Coefficients estimation

[E-mily]

Hello everyone, I'm new here and hope you can give me a hand with my problem.

Why do the p-values of my MEAN's coefficients change when I estimate a (G)ARCH model?

Shouldn't Eviews be only working with the residuals of that MEAN?

I hope I've been clear

[startz]

The "H" in ARCH stands for heteroskedasticity. That affects standard errors.

[E-mily]

I see startz, thanks.

So, if I want to choose the equation of the mean, I should do that with the ARCH method (in Estimation Setting) and not choose the equation for the mean FIRST(using ols or arma method) and then estimate the garch.

[startz]

That's right. The mean equation and ARCH are estimated together.

[E-mily]

Ook, thanks :*

It is a bit odd, if I might, anyway. Because if there is autocorrelation in the series you may want to correct it BEFORE you run your garch model, otherwise the estimation may be altered.

[startz]

Nothing stops you from correcting for serial correlation in the mean equation during ARCH estimation. Just include AR(1) (or whatever) in the specification.

[E-mily]

I know that ARCH family is used to model time-varying volatility, but would it be okay if I just developed it in order to see the significance and "the impact" of an independent variable over the dependent one?

I could do it with an ordinary regression, but I would have to face heteroskedasticity which could alter its parameters and may lead to false results.

[startz]

Totally fine

[E-mily]

@startz

Just to be perfectly clear, is there any difference between an ARMA and an ARMA-GARCH model as far as the mean equation is concerned?

[startz]

The difference is that ARCH gives a GLS-like estimate correcting for time-varying heteroskedasticity, which seems to be what you want.

[E-mily]

Does that mean that the coefficients estimation is more precise?

[startz]

Yes, it does.