[R] auto.arima from the Forecast package

ludovic.theate ludovic.theate at gmail.com
Sat Dec 17 15:47:02 CET 2011


Hi,

I've got a little problem using auto.arima.

I run the following command 

auto.arima(drivers,ic="aic",d=1,D=1,max.order=10,max.p=5,max.q=5,max.P=5,max.Q=5,stepwise=FALSE,allowdrift=FALSE)

and I get the following output :

Series: drivers 
ARIMA(0,1,1)(5,1,1)[12]                    

Coefficients:
          ma1     sar1     sar2     sar3     sar4     sar5     sma1
      -0.6421  -0.1341  -0.2063  -0.1076  -0.2361  -0.2205  -0.7387
s.e.   0.0718   0.1273   0.1061   0.1063   0.0938   0.1029   0.1312

sigma^2 estimated as 16916:  log likelihood=-1137.18
AIC=2290.37   AICc=2291.21   BIC=2315.87

So the fitted model has an AIC of 2290,37. But if I use the TRACE option, I
see that R analysed models like
 ARIMA(5,1,1)(3,1,0)[12]                    : 2175.377
 ARIMA(5,1,1)(3,1,1)[12]                    : 2176.402
 ARIMA(5,1,1)(4,1,0)[12]                    : 2155.935

which obviously have an inferior AIC... Since R is supposed to choose the
model minimizing the AIC criterion, I don't understand why he chose that
model. Is there something I misunderstood in the behavior of auto.arima ?

Thank in advance for any help...

Ludovic

--
View this message in context: http://r.789695.n4.nabble.com/auto-arima-from-the-Forecast-package-tp4208430p4208430.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list