R help - May 2008 - How to remove autocorrelation from a time series?

Dear R users,
someone knows how to remove auto-correlation from a frequencies time series?
I've tried by differencing (lag 1) the cumulative series (in order to have
only positive numbers) , but I can't remove all auto-correlation.
If it's useful I can send my db.

x <-             # autocorrelated series
new1<-cumsum(x)
new2<-diff(new1,lag=1,differences = 1)
acf(new2)     # remains autocorrelated 

Thanks in advance
Ciao
Claudia


      ___________________________________ 
Scopri i
http://www.ymailblogit.com/blog/

Claudia D'Aniello <kikki_80 <at> yahoo.it> writes:
> 
> Dear R users,
> someone knows how to remove auto-correlation from a frequencies time
series?
> I've tried by differencing (lag 1) the cumulative series (in order to
have
only positive numbers) , but I
> can't remove all auto-correlation.
  Take the residuals from an ARIMA fit?  (see arima(),
residuals() ) -- this won't be automatic, you'll have
to figure out the appropriate order model to fit ...

  good luck

  Ben Bolker

Claudia D'Aniello <kikki_80 <at> yahoo.it> writes:
> 
> Dear R users,
> someone knows how to remove auto-correlation from a frequencies time
series?
> I've tried by differencing (lag 1) the cumulative series (in order to
have
only positive numbers) , but I
> can't remove all auto-correlation.
> If it's useful I can send my db.
> 
> x <-             # autocorrelated series
> new1<-cumsum(x)
> new2<-diff(new1,lag=1,differences = 1)
> acf(new2)     # remains autocorrelated 
> 
This might come close to what your want

ar = arima(lh, order = c(1,0,0))
plot(ar$residuals)


Dieter