Contrary to the usual assumption made in most oligopoly models, relations among firms are seldom of a wholly cooperative or noncooperative type: in many situations, they compete in some fields, while they cooperate in others. An important example is the case of cooperative research efforts bringing fierce competitors together. Two types of agreement are observed. First R&D cooperation can take place at the so-called “precompetitive stage”: companies share basic information and efforts in the R&D stage but remain rivals in the market-place.1 A second type of agreement involves an extended collusion between partners, creating common policies at the product level. The usual justifications of this extension are the difficulties of protecting intellectual property. The idea is then to allow partners who have achieved inventions together, to also control together the processes and products which embody the results of their collaboration, in order to recuperate jointly their R&D investments.2 What could be expected from these types of agreement is a reduction in R&D expenditures, because of less wasteful duplication, and a reduction of total production, because of more ∗Reprinted from The American Economic Review, 78(5), 1133-1137, 1988. †Center for Operations Research & Econometrics, 1348 Louvain-la-Neuve, Belgium. We are grateful to Jean

Our purpose in this note is to revisit the popular monopolistic-competition model of Avinash K. Dixit and Joseph E. Stiglitz ( 1977) and to stress the fact that the variant of this model used in the recent macroeconomic literature is significantly different from the original. In particular, by taking n as the number of active monopolists, the recent discussion of Dixit and Stiglitz (1993) and Xiaokai Yang and Ben J. Heijdra (1993) about the the advantages of neglecting terms of the order 1/n in the computed elasticities, is significantly affected by the choice of the variant of the model. The basic presented in Section I, has been used from the start by Dixit and Stiglitz to study optimum product diversity. It is a simple general equilibrium model with n monopolistic goods and a numeraire good, which can be interpreted as labor (or leisure) time or as the aggregation of all the other goods in the economy. The variant of the analyzed in Section II, was independently developed by several authors for different simple applications in macroeconomics.1 It is an model, that includes an additional good, interpreted as labor time but not taken as the numeraire. More importantly, the enlarged model does not lead to a general equilibrium analysis until the wage rate, taken as given in a first step, is adjusted competitively or strategically. It is for the basic model that Yang and Heijdra (1993) (YH) give an alternative computation method taking into account the priceindex effect of individual pricing decisions. This effect had been neglected in the original paper of Dixit and Stiglitz (1977) (DS), who were only concerned with the large n case (ensured by low fixed costs and imperfect substitution between the monopolistic goods). Limiting their model to the special case of a unitary elasticity of substitution between the monopolistic goods and the numeraire good, YH obtain an explicit solution. But YH's solution is still an approximation, because it neglects the indirect effects that feedback has on pricing decisions. We will show that, in the enlarged taking into account this income-feedback effect allows for an explicit solution and simplifies calculations. But in the variant, some meaningful cases are incompatible with free entry and thus prohibit the use of DS's approximation. However, as we conclude in Section III, this is not to say that their approximation should never be used. On the contrary, the approximation hypothesis is a very useful part of Dixit and Stiglitz's contribution.