Awkward MOMENTS OF SIMULATED VARIABLES for interpretation

Hi all,
After running the result out with help from the forum, I came across new difficulties when trying to interpretate it. The means reported in the MOMENTS OF SIMULATED VARIABLES of most variables are extremely different from what I expect or not the same at all with the model economy.

1.When doing stoch_simul with periods option on, the means of simulated variables looks strange. Some are very big, some are negative, some are close to zero. More importantly, some bounded variables such as interest rate are in tens or hundreds.
2. When doing stoch_simul with periods option off, the means of theoretical variables are all zeros. What does it mean?

So, my questions concern:

  1. Do the MOMENTS OF VARIABLES need to be as realistic as possible? i.e.,whether it is a way for testing the fitness of the model to the real economy?
    2.If the moments of variables already display bad results such as mine, can I still do interprtation on the IRF of the result or the model has to be modified before going on further?

Many thanks for your help.

First, check whether you simulate at second order and if yes, use the pruning option. Theoretical moments of 0 indicate a problem. You might want to use 4.4 and run model_diagnostics. If the problem persists, you might want to post the mod-file.

  1. Yes, moments should be realistic (although that is often hard to judge given that empirical counterparts are often directly equivalent to model variables)
  2. There seems to be a problem that you should fix before going on. However, when the IRFs are also obviously, this might help you finding the problem.

Thank you ,jpfeifer.

I just tried order=2 and pruning,and the theoretical moments are still all 0s. The results come out together with "Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND =1.781720e-23. " Then I ran model_diagnostics as you suggest, got the diagnostics of colinear equations report. So I replaced one colinear equation with other equation, but got the similar warning information.Then I tried model_diagnostics again, but thie time nothing displays.So, I am wondering what’s wrong with the model.

By the way, it seems that the simulated means are very sensitive to the number of periods. I s this a common feature of such kind models? When adding periods option with a huge number such as 210000, the simulated means will look much better than when periods are only 2100,for instance. So, is it a fairly good way to fix the means of the simulated variables to the desirable level by searching the “right” number of periods?

I am appreciated for your help and advice, indeed.



Please post the most recent mod-file

By the way, the model put into DYNARE is log-linearized. Someone said that the theoretical moments should be zeros for log-linearized model. Is it true? If so, does it mean that all evaluations such as means of the simulated variables are not in levels but in deviation from steady state? For example,if C=1, does it mean that the variable is positively above its steady state ?
I have post the mod-file.
I am sincerely looking forward to your answer and help. Many thanks.

If you loglinearized your model, it is in deviations from steady state. C=1 will mean that consumption is 1 percent above steady state - if you specify your shocks correctly.
This brings us to the problem with your mod-file. When you say e.g.

you are saying that this shock has a standard deviation of 359% (if the shock is in percent as is usually the case). For first order approximations, it does not matter whether you specify 3.59% as 3.59 or 0.0359. At second order it does and you must use the latter.

Lastly, if your model is loglinearized, all your first moments should by approximately zero in simulations. That does not hold true for second moments. They should not be 0.

Many many thanks, jpfeifer. Your easy-to-understand interpretation really help me solve the doubts. I am grateful for what your kindheart and inspiration.

On a (somehow) related note - it often happens that oo_.endo_simul comes out completely empty when running 2nd order approximations (with pruning). However, Dynare does not “complain” about anything. What is the likely problem? Thanks.

Did you specify the periods option? If yes, please send me a mod-file.

Thanks for taking the time to share this, I feel strongly about it and love learning more on this topic.