fGarch: possible bug(s)
Yohan Chalabi
chalabi at phys.ethz.ch
Mon Feb 4 10:57:05 CET 2008
>>>> "MM" == michal miklovic <mmiklovic at yahoo.com>
>>>> on Sat, 2 Feb 2008 08:28:16 -0800 (PST)
MM> In the summary function of fGarch, I would suggest that
MM> standardised residuals are computed as:
MM>
MM> r.s = object at residuals/object at sigma.t
MM>
MM> because r.s = object at residuals/sqrt(object at h.t) is wrong
MM> for aparch models, in which it does not hold that sigma =
MM> h^(1/2). The correct expression that holds in general is:
MM> sigma = h^(1/delta).
MM>
MM> In fGarch 260.72 sigma.t is calculated as: (.series)^deltainv,
MM> where deltainv = 1/fit. However, fit does not get updated
MM> with coefficient estimates and thus is always equal to 2. This
MM> leads to wrong calculation of sigma.t, which is always equal
MM> to h^(1/2).
MM> Has this problem been solved in the development version of
MM> Rmetrics (fGarch)?
Hi Michal,
I completely agree with you. I made the appropriate changes to the dev
branch of Rmetrics.
thanks for your helpful comments,
Yoahn
More information about the Rmetrics-core
mailing list