SEMAINE D'ÉTUDE SUR LE ROLE DE L’ANALYSE ECONOMETRIQUE Etc. 116
time series under analysis have the same structure of interrela-
tions in the observation range and the prediction range. As
applied to CC- and ID-systems the Janus quotient can be
adapted in various ways so as to take into account the distinc-
tion between genuine forecasts from the model and ancillary
forecasts of the exogenous variables. A specific use of the
Janus quotient is as a danger signal against overfiting, that
is, the pitfall of estimation techniques where illusively small
residuals are obtained because the parameters to be estimated
are numerous relative to the available observations. This aspect
of the Janus quotient comes to the fore when ID- and BEID-
systems are estimated by the two-stage method of least squares.
The predetermined variables then pile up with unknown coef-
ficients in each relation of the reduced form; hence unless
special precautions are taken the reduced form will be over-
fitted, with risk to run into the pitfall to conclude that the
current endogenous variables coincide with their conditional
expectations, and that the ID-system coincides with the cor-
responding BEID-system.
The topic of the report is an area of active interest to my
research seminar. I am greatly indebted to Professor ENDERS
A. ROBINSON, with whom I collaborate at present in conducting
the seminar, and to Messrs. A. Gapp, E. LYTTKENS, S. MARTI-
NELLE and G. STojkovic for allowing me to incorporate
unpublished research results in the report, as indicated bv
specific references to their work.
. THREE LEVELS OF GENERALIZATION IN DYNAMIC MODEL
BUILDING (2)
We shall consider three types of approach which (a) have
the form of simultaneous equation svstems: (b) are designed to
(?) The reader is assumed to have some orientation in the literature of
dvnamic econometric models. For textbook treatments of CC- and ID-
system, see Refs. 10 and 11. The present exposition leans heavilv on Refs.
t2 and 12.
[2] Wold - pag.
5
