I'm estimating a VAR(4) model with level volatilities (stationary checked), and the impulse response graph shows me that the responses of all variables in the system don't seem to go away, they all remain at a constant level (positive) after the first/second period (see attachment). I wonder why is this the case and it doesn't seem right to me (am I right?). For a comparison I also estimated a VAR with differenced volatility, and this time the IRF appears rather normal (see attachment). But what bothers me is that when I look into the difference between these two VARs in terms of model checking wanting to find the reason that can account for the difference output in IRF, I found that there's no difference between these two models in terms of AR roots check and residual tests (none of them passed autocorrelation tests, and I have tried different lags and it's the same, but since my goal is to learn about the inter-relationship between these variables, I proceed as it it any ways). So anyone can help with the explanation and implication of the bizarre IRF graph please? Much appreciated!
- irf_level.PNG (9.76 KiB) Viewed 1939 times
- irf_dif.PNG (8.76 KiB) Viewed 1939 times