1

I currently have the following formulas:

$AIC = \ln(\sigma^2)+2(p+q+1)/T$

$BIC = \ln(\sigma^2)+ln(T)(p+q+1)/T$

This does not work for every $ARMA(p,q)$ model. How do I amend the formula depending on the number of parameters?

M47145
  • 4,106
Haych
  • 121
  • What do you mean saying that this doesn't work for every $ARMA(p,q)$ model? You need to fit all possible models for all $p\leq P$ and $q\leq Q$ for some maximal $P$ and $Q$ that you can afford, take $\sigma^2$ estimate from every model and choose the model that gives $\min_{p,q}AIC(p,q,\sigma^2)$ – Slowpoke May 08 '16 at 21:12

0 Answers0