Suppose I have one time series and two competing models that describe it. Model 1 is ARMA$(p_1,q_1)$, model 2 is ARMA$(p_2,q_2)$-GARCH$(r,s)$.

I obtain AIC values of model 1 and model 2. I would like to conclude that the model with the lower AIC value is the prefered model. However, I am not sure whether I can compare AIC values coming from two different classes of models such as ARMA and ARMA-GARCH.

My intuition is the following: if we find a model that nests both model 1 and model 2, we should be able to use AIC to compare the former two. Indeed, ARMA$(p_1,q_1)$ and ARMA$(p_2,q_2)$-GARCH$(r,s)$ are both nested within ARMA$(\max\{p_1,p_2\},\max\{q_1,q_2\})$-GARCH$(r,s)$ and may be obtained from the latter model by applying zero restrictions on coefficients.

However, I am not sure whether my approach makes sense. Could someone please elaborate on this issue and answer the question: can we, or can we not, use AIC to compare ARMA models with ARMA-GARCH models?


1 Answer 1


Yes. AIC can be used to compare an ARMA model to an ARMA-GARCH model.

GARCH part is adding the variance equation. Specifying GARCH does not change the dependent variable and the loglikelihood is comparable. Hence AIC.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.