After changing a parameter Blanchard & Kahn conditions are not satisfied

[catnip1997]

calib1.m (3.5 KB)
check.m (58 Bytes)
checkss.m (1.5 KB)
copy2.mod (6.1 KB)

Hi all, I am copying a paper, and I realized the original code have a mistake as beta, the utility depreciation factor will be larger than 1 if lambda_z, the growth of technology is set 1.02. However, the model worked in this case.
After I changed lambda_z to 1,as copy2.mod, the beta is less than 1, which makes sense, however in this case, it shows that
错误使用 print_info (line 32)
Blanchard & Kahn conditions are not satisfied: no stable equilibrium.
I’ve wrote a code (try run checkss) to examine the steady states in calib1.m and it seems correct. I keep wondering why. This question has been bothering me for months and it is related to my graduation paper. Hope to get your help soon.

[jpfeifer]

I am not very familiar with the paper you are looking at. But have you checked whether that is the only issue. Usually, only the growth adjusted discount factor appears in the detrended model, e.g. in the Euler equation:

beta/(pi_ss*lambda_z)

That ratio seems unaffected by your change. But what is different now is the Phillips curve, which seems to be the only place where beta shows up separately:

v_ss*v - Omega_p/theta_p*C_ss/Y_ss*(pi-beta*pi(+1)) = 0;

Is that correct?

[catnip1997]

Bench.mod (5.4 KB)
calib.m (3.7 KB)
CLS_Appendix.pdf (269.1 KB)
Thank you I’ve checked the equation you’ve mentioned and it seems right. I’ve attached the original code and Appendix of the authors, and you can check A3 in Appendix here,

I believe there is supposed to be betaR/lambda_zpi=1 for steady state, am I right? However in their code(see calib.m) you will find those parameters are all given and it is not satisfied. I think it is a mistake.
So, In my code(copy2.mod) I tried to calibrate beta as lambda_z*pi/R and replace ramsey_policy with a simple taylor rule as my tutor told me. Now everything seems really strange, and I don’t even know where the problem is…

[catnip1997]

I checked the equation you mentioned again by replacing the code of Philips curve with the code of the authors and it seems the same. I’ve checked my linearzation multiple times and I think there is not supposed to be anything wrong with it…

[jpfeifer]

1. I haven’t read the paper, but usually \beta\frac{R}{\lambda_z\pi}=1 for (A3) to hold with \psi_t=\bar \psi.
2. I am still not sure their original equation A(13) is correct. I still think the detrending is wrong.

[catnip1997]

Thank you for your reply. Unfortunately, my tutor told me this model might be wrong from the beginning (which means the original essay might have some mistake) and it would definitely take lots of time to fix this. As I have to finish my graduation paper ASAP so he advised me to give up my study on DSGE and try do my research with VAR. I feel really bad these days for the time I’ve spent on my research, so I didn’t reply you earlier. Anyway, thank you again for your help.

[jpfeifer]

I am sorry to hear that. But that sounds like a big problem with supervision. See e.g.

Replication of "Countercyclical Capital Regulation in a Small Open Economy DSGE Model" of Matija Lozej, Luca Onorante, and Ansgar Rannenberg

That depends on your background. If you have not worked with DSGE models before, then I would plan with at least 6 months. You will also need proper supervision, i.e. someone to turn to in case you encounter problems. If that is not assured, I would not try to do this.

and

Help me how to learn DSGE model

Here is my standard answer in such cases: Students are not supposed to work on models that complicated without intense supervision. I have seen too many cases like this miserably fail. Most of the time, the goals set are too unrealistic given the prior training and the time frame envisioned. The paper you have in mind is rather infamous for being extremely hard to replicate.

So any grading should take that failure of supervision into account.