Justiniano and Preston 2010 - Replication attempt

Hello, I am trying to replicate the results from Justiniano and Preston (2010) paper “Monetary Policy and Uncertainty in an Empirical Small Open-Economy Model” with the data set they provide for Australia, but with now success.

When I try to obtain the mode, this is not even closer to the values that they obtain in their estimations. I am now not sure I am writing the measurement equations properly. Could you please help me out with this? I would really appreciate it.

I upload the dynare mod file and the data set I am using that is the one they uploaded.

DSOEM.mod (8.1 KB)
justiniano_data_400.mat (6.5 KB)

Thank you very much.


I cannot spot anything immediately suspicious. Investigating the mode_check-plots shows that the optimizer runs to the indeterminacy region. The big question is why?

Hi Johannes, thank you for the answer. I even tried with the data the authors provide for Canada and it does the same. They state in a footnote of their paper that they use four observations before the start of the sample dates to deal with the initialization of the Kalman filter, so I was wondering whether this could be the source of the problem, since I am not doing that now.

I also tried the two different specifications of the Taylor rule that you can see in the mod file, since in their paper they state that the responses in their equation 26 are not multiplied by (1 - \rho_{i}) as in (23). But when you see both equations, they turn to be the same when written in the paper, so I assumed that there was a typo.

Dear Marcos,

You can easily check if the initialisation of the filter explains the observed difference with the presample option of the estimation command.


Dear Stéphane,

I tried it, but no success, so it seems that the observed difference does not come from that.


I have tried to solve the optimizer problem. When I consider a Taylor rule where the interest rate smoothing parameter affects multiplies both the previous period interest rate and inflation, output, output growth and change in the nominal exchange rate, the optimizer does not go to the indeterminacy region. However, even though most of the modes are values close to the median of the posterior that the author show in their paper, the size of the TFP shock in the domestic economy and the preference shock are way higher. I don’t really know where this can come from. I double checked the equations of the model, the measurement equations, the data and I can’t come to an idea of where this divergence can come from.

I update an “updated” version of the mod file and data.

Thank you
Australian_Data.mat (6.6 KB)
DSOEM.mod (8.2 KB)

My impression at this stage is that only a step by step comparison with the original results can help. For example, do simulation with the parameters at e.g. the mode return the same IRFs?