Hi,

I couldn’t find anything on this subject.

I’ve written the technology process as

```
a=rho_a*a(-1)+rho_a2*a_m1(-1)+e_a;
a_m1=a(-1);
```

where

```
rho_a=1.9;
rho_a2=(1-rho_a_A)-.001;
```

The eigenvalues of this AR(2) block are

Yet Dynare returns

While having a standard AR(1) process generates a saddle-path solution.

Am I missing something? Is this a known issue?

Thanks

Gianni

PS: I’m using version 4.4-unstable

Sorry, but I cannot replicate the issue. What exactly is rho_a_A in your definition of rho_a2? Is it supposed to be rho_a? Could you run the attached file on your machine and see whether it runs?

lombard.mod (390 Bytes)

Sorry, it was a typo. You correctly guessed it was rho_a.

As for the possibility of reproduce the problem, it is not so simple. That is why I was checking for

similar experiences. If you simply write the AR(2) in Dynare, it will solve. The problem, as I mentioned

is that in a larger model you go from being able to solve to no solution.

I’ll try to post a specific example, if I find a minute to wrap things together

Cheers

Gianni

I see. Have you checked whether adding an additional variable just decreases the numerical accuracy and the qz_criterium now is too narrow?