Skip to content

output processing / ARMA order identification

2 messages · bogdan romocea, Brian Ripley

Original
bogdan romocea
# 
Dear R users,

I need to fit an ARMA model. As far as I've seen, EACF (extended ACF)
is not available in R. 

1. Let's say I fit a series of ARMA models in a loop. Given the
code/output included below, how do I pull 'Model' and 'Fit' (AIC)
from each summary() so that I can combine them into an array/data
frame to be sorted by AIC?

2. Apart from EACF, are you aware perhaps of another function in R
that can help solve the issue of ARMA order identification?

Thank you,
b.
+ include.intercept = TRUE, series = NULL)
Call:
arma(x = var, order = c(1, 1), lag = NULL, coef = NULL,
include.intercept = TRUE,     series = NULL)

Model:
ARMA(1,1)

Residuals:
     Min       1Q   Median       3Q      Max 
-686.092  -68.499    4.024   65.531  509.171 

Coefficient(s):
           Estimate  Std. Error  t value Pr(>|t|)    
ar1        0.990653    0.003724  265.987   <2e-16 ***
ma1       -0.019562    0.030110   -0.650   0.5159    
intercept 90.940774   36.914682    2.464   0.0138 *  
---
Signif. codes:  0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 ` ' 1 

Fit:
sigma^2 estimated as 14193,  Conditional Sum-of-Squares = 17116373, 
AIC = 14983.22
Brian Ripley
#
On Mon, 25 Oct 2004, bogdan romocea wrote:

            

      
      
      
    

        
But PACF is as well as ACF.

            

      
      
      
    

        
Look at the examples in tests/ts-tests.R.  Or read the arima() help page 
more carefully, as it tells you where the results are (and they are not in 
summary(), which does not work for arima fits!).

            

      
      
      
    

        
It's a rather old-fashioned idea.  Just fit all the target models and 
compare their performance.