Model-implied regression coefficients

Hello everyone,

Can you do model-implied regression based on simulated dynare results?


Why not? But without context, it is hard to give a well-founded answer.

Thanks for your reply. For example, only state and exogenous variables are stored in oo_.dr.ghx and oo_.dr.ghu and presumably regressing impulse responses on oo_.dr.ghu is a bad specification?

Why would you want to run that particular type of regression?

I wanted to show that my model is consistent with data results. I have OLS regression and wanted to present an equivalent regression from the model.


You need to simulate your model, setting periods equal to a positive value in stoch_simul. The simulated data will be in oo_.endo_simul for the endogenous variables (each row is for a different endogenous variable, the variables are ordered as in M_.endo_names) and oo_.exo_simul for the exogenous variales (each column is for an exogenous variable, and the variables are ordered as in M_.exo_names) but you don’t need them (unless you want to compare the true innovations with the OLS residuals). Using the data in oo_.endo_simul, you can build the vector y and matrix X of the regression and compute the OLS estimator: \hat\beta = (X'X)^{-1}X'y.

We have developed codes to do that in dynare but they are not yet public. You can have a preview about the implementation in this post:



Hi Stéphane,

This is very helfpul, many thanks for the information!