Non-DSGE models


Curious to know if Dynare can be used for non-dsge macroeconomic models? The Bank of Japan’s Quarterly Economic Model is a good example of the style of model I am interested in, which can be found here:

The authors of the model/paper describe it as large scale semi-structural model. Looking through the paper, there are forward looking variables, but seeing that the model (which is provided by the authors) is written in Eviews, my guess this would use extended path (Fair-Taylor) if these are model consistent expectations. Putting that to one side, the real question is, can Dynare be used to solve macroeconomic models that are built up of a number of error correction models / other reduced form econometric models with or without model consistent expectations?

Could Dynare be used for solving such a model, if parameters were estimated and in another software and written to the .mod file?

Apologies if this has been covered before, I am new to Dynare and couldn’t find any examples when searching “semi-structural model” which fit the bill. The closest I could could find was the work by Douglas Laxton, which is a little different - but does suggest that this type of model should be possible.

Any examples would be appreciated.


1 Like

The short answer is yes. Dynare allows you to use any nonlinear setup in a perfect foresight context. In a stochastic context, you can easily handle linear Gaussian models. So combining an RE model with econometric linear state space models like VARs is not an issue. An example of this is the VAR in

I would add that you can simulate any nonlinear model whose expectations are not model-consistent (as long as those expectations are a function of past variables, i.e. the model is actually backward, like e.g. with adaptative expectations).

However there is currently no facility for estimating such models (though that may be added the future).

For models with model-consistent expectations, @jpfeifer pretty much summed it up. I would add that any nonlinear model can also be simulated using the extended path (Fair-Taylor) method.

Thanks both. Seems like this will work just fine! Estimation of parameters is not an issue at the moment.