An improved model to forecast tail risk to macro indicators

Quantile regression has become widely used in empirical macroeconomics, in particular for estimating and forecasting tail risks to macroeconomic indicators. Carriero (Queen Mary University of London), Clark (John Hopkins University) and Marcellino (Bocconi University, Baffi Centre, in photo) examine various choices in the specification of quantile regressions (QR) to forecast tail risk to macroeconomic indicators, such as GDP growth, inflation, and unemployment in the United States and GDP growth in some other advanced economies.

The paper “Specification Choices in Quantile Regression for Empirical Macroeconomics”, published in February 2025 in the Journal of Applied Econometrics, compares the accuracy of quantile forecasts at different horizons and using different models. Carriero, Clark and Marcellino find that  Bayesian quantile regression performs best. The gains to Bayesian shrinkage are relatively large and consistent (across applications, horizons, and quantiles) for applications featuring more than a few predictors.

Absolute forecast performance also favors Bayesian quantile regression over frequentist quantile regression, with forecast optimality typically rejected for simple frequentist quantile forecasts and not for Bayesian quantile forecasts.

Based on the paper results on forecast accuracy, newer analyses of risks to GDP growth, inflation, and unemployment would be better based on Bayesian quantile regression rather than the classical quantile regression that is most commonly used in macroeconomic applications.