Hi everyone,
I’m an undergrad student and I’m replicating the Smets and Wouters model for the Euro area with updated data for my final project.
I have written the model following the original paper and the corrections suggested by prof. Uhlig as in this lecture. When working with the calibrated parameters everything works fine, as IRF are identical to the ones provided by prof. Uhlig in his toolkit.
When trying to estimate the parameters some issue arise. First of all, I am using an updated version of the time series indicated in the paper; all real variables are converted into logs and detrended by a linear trend with intercept so to have zero mean. Inflation and nominal interest rate are transformed according to Pfeifer 2013 and detrended the same way as real variables. Interest rate is not converted into logs, beacuse of the presence of negative values. For the estimation, values provided by the authors and dynare defaults mode are used.
Now, when estimating values with data from 1995Q2 to 2023Q2 things get really nasty, as IRF act erratically (e.g. responses to a productivity shock have wrong sign), not according to theory; and the variance is attributed only to a few shocks.
Observing the time series I decided to try and run again the estimation with data from 1995Q2 to 2019Q4 (as in 2020-2021 data are very volatile) and then IRF started behaving according to theory.
I also tried to change some parameters of the estimation function like mode_compute, but the results didn’t really change. I also had a look to the guide and at prof. Viegi code, but with no comfort.
What could be wrong here? Should I just use data up to the 2019? Is it really that the issue? thank you for your time, Marco.
Here are the files:
smets_wouters_2003.mod (15.3 KB)
data_1995Q2_2019Q4.csv (13.3 KB)
data_1995Q2_2023Q2.csv (15.1 KB)