B1915
Title: Bayesian dynamic calibration of models predictions
Authors: Roberto Casarin - University Ca' Foscari of Venice (Italy) [presenting]
Abstract: The aim is to propose a calibration model that combines and dynamically calibrates predictive densities. While the weights are statically estimated, the time-varying calibration is introduced giving observation-driven dynamics to the parameters of the calibrating function which is driven by the score of the assumed conditional likelihood of the data-generating process. The model is very flexible and can handle different shapes, instability and model uncertainty in the data-generating process density. Its effectiveness is shown on various simulated datasets. Two empirical applications are also introduced, one on financial index density forecasts and one on short-term wind speed predictions. Both the simulations and the empirical applications document the large instability of individual model performance compared to the properties of the combined and calibrated forecasts, favouring the model in terms of predictive accuracy.
