When is it appropriate to use DSGE models for forecasting?

Hi everyone,

I hope this message finds you well. I have a question regarding the use of DSGE models for forecasting. I understand that DSGE models minimize one-step-ahead forecast errors during estimation, but I’m wondering about their broader use for forecasting:

Is it true that DSGE forecasts rely on the assumption of a stable economic structure? If the structure changes (e.g., due to regime shifts or major shocks), are forecasts invalid?
Beyond structural changes, are there other cases where DSGE models are unsuitable for forecasting (e.g., extreme shocks, multiple shocks, non-linear dynamics)?
I’d appreciate any insights on when DSGE models are appropriate for forecasting and how to handle structural changes.

Thanks!
Yamoi

  1. Forecasts in such cases will not be invalid, but rather may perform poorly relative to the actual data realizations.
  2. Fundamentally, forecasting with DSGE models assumes that the model solution is the data generating process. Whenever that assumption is invalid, the the forecasts will deviate from the data for other reasons than just unforeseen shocks. That may be because the model solution is linear but the DGP is not, because the model assumes no structural breaks but there are some in data, etc.

That’s very helpful. Thank you, Prof. Jpfeifer.