I am very sad if no one have interest on my question.

If my question is not so easy to answer, can someone give me a standard answer ? I can check it by myself.

Thanks and pls!

your step 1 to step 3 looks like right. when you know the first difference is stationary, you should regress (1-L)LNDAX

on its lagged value. according your notation, LNDAX itself is not stationary, so your taking the new sequnece E

is still nonstationary. you should plot you data with time at first and you can find something at first before you test the

unit root.

Thanks a lot!!!

Another question, is unit roots test only for ARMA necessary_?

q1:

ans:the E here is nostationary,at first,you test that LNDAX has unit root,but you use the nonstationary series E when you create the ARMA model ,I suggest that you should diff E at first order..

q2:

ans: as to different series,ARMA is not always better than ols.when you diagnose your model,you should check :

first: the residual series

second:the effect on forecasting

IF you find the series has unit roots(nostationary series),you should test the unit root until the series differenced has no unit root.then you model on the diffed series.

that's all.

[此贴子已经被作者于2005-7-10 22:24:26编辑过]

Thanks for all practical suggestions!