Can SW07 RMSE exercise be reproduced with dynare_sensitivity

Dear all,

I try to compute the root mean square error (RMSE) for a couple of variables using a successively expanding sample, equivalently to Smets and Wouters 2007.

As I understand SW07, they estimate the model over a sample, compute a forecast, and store the forecast error. Then they expand the sample, estimate the model, and so on. Finally they compute the RMSE. The model needs to be estimated for every sample which is quite computationally intensive.

I came across the dynare_sensitivity command which has an option for RMSE. Yet, I do not fully understand it. Can the SW07 exercise be replicated using this command?

Thank you for your help!

Best, Christian

That might be doable, but would be cumbersome. What you would need to do is run a recursive simulation using

nobs=sampleendpoint1:sampleendpointT

with the

option enabled. That will result in the estimation over the expanding sample, generating the required forecasts. You can then easily compute the RMSE yourself.