Using the cost function to generate Marshallian demand systems

Limited data means that prior structure is needed when working with large demand systems. The cost function is a convenient vehicle for generating demand systems incorporating such structure. While the cost function directly yields Hicksian demand functions they will not usually have an explicit representation as Marshallian demand equations i.e. in terms of the observable variables. With fast hardware and modern software, however, this need not hinder the estimation of the (implied) Marshallian demand equations. This paper develops the formal theory for using cost functions in this context, and reports on initial trials on the operational feasibility of the method. First version received: September 1997/final version accepted: July 1999