The RATS Software Forum

I used box jinkins to find the appropriate model for my stock returns. BJ suggested arma 1-1. I tried to test it for remain arch effects by using this code below. I am not sure im doing this correctly, has i also checked the 4 other, AR1 AR2 MA1 ARMA 1-1 ARMA 2-1 and 1-2 they all came back the same:

Chi-Squared(5)= 263.444740 with Significance Level 0.00000000 Same result of all the models.

Any help would be much appreciated
Thanks

*ARMA(1,1)
boxjenk(constant,ar=1,ma=1) RAIB / resids
cor(partial=pacf,qstats,number=12,span=4,dfc=%nreg) resids
compute aic=%nobs*log(%rss)+2*%nreg
compute sbc=%nobs*log(%rss)+%nreg*log(%nobs)
display 'aic=' aic 'sbc=' sbc

SET resids2 = resids**2
LINREG(NOPRINT) resids2
# CONSTANT resids2{1 TO 5}
COMPUTE CHISTAT = %NOBS*%RSQUARED
CDF CHISQ CHISTAT 5

An ARMA model is designed to eliminate serial correlation but isn't designed to eliminate ARCH effects, so the fact that the residuals show ARCH isn't necessarily a surprise.

So i didn't make an mistake in my code, and the end result is that there is arch effects present, so a garch model is required?

That's correct. It seems fairly clear that you need a GARCH model, possibly with an AR or ARMA mean model.