I am looking to ultimately generate Impulse Response Functions and plotting them for a set of variables. These variables are all non-stationary in levels when a lag order of 5 is selected. They are also cointegrated at this order. I believe this satisfies the prerequisites to estimate a VECM, which I could then convert to VAR and generate IRFs.
My issue however, is that when I plot these impulse response functions with the above variables, the response variables do not always return to long run trend I.e. return to zero; they can continue to incline or decline after many periods. Do you have any idea why this would be the case?
In addition, I have reason to believe that a lag order or 5 is too high for my model, and k=2 would be more sufficient, however when I set the lags to 2 some variables become stationary in levels and others remain non stationary, and all are stationary in first differences. Levels are still cointegrated. In this case, can I still use the VECM, and if I use the VECM do I use the differenced variables in this estimation or input the level variables (some non-stationary, some stationary) in it? Differenced variables from VECM to IRF seem to provide responses that return to long run trend better than levels.
