How Reliable Are VAR Estimates of Responses to Monetary bPolicy Shocks?

We study the reliability of the estimated responses of major economic aggregates to monetary policy shocks. We investigate the bias of the estimated impulse response functions and the reliability of their qualitative features such as their sign and shape and the timing of peaks and troughs. We also study the coverage accuracy of the corresponding pointwise confidence intervals. This simulation study is the first of its kind to focus on large-dimensional VAR models with many lags. Our data generating processes are based on VAR models estimated by leading practitioners in the field. In contrast, existing studies have largely focused on artificial bivariate VAR(1) models without economic interpretation. We find that in general qualitative features of impulse response functions are not estimated reliably. There are only minor differences in coverage accuracy between the intervals proposed by Kilian (1998a) and Sims and Zha (1995). Both intervals tend to be more accurate than the asymptotic delta method, but somewhat erratic. Poor coverage accuracy is associated with bias in the impulse response point estimates. In particular, impulse response point estimates and confidence intervals both tend to understate systematically the magnitude of the true output response to a monetary tightening.