s-news
[Top] [All Lists]

arima.mle vs gls

To: s-news@lists.biostat.wustl.edu
Subject: arima.mle vs gls
From: "Jeannine St. Jacques" <jmcheval@sympatico.ca>
Date: Thu, 05 Feb 2009 15:25:45 -0600
User-agent: Thunderbird 2.0.0.19 (Windows/20081209)
Dear all

I am running the student S-plus version 8 on Windows XP. I am trying to figure out why I get such different results with gls and arima.mle (using first 10 data points for conditioning). They are not even close which makes me wonder which is right. I also notice very big differences in results between 0 mean, var 1 standardized regressors as inputs (ie xreg) and non-strandardized xreg inputs. Also why does fitted.arima return with the last data missing? That is, I input arima.mle with 46 data points and fitted.arima returns very reasonable looking fitted values for the first 45 points only? As well, if I fit the regression manually using the values I input in xreg and the returned reg.coef, I do not get what fitted.arima returns. This leads me to suspect that arima.mle transforms xreg inputs before feeding them into the regression, and returns the regression coefficients for the transformed xreg. Any insight is appreciated, recommended books, etc.

Thank you, frazzled

<Prev in Thread] Current Thread [Next in Thread>
  • arima.mle vs gls, Jeannine St. Jacques <=