Estimation weird mode_check plot and posterior

Hi professor, I was trying to estimate parameters, but the mode_check plot and posterior distribution seems to be weird. Is the problem probably about the prior or model equation? How should I solve this problem. Could you please give me any clue? Thank you so much for your help.


Your estimation is hitting various bounds. You need to find out why. Most likely, there is a problem with the observation equations. Did you set prior_trunc=0?

Yes, professor, I set prior=trunc=0. Is it possible that I didn’t correctly set the dependent parameters? I read your paper “A Guide to Specifying Observation Equations for the Estimation of DSGE Models”. It said the error might arise if the dependent parameters not correctly set.

There are a lot of possibilities, but without the codes it is impossible to tell.

Thank you so much for your reply. I uploaded my mod files in another question. Estimation problems. Could you please give me any clue? Thank you for your help.

It seems you are matching non-zero steady state variables in the model to demeaned data. Thus, you are forcing shocks and persistence to account for the mean.

Thank you professor so much for your help. Your suggestion is so inspiring. I estimated the model based on 8 data. 4 of them are inflation rate for two countries and interest rate for two countries. The other 4 data are output growth rates and housing price growth rates which should be zero as variables in model. Do you mean that I shouldn’t apply hp-filter on inflation rate and interest rate so that they would not be demeaned? Am I correctly understanding?

Professor, I have another follow up question. Should I just stop the estimation if the “New value of jscale” always keeps very small, like 0.0000374, till the end? Does that mean that something goes wrong with the estimation?

Professor, thank you so much for your help. With your suggestion, the estimation improves a lot.

But there is still some red dots on mode_check plot. I am really confused with one question. As for the parameter phics, I set the range between [6.5, 10] as following picture shows.

But the the results reported that phics at some values, like 4.004 out of my prior range, not satisfy the lower bound. Why this happen given I already set the range. Could you please give me some clue?

  1. You need to be consistent. If you filter the rates, then you need to match them to demeaned variables in the data.
  2. The value of jscale should not be tiny. That after means you ran into some bound.
  3. mode_check will try an interval around the detected mode and only then check whether it satisfies specified bounds. That is nothing to worry about (except for the fact that the mode seems to be right at the bound).