Dear Dynare Community,
I am trying to use dynare’s stoch_simul command with the periods option set to a positive value to simulate time-series data to be used in another project.
Specifically, I am using Dr. Pfeiffer’s mod file for the Smets and Wouters (2007) model (link: DSGE_mod/Smets_Wouters_2007/Smets_Wouters_2007_45.mod at master · JohannesPfeifer/DSGE_mod · GitHub ) but instead of estimating I stochastically simulate the model.
The problem I encounter is that irrespective of the shocks’ specification the data contained in oo_.endo_simul is always linearly dependent, i.e. the rank of the data’s covariance matrix is always nearly zero.
Is there any way to adress this issue and render the simulated data less linearly dependent?
Thank you very much!
Hi,
The singularity is expected when considering all variables of the model.
First, the model is linear, and there are fewer sources of randomness (varexo) than endogenous variables (var), leading to stochastic singularity when examining all the endogenous variables. This is why, for likelihood-based estimation, it is necessary for the number of observed endogenous variables to not exceed the number of shocks.
Second, if you include all variables in an accounting identity among the endogenous variables you are analyzing (for example, Y = C + I + G), deterministic singularity will occur. In this case, it is essential to exclude at least one of those variables (or add a shock in the identity, but I suppose you do not want to do that).
Best,
Stéphane
Thank you Stéphane! I understand the issue better now and will try to adress it by focusing on a smaller subset of variables.