Weird IRF with kinks and oscillatory return to steady states

Dynare contributions and examples
1

Hi everyone,
I’m running my codes (attached) in Dynare with 1st order linear approximation, but I get very weird shaped IRFs that have lots of kinks, and some of them go back to steady states more than once. My suspicion is that I am not using dynare properly, but I couldn’t figure out which part goes wrong. Could anyone point to the right direction for me?
Or would log-linearising them manually before running them in Dynare generate more sensible IRFs (smooth or hump-shaped)?
Thank you very much!
wxirf.mod (6.57 KB)

2

Or this kinked IRF could be a model feature?

3

This looks like a model issue. Is it on purpose that your model has a unit root?

4

Hello jpfeifer,
The unit root is not on purpose. I’m new to Dynare, so I’m not sure what goes wrong. I checked the FOCs and market clearing conditions are correct. Could it be timing-convention problem or should I log-linearise the system? (I’ve got over 40 nonlinear equations, log-linearisation would prove very difficult)
Would appreciate some hints!

5

It could be a timing issue. Log-linearizing does not help. As the unit root is not on purpose, there still must be a mistake in your model.

6

The mistake should be a timing mistake, right?

Thanks.

7

Maybe. It could be anything from a wrong equation, a typo in parameter initialization, a timing mistake…

8

I am experiencing the same problem. My code is attached. The IRF are weird and returning to the steady state.
Anyone could help me?
Thanks.
BNDES_K.mod (5.0 KB)

9

There must be a mistake in your model. Dynare 4.5 says:

ERROR: If the model is declared linear the second derivatives must be equal to zero.
       The following equations had non-zero second derivatives:
       * Eq # 16
10

Hi, Professor, thank you for the hint!
I got the idea, but, I didn’t get the way out in order to solve the problem.
Could you explain with examples in order to get it?
Thanks a lot!

11

It means that equation 16 cannot be correct in your mod-file, because it is not linear in the variables.

Qss*Q*(1 + chi*(I - I(-1)))

has a product of Q and I

13

Hi, Jpfeifer! Thanks for helping!
I looked for the Eq. 16 and understood what was wrong. I corrected the archive, but he problem is the same! The IRFs have zig-zag shape.
BNDES_K.mod (5.2 KB)

14

I can only give you the standard advice: simplify your model as far as possible and check where the problem arises.

15

Thank you for the advice! Another question: when I have BK condition verified, like
There are 8 eigenvalue(s) larger than 1 in modulus
for 8 forward-looking variable(s)

and

The rank condition ISN’T verified!

what does it mean? Probably I have something related to the parameters?

16

No, usually it is not about parameters, but about a problem with the model setup. Please run model_diagnostics

17

Pfeifer,
I am putting my model in a basic setup. In order to do that, I am in trouble with Phillips curve. The models I have they put Phillips curve in linear way. If I want to put my model in exp() way, how could I do that? Do you have some references to build a Phillips curve in a exp() way? Thanks again!!!

18

As always, do not do any exp()-substitution until your model runs.