Skip to content

time series regression

6 messages · bereket weldeslassie, Mark Leeds, Thibaut Jombart +1 more

Original
Mark Leeds
# 
Your entering into a complex danger zone here because you really need to
check first if all the dependent and independent variables are stationary.
Otherwise, your lm results are meaningless ( you're estimation a spurious
regression ).  I would look at Bernhard Pfaff's yellow book or any other
decent time series econometrics text ( hayashi, hamilton )  for more on
this topic. It's a quite complex problem you
are working on so you need to get familiar with the cointegration/unit root
concepts, if you aren't already.



-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of bereket weldeslassie
Sent: Thursday, March 20, 2008 11:54 AM
To: r-help at stat.math.ethz.ch; r-help at r-project.org
Subject: [R] Fwd: time series regression

 Hi Everyone,
One more information to my question. I am trying to do a time series
regression using the lm function. *My intention is to investigate the
relationship between a dependent time series variable and several
independent time series variables.* According to the durbin watson test the
errors are autocorrelated. And then I tried to use the gls function to
accomodate for the autocorrelated errors. My question is how do I know what
ARMA process (order) to use in the gls function? Or is there any other way
to do the time series regression in R? I highly appreciate your help.
Thanks,
Bereket


______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Thibaut Jombart
#
bereket weldeslassie wrote:
Hi,

Temporal autocorrelation seems to be a contagious process... even for 
emails. I received yours three times in 5 minutes.

Concerning your question, I am no expert in time series, but you may 
also try ordinary least squares after 'removing' autocorrelation. This 
can be achieved by regression onto a lagged variable (see lag.listw in 
spdep, which can also be applied to temporal context), or onto 
eigenvectors of a temporal proximity matrix.

Cheers,

Thibaut.