Dear Professor Pfeifer,

I set a DSGE model and run the Bayesian Estimation. The mode plots and the convergence of the two chains all looks OK. I thought the estimation results are good until I compare the distance between the model-implied moments and their empirical counterparts. The empirical moment are not in the 95% probability interval of model-implied moment as I wish.

Then,

(1) Does this mean that there is some problem with my model setting ?

(2) How should I improve this ? Is there any method to adjust the model-implied moment and make it close to the empirical ones ?

(3) Could I get better results if I try simulated method of moments estimation ?

Thank you for your precious time !

Best regards

That usually happens if the model has a hard time fitting the data. While the model has uncorrelated shocks, they may be correlated in sample. This is typically a sign of model misspecification, but not unheard of.

Dear Professor Pfeifer,

Thank you for your reply!

(1) Do you mean that the only way to improve is resetting the model ? However, the problem still exists after I simplified the model to the best of my ability. Is there any other possible mistake that may cause this type of problem ?

(2) When doing the comparison between model-implied moments and their empirical counterparts, I use posterior mean and command

“stoch_simul (order=1,irf=0,periods=2500000)”

and then calculate with 10,000 simulations of the same length as the data. For every simulated sample, I do the first-difference of every interested variable(e.g. y c i pi ) and calculate the standard deviation. Am I doing it correctly ?

Also, I see some paper use “order=3,pruning” in stoch_simul command. What’s the difference between them? I tried “order=3,pruning” and get very huge value of moments.

Thanks again for your time! I really appreciate your kindness.

Best regards

If the problem persists despite a correct comparison, then you can only modify your model.

Regarding the comparison, you should compare the theoretical moments (periods=0) of the observed variables to the ones in the data.