Seminario Statistica Matematica

[silveri@vaxiac.iac.rm.cnr.it (Giuliana Silveri)]

**********************************************
      CON PREGHIERA DI DIFFUSIONE, Grazie!
**********************************************

\magnification =1200

\centerline{{\bf ANNO DI STUDIO IAC 1996}}
\vskip 3truemm
Nell'ambito dell'Anno di Studio 1996, dedicato al tema generale
'Recenti Applicazioni della Statistica', in preparazione del 
Workshop su {\it Curve Estimation, Superresolution and
Applications} il prof. {\bf Nils Lid HJORT} (Universit\'a
di Oslo, Norvegia) sar\'a ospite dell'Istituto per le Applicazioni 
del Calcolo del CNR, viale del Policlinico 137, Roma (tel. 06 
88470241) dal 26 febbraio al 2 marzo 1996. Il professor Hjort
terr\'a due seminari, luned\'i 26 febbraio alle ore 14.30
e mercoled\'i 28 febbraio alle ore 16.00. 


\vskip 3truemm
SEMINAR TALK I: 

\vskip 3truemm
{\bf `New methods for hazard rate estimation'}

\vskip 3truemm
\noindent
Abstract: 
The traditional way of estimating a hazard rate function 
is to kernel smooth the Nelson--Aalen estimator. 
This method goes back (at least) to papers by
Ramlau-Hansen, Tanner and Wong, and Yandell, 
published in the same issue of Annals of Statistics in 1983,
and its properties are well understood.
In my talk I will present a little selection of
competitors that in various ways may have better performance.  


\vskip 3truemm
\vskip 3truemm
SEMINAR TALK II: 

\vskip 3truemm
{\bf `Local likelihood density estimation'} 

\vskip 3truemm
\noindent
Abstract: 
Local polynomial and more generally local likelihood ideas
are playing an important role in the field of nonparametric
regression, as witnessed by recent books by Wand and Jones (1995),
Fan and Gijbels (1996) and Simonoff (1996), for example. 
But such ideas do not immediately carry over to density estimation. 
Recently such theory {\it has} been developed, however, 
featuring in particular a natural construction
of a local likelihood function for non- and semiparametric 
density estimation. The idea is to work with any
parametric family $f(t,\theta)$ as a local approximation
to the true density $f(t)$ for $t$ in a neighbourhood of 
some given $x$, and then decide on the locally best $\hat\theta_x$
from data. The result is the density estimator 
$\hat f(x)=f(x,\hat\theta_x)$, for example a `running normal'
with a local $\hat\mu_x$ and $\hat\sigma_x$. 
My seminar talk will present the basic motivation and results
for such constructions, and illustrate with several 
special cases. 

\bye