[R] Fitting an ARIMA model to a time series

[Wouter van Rijsinge]

R 1.6.2 on windows XP (and windows 2000)

Dear Readers,

I have to fit an ARIMA model to a blood pressure series to make predictions 
with it. But since I don't have a blood pressure data set yet I have to 
work with self made data. So I have created an AR( 2 ) series with the 
following code:
series <- list()
series$series <- arima.sim(n=2100, model=list(ar = c(0.6, 0.1)), sd=1)  + 120
It is based upon the assumption that the blood pressure will be a value 
around a mean of 120 with only minor fluctuations. Only when something goes 
terribly wrong it will have a trend up or down.

I now try to predict new values every interval of size 20 with data from a 
sliding window of size 100 with the following code:
windowSize <- 100
lagSize <- 5
for(i in 1 : 100)
{
	index <- windowSize + ((i-1)*20)
	model <-  arima(series$series[ ( ( (index-windowSize) + 1) : index ) ], 
order = c(2, 1, 1), method = c("ML")  )
	predictions <- arima.sim(n=5, model = model)
}

The problem is that whatever order I use for the model estimation I always 
end up with the error: "ar part of model is not stationary". I have tried 
to fit totally different series with less random fluctuations and more 
trend etc. but it also doesn't work. I don't know any solution for this 
problem anymore. So I have the following questions:
1. Is there anybody who can tell me how to (automatically) fit an ARIMA 
model to the data and make predictions with it?  It of course has to work 
with future blood pressure series too. Even with blood pressure series with 
trend because in the future I will have to fit it to data from the 
Intensive Care.
2. Or is the model not compatible with the (simulated) blood pressure data 
and should I just stop trying?

I hope someone has a solution for this problem because I can't find one.

Greetings,

	Wouter