A necessary condition for decentralization and an application to intertemporal allocation (Q749421)

This paper is concerned with the question of whether a time sequence of resource allocation can be realized by means of a decentralized (privacy- preserving) mechanism, realistic but not necessarily perfectly competitive. Special attention is given to the simple class of intertemporal allocation models studied by Malinvaud and Koopmans (discussed in detail in section 3), in which the goal function (i.e., the welfare criterion) is the discounted sum of utilities of consumption over a finite or an infinite horizon. In section 1, the basic concepts are defined. A mechanism is defined to consist of a message space M; a verification function G, which is defined over the feasible message space; and an outcome function, which is a mapping from M to the allocation space. The goal function, which defines a welfare measure, is a mapping of the set of admissible economies into the allocation space. A temporal process is a sequence of mechanisms over the horizon of the intertemporal allocation problem. A temporal process generally requires knowledge of the future in order to make an allocation. A more realistic assumption is to preclude verification by future agents. Such a temporal process is said to be evolutionary. Section 2 derives the general necessary condition that must be satisfied by a goal function if it is to be realized by a decentralized mechanism. Specifically, Theorem 2.1 shows that the goal function cannot be realized by a mechanism in which the range of the verification function has dimension k, unless the allocation space and the set of the admissible economies also have a dimension of k. Section 4 shows that if a decentralized mechanism realizes a T-horizon intertemporal allocation of investment, then the verification function must have a dimension of at least T/2. As T approaches infinity, the dimension of the range of the verification function must approach infinity. Section 5 shows that if the verification function G satisfies a strong solvability condition, then the dimension of the message space must be at least as large as the dimension of the range of the verification function. The results of sections 4 and 5 together imply that the goal function of an infinite-horizon allocation process cannot be realized by a decentralized evolutionary mechanism which uses a finite number of verification conditions or a finite dimensional message space.