Big values of AIC in ETS Smoothing

[kiber_master]

Hello!

It seemes to me AIC value in ETS Smoothing is too big and doesn't fit standart formula. May the formula fo AIC (SC, HQ) differ for this model?

The most significant part of AIC value is LogL divided by number of observations, so, absolute value of AIC should be less than LogL, but it is not so.

In attached example

Included observations: 72

Compact Log-likelihood -582.2016
Log-likelihood -530.4052
Akaike Information Criterion 1168.403
Schwarz Criterion 1172.957
Hannan-Quinn Criterion 1170.216
Sum of Squared Residuals 10556661
Root Mean Squared Error 382.9103
Average Mean Squared Error 146243.3

[trubador]

Yes, they are not divided by the number of observations in this case:
AIC = -2*Compact LogL + 2*Number of estimated parameters = -2*-582.2 + 2*2 = 1168.4
Note that, parameter vector also includes the estimated initial states.

[kiber_master]

Why does the formula differ from common case?