Can anyone help - dynare data estimation problem

Estimation ML/Bayesian estimation
1

Dear Prof Jpfeifer,

I m trying to estimate my model using bayesian method. but dynare cannot proceed at all. However, when i do stochastic simulation without data, it can proceeds smoothly without any problem.
I attached zip file for your attention.

thank you so much
Mymode21.rar (17.3 KB)

2

When I faced this problem in the past, it was because some of my data were not stationary. You may want to check. Also, you can use the stochastic model to generate artificial data and use that for the Bayesian estimation. If you have no problems using the artificial data, then the problem is with your real data. Perhaps some of them are not stationary.

3

@yenyen What is the error message you are getting?

4

here it is.
tq
Mymode21.log (124 Bytes)

5

one more attachment on errors from matlab

thanks a lot
dynare_error.pdf (2.6 MB)

6

Hi yenyen,

with the estimation command

estimation(datafile=‘data7505.xls’, mode_compute=4, mh_replic=2000, mh_nblocks=1, mh_jscale=0.45);

it works for me.

Best

7

The problem is that the sheet name must start with a letter as of right now. Changing the name to a7505 solve that problem. With the attached files I can run the estimation, but there is a problem with mode-finding. The red dots indicate that estimation ran to the boundary of the indeterminacy region.
data7505.xls (43.5 KB)
Mymode21.mod (6.2 KB)

8

Thanks so much Professor.
Since i m quite new to dynare environment, i dont quite understand “there is a problem with mode-finding. The red dots indicate that estimation ran to the boundary of the indeterminacy region”.
Can you suggest how to go about this problem?
Thanks again and appreciate your help.

9

Just forget to ask you, if this problem didn’t solve can i still proceed to do stochastic simulation, irf, etc?

thanks a lot

10
  1. You need to find out why the model assigns such a high likelihood to parameters close to the determinacy region. Sometimes there is an economic mechanism at work. Sometimes there is a problem in the data treatment.
  2. With simulations, you pick the parameters. That is fine. The problem is with estimating the parameters.
11

thank you so much

12

Hi Professor,

It is something to do with values I assigned in the estimated_params in the model, for example
Cupsilon, beta_pdf, 0.5, 0.05;
eta, gamma_pdf, 0.5, 0.1;
thetaF, beta_pdf, 0.8, 0.1;
and so on.

Thank you very much.

13

Hello,
Did you encounter any problem when running the estimating parameters process
especially what Professor Jpfeifer pointed out in my reply?

thanks very much

14

Not really. You can see in the mode_check-plots that the prior hardly changes the likelihood as posterior and likelihood have almost the same shape.

15

thank you very much

16

i am also facing a same problem

18

S
sytest2.rar (646.8 KB)
sytest2.mod (6.3 KB)
sydata.xlsx (20.2 KB)
orry for interrupting again. I m trying to fix the problem but could not get it done.
would you mind take a look again into my model? I attached herewith the error message and dynare code and data.
thank you very much for your help

19

It seems your data is not demeaned, but your model variables have steady state 0

20

Dear Prof.
Thanks for pointed the errors.

To demean data:
i follow this few steps

  1. seasonally adjust the data (y)

  2. take the log of the data (ly)

  3. apply the one-sided HP-filter

lyh =ly - one_sided_HP_filter(ly)

next step to demean lyh

lyh = detrend (lyh(2:end),0); %% (x,0) which means removes the mean value from x

Is this the correct steps?

thanks very much

21

The one-sided HP-filter should already take care of the mean. From what I saw, the problem are the interest rates that should not be filtered.