Using time-varying VARs to diagnose the source of ‘Great Moderations’: a Monte Carlo analysis

In this paper, we assess the ability of time-varying VAR models to correctly diagnose the source of ‘Great Moderations’ generated in simulations of a learning model. We find that, in general, they can. For example, in data sets with Great Moderations generated by good policy, the VAR correctly identifies a downward shift in the policy disturbance. And it shows that if the policy behaviour associated with the latter part of the sample (during which policy is conducted well) are applied to the earlier part of the sample, the implied variances of output, inflation and interest rates would have been much lower. An important caveat to our results is that they appear to be sensitive to the method used to identification of monetary policy shocks. When we identify monetary policy shocks using a Cholesky decomposition, the VAR provides quite clear evidence in favour of the correct explanation for our simulated Great Moderations When sign restrictions are used to identify the monetary policy shocks, conclusions from the counterfactual experiments are less precise. The contrast between our results and previous work based on Monte Carlo evidence using RE models suggests that the ability of VARs to correctly diagnose the source of the Great Moderation may be dependent on the nature of the expectations-formation process in the private sector.