COVID-19, US macroeconomic tail risk, and inflation forecasts

The COVID-19 pandemic undermined well-established macroeconomic dynamics and created outliers in the distribution of main macroeconomic variables. The inclusion of extreme observations in the data substantially distorts the estimated parameters and out-of-sample forecasting from standard Bayesian Vector Autoregressions (BVARs). To capture these tail events, we use a newly developed BVAR model, with a generalized hyperbolic skew Student’s t-distribution and stochastic volatility for the innovations, which considers both skewness and heavy tails. Our empirical study, based on 12 key US macroeconomic variables ending in 2023:Q4, shows that the pandemic created macroeconomic tail risk due to a simultaneous shift in the tail fatness of macrovariables. The COVID-19 shock generated a long-lasting increase in stochastic volatility for the real GDP, inflation, and the Fed rate. By contrast, the increase in volatility for the labor market was transient. We also find that inflation responded differently to shocks in the real output, monetary policy, and tightness in the labor market during the pandemic as compared to the pre-COVID period. Our analysis reveals a changing behavior of the Phillips curve, which steepened at the beginning of the pandemic and flattened again by its end. Finally, we compare the inflation forecasts by the proposed BVAR model with those generated by a battery of competing models. We find evidence of added forecastability in point and density forecasts of the inflation rate during the pandemic period and for multiple subsequent periods.