The vector autoregressive (VAR) model is one of the key models for macroeconomic forecasting and structural analysis. However, the emergence of COVID-19 has rendered the estimation and inference of VARs in the post-pandemic period challenging and has highlighted the need for outlier-robust models. The outlier-robust VAR model assumes that the reduced-form error is a sum of the regularized error and the outlier. Then, the PLS estimator of the VAR coefficients is obtained by imposing an L1 penalty on the outlier. It is demonstrated that the robust estimator of the VAR coefficients is equivalent to the multivariate Huber’s M-estimator of the VAR coefficients. Subsequently, the asymptotic normality of the robust estimator is established on the foundation of the asymptotic normality of the multivariate M-estimators. To illustrate the methodology, we consider a five-variable VAR of macroeconomic variables, using monthly data from March 1959 to June 2024 sourced from the FRED-MD database. The results demonstrate that the out-of-sample forecasts for the
post-Covid period from the robust VAR are generally more accurate than those from the standard VAR and the outlier-augmented stochastic-volatility Bayesian VAR proposed by Carriero, Clark, Marcellino, and Mertens (2022).
Hyoseok Kim
Job Market Candidate
Fields: Econometrics, Macroeconomics
Main Advisor(s):
Serena Ng
and Jushan Bai
Advisor(s):
Abstract:
