[R-SIG-Finance] I want to find the best arima model using AIC criterion, but having problems
Matthieu Stigler
matthieu.stigler at gmail.com
Mon Jan 17 14:40:58 CET 2011
HI think too for this purpose auto.arima model might be good.
Concerning your loop, the problem is not in your loop I believe, but
simply that numerical procedure in arima() does not converge (i have
seen this problem quite often). You might try using rather method=CSS
(the usual least square).
Le 17. 01. 11 13:26, Paul Teetor a écrit :
> Have you considered using the auto.arima function in the 'forecast' package? It
> searches for the best ARIMA model using the AIC, AICc, or BIC criteria. It might
> save you a lot of aggravation.
>
> Paul Teetor, Elgin, IL USA
> http://quanttrader.info/public
>
>
>
>
> ________________________________
> From: BSanders<adam at mycostech.com>
> To: r-sig-finance at stat.math.ethz.ch
> Sent: Sun, January 16, 2011 10:42:14 PM
> Subject: [R-SIG-Finance] I want to find the best arima model using AIC
> criterion, but having problems
>
>
> I'm running a loop for trying to find the best ARIMA seasonal model. I'm
> experimenting with different orders of AR, MA and differencing.., but when I
> run my loops, I'm getting an error such as
> "Error in optim(init[mask], armafn, method = optim.method, hessian = TRUE,
> :
> non-finite finite-difference value [3]
> "
> and R exits the loop, however, it isn't finished. Is there a way to make R
> continuing what it should?
>
> My code looks like this below :
>
> count = 0
> for(i in 0:7){
> for(j in 0:2) {
> for(k in 0:1) {
> for(l in 0:1) {
> for(m in 0:2) {
> for(n in 0:1) {
> for(o in 1:2) {
> model = arima(y, order=c(i,j,k), seasonal=list(order=c(l,m,n), period=7*o))
> if (count == 0){
> aicmin = model$aic
> bestmodel = model
> }
> if ((count != 0)&& (model$aic< aicmin)){
> aicmin = model$aic
> bestmodel = model
> diff1 = j
> diff2 = m
> seas = o*7
> }
>
> count = count+1
> }
> }
> }
> }
> }
> }
> }
> print(bestmodel)
> print(count)
> print(diff1)
> print(diff2)
> print(seas)
>
More information about the R-SIG-Finance
mailing list