Title: A score-driven model for GDP-at-risk
Authors: Davide Delle Monache - Bank of Italy (Italy)
Andrea De Polis - University of Warwick (United Kingdom) [presenting]
Ivan Petrella - University of Warwick (United Kingdom)
Abstract: A fully parametric model is proposed to characterize the predictive density of GDP growth. In our trend-cycle model, the disturbances follow an Epsilon Skew-t distribution with time-varying moments (location, scale and shape), whose dynamics is driven by the score of the predictive likelihood and possibly by additional exogenous information. When we include financial condition indices as additional drivers in the updating processes, we observe significant improvements in the out of sample predictive ability for different horizons reflecting the model's ability to pick up in a timely manner changes in the shape of the forecast density. We also recover most of the stylized facts about GDP growth documented in literature. Particular attention is devoted to GDP vulnerability as proxied by the asymmetry of the predictive distribution. We find that financial tightening are robust drivers of left skewness of the predictive distribution, ultimately sharpening economic growth predictions at the onset of recessions.