[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
AVVISO DI SEMINARIO
=================================================================
Forum per la Societa' Italiana di Statistica
AVVERTENZA
il nuovo servizio viene offerto in via sperimentale
dal Dipartimento di Scienze Statistiche di Perugia
con il software Majordomo 1.94.5
istallato su una Workstation UNIX (AIX 4.3.3)
Ci scusiamo per eventuali disguidi !
=================================================================
DIPARTIMENTO DI SCIENZE STATISTICHE
Via S. Francesco, 33 - 35121 Padova
tel. 0498274168
Il Prof. Nils Lid Hjort
Department of Mathematics
UNIVERSITY OF OSLO
terrą il seguente seminario:
MAXIMUM SIMULATED LIKELIHOOD ESTIMATORS
4 Ottobre 2000
ore 11.30-12.30
Aula B2, Ca' Borin
Summary
Suppose summary statistics $T=(T_1,\ldots,T_p)$ have
been observed, perhaps computed from a difficult raw data set,
and that it is required to fit a certain model to these,
parameterised by $\theta=(\theta_1,\ldots,\theta_d)$,
where $p\ge d$. I focus here on cases where the complexities
of the model or of the data collection make it impossible to
calculate the likelihood $L_T(\theta)$. I propose a simulation-based
method for approximating $\theta^*$, the ML estimator based on $T$.
It consists in simulating many realisations $T^*$ for many
values of $\theta$, and then using techniques of density estimation
and nonparametric regression to estimate the real maximiser
of $L_T(\theta)$. Similarly Bayes estimators may be approximated
via simulations.
Using such methods one may in principle approximate ML estimators
in any parametric models, as long as it is possible to simulate
realisations of the $T$ in question from any given parameter value.
In particular, ML estimators may be computed without deriving
or caring about formulae for the likelihood or its derivatives
at all. Estimation uncertainty may also be addressed, via parametric
boostrapping.
Illustrations of the technique will be given.
Cordiali saluti,
Carlo Gaetan
Carlo GAETAN e-mail:gaetan@stat.unipd.it
Dipartimento di Scienze Statistiche phone : ++39-049-827-41-68
Via S. Francesco, 33 ++39-049-827-41-80 (desk)
I-35121 PADOVA fax : ++39-049-87-53-930
ITALY