Hi, my final task is calculate VaR using egarch to filter the residuals.. the leverage parameter in egarch is significant, but the sign bias test isn't significant at all.. can I still use egarch? Thanks..
Hi, if you perform the test for residuals obtained from EGARCH model, the result is ok. It shows that you have no remaining asymmetry in the residuals. On contrary - if you estimate GARCH model, perform the test, and it gives you conclusion that there is no asymmetry in the residuals - why do you even consider the asymmetric model?
When I check the sign bias test from standard GARCH, all of the p-value isn't significant.. but the leverage parameter in EGARCH is significant.. sorry I little bit confused.. my final task is focused on EGARCH @Agata
Sign bias test is proposed by Engle and Ng. The null hypothesis said that the positive and negative shocks have same impacts on the volatility.. there are 2 other tests which are positive size bias test and negative size bias test @Srikanth
Well, so I still do not understand why do you HAVE TO use the EGARCH model if the tests do not suggest any asymmetry. But anyway - I would suggest to run the tests on your residuals from EGARCH model and see if they are OK. Then, I would also check the stability of your "leverage" using the Nyblom test. Eventually, you can compare the information criteria from GARCH and EGARCH models. If the ICs suggest that your EGARCH model is better or equally well - I think that you can go ahead with it.
Besides, in the EGARCH models parametrization you have standardized residuals, while in the sign-bias test the not-standardized ones. I am not sure, but maybe this is the reason why you have not significant result in the SB test, while significant "leverage" in the model. Maybe someone can correct me?
The only reason why I still use EGARCH is my supervisor 'force' me to deal with asymmetric GARCH in my final task.. because standard GARCH often used here..
Okay, I have check the AIC and log likelihood from EGARCH and standard GARCH.. and the result is EGARCH better than GARCH because have smaller AIC and higher value in log likelihood..
And for the sign bias, maybe your hypothesis is true, because sign bias doesn't use standardized residual (correct me if i'm wrong) ..
Thanks for your help Agata.. I appreciated it..
"On the other hand, the negative size bias test statistics, as well as the joint test statistics, are significant for all models"...………., it means that all the model have some problem in capturring the impacts of news on volatility (Engle & Ng, 1993)
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size