@Kodiologist @Kodiologist thanks. also sorry for the late response :)

@Kodiologist: any opinion on this question: stats.stackexchange.com/questions/214961/… ? You asked me once why I would make up my own test (significant difference in MSPE) when I could simply use AIC. That led me to think...

@Erosennin I took a shot.

@Kodiologist: Great! Thanks a lot! I'll write a small comment now!

De nada.

@Erosennin: Have you seen AIC: Myths & Misunderstandings?

There B&A say quite categorically that AIC doesn't assume a true model.

Though later on they say about the penalty term being an approximation.

@Scortchi: Yes, I've read that. I've posted many questions on this topic previously, and B&A has been a lifesaving source of information. That is why I categorically dismissed the first link provided in my last question, and in my own arrogance turned a blind eye to the interesting part about the penalty term being an approximation :)

e.g. stats.stackexchange.com/questions/205222/…

This post stats.stackexchange.com/questions/215244 was migrated to SO, but then the migration was "rejected" and it came back as closed. Why was the migration rejected?