I have residuals that I got from an ARMA-GARCH model via rugarch package:

ug_spec <- ugarchspec(distribution.model="std", mean.model=list(armaOrder=c(1,0)),variance.model=list(garchOrder=c(1,1)))
ugfit=ugarchfit(spec = ug_spec, data =x)

the as.numeric(unlist()) is just to get the part of the result I want. First off, was that the correct way to get the residuals off a data that has been fitted to the ARMA(1,0)GARCH(1,1) model?

I have to now check the residuals for any serial dependence, and decided to use Box.test with the Ljung-Box type:

   Box.test(residue, lag=5, type="Ljung", fitdf=1)

I set fitdf=1 as per the instruction in the vignette to place p+q as the fitdf. The result is this:

Box-Ljung test

data:  residue
X-squared = 15.548, df = 4, p-value = 0.00369

The p-value < 0.05, which doesn't make sense since I've previously detrended and de-seasonalized the data before fitting it to the ARMA-GARCH model (best fit by BIC). Does this mean that there is, indeed, serial autocorrelation even though I've done all I was supposed to in pursuit to get rid of it, or did I make a mistake in the usage of Box.test?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.