[R] Error with arima()

[Prof Brian Ripley]

It means that the CSS estimates from the model are invalid.  That's 
possible as CSS does not enforce validity.  It probably means the model is 
inappropriate.

On Thu, 17 Jun 2004, Dan Bebber wrote:

> Could someone please give a brief explanation, or pointer to an explanation,
> of the following error:
> 
> > arima(ts.growth, order = c(1,0,0),include.mean=T)
> Error in arima(ts.growth, order = c(1, 0, 0), include.mean = T) :
>         non-stationary AR part from CSS
> 
> and why it does not arise with
> 
> > arima0(ts.growth, order = c(1,0,0))

Not the same algorithm.  These algorithms only find a local minimum.

You are asking the question the wrong way round: unless you know there is 
uniquely defined estimator, why would you expect the same results?  
Especially if the model is a poor fit.

-- 
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