Skip to Content
Home » Questions » R parameters and AIC of function arima() vs auto.arima()

R parameters and AIC of function arima() vs auto.arima()

Submitted by user1189816 on February 4, 2012

I've got a problem with estimating parameters using functions auto.arima() and arima(). The problem is that when I use auto.arima and get the model with lowest value of AIC, then I try to put appropriate orders p,q,P,Q into function arima() with the same data, I get different value of parameters and different value of AIC. Here is the example:

> auto.arima(ddlogvalue, max.p = 3, max.q = 2, max.P = 2, max.Q = 2, start.p = 1,
+            start.q = 1, start.P = 2, start.Q = 2, max.order = 5, ic = "aic")
Series: ddlogvalue 
**ARIMA(0,0,1)(2,0,2)[12] with zero mean**     
Coefficients:
          ma1    sar1    sar2     sma1     sma2
      **-0.6565  0.0568  0.0042  -0.5480  -0.2353**
s.e.   **0.0615  0.1184  0.0739   0.1263   0.1065**
sigma^2 estimated as 0.0008404:  log likelihood=479.54
**AIC=-946.89   AICc=-946.5   BIC=-926.36**
> a = arima(ddlogvalue, order = c(0,0,1), seasonal = list(order = c(2,0,2)),
+  include.mean = FALSE )
> a
Series: ddlogvalue 
**ARIMA(0,0,1)(2,0,2)[12] with zero mean**     
Coefficients:
          ma1    sar1     sar2     sma1    sma2
      **-0.6297  1.3138  -0.3953  -1.9657  0.9989**
**s.e.   0.0572  0.0723   0.0891   0.2322  0.2344**
sigma^2 estimated as 0.0008351:  log likelihood=463.5
**AIC=-914.99   AICc=-914.61   BIC=-894.47**

I'll be glad if anyone could explain me why estimated parameters and value of AIC differ in both cases.