[R] lm and time series: simple question

[Prof Brian Ripley]

On Thu, 16 Jun 2005 mwh at indiana.edu wrote:

> This question is partly about R and partly out of my ignorance about 
> time series.
>
> I want to regress one time series on another, taking into account the
> autocorrelation (in an AR1 model) within each series.  I am interested in how
> the standard error changes when the acf is taken into account.

This does not happen with least-squares fitting as done by lm. You can use 
arima or gls (in package nlme).  Note that both assume a model for the 
residuals, not for the series themselves.

You could also make a joint model of the two time series.  That is 
probably not what you want.

> I've made both of my datasets into ts objects and used the basic lm 
> function (with na.action=NULL) to no effect (i.e. the resulting standard 
> error is the same as if they were not times series).  I've also looked 
> at binding the two series together with ts.union or ts.intersect, but 
> then I am left with a single object, and don't understand how to regress 
> one of the components of this onto the other.

-- 
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