What is a good DSGE model?

The answer to this question maybe is, “anything the referee accepts”:slight_smile: .

Sorry to ask such a basic question here, but my macroeconomics courses over the years have looked something like this (besides general numerical methods):

  1. Professor solves the slow model, discuss advantages and criticisms, done.
  2. Ramsey model and its variants, advantages, criticisms, done.
  3. OLG model and its variants, advantages, criticisms, done.
  4. RBC model and its variants, advantages, criticisms, done.
  5. NK model and its variants, advantages, criticisms, done.
  6. CEE and its variants, advantages, criticisms, done.
  7. SW and its variants, advantages, criticisms, done.
  8. etc.

Each model for sure achieves its objective (whatever it is the authors wanted to show), but other implications and predictions are not always so great (even Nobel winning RBC :)).

So my guess is, focus on your objective, ignore the other weird implications and predictions of your model that you cannot do anything about (fix the ones you can). In summary, leave it to the referee to judge.

Or perhaps there is some acceptable threshold, like the model should predict 60% of original facts? Maybe 75%, 85%?

It is hard to get a single score like that to say a model is good or better as they do in machine learning. But yeah, what do reviewers consider a good DSGE model? If it can achieve its purpose even though it has other weird predictions and implications?

That depends on what you are trying to do with your model. If your goal is forecasting (as the comparison with machine learning suggests) then there are clear criteria like RMSE comparisons.

If you are not doing forecasting, then it’s usually about economic insights. Your model should help you to understand some facts that could not be understood by previous models.

1 Like