Daan Taks <[EMAIL PROTECTED]> wrote:
>I have a question about my residuals. When testing for autocorrelation
>I come to the conclusion that the models (garch, Egarch, GJR a.k.a.
>Tarch) remove the correlation from the squared standardized residuals
>but not from the standardized residuals.
Just standardizing residuals should not remove the autocorrelation
from them. That would require some kind of filtering.
Even filtering might not remove all autocorrelation.
For example, in an AR(1) model estimated with conditional ML,
the filtered residual is e = u - rho*u(-1) .
The numerator of the first order autocorrelation
of the filtered residual is (u - rho*u(-1))'(u(-1) - rho*u(-2)).
The first order condition for estimating rho is
(u - rho*u(-1))'u(-1) = 0. This is not enough to make the numerator
of the first order autocorrelation zero (it doesn't handle the
rho*u(-2) part).
>Are my models misspecified??
It's not possible to say, just on the basis of this.
>I use returns from the FTSE, the DAX, and the S&P. These returns are
>(heavily) correlated, should a garch model remove the correlation of
>the returns? Or should it only remove the correlation of the squared
>returns??
This is a different question (unless there are no RHS variables),
since presumably you are talking about the dependent variable in the
model, and not the residual.
Clint Cummins
TSP International
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at
http://jse.stat.ncsu.edu/
=================================================================
