[Ml-stat-talks] Statistics Seminar at ORFE
jbradic at princeton.edu
Tue Sep 21 10:59:58 EDT 2010
Tomorrow (September 22nd) we will be having a talk by Davide La
Vecchia from University of Lugano with the title "Higher-Order
This is a joint work with Elvezio Ronchetti and Fabio Trojani.
The talk will be in the Statistics Lab, room 213 in Sherrerd Hall.
An abstract of the talk is included.
We investigate higherorder robustness for statistical functionals by
the linear term (Inﬂuence Function) in the von Mises expansion and
we provide the
conditions needed to ensure the stability of the asymptotic bias to
higher order. We
then deﬁne a class of admissible M-functionals which have second-
Our functionals allow for a strong control of the bias (B-robustness)
and the variance
(V-robustness), while preserving high eﬃciency at the reference
model. We also
investigate the link between the second-order robustness and the
stability of the
saddlepoint approximation of the density of M-estimators. Monte Carlo
provides evidence that our estimators improve both MLE and ﬁrst-
order optimally B-
robust M-estimators, in moderate to small sample sizes and for
from the reference model. An empirical analysis of the extreme
negative returns of
the Nikkei 225 illustrates the methodology in a real-data application.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ml-stat-talks