26

PONTIFICIAE ACADEMIAE SCIENTIARVM SCRIPTA VARIA - 2§

3) Each relation in the reduced form is an eo ipso predictor
subject to random disturbance.

1.4. Incentives for the generalization from VR- to CC- and
ID-svstems.

1. Aggregation over time. Vector regression (6) is of old
standing in dynamic model building, and especially in the
natural sciences a good many dynamic theories can be quoted
that make use of this type of approach. An important feature
is that if the data so permit, the time unit can be chosen very
small; hence, in principle, VR-systems (6) cover also the approch
 of differential equation systems of any order. Model (6)
thus lies near at hand in situations where the data are registered
continuously (recording barometres, seismographs, etc.), or,
more generally, are registered periodically with time intervals
that are short relative to the changes in the variables between
the recordings.
CC-systems (10) and ID-systems* (18) are recent innovations,
 both emerging in econometrics in the decade 1935-1945.
It is not by chance that this generalization was initiated in
econometrics, an area where theoretical model building had
been well developed since long ago, and where by long tradition
 a large part of the available time series data had the form
of annual aggregates. Hence there was — and is — a twofold
incentive for the generalization from (6) in the direction of (10)
and (18). One was that annual data were known to involve
lots of information in the form of interrelations between economic
 factors observed in one and the same year, a source of
information that cannot be exploited in the VR- approach (6).
And the very existence of large masses of annual data reinforced
 the incentive for the generalization at issue.
2. The chain principle of explanation and forecasting. More
recently, another incentive for the generalization has come to

2°

Wold - pag. 12