Improving Sharpness and Calibration in Density Forecasting Combination using Bayesian Large Models

R. Morales Arsenal, M. Á. Gómez Villegas

Forecasting with large sets of data is one of the most important fields in Macroeconomics. In order to enlarge the information set we use two models: 1) Bayesian Global Vector Autorregresive Models (combining BVARs and Global VARs models) and 2) Bayesian Neural Networks (BNNs). In both models, we use different prior specifications to obtain the density forecast applied to the HICP in the euro area. In a second step we combine the two obtained density forecasts from the two models used in order to obtain the final density forecast. For this task, we use bayesian techniques in a similar way to the one used in Hall and Mitchell (2004). For the evaluation and calibration of density forecasts we use: 1) the probability integral transform (PITs) and 2) scoring rules. The final result shows strong evidence in favor of the combined approach instead of the individual ones in terms of calibration and sharpness.

Keywords: Density Forecast Combination Bayesian Models Calibration and Sharpness.

M. E. Castellanos Nueda, G. García-donato Layrón, S. Cabras

R. Sillero Denamiel, J. M. Marín, P. Ramirez Cobo, F. Ruggeri, M. Wiper

P. Barbillon, A. Forte Deltell, R. Paulo