You are on page 1of 2

[Must the Diffusion Index Lead?

]: Rejoinder
Author(s): Stefan Valavanis
Source: The American Statistician, Vol. 11, No. 4 (Oct., 1957), p. 17
Published by: Taylor & Francis, Ltd. on behalf of the American Statistical Association
Stable URL: http://www.jstor.org/stable/2681587 .
Accessed: 25/12/2014 23:32

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Taylor & Francis, Ltd. and American Statistical Association are collaborating with JSTOR to digitize, preserve
and extend access to The American Statistician.

http://www.jstor.org

This content downloaded from 128.235.251.160 on Thu, 25 Dec 2014 23:32:12 PM


All use subject to JSTOR Terms and Conditions
But there are various ways of learning more about economic time series. Many other tests are available to
these matters, not a single way. Something can indeed be guide one's judgment as to whether results are to be
learned by constructing simple cases, and by assuming taken seriously-their consilience with other data, or
certain mathematical properties exist, and by inferring with similar data from a different time or a different
what would happen if the real world were like that. place; their repetitive character; their reasonableness in
These tools have long had a respectable place in eco- the light of knowledge of economic processes and poli-
nomic research, and I expect they always will. This is cies. These tests may not yield estimates of mathe-
especially true when one is exploring the implications of matical probability, yet I wonder if this is really a differ-
a new statistical device, such as a diffusion index. For ence. For the probabilities computed from a mathe-
example, perhaps Mr. Valavanis' simple mathematical maticized system must also be "interpreted"in the light
examples will prevent many from jumping to the conclu- of the known simplifications built into the model. They
sion that exploring the properties of diffusion indexes may be precise for the model, but in view of the simpli-
constructed from economic data is of no interest because fications that are almost inevitably required to obtain
the diffusion index is a simple proxy for the rate of results, we know in advance that the model is not a
change in the aggregate. precise imitation of "reality." A judgment as to how
Nevertheless, I am persuaded that this is not the seriously the results should be taken is called for in
only way to make progress in economic research. A the one case as in the other.
great many of the findings of Mitchell and others about
business cycles, for example, have stood the test of time Those who place a heavy weight on mathematical
and provided a valuable understanding of economic de- probabilities, and those who are impressed by tests whose
velopments in a period subsequent to that from which probability they do not know how to evaluate can there-
they were derived, yet few if any have depended on the fore work side by side, and each can take the other's
investigator's knowledge of the "stochastic properties" of results "seriously."9

REJOINDER STEFAN VALAVeANIS


Harvard Univer-sity
To look for quick, cheap and simple shortcuts to cal probability" of his results is essentially the same
economic forecasting is a reasonable research endeavor. thing as judgment based on consilience etc., only-I be.
It is perfectly justified to record and publish which indi- lieve-clearer and more systematic.
cators and which combinations of them have in the past When I consider the treatment that a time series
tended to lead. The diffusion indexes so far published receives in the hands of Dr. Moore and his associates, I
by the National Bureau do show interesting leads over find it extremely well adapted to the stochastic test that
the sum of their components, both within the sample I favor. Every series that enters this or that diffusion
period and beyond it. index has already been analyzed into its several com-
Two considerations disturb me, however. First, only ponents, when it was being screened for conformity etc.,
the successful diffusion indexes get shown. Second, the early in its elaboration. In fact all the ingredients of
unsatisfactory indexes contain information about the the recipe in the last formula of page 15 are ready on
reliability of the satisfactory ones. hand. What is missing is a chef to tell us beforehand
I don't know how many indexes Dr. Moore or anyone what sort of pudding to expect from them-whether, for
else discards for every one that he adopts, but the more instance, so many raisins per slice are just a freak, or
he discards the less should he defend the diffusion index something inherent in the logic of baking the mixture.
in general. It is a perfectly wise research strategy to Likewise, we can combine the various (known) m, M, T, e
try many things and discard many. However, the characteristics of each series into beforehand knowl-
belief-worthiness of what is retained is not independent edge about the covariation of the aggregate S and the
of what was thrown out. This is why I like "stochastic corresponding diffusion index.
specification"; with it you can tell success by chance To derive the covariation is staggeringly complicated
from success by genuine discovery. Now, Dr. Moore's for the general case, but quite simple, though tedious,
"consilience with other data", "repetitive character", for special numerical values of mi, Mi, etc. Any mechani-
and "reasonableness in the light of knowledge of eco- cal computer can quickly generate the appropriateproba-
nomic processes and policies" are precisely the stuff from bility tables by merely combining in all possible ways a
which stochastic specification is worked out-only per- few representative values of the disturbances, or irregu-
haps more systematically. "Interpreting the mathemati- larities, which I have denoted by u (t).

17

This content downloaded from 128.235.251.160 on Thu, 25 Dec 2014 23:32:12 PM


All use subject to JSTOR Terms and Conditions

You might also like