# [R-SIG-Finance] Seasonsal ARIMA

rkevinburton at charter.net rkevinburton at charter.net
Sat Sep 20 05:55:16 CEST 2008

No I have 52 observations per year (which is once a week) so isn't the periodicty 52? The original data had 365 observations per year and arima could not handle that. I tried splitting the year in half (having 182 observations per period/half year) and then splitting the year into quarters (91 observations per year) but that all proved too much for arima to handle (seasonal wise). So the next logical step was to split the data into weeks. Summing the daily data so that there are 52 observations per year. This kind of works in that there are not as many errors. I still get errors from 'optim' but I think eventually with trial and error I will be able to get rid of those few.

Hopefully I have made it a little clearer on what I am trying to do.

Kevin
---- Jae Kim <jh8080 at hotmail.com> wrote:
>
> Hi,
>
> If you have daily 7-day a week data, you may have weekly, bi-weekly or monthly seasonality. That is, 7-day cycle.
> This means that the value of your periodicity may have to be 7.
>
> But I guess you can inspect the SACF to confirm the presence of this periodicity.
>
> JHK
> > Date: Fri, 19 Sep 2008 14:03:59 -0700> From: rkevinburton at charter.net> To: r-sig-finance at stat.math.ethz.ch> Subject: [R-SIG-Finance] Seasonsal ARIMA> > First off let me say that from 'R' calling arima(xdata, order=c(1,1,1), seasonal=list(order=c(2,1,1), period=12) (an ARIMA(1,1,1)X(2.1.1) 12 model) works just fine (provided that data is monthly). But for my data I have daily data so there are 365 obeservations per year and I have about 4 years of data. With this data and replacing 12 with 365 gives me an error indicating that I cannot set the lag above 350. So I gradually worked my way down and now I have 52 observations pwr year and I am calling arima like:> > arima(x, order=c(1,1,1), seasonal=list(order=c(2,1,1), period=52) > (an ARIMA(1,1,1)X(2.1.1) 52 model)> > Now I am getting an error that I don't understand. It must have something to do with the practicle/numeric limitations of the implementation. > > <simpleError in optim(init[mask], armaCSS, method = "BFGS", hessian = FALSE, control = optim.control): initial value in 'vmmin' is not finite>> > So my question to this group is, "what are the practicle limits for using arima to fit data to a seasonal model?" One, I found that I cannot specify a lag larger than 350 either specifically or implicitly with the model that I am building. If the fit takes longer than say 5 minutes I would say that it is impracticle and I need to look for a different solution. Obviously if I get numerical errors like shown above that would be another practicle limitation of the function. So darwing on the experience of this group rather than me going through trial and error what would be the practicle limits of arima? Can these limitations be overcome by possibly another model or other parameters? > > Let me give an example. If I want to brute force find 100! at first the response would be that it is impossible, that big of number can't be represented. But there have been a number of solutions to make findiing 100! "practicle". I am looking for the same kind of advice with arima. > > Thank you.> > Kevin> > _______________________________________________> R-SIG-Finance at stat.math.ethz.ch mailing list> https://stat.ethz.ch/mailman/listinfo/r-sig-finance> -- Subscriber-posting only.> -- If you want to post, subscribe first.
> _________________________________________________________________
> It's simple! Sell your car for just \$40 at CarPoint.com.au
> http://a.ninemsn.com.au/b.aspx?URL=http%3A%2F%2Fsecure%2Dau%2Eimrworldwide%2Ecom%2Fcgi%2Dbin%2Fa%2Fci%5F450304%2Fet%5F2%2Fcg%5F801459%2Fpi%5F1004813%2Fai%5F859641&_t=762955845&_r=tig_OCT07&_m=EXT