[R] Fitting Arima Models and Forecasting Using Daily Historical Data

[Prof Brian Ripley]

On 09/09/2013 20:36, Paul Bernal wrote:
> Hello everyone,
>
> I was trying to fit an arima model to a daily historical data, but, for
> some reason, havent been able to.
>
> I basically have 212 observations (from 12/1/2012 to 06/30/2013) containing

Those dates are not in a standard format: ISO 8601 is preferred.

> the number of transits for a particular vessel.

Which seems to be less than a year and you are trying to fit a 
seasonally differenced model with period 365.  So I guess there are no 
pairs of observations a year apart.

This is not really an R issue: you need to get basic advice on 
time-series modelling.

> The following messages are produced by R:
>
> dailytrans.fit<-arima(dailytrans$transits, order=c(0,1,2),
> seasonal=list(order=c(0,1,2), period=365), include.mean=FALSE)
> Error in arima(dailytrans$transits, order = c(0, 1, 2), seasonal =
> list(order = c(0,  :
>    too few non-missing observations
>> dailytrans.fit<-arima(dailytrans$transits, order=c(0,1,2),
> seasonal=list(order=c(0,1,2), period=200), include.mean=FALSE)
> Error in makeARIMA(trarma[[1]], trarma[[2]], Delta, kappa) :
>    maximum supported lag is 350
>
> Then I tried the auto.arima function but the following happened:

Which is not part of R.

> fit<-auto.arima(dailytrans$transits)
>
> Warning in if (class(fit) != "try-error") offset <- -2 * fit$loglik -
> length(x) *  :
>
>    the condition has length > 1 and only the first element will be used
>
> If anyone could give me some guidance I will really truly appreciate it,
>
>
> Best regards,
>
>
> Paul



-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595