[R] can Box test the Ljung Box test say which ARIMA model is better?

[Spencer Graves]

	  First, have you made normal probably plots (e.g., with function 
qqnorm) of the data and of the "whitened residuals" from the fits of the 
different models?  If you've got outliers, they could drive strange 
results;  you should refit after setting the few outliers to NA.  You 
didn't tell us what software you used, but the following seemed to work 
for me:

qqnorm(as.numeric(lh), datax=TRUE)
lh100 <- arima(lh, order = c(1,0,0))
qqnorm(as.numeric((resid(lh100)), datax=TRUE))
lh.tst <- lh
length(lh)
# Remove observation 20
# pretending it was an outler.
lh.tst[20] <- NA
lh.tst100 <- arima(lh.tst, order=c(1,0,0))
resid(lh.tst100)
# NOTE:  redid(...)[20] is NA

	  What are the AIC values?  I think most experts would suggest making 
the choice based on the AIC, especially if both models passed the 
Box-Ljung test.

	  In my opinion, the best work I've seen relevant to your question 
talks about Bayesian Model Averaging.   RSiteSearch("Bayesian model 
averaging for time series") led me to the following:

http://finzi.psych.upenn.edu/R/Rhelp02a/archive/70293.html


	  hope this helps.
	  Spencer Graves

Michael wrote:
> two ARIMA models, both have several bars signicant in ACF and PACF plots of
> their residuals,
> but when run Ljung Box tests,
> both don't show any significant correlations...
> 
> however, one model has p-value that is larger than the other model,
> based on the p-values,
> can I say the model with larger p-values should be better than the model
> with smaller p-values?
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html