I have for a number of variables historical decomposition which don’t look nice. Basically their common feature is that it seems that some shocks are needed in order to counterbalance the others…
I have attached an example, inflationHist_decomp_Inflation.pdf (1.2 MB)
While for some other variables such as GDP, the problem isn’t apparent… Hist_decomp_GDP.pdf (1.2 MB)
I am wondering what steps could I explore in order to improve it ?
The graphs don’t strike me as that unusual. You seem to have a very strong prior on how the shocks have to look like. Did you have a look at the variance decomposition to see whether the shocks you expected are important?
Dear professor Pfeifer, thank you for your answer.
I did have some priors about what to expect since I have been working with different version of the same model for a while. I was searching for some signs that could help me to spot some mistakes in the model and since those graphs were significantly different from previous versions, I was wondering if they could help me spot what went wrong.
I did try to look at the variance decomposition to find some potential problems, but I have noticed that working with this particular model, it is changing quite a lot with any small changes that I make in the model (I also estimate the trends). On the other side, I found the qualitative insight given by the historical decomposition to be more stable across model specifications and therefore often more informative to spot issues.
After reviewing the code, I did find a mistake (I confused two variables in one equation) and now the decomposition makes some economic sense to me InfDiffVarDec.pdf (1.2 MB)
. (after some additional respecifications of the model too)