Misspecifications
of econometric models can lead to biased coefficients and error terms, which in
turn can lead to incorrect inference and incorrect models. There are specific
techniques such as instrumental variables which attempt to deal with some individual
forms of model misspecification. However these can typically only address one
problem at a time. This paper proposes a general method for estimating
underlying parameters in the presence of a range of unknown model
misspecifications. It is argued that this method can consistently estimate the
direct effect of an independent variable on a dependent variable with all of
its other determinants held constant even in the presence of a misspecified
functional form, measurement error and omitted variables.
Most econometric relationships are subject to specification errors
arising from the following three problems: (i) the true functional forms of
economic relationships are usually unknown, (ii) econometric models cannot be
specified without omitting some relevant explanatory variables, and (iii) data
on economic variables contain measurement errors. Consequently,
misspecification of models is difficult to avoid. There are specific techniques
which attempt to deal with these problems, usually one at a time. Instrumental
variables are an obvious example of a technique designed to deal with
measurement error. But this technique cannot deal with a misspecified
functional form or omitted variables. Similarly the non-parametric estimators
such as neural networks or nearest neighbor estimation are designed to deal
with an unknown functional form. These techniques can not however cope with
measurement error and they also typically require very large data sets. This
paper sets out a new approach to estimation which can deal with all three of
these problems at the same time and which is practical in relatively small
samples[1].
[1]
P. A. V. B. Swamy,
George S. Tavlas, Stephen G. Hall And George Hondroyiannis.
Estimation of Parameters in
the Presence of Model misspecification and Measurement Error p-1-3
Hiç yorum yok :
Yorum Gönder