Forecasting and Modeling Macroeconomic Vulnerabilities in CESEE

ABSTRACT This paper develops a nonparametric multivariate model for assessing risks to macroecononomic outcomes in three major CESEE countries. Our model builds on Bayesian additive regression trees (BART) that remains agnostic on the relationship between the macro series and the lags thereof. Our model produces predictive distributions that exhibit non‐Gaussian features such as heavy tails, asymmetries, or multi‐modalities, making them suitable for policy analysis in extreme environments. We show that our BART model yields tail forecasts of output growth, inflation, and financial risks that are often more precise than the ones of a linear benchmark model. We then move on to analyze how the tails of selected macro series react to domestic and euro area–based financial condition shocks.