density forecasts Authors:  Leopoldo Catania - Aarhus BBS (Denmark) [presenting]
Tommaso Proietti - University of Roma Tor Vergata (Italy)
Abstract: Point and density forecast combinations based on some optimality measure are becoming increasingly popular for pooling information coming from different sources. We propose a new way of combining point and density forecasts allowing for time varying weights. Our method allows for model misspecification and does not require that the true data generating process belongs to the available set. Concerning the density combination schemes, our combined predictive density dynamically approximate the true unknown density in a Kullback--Leibler sense using a recursion based on the score of the implied mixture conditional density. Similarly, for the point forecast combination schemes we propose, the weights are updated using the gradient of the user defined loss function. The relevance of the new combination techniques is illustrated by several Monte Carlo experiments and an empirical application in time series econometrics.