negative coefficient for alpha+beta in CGARCH

[tpr3396]

Hello guys,
Can someone please help me with the following question?
When I estimate CGARCH model in five exchange returns, for three of series, I got negative coefficient on either alpha or beta, so alpha+beta<1. and both of them are insignificant as well. I tried both student-t and normal distribution, but the problem still there. How can I fix this? Thanks a lot

[tpr3396]

If I change the sample size, and make it longer, the alpha and beta will be positive and significant. Can anyone explain this? Thanks

[trubador]

Since this issue is related to techincal details of optimization, the explanation you seek is not specific to EViews. Please do not forget that ARCH models are conditional estimations of volatility and therefore depend on the data (along with other empirical issues) at hand. You should study ARCH models in more detail, before carrying out such estimations in any software.