GARCH and EGARCH

#1
I'm using Stata 11 to estimate GARCH and EGARCH model.
For GARCH equation, I have the problem that the coefficient of ARCH or the error term
is more than 1(it's around 5).
For EGARCH, I cannnot get the results. It shows this sentence'flat log likelihood encountered, cannot find uphill direction'.
I already tried to solve many times but it doesnt work.
What should I do to fix these two problems??
Thank you so much
 

vinux

Dark Knight
#2
I'm using Stata 11 to estimate GARCH and EGARCH model.
For GARCH equation, I have the problem that the coefficient of ARCH or the error term
is more than 1(it's around 5).
For EGARCH, I cannnot get the results. It shows this sentence'flat log likelihood encountered, cannot find uphill direction'.
I already tried to solve many times but it doesnt work.
What should I do to fix these two problems??
Thank you so much
Probably you need to check the stationarity assumption. Does the series is stationary? Is the mean part (ARMA) significant?
 
#3
I already checked for the stationarity and ARCH effect, and everything is ok. ARMA is also significant. So now I don't know what the problem might occur.
 

vinux

Dark Knight
#4
I guess there is something wrong somewhere. I don't think the issue is with integrated Garch process. Sum of alpha ( error) and beta ( previous variance) in GARCH can be very close to one. But alpha coefficient should be less than one. Normally beta coefficient larger than alpha.

Could you report the ADF statistic? Also check visually, the time series plot., Is there any outliers?
 

vinux

Dark Knight
#6
Depends. If there is a explanation associated with the outliers then, you need to put dummy variable and add it as a covariate.
Quick approach is drop the observations.