# The Jacobian or the dynamic model contains Inf?

I get the following the error message when I run my code:
“The Jacobian or the dynamic model contains Inf. For more information, use options_.debug” Then when I use 'options_.debug" I get ans=0. I am not sure what’s going on. Can anyone help?
I checked my equations for mistakes (many times) and I am pretty confident they are all correct. And the steady state is well defined.

I get

[quote]STOCHASTIC_SOLVER: The Jacobian of the dynamic model contains Inf. The problem is associated with:

Derivative of Equation 17 with respect to lag of Variable Ai (initial value of Ai: 0)
Derivative of Equation 23 with respect to lag of Variable Ai (initial value of Ai: 0) [/quote]

You are dividing by Ai, which has steady state 0, but it is not in exp in some equations, resulting in a division by 0.

Thank you! I rewrite my equations and still made those mistake.
One last quick question: I know that changes in variables have to be over a certain threshold for dynare to display their IRF, how and where can I change that? Thanks in advance.

Cheers,

See the manual on

``irf_plot_threshold``

Thank you very much.

The process for your exogenous variables is wrong:

``````Derivative of Equation 31 with respect to lag of Variable Al  (initial value of Al: 0)
Derivative of Equation 34 with respect to lag of Variable upsilon  (initial value of upsilon: 0)
``````

For example, `Al` is already in logs, so

``````Al = Al_star^(1-rhoal) *Al(-1)^rhoal   *exp(sigmaal*eal);
``````

will not work, because it assumes it to be in levels.

Why have all your shocks variance 1?

1. Please upgrade to Dynare 5.4. Use this steady state file:
2. Your observation equations are wrong. Your are matching mean 0 data to level variables still containing the steady state.

Thank you Professor, So How should I set the observation equation to match my model? I did not found answer in ”A Guide to Specifying Observation Equations for the Estimation of DSGE Models" I am so sorry.

Most probably, you simply need to demean your data, but without more information on what you did in your data, it is impossible to tell.

I previously filtered the data by HP. So now I just need to log the original data and then remove the mean, without any other operations

1. The HP-filter should result in demeaned data. But it’s not in your case, which begs the question what went wrong.
2. Using two-sided HP filtered data for estimation is problematic.
3. You are supposed to log the data before filtering.

professor，It is still wrong. I took the logarithm of the raw data and then did one-side HP filtering。

Without the files it is impossible to tell.

Search the forum for stochastic singularity. It is very likely that your model implies an exact linear combination between the observables.

Thank you, professor