In economics the time series data are not available in sufficiently short time units in comparison with the adjustment speed involved in the relationships. Very frequently this forces us to the analyses in which some instantaneous relations cannot possibly be ruled out of consideration.1) A vector autoregressive modelling is proposed as an opposition to the simultancous equation approach, but the former is really a special case of the latter when instantaneous relations are involved. The conceptual framework of the simultaneous equations should be useful for any attempt to clarify the vector autoregressive modelling methodology.
Such attempt has revealed quite surprisingly some serious imprudence in the simultaneous equations methodlogy. Econometricians have long had a tendency to forget the point that the simultaneous equations system is a part of what Koopmans [18] called the complete equations system, the one that contains all equations for all variables. Recently Richard [21] and Engle, Hendry, and Richard [6] reinvestigated the concept of exogeneity. In the present paper we shall question both the identifiability and the pragmatic usefulness of the reduced form, and present a few methods for the policy analyses not using the reduced form.
The contents of the paper may be summarized as follows. (a) In sections III and VIII is shown the point that the Granger causality and the Sims exogeneity are each a property of the parameters that can be uniquely determined by the observation alone, and hence immune from the identification problem. (b) The condition which enables us to concentrate the estimation effort into the endogenous equations is called the weak exogeneity in Richard [21] and Engle, Hendry, and Richard [6]. A sufficient condition given there for the weak exogeneity is the a priori condition that the feedback is known not to be instantaneous. This is in section IV, and the parts (a) and (b) mentioned above form the expository portion of the present paper. (c) It is a grave fallacy to assume that the reduced form parameters are identified without an aid of a priori conditions. Their identifiability requires the predeterminedness, which is an a priori condition though weaker than the feedback being a priori known not to be instantaneous. Moreover, in order for the reduced form to produce valid predictions consistently over multi-periods it is required that the predeterminedness and the Granger-noncausality have to hold. These are in section V. (d) When the feedback is a priori believed not to be instantaneous the absence of feedback in all lags is a refutable property, a new condition in the study of the identification problem. (e) The policy simulation can be made without the reduced form, in fact, without any a priori conditions. A simple least squares method is also suggested for the case in which the dimensionality of the universe is large in relation to the length of available time series data. These are in section VII. (f) The distributed lag regression such as run at the Federal Reserve Bank of St. Louis can be derived without any a priori condition such as the right hand side variable being exogenous. This is in section VIII. (g) In section IX is sketched for a contrast what should be done with the a priori condition that the feedback is known not to be instantaneous. It is only with the weak exogeneity that policy rules can be properly analyzed. However the reduced form is not likely to be of much use for policy simulations. This is because the Granger-noncausality condition is unlikely to hold, as the policy variables are manipulated in view of the past values of target variables. (h) The prevailing opinion about the prefiltering is that it leaves the causality property invariant. The opinion is criticized in regard to the nonstationary process.
Although the vector autoregressive model is used throughout the paper, the essense of the paper, (c), would hold even in the nonstationary processes.

View full abstract