forecasting power of an estimated model

[anitazhang]

Dear all,

I am writing to ask if the dynare provides any statistics that reflect the forecasting power of an estimated model.

In particular, when I use the "estimation" command to estimate the model using a Bayesian estimation method, I notice that marginal densities are reported to reflect how well the model fits the data. But I could not find statistics such as forecast error that may reflect the forecasting power of the model. are such statistics reported somewhere? or, what options should I specify to obtain such statistics?

[jpfeifer]

No, Dynare currently does not do that. You would judge the forecasting performance using an out-of-sample forecasting exercise, e.g. by computing the RMSE.

[dominik]

Hi
I have come across a similar question.
assume I wanted to compare two models. Say one has a bad fit and one has a good fit. And say I'm interested in ex post explaining the behaviour of only one variable over the period that I have data for. Would I then not want to look exactly at what anitazhang proposed? I guess one way to get this number would be to define an auxiliary variables EY=Y(+1) and DY=EY(-1)-Y and look at the variance of the latter. If im interested in longer forecast horizons and more variables this would imply many auxiliary variables, but it should work, right?
EDIT: essentially i would like to get the level of the forecast error variance at different horizons for certain variables.