Variance Decomposition Differences

Estimation ML/Bayesian estimation
1

Hi Dynare Community,

I have just estimated a model. All the convergence statistics look very good. Right after my estimation command, I have a stoch_simul command. So when the .mod file runs, my understanding is that the stoch_simul uses the posterior means of the estimated parameters from the estimation to generate the IRFs etc. I assume it also uses this set of parameter values to produce the variance decomposition.

My issue is this: the variance decomposition that is produced immediately after estimation + stoch_simul looks very different from the variance decomposition that is produced when I run stoch_simul after manually loading the estimated parameter values. Specifically, I use the same .inc file with same model equations, but right after the shocks block, I use:

x=load('Baseline_Estimation_results.mat');
M_.params=x.M_.params;
steady;
stoch_simul(hp_filter=1600, order=1, irf=25, relative_irf) Pie Y inv L C W;

The variance decomposition that is now produced is wildly different. The IRFs look identical to the IRFs after estimation + stoch_simul, so the correct parameter values are being loaded. What am I missing?

Thanks in advance for your time!

2

Did you estimate elements of the covariance matrix? If yes, you did not set M_.Sigma_e and potentially M_.H for measurement error.

3

You are absolutely right, Prof. Pfeifer – I did forget to update the covariance matrix, so the variance decomposition is reflecting the variance of the shocks which I had calibrated, rather than those I estimated. I don’t have any measurement errors in my observables, so hopefully I don’t need to worry about that.

Thank you for the quick response!!