[RsR] Robust location estimator - an interesting application in finance
Martin Maechler
m@ech|er @end|ng |rom @t@t@m@th@ethz@ch
Mon Sep 28 13:08:25 CEST 2009
>>>>> "ARu" == Andreas Ruckstuhl <rkst using zhaw.ch>
>>>>> on Mon, 28 Sep 2009 09:23:56 +0200 writes:
ARu> Dear Ajay
ARu> I agree on what Matias said.
ARu> In addition, I wondered whether you should not better use a "direct"
ARu> robust location estimator instead of the robust location estimator based
ARu> on the regression case. For example, there is R function huberM() which
ARu> returns a robust M-estimation of location based on a safe way of
ARu> calculating the robust scale estimator. An additional advantage of using
ARu> huberM() is, that the resulting M-estimator has breakdown-point 0.5
ARu> which is (much) higher than that of a trimmed mean.
ARu> To sell the robust M-estimator, you can talk of "an estimator which
ARu> down-weights outliers automatically according to their outlyingness".
ARu> (The M-estimator can also be modified that it removes very distant
ARu> outliers completely by using a redescending psi-function.)
Yes, indeed, thank you, Andreas,
I was also going to propose using huberM(),
which I had written a few years ago, exactly for dealing with
the "degenerate" case of having more than half of the
observations being equal.
Nonetheless, Ajay, I'm grateful for your bug report!
Regards,
Martin Maechler, ETH Zurich
More information about the R-SIG-Robust
mailing list