I am trying to learn ARDL technique. I have two I(1) annual variables A and B. According to AIC, I found that the optimal combination of lag length is (2,3) for A as dependent variable and (3,1) for B as dependent variable.
However, the model suffer from the problems of serial correlation and non-normality. For B as dependent variable, there is also the problem of misspecification problem. In addition, some of the coefficients are not significant.
1. My question is mainly about how to resolve the serial correlation problem, which is emphasized by Pesaran's orignial paper. I tried to add more lags. However, it seems not an effective remedy. Any one please shed some light on this issue? I have not come across any paper using ARDL facing this issue yet.
2. And for the insignificant coefficient issue, should I use the general-to-specific approach? What did is that I tried to delete the lags with insignificant coefficients. My question is that when we use general-to-specific approach, is our only concern on the significance of the coefficients? What about the diagnostic tests on the residuals? Do we check these tests? Moreover,what happen if we delete all the lags except the A(-1) and B(-1), of which the coefficients are still insignificant?
Thank you very much in advance!!!
For econometric discussions not necessarily related to EViews.
1 post • Page 1 of 1
Who is online
Users browsing this forum: No registered users and 10 guests