Website CV Paper
Fields
Econometrics, Finance, Macroeconomics
Job market paper
Escaping the Gravity of the Mean: Quantile Bayesian Predictive Synthesis for Macro-financial Tail Risk
While central banks closely monitor downside macro-financial risks, standard dynamic forecast combination methods available to them typically optimize for average conditions rather than distribution tails. This paper introduces Quantile Bayesian Predictive Synthesis (QBPS), a dynamic framework that combines forecasts through time-varying linear weights optimized directly for specific quantiles. First, I propose a novel quantile likelihood parameterization that separates tail shocks from baseline volatility, alleviating numerical instabilities when targeting extreme distribution tails. Second, I introduce an observation-driven mechanism that adaptively regulates how quickly combination weights update, accelerating adjustments during sudden economic transitions. Applied to quasi-real-time US Growth-at-Risk forecasting using financial indicators across 1- and 4-quarter horizons, QBPS delivers competitive predictive accuracy overall, while outperforming standalone machine learning forecasts, standard mean-based combinations, and established Growth-at-Risk benchmarks during recessions. Finally, dynamic weight decompositions enable real-time risk attribution: at short horizons, the decomposition highlights intermediary leverage as the primary leading indicator for downside risk, while at extended horizons, the predictive contribution of financial indicators diminishes.