How to fix a wrong Historical Shock decomposition?

Hello Johannes! :slight_smile:
I am trying to estimate the model of Iacoviello and Neri (2010), but I find the historical shock decomposition very bizarre. Basically, what the graphics show a very thick and symmetric (up and down) historical shock decomposition. Besides, on some variables, the graph shows a trend; which is estrange because the trends for these observables were specified in the model.
I am attaching a word document with all the relevant graphs, as well as the dataset and the model file. If you can take a look at it and give me some suggestions, I will be very much obliged.
Thank you in advance.
Yours cordially,
S.T. (435 KB)

This looks very much like a problem in mode-finding. Your β€œmode” is stuck at the boundary of the parameter space. That in particular means that the parameters determining the trend are not correctly estimated. In that case, there will be a part of the trend left in the data. That is what you observe in the shock_decomposition. You should try different mode-finders (try 5 for a change). Have you made sure that your observation equation does not only capture the slope of the trend, but also the intercept, i.e. the constant?
Also, have you verified all parameters are identified? Lastly, could it be that some of your data still suffers from seasonality patterns?

Hello Again Johannes,
Thank you very much for your insightful suggestions.
I will test them and I will tell you how did they go.
Yours cordially,