Derechos:
Atribución-NoComercial-SinDerivadas 3.0 España
Resumen:
Fixed effects in discrete choice models has been a challenge to econometricians
from its existence. These unobservable heterogeneities are so important since
their impacts can be seen clearly from the behavior of agents being studied. This
has been consolidFixed effects in discrete choice models has been a challenge to econometricians
from its existence. These unobservable heterogeneities are so important since
their impacts can be seen clearly from the behavior of agents being studied. This
has been consolidated by lots of studies and simulations including mine. However
their existence prevent us from identifying models without restrictive assumptions
about them. It is also hard to get rid of fixed effects since they enter the
model not in a linear additive way and the outcomes are not continuous, therefore
extant difference methods do not apply to discrete choice models with fixed
effects. To have flexible specification on the fixed effects, it seems that partial identification
is more practicable. There do exists some idea about set identification
for discrete choice models and even some estimation methods were proposed for
logistic-alike discrete choice models, whose key feature is that all model deduced
conditional choice probabilities are well formulated in closed form expressions.
For reasons people may want to have discrete choice models with disturbance
other than extreme type I distributed one to overcome some of its implications,
e.g the property of independence with irrelevant alternatives among others. The
challenge to meet such requirement is that the key feature of closed form expressions
does not hold anymore, and techniques like simulation should be used. My
PhD thesis provides the foundation and framework on how to practice the simulation
based estimation for discrete choice models with rather flexible fixed effects.
This framework is both theoretical and practical, I show how to construct the simulation
based estimation and study conditions about both the property of model
and practice of simulation under which the estimator is consistent. This object
is achieved in two steps. I first develop the theory for static discrete choice models
where outcomes of behavior does not depend on previous outcomes. In this
case specification of disturbance could be rather free and even serial correlation
could be included. Later on, I extend the framework to dynamic discrete choice
models, where current behavior depends on some state variables which depend
on previous behavior in turn. In dynamic models, specification for disturbance is
still free except that serial correlation could not be allowed. These two steps consist
of the first and second chapters, in both chapter a numeric example is given
which shows how well the simulation based estimator works. In the last chapter I turn to the real data and apply my method to the problem of career decision of
young men. Essentially this is a typical application of dynamic programming discrete
choice model, which means individual’s object function is the lifetime utility
and it depends on both previous behavior and future states and what individual
should decide is not only the current behavior but also future actions. By introducing
a reduced form of the future utility I succeed in fitting this problem into
the framework of dynamic discrete choice model with fixed effects.[+][-]