Although SIC and SICf are asymptoti

Although SIC and SICf are asymptotically equivalent, the simulation results indicate that their selection properties are different for small to moderate sample sizes.Models chosen by SIC often tend to be larger than necessary.In many applications, the additional term in SICf corrects this propensity towards overfitting without overcompensating and leading the criterion towards choosing undersized models. Since underfitting is often regarded as a more serious error than overfitting, the fact that SIC f often guards against the latter without leaning towards the former makes the variant an attractive alternative to SIC.