Serial correlation in EGARCH Model

[John S.]

Hello,

I am running an EGARCH (1,1) model on Eviews 8 and I´ve serious issues with serial correlation in the mean equation.

I have two questions:

1. The p-values of the Correlogram are not valid. Is it an ordinary problem in Eviews-ARCH Models or is something special with my estimation or my data?
Because this warning occurs every time I am checking the residuals of an ARCH/GARCH/EGARCH Models.

2. I have tried to remove the serial correlation with lagged dependent variables and with AR and MA terms, but still every mean equation is serial correlated. Any further advice?


Thanks for your help!

[trubador]

Make sure your dependent variable is stationary.

[John S.]

I´ve checked my dependent variable with ADF and KPSS tests. Both tests rejected non stationary Data.

[trubador]

If you need further advice, you should share your data along with your model. There may be plenty of reasons why the model diagnostics are not so well.

[John S.]

I want to estimate an EGARCH(1,1) model. I have tried to put every information inside the file.
The basic mean equation should be: y = c + c1*x1 + … + c4*x4 + c5*d1 + … + c14*d9

[trubador]

1) Variance of dependent variable prior to 2009 is very high compared to rest of the data (you can try to fit two separate models).
2) Approx. 1/4 of the data are full of zeros (investigate the reason and make sure they are valid observations).
3) It seems the dependent variable corresponds to a change in level (you can try percentage or log change).
4) Residuals of models indicate a seasonal variation at lags 5, 10, 15, etc. (seasonal differencing or SAR lags might help)
5) You do not have to worry about the significance of serial correlations as long as they are practically very small.

[John S.]

Thank you very much for the Input.

But for what reason can I ignore the p-value?