Hello,
I'm running a number of arima models using the "arima" function. Often,
when lag length gets too high, these model don't converge and an error
message appears as this:
> reg <- arima(y,order=c(7,0,7),xreg=isr)
Warning message:
In arima(y, order = c(7, 0, 7), xreg = isr) :
possible convergence problem: optim gave code=1
In this case, when you print the results from the arima model, the error
message appears again:
> reg
Call:
arima(x = y, order = c(7, 0, 7), xreg = isr)
Coefficients:
ar1 ar2 ar3 ar4 ar5 ar6 ar7 ma1
ma2
-0.3356 0.0282 -0.0124 0.0604 0.1608 0.7949 0.2286 0.4461
0.0099
s.e. NaN 0.2140 0.1711 0.2545 0.1547 0.0725 NaN NaN
0.1842
ma3 ma4 ma5 ma6 ma7 intercept isr
0.0841 0.0773 -0.054 -0.7227 -0.1322 6.1433 -0.0818
s.e. 0.1911 0.2689 NaN 0.1774 NaN 5.0604 0.2110
sigma^2 estimated as 4.542: log likelihood = -651.68, aic = 1337.36
Warning message:
In sqrt(diag(x$var.coef)) : NaNs produced
I'm writing a loop to run several of these models and I would like to
include an if statement with instructions in case there is a possible
convergence problem. Is there any way to automatically detect a warning
like this without visually inspecting each result? I tried
warnings(arima(y,order=c(7,0,7),isr), but I don't think that will get me
what I want.
The contents of an arima object are below. I was hoping one of them would
include error messages, but I haven't been able to find it.
> summary(reg)
Length Class Mode
coef 16 -none- numeric
sigma2 1 -none- numeric
var.coef 256 -none- numeric
mask 16 -none- logical
loglik 1 -none- numeric
aic 1 -none- numeric
arma 7 -none- numeric
residuals 299 ts numeric
call 4 -none- call
series 1 -none- character
code 1 -none- numeric
n.cond 1 -none- numeric
model 10 -none- list
Thanks in advance for any help.
-Mitch
______________________________________________
R-help@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.
