Constructing complete demand systems is an example of the ideal type of econometrics. It uses a well-developed theory, an extensive set of data and a refined estimation technique. In particular, a strong relationship between theory and estimation distinguishes it from other empirical economic studies. This strong relationship is at the same time a source of possibilities and a source of problems. Here, the focus will be on some problems caused by the fact that the theory is about individual behaviour while the data is about aggregate behaviour and by the fact that the theory is not much help in specifying the functional form of the system, which is obviously needed for estimation. These two problems are interrelated to a certain extent and have not yet been solved satisfactorily.

Although dynamic models are now an important part of applied demand analysis, some of their fundamental concepts still suffer from a substantial lack of clarity. A particular case in point is the distinction between stock adjustment and habit formation in the Houthakker-Taylor model (1970). Empirical considerations prompted Houthakker and Taylor to specify a single state variable for each good and then to conclude that a good is habit forming or stock adjusting depending on the state variable's sign. However, besides having an obvious drawback of being ex post, this procedure is subject to the further criticism that, by treating expenditures for nondurables and services in the same way as expenditures for durables, it implicitly views habit formation and stock adjustment as arising from symmetrical processes. This is unfortunate, for stock adjustment and habit formation reflect intrinsically different phenomena, and to elaborate upon this, without the constraints imposed by thoughts of immediate empirical application, is one of the primary purposes of this paper.

When one writes a paper, one hopes for a large hearing. I have reduced my illusions, as to my audience, to those few who have read either Houthakker and Taylor [(1970, chap. V)] of Phlips (1972) or Taylor and Weiserbs (1972). Probably only they could be interested in a presentation of the theory underlying these studies and/or in some improvements in the empirical comparison realized by L. Taylor and myself. There is nothing here for a revolution in the theory of the consumer but only an attempt to ornament the last one.

L'introduction des stocks dans la fonction d'utilité, telle que le suggèrent Houthakker et Taylor [1970, chapitre 5] permet la construction et l'estimation de systèmes de demande complets dynamiques et théoriquement plausibles. Ainsi sont apparus les modèles quadratiques additifs d'Houthakker et Taylor, obtenus en dynamisant la fonction d'utilité quadratique de Gossen et une version dynamique du modèle linéaire des dépenses, élaborée par Phlips [1972] en dynamisant la fonction d'utilité de Stone-Geary.

The task of re-stating the theory of the true cost-of-living index when preferences are subject to taste change was undertaken by F. Fisher and K. Shell in their contribution to the Hicks' Festschrift [Fisher-Shell (1969)]. It was succeeded by a paper, co-authored by L. Phlips and R. Sanz-Ferrer (1972), that proposed an index different from the one elaborated by Fisher and Shell. They also expounded an algorithm for calculating the two indexes with the help of Phlips' dynamic linear expenditure system [Phlips (1972)].

The purpose of this paper is to extend the work of Fisher and Shell in three directions. First, taste change is parametrised in the utility function in such a way as to allow its application to Phlips' dynamic system. Secondly, a broader definition of the true cost-of-living index is provided. Thirdly, the influence of taste change on the index is less restrictive than in the Fisher-Shell case without, we believe, incurring the reproach of cardinality. It is patterned on the Phlips-Sanz-Ferrer approach.