For as long as there have been economists there have been debates
over the proper methodology for economics. Econometricians have by
contrast avoided such arguments, preferring to get on with the job.
Generally, discussions of econometric methodology have merely reflected
the controversy between "classical' and "Bayesian' statistics.
In some ways this has been unfortunate as there are at least three
important benefits which might result from the adoption of an explicit
methodology. First, it could provide a set of principles to guide and
improve applied research. Second, it could help codify and organise
econometric knowledge, which would greatly facilitate teaching. Finally,
it could encourage researchers to report their results in an informative
and succinct style.
In the last few years interest in the methodology of econometrics has
increased, and distinctive styles of doing econometrics have emerged.
Four major ones might be distinguished. A Bayesian approach to
econometrics has long been advocated by Arnold Zellner. Time-series
approaches have been exemplified in the work of Christopher Sims. An
"LSE' methodology can be found primarily in the work of Denis
Sargan and David Hendry. Finally, Edward Leamer has set out a collection
of methods that have been described as "vestigial Bayesianism'.
In Discussion Paper No. 39 Michael McAleer , Paul Volker and CEPR
Research Fellow Adrian Pagan explore the effectiveness of Leamer's
methodology as a guide for applied research and as a reporting style.
They argue that Leamer's methodology performs very poorly in these
roles. To substantiate their claim they examine one of the techniques
used by advocates of this methodology: Extreme Bounds Analysis (EBA).
Suppose one is interested in measuring the effect of interest rates on
the demand for money. Extreme Bounds Analysis would begin with a fairly
general model of money demand which included interest rates, as well as
other variables. This general model can be "simplified' in various
ways, for example by excluding one or more explanatory variables.
Extreme Bounds Analysis is concerned with the largest and smallest
values of the estimates of the interest rate coefficient when various
simplified models are estimated.
Suppose the estimated coefficient varies greatly over the range of
simplified models. Inference concerning the coefficient is then said to
be be "fragile' or "unreliable', since the coefficient
estimate obtained appears to be sensitive to the precise specification
of the model used. McAleer, Pagan and Volker demonstrate that such
analysis based on extreme bounds is directly related to traditional
tests of hypotheses that particular simplifications of the model are
compatible with the data. The extreme bounds methodology, according to
the authors, merely presents in a different format the same information
as does conventional regression analysis. Pagan and his coauthors argue
that the most disturbing feature of the extreme bounds technique is that
it also provides much less information about other aspects of the
model's adequacy. They argue that this "vestigial Bayesianism' is
therefore inadequate both as a research guide and as a reporting style.
The Discussion Paper concludes by examining the best known application
of the EBA technique - a demand for money study by Thomas Cooley and
Steven Le Roy. Cooley and Le Roy found that the impact of interest rates
on money demand is "ill-determined'. Pagan, McAleer and Volker
argue that Cooley and Le Roy arrived at such a conclusion only as a
result of their adoption of this inadequate methodology, and that the
misgivings expressed about the theory of vestigial Bayesianism also
apply in its practice.
What Will Take the Con Out of Econometrics?
Michael McAleer, Adrian R Pagan and Paul A Volker
Discussion Paper No. 39, January 1985 (ATE)