Statistics Seminars 2002-2003,
Department of Economics, Pompeu Fabra University
Schedule.
Thursday, October 10, 17:30--19:00, room 20.237.
Andrew B. Nobel
(Department of Statistics,
University of North Carolina, Chapel Hill)
A Brief Overview of Cluster Analysis and its Application
to Gene Expression Arrays
Abstract:
In this (informal) talk we will present a brief overview of cluster
analysis, and discuss its application to the diagnosis of breast
cancer using gene expression (micro) arrays. The talk will present
some common clustering algorithms, and describe several statistical
problems that arise in the application and interpretation of
clustering methods.
Thursday, November 14, 17:30, room 20.175.
Tamás Linder (Queens University, Canada)
Learning-theoretic methods in clustering and data compression
Abstract:
In k-means clustering a set of k points in R^d are to be chosen so
that the expected squared distance of the closest of these points to a
random vector X is minimized. The function that maps X to the closest
point is called a vector quantizer in the data compression literature
and is of independent interest there. In this talk I will give an
overview of classical and recent result on learning an optimal set of
cluster points from training data drawn from the distribution of X.
Techniques from statistical learning theory are heavily relied on in
these analyses which provides an interesting link between this
principle and data compression theory.
Thursday, November 21, 17:30, room 20.175.
Joan del Castillo (Universitat Atonoma de Barcelona)
Exponential models and their applications to Finance
Abstract:
The Black and Scholes model is now the basic tool to understand the
evolution of financial markets. However, their hypothesis of normality and
randomness for the asset returns are far from being perfect. The analysis of
empirical data often shows heavy tailed distributions and this has a
strong
effect in risk measures.
Lévy processes allow a flexible model for the marginal distribution of
returns, consistent with the continuous time structure suggested by the
hypothesis of market efficiency. Most of the more popular Lévy
processes
in
Finance are closely related to exponential models.
In this context, I will present the paper (joint with Anna López)
"Saddlepoint approximation in exponential models with boundary
parameters".
The paper shows that high-order asymptotic methods can be used for
likelihood ratio testing in exponential models with boundary parameters. The
result can be applied to testing Gamma against Generalized Inverse
Gaussian
Lévy processes.
Thursday, December 5, 17:30, room 20.175.
Baard Storvik (Norwegian Computing Center, Oslo, Norway)
Variable weight and location kernel density estimators
Abstract:
This talk introduces a general class of kernel
type density estimators. The ordinary kernel method is
generalized by allowing different
weights attached to the kernel and different locations within the
kernel.
This generalization makes it possible to construct new estimators having
better properties than previously proposed estimators. Some members of
the
class may even reduce both the asymptotic bias and variance
simultaneously, when compared to the kernel estimator.
We will also shed some light on the difference between
using a
different weight and location for each sampled data point versus using
weight and location as functions of the point of estimation. Theoretical
properties of the estimators are
investigated. A comparison study with other competitive estimators is
carried out, in moderate sample cases. It indicates that some of
the new methods provide dramatic advantages over recently proposed
estimators and the ordinary kernel method, in terms of minimizing the
mean integrated squared error w.r.t. the smoothing parameter.
Thursday, February 20, 17:30, room ??.
Karl Petersen (University of North Carolina, Chapel Hill)
Symbolic Dynamics and 0,1 Laws
Abstract:
The Kolmogorov 0,1 Law says that for an i.i.d. process, events in the
remote future (or past) must have probability 0 or 1. A strengthening due
to Hewitt and Savage says that an even finer sigma-algebra is trivial.
Viewing these statements in the context of ergodic theory leads to
surprising improvements and interpretations as well as fascinating open
problems in number theory and combinatorics.
Thursday, March 20, 17:30, room 20.183.
Leopold Soegner
(Technical University of Vienna)
Bayesian Estimation of the Heston stochastic
volatilty model.
(based on joint work with S. Fruewirth-Schnatter)
Thursday, May 29, 12:00, room 20.137.
(University of Amsterdam)
Estimation of the Kronecker and inner products of two mean vectors in multivariarte analysis
Thursday, June 19, 17:30, room 20.237.
Frederic Udina and
Gábor Lugosi
(UPF)
Universal strategies for sequential investment
Abstract:
We present various strategies for stock-market investment
whose capital, on the long run, achieves the best possible rate
of growth under the only assumption that the market is
stationary. Experiments on past NYSE data show a remarkable
behavior of some of these methods.