wilmott.com

Hello,I was wondering, if we run a regression where we also specify a GARCH(1,1) for the error variance, and we find that both the arch and garch terms are insignificant, does that mean there is no advantage from using the garch specification? Or are there still benefits from allowing a non constant variance? What if the constant in the conditional variance equation is significant but the arch and garch terms aren't? Does that mean assuming a constant error variance is more appropriate?Thanks!

You should carry out a likelihood ratio test to check your hypothesis.

You mean a test for arch effects?

So, if I find no arch effects, there's no advantage whatsoever from using a garch specification? If only the constant term in the conditional variance equation is significant, does that suggest assuming a constant variance is more appropriate?

Sometimes perfectly good theory is hard to establish from statistical tests.Take these two statements, which are both Arch/Garch effects:1. Higher equity market volatility is associated with higher expected returns, at least up to a point.2. Higher equity market volatility is associated with lower realized returns.I believe both of these, but the second obfuscates statistical tests of the first.

When writing a paper, yes. When playing with your money, you will need to know whether the error you get is from a sensible model (misspecification).