Avoiding pitfalls in using structural VARs to estimate economic models

Structural Vector Autoregressions with a differenced specification of hours (DSVAR) suggest that productivity shocks identified using long-run restrictions lead to a persistent and significant decline in hours worked. Economists have interpreted this evidence as showing that standard business cycle models in which a positive technology shock leads to a rise in hours are inconsistent with the data. In this paper we argue that such a conclusion is unwarranted because model's data and actual data are not treated symmetrically. To illustrate this problem, we estimate and test a flexible-price DSGE model with non-stationary hours using Indirect Inference on impulse responses of hours and output after technology and non-technology shocks. We find that, once augmented with a moderate amount of real frictions, the model can mimic well impulse responses obtained from a DSVAR on actual data. Using this model as a data generating process, we show that our estimation method is less subject to bias than a method that would directly compare theoretical responses with responses from the DSVAR.