|
|
"BC
BOOTSTRAP CONFIDENCE INTERVALS FOR RANDOM EFFECTS
PANEL DATA MODELS".
Por:
Andrés Carvajal (Brown University - Department of
Economics)
Abstract:
We study the application of bootstrap procedures
to the problem of constructing confidence intervals
for the coe .cients of random e .ects panel data
models, based on GLS point estimation. The central
problem is the one of adequately resampling from
the estimated residuals of the model, avoiding
violations of the structural features of the random
shocks.
1 Introduction
One of the most important tools in microeconometrics,
as well as in other fields of econometrics, is
the use of models that combine time series and
cross-sectional data, or panel data models. On
the other hand, bootstrap procedures to evaluate
the accuracy of summary statistics, or for inference
problems in general, have gained popularity. Although
they are computationally more expensive than standard
methods, they can be applied to almost any statistical
problem, do not pose problems when the statistician
transforms her or his parameters, are usually
more accurate than the standard intervals 1 and
do not require having to assume particular probability
distributions 2 .
In this paper, we study the application of bootstrap
procedures to the prob-lem of constructing confidence
intervals for the coe .cients of random e .ects
panel data models, based on GLS point estimation.
The central problem is the one of adequately resampling
from the estimated residuals of the model, avoid-ing
(important) violations of the structural features
of the random processes.
The paper is organized as follows: in the following
section we introduce the random e .ects panel
data model; then, we study the generalities of
the boot-strap procedures, paying particular attention
to the resampling problem for a random e .ects
panel data model. In particular we concentrate
in the problem of resampling the estimated residuals
in a coherent way that avoids the impo-sition
of false restrictions on the structure of the
random shocks. We propose four alternative resampling
plans; after that, we introduce an experiment
that tests the proposed plans; once we analyze
the results of our experiment, we state some conclusions
that are to be taken as a preliminary approach
to the problem.
Descargar documento
completo (formato pdf)
|
|
