[Todos] Seminario de Probabilidad y Estadística Matemática

[Pablo Groisman]

Seminario de Probabilidad y Estadística Matemática.

PROXIMO ENCUENTRO: Miércoles 20 de julio, 12:00hs.
EXPOSITOR:  Carlos Lamarche, University of Oklahoma
TITULO: Robust Penalized Estimation for Panel Data Quantile Regression.
LUGAR: Instituto de Cálculo, 2do piso, Pabellón 2.

RESUMEN: We present recently developed methods for estimating a quantile
regression panel data model. We restrict our attention to penalized
estimators for panel data. An $\ell_1$ penalty serves to shrink a vector of
individual specific effects toward a common value and a tuning parameter
$\lambda$ controls the degree of this shrinkage. Existing approaches in the
literature for $\lambda$ selection could lead to incorrect inference under
non-classical assumptions. This paper offers a robust alternative for
$\lambda$ selection, which is based on a method that does not rely
on Gaussian conditions, existence of second moments and spherical errors. It
is shown that the class of estimators is asymptotically unbiased and
Gaussian when the individual effects are drawn from a class of zero-median
distribution functions. The optimal tuning parameter, $\lambda$, can thus be
selected to minimize estimated asymptotic variance. We carry out several
Monte Carlo studies to investigate the finite sample performance of the
method in comparison with other candidate methods. A few empirical applications
illustrate the use of the approach.


_______________________________________________________________________
Para ver el calendario del seminario
http://www.google.com/calendar/embed?src=987brtcpho5tt3ch2ud0oobkds%40group.calendar.google.com&ctz=America/Argentina/Buenos_Aires

Para mas información sobre el seminario
http://mate.dm.uba.ar/~drodrig/seminario/

Para recibir información sobre el seminario, suscribirse a la lista
Kolmogorov
http://mate.dm.uba.ar/mailman/listinfo/kolmogorov
____________________________________________________________________
------------ próxima parte ------------
Se ha borrado un adjunto en formato HTML...
URL: http://mail.df.uba.ar/pipermail/todos/attachments/20110718/0fe5db0e/attachment.html