possible bug(s) in fGarch

[Yohan Chalabi]

michal miklovic wrote:
> Hi,
>
> I estimated a garch model and obtained strange and self-contradictory 
> results when I checked the model using the plot function. I ran the 
> following code:
>
> library(fGarch)
> garchmodel <- garchFit (~ arma(0,3) + garch(1,1), data, 
> init.rec="uev",include.mean=FALSE, algorithm="sqp")
> summary(garchmodel)
> plot(garchmodel)
>
> Selections 7 ('Residuals') and 9 ('Standardised Residuals') of the 
> plot function produce the same graph. Other selections appear to use 
> 'Residuals' instead of 'Standardised Residuals', which leads to 
> results contradicting the Standardised Residuals Tests from the model 
> output. For example, the Ljung-Box test for squared standardised 
> residuals does not indicate any remaining garch structure but the 
> selection 11 ('ACF of Squared Standardised Residuals') shows 
> significant autocorrelations at the first four lags. Selections 10 
> ('ACF of Standardised Residuals') and 13 ('QQ-Plot of Standardised 
> Residuals') are similarly affected. On the other hand, I think the 
> Standardised Residuals Tests are computed from 'Standardised 
> Residuals', which is correct. I enclose the model output and the ACF 
> of Squared Standardised Residuals in attachment.
>
> I would like to point out that the graph from selection 11 ('ACF of 
> Squared Standardised Residuals') has wrong title, the word 'Squared' 
> is missing in the title.
>
> I would also like to ask what is the degree of freedom of the ARCH-LM 
> test. In the model that I estimated the degree of freedom appears to 
> be 12. Is it so? If yes, why did you choose 12? It is a bit too high, 
> in my opinion.
>
> Thank you for your answer.
>
> Best regards
>
> Michal Miklovic
>
> ------------------------------------------------------------------------
> Be a better friend, newshound, and
> ------------------------------------------------------------------------
>
> _______________________________________________
> Rmetrics-core mailing list
> Rmetrics-core at r-project.org
> https://stat.ethz.ch/mailman/listinfo/rmetrics-core
Hi Michal,

thanks for you feedback. There was a typo in the function which 
calculates residuals and it always returned non-standardized residuals. 
It is now fixed on the dev branch of Rmetrics.


regards,
Yohan