SEMAINE D'ÉTUDE SUR LE ROLE DE L’ANALYSE ECONOMETRIQUE ETC.
15¢
and then applies the chain principle in the same way as in
(82)-(83) to forecast y,~W¥,. We see that from the point of
view of forecasting, the deterministic component ¥, of the model
is, in principle, equivalent to an exogenous component.
Furthermore, the approaches (82) and (86) extend to mul
tivariate models; Ref. 46. The corresponding predictive de-
composition (82a-b) will then yield a representation in the
form of a CC-system. The minimum-delay property extends
to multivariate predictive decomposition. Hence the ensuing
forecast variance (84) is smaller than in other linear forecasting
models, such as ID- or BEID-systems.
The application of the JANUS quotient extends to multi-
variate systems, and in particular to VR- CC- ID- and BEID-
systems. By suitable adaptations the JANUS quotient can test
the entire system or a specific behavioural relation. Ii can
also, as illustrated below, be adapted so as to focus on the
extrapolation aspect of specific relations by removing those
forecasting errors which arise from imperfect forecasting of the
explanatorv variables.
ILLUSTRATIONS
1) Market model for pork. US 1932-1056.
Ref. 20).
(G. STOTKOVIC.
We shall consider the recalculated model as given by
(36a-c). For the demand relation the fit in the observation
range is moderatelv close.
S(VY=0.60 sid)
The Janus quotient (76) as calculated for a forecast span of
three years, treating the explanatory variables as known ex
post. gives the vearlv valnes
<1 Wold - pag. +45