Session 6: Tail risk

Chair: Farrukh Javed

Andrea De Polis: Modeling and Forecasting Macroeconomic Downside Risk

Abstract: We model secular trends and cyclical changes of the predictive density of US GDP growth. A substantial increase in downside risk to US economic growth emerges over the last 30 years, associated with the long-run growth slowdown started in the early 2000s. Conditional skew- ness moves procyclically, implying negatively skewed predictive densities ahead and during recessions, often anticipated by deteriorating financial conditions, while positively skewed distributions characterize expansions. The modelling framework ensures robustness to tail events, allows for either dense or sparse predictor designs, and delivers competitive out-of-sample (point, density and tail) forecasts, improving upon standard benchmarks.

Benny Hartwig: Bayesian VARs and Prior Calibration in Times of COVID-19

Abstract: This paper investigates whether a heavy-tailed or time-varying volatility error structure is better suited in dealing with the abnormal COVID-19 observations in a Bayesian VAR and discusses pitfalls of using mechanical prior updates. This paper presents evidence that the COVID-19 shock is better captured as a rare event rather than a persistent increase in volatility. Not accounting for heavy-tailed errors may lead to imprecise density forecasts during the pan demic. This paper shows that mechanical updates of prior distributions which depend on scale estimates – such as the commonly used Minnesota prior – may be another source of parameter instability. To mitigate this sensitivity, a COVID-19 robust prior calibration strategy is put forward.

Michael Pfarrhofer: Investigating Growth at Risk Using a Multi-country Non-parametric Quantile Factor Model

Abstract: We develop a Bayesian non-parametric quantile panel regression model. Within each quantile, the response function is a convex combination of a linear model and a non-linear function, which we approximate using Bayesian Additive Re- gression Trees (BART). Moreover, the p th quantile response depends on a factor that summarizes the available cross-sectional information at that quantile. The non-parametric feature of our model enhances flexibility, while the panel feature, by exploiting cross-country information, increases the number of observations in the tails, providing more accurate estimates of them. We develop Bayesian Markov chain Monte Carlo (MCMC) methods for estimation and forecasting with our quantile factor BART model (QF-BART), and apply them to study growth at risk dynamics in a panel of eleven advanced economies.