| NEWS | R Documentation |
Georgi N. Boshnakov
Fixed issue 6061 raised by William Scott, who also supplied examples.
The quantile function, qsnorm, was wrong around
0.5. The error was in .qsnorm. For now its version before
the fix is kept as .qsnorm_orig. Basically, branching was
done w.r.t. p = 0.5, which is correct only for the symmetric
case, \xi = 1, and should be 1/(1+\xi^2) instead. More
details in the source code. The error was affecting the central
part of the distrbution with the interval becoming larger for
\xi further away from 1.
The cdf, psnorm, had an error at a single point,
coinciding with the wrong value for p = 0.5 returned by
qsnorm(0.5) before the fix. The result was that
psnorm(qsnorm(0.5)) was returning 0.5, falsely giving
reassurance that qsnorm(0.5) was correct.
Not mentioned in issue 6061 but the same problems held for
the other skewed distributions: qsstd, psstd,
qsged, psged. The original versions of the relevant
internal functions are kept for now with a suffix _orig, as
above: qsstd_orig, psstd_orig, qsged_orig,
psged_orig.
Edited the documentation of "garchSpec" and
garchSim. It was somewhat incomplete and contained
leftovers, apparently from old versions of the functions.
Documented the datasets. Previously the help page for them was a placeholder, without the names of the available datasets. There is no information about the time span of the data or how the returns were calculated.
Packages timeSeries, timeDate and fBasics
are no longer in Depends, but only in Imports and hence no
longer automatically attached to the search() path whenever
fGarch is.
This may require updates in your code, e.g., adding
stopifnot(require("timeSeries"))
as it has been done in our own fGarch's examples and tests.
.gogarchFit() is at least mentioned in the documentation.
ad interim: Martin Maechler
Added registration of compiled functionality for speed up and as good practice.
Removed all Depends: entries and checked exactly which
parts of packages, notably fBasics, timeDate, and
timeSeries, are needed and imported only these.
Eliminated warning about 'length > 1' character formula in
garchFit(), i.e., .garchFit().
Replaced the error-prone checking for 'class()' equality by 'inherits(*, <class>)'.
Exporting practically everything seems “wrong” (according to MM):
Several .<some> functions have no documentation and
hence should either be (renamed and) documented or no longer be exported.
a data argument should never have a default: hence removed
from garchFit().
.
.
Startup message removed
Incorporate fixes by CRAN team (Brian Ripley?)
Checks and adaptions for R 3.4.2, e.g., ‘DESCRIPTION’, ...
.
Changes up to April 2013, by Yohan Chalabi, Diethelm Wuertz, Pierre Chausse and Martin Maechler are all in file ‘ChangeLog’.