Comparing Impulse Response Estimation across VAR and Local Projection Methods

Abstract This paper examines the estimation of impulse responses in macroeconomic analysis using Vector Autoregression (VAR), Local Projections (LP), and Bayesian Network Vector Autoregression (BNVAR) models. Impulse response estimation is crucial in evaluating the effects of economic shocks on macroeconomic variables. While traditional VAR and LP models have been extensively used, the BNVAR framework introduces a novel approach by incorporating network structures to capture variable interdependencies. We conduct extensive bootstrap simulations on a well-known macroeconomic dataset, constructing various data-generating processes (DGPs) to compare the performance of these methods. Our results show that the BNVAR model provides more efficient pointwise impulse response estimates and better identifies causal dependencies than conventional VAR and LP methods. Moreover, we compare BNVAR with Granger causality and find that BNVAR offers improved accuracy in determining the relationships among economic variables.