Can Long-Run Restrictions Identify Technology Shocks?

Gali's innovative approach of imposing long-run restrictions on a vector autoregression (VAR) to identify the effects of a technology shock has become widely utilized. In this paper, we investigate its reliability through Monte Carlo simulations of several relatively standard business cycle models. We find it encouraging that the impulse responses derived from applying the Gali methodology to the artificial data generally have the same sign and qualitative pattern as the true responses. However, we highlight the importance of small-sample bias in the estimated impulse responses and show that the magnitude and sign of this bias depend on the model structure. Accordingly, we caution against interpreting responses derived from this approach as ``model-independent'' stylized facts. Moreover, we find considerable estimation uncertainty about the quantitative impact of a technology shock on macroeconomic variables, and a corresponding level of uncertainty about the contribution of technology shocks to the business cycle