Message-ID: <Pine.LNX.4.44.0512081717540.12177-100000@topcat.sfos.uaf.edu>
Date: 2005-12-09T02:52:05Z
From: gaffigan@sfos.uaf.edu
Subject: R-help: gls with correlation=corARMA
Dear Madams/Sirs,
Hello. I am using the gls function to specify an arma correlation during
estimation in my model. The parameter values which I am sending the
corARMA function are from a previous fit using arima. I have had some
success with the method, however in other cases I get the following error
from gls: "All parameters must be less than 1 in absolute value". None
of the parameters (individually) are greater than or equal to 1.
Please copy the code below into R to reproduce the error. Thanks.
Is my logic incorrect? In the corARMA function, there's a call to
pre-compiled C code with the name "ARMA_unconstCoef". Is the source
code for such compiled code freely available for download?
Thanks for your suggestions.
Sincerely
Steve Gaffigan
data=read.table("http://ak.aoos.org/data/sample_070989.dat",header=T)
attach(data)
mod.ols=lm(obs~model)
mod.sma=arima(residuals(mod.ols),order=c(0,0,1),seasonal=list(order=c(0,0,2),period=12))
theta.1=mod.sma$coef[1]
THETA.1=mod.sma$coef[2]
THETA.2=mod.sma$coef[3]
ma.coefs=c(-theta.1,double(10),-THETA.1,theta.1*THETA.1,double(10),-THETA.2,theta.1*THETA.2)
library(nlme)
mod.gls=gls(obs~model,correlation=corARMA(q=25,value=ma.coefs,fixed=T),method="ML")
detach(data)
--
Alaska Ocean Observing System
School of Fisheries and Ocean Sciences : University of Alaska Fairbanks