Imposing restriction on GARCH(1,1)

[Nontherole]

Hi!
I'm having hard time imposing restriction on GARCH(1,1), for non-negative, non-degenerate and covariance stationary condition. More specifically, in variance equation I want to have omega>0, 0<landa<1, 0<=teta<1.Applying IGARCH is not proper as it removed the constant.Also I went through some posts in this forum found some http://forums.eviews.com/viewtopic.php? ... estriction. Yet, codings are not clear about how to apply in GARCH context.

Thanks for help
Regards,

[trubador]

I am not sure which coefficients you are referring to, but GARCH estimation does not require explicit restrictions in EViews. If regular estimation yields coefficents that do not satisfy these conditions, then the problem might be elsewhere and there is no way to detect it without seeing the actual data.

[Nontherole]

Hi trubador,
Here, I attached the workfile. Please note that the coefficient for RESID(-1)^2 is negative and I'm pretty frustrated.Beside low frequency data, no matter what I do (adjusting the "starting coefficient value"),it wont change. Your kind comment on this would be great help to me. Sorry for late response i was wrestling with data.
Regards,

[trubador]

Low frequency data is the problem. As the p-values of coefficents indicate, there is no significant GARCH effect in residuals. It is quite difficult to detect and identify heteroscedastic behaviour in such short samples. It may simply be the result of the variation of residuals. Before conducting such exercises, I suggest you to check residual diagnostics of the mean equation first (i.e. correlogram, histogram, heteroscedasticity, etc.)

[Nontherole]

Yes, I also thought about this low frequency data problem but that's what I have. I also checked the residuals, unfortunately there was weak correlation,In fact, they don't have time to wary enoghtto induce any GARCH effect.BTW,Thanks for feedback and one last question though, is there any other comparable model (with GARCH)I could use instead (in this low obs context?)

[trubador]

You are missing the point here. It might be what you have, but it's also what you can get out of it. I do not know how your hypothesis is defined, but the usual diagnostic tests tell you that there is no heteroscedasticity in your data. You "do not have to" build a GARCH or any other volatility model to prove otherwise. By the way, I suspect that the dependent variable itself is also a volatility indicator.