Re: [Gretl-users] Tests for GARCH models
On Thu, 7 May 2009, yinung at Gmail wrote:
In general, after estimating a GARCH, the diagnosis should be on
"standardized" residuals rather than on "regular" residuals from the
mean equation.
I've now added an option --stdresid to the garch command (and also
a corresponding GUI checkbox). This standardizes the residuals.
However, the current test (named ARCH) with a GARCH model via GUI is
on "regular" residuals from the mean equation.
The regular ARCH test shouldn't be available for GARCH models;
that bug is now fixed.
In addition, the tests on "normality of residual" and
"autocorrelation" is now disabled with a GARCH model. Are they
possibly enabled (and on the "standardized" residuals)?
The normality test is now enabled. I'm not sure what an
autocorrelation test would look like for a GARCH model, but you
have the Box-Pierce Q statistic that's given with the residual
correlogram.
Allin.