[R] User error (was arima.sim bug?)

[Prof Brian Ripley]

On Sun, 2 Oct 2005, Kemp S E (Comp) wrote:

> Hi,
>
> I am using the arima.sim function to generate some AR time series. 
> However, the function does not seem to produce exactly the same time 
> series when I specify the innov parameter. For example
>
>> r <- rnorm(300)
>> x <- arima.sim(300, model=list(order=c(1,0,0),ar=c(.96)), innov=r, n.start=10)
>> y <- arima.sim(300, model=list(order=c(1,0,0),ar=c(.96)), innov=r, n.start=10)
>
>> x[1:10]
> [1] 3.194806 4.214894 5.168017 7.925152 8.810817 9.131695
> [7] 7.521283 8.266911 8.923429 9.651293
>
>> y[1:10]
> [1] -0.7202632  0.4564274  1.5598893  4.4613486
> [5]  5.4855660  5.9394547  4.4567320  5.3249417
> [9]  6.0991390  6.9399748
>
> Given the fact that I have provided the innovations shouldn't the time 
> series be exactly the same?

No.  Hint: where does the randomness for the burn-in come from?

-- 
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