On economies of scope in communication

A classic puzzle in the economic theory of the firm concerns the fundamental cause of decreasing returns to scale. If a plant producing product quantityX at costC can be replicated as often as desired, then the quantityrX need never cost more thanrC. Traditionally the firm is imagined to take its identity from a fixednon-replicable input, namely a ‘top manager’; as more plants or divisions are added, the communication and computation burden imposed on the top manager (who has information not possessed by the divisions) grows more than proportionately. Decreasing returns are experienced as the top manager hires more variable inputs to cope with the rising burden. Suppose it turns out, however, that when the divisions are assembled, and are given exactly the same totally independent tasks that they fulfilled when they were autonomous, then asaving can be achieved if they adopt a joint procedure for performing those tasks rather than replicating their previous separate procedures. Then the top manager's rising burden must be shown to be particularly onerous—otherwise there may actually beincreasing returns. We show that for a certain model of the information-processing procedure used by the separate divisions and by the firm, there may indeed be such an odd unexpected saving. The saving occurs with respect to the size of the language in which members of each division, or of the firm, communicate with one another, provided that language is finite. If instead the language is a continuum then the saving cannot occur, provided that the procedures used obey suitable ‘smoothness’ conditions. We show that the saving for the finite case can be ruled out in two ways: by requiring the procedures used to obey a regularity condition that is a crude analogue of the smoothness conditions we impose on the continuum procedures, or by insisting that the procedure used be a ‘deterministic’ protocol. Such a protocol prescribes a conversation among the participants, in which a participant has only one choice, whenever that participant has to make an announcement to the others. The results suggest that a variety of information-processing models will have to be studied before the traditional explanation for decreasing returns to scale is understood in a rigorous way.     

Communication costs in the performance of unrelated tasks: Continuum models and finite models

We consider two two‐person organizations, called A and B. Each organization faces a changing environment; an environment has two components and each of them is privately observed by one of the organization's two members. Each organization's task is to respond to the current environment by taking a correct action; the correct action is a known function of the environment. However, the task of A is totally unrelated to the task of B: if A knew B's current environment and B's current correct action, that would tell A nothing at all about its own current correct action (and vice versa). Now suppose that each organization performs its task by a sequence of message announcements that stop when an “action‐taker”; has just enough information about the two members’ private observations so that he can take the correct action. Suppose we measure the effort this requires by the size of the set of possible message announcements. Then a compelling conjecture says that there can be no saving in total effort if we merge the two organizations into a single four‐person organization in which a single action‐taker takes both actions.

The conjecture turns out to be true when the possible messages form a continuum whose size is measured by its dimension, provided the message‐announcing procedure obeys suitable regularity conditions. When we turn to a model in which the number of possible messages is finite, the situation is different. While a certain general proposition about coverings and projections is the main tool in proving the “continuum”; conjecture, the finite analog of that proposition is (surprisingly) false. The finite version of the conjecture holds, on the other hand, when one adds a certain regularity requirement ("contiguity") to the message‐announcement procedure. The truth of the finite conjecture without such a requirement remains open.     

Finite allocation mechanisms: approximate Walrasian versus approximate Direct Revelation

Summary. This paper obtains finite counterparts of previous results that showed the informational efficiency of the Walrasian mechanism among all mechanisms yielding Pareto-optimal individually rational trades in exchange economies while using a continuum of possible messages. In particular, we develop finite counterparts of the superiority, with respect to message-space dimension, of the Walrasian mechanism over Direct Revelation (DR). We measure a finite mechanism's cost by the number of its (equilibrium) messages. Our two main results are as follows: (1) For exchange economies we find that the overall (maximum) error of a (sufficiently fine) approximate Walrasian mechanism is less than the overall error of a not-more-costly approximate DR mechanism whose equilibrium outcomes are trades that are (approximately) Pareto optimal and individually rational; more generally, approximate Walrasian mechanisms are superior, in the same sense, to approximations of any continuum mechanism whose outcomes are Pareto optimal individ ually rational trades and whose message space has higher dimension than that of the Walrasian mechanism. (2) As we increase without limit the dimension of the set of environments (characteristics) defining our class of exchange economies, the extra cost of DR approximations relative to Walrasian approximations, when both achieve the same overall error, also grows without limit. Thus the informational superiority of the Walrasian mechanism emerges again when we approximate it and take the finite number of messages in the approximation as our cost measure. Received: June 16, 2002; revised version: July 22, 2002 RID="*" ID="*" The second author is grateful for support from National Science Foundation grant #IIS-0118600. Correspondence to: T. Marschak     

Information Technology and the Organization of Firms

This paper seeks to understand why improved information technology (IT) might strengthen the case for decentralization, as recent empirical work suggests. We study a firm with a headquarters and two managers, each of whom gathers information about her changing local environment. The firm earns a gross profit that depends on actions taken as well as the current local environments. More information permits better actions, and information‐gathering costs drop as IT improves. When the firm is centralized, information‐gathering expenditures are first best, but after the firm decentralizes, each manager becomes a self‐interested player of a “sharing game” in which she collects a share of gross profit and bears the cost of her chosen information‐gathering activities. The firm's actions are determined by the information gathered at the equilibria of the game. As a result, the firm experiences a decentralization penalty, namely the change in net profit (gross profit minus informational costs) after decentralizing. If the penalty is small, then it is outweighed by the advantages of decentralizing—the vanishing of monitoring costs and perhaps the improved motivation of a decentralized manager's staff. To gather information a manager chooses (once and for all) a partitioning of her possible local environments and then searches to find the set in which her current environment lies. Our main measure of a manager's information cost is a technology parameter, θ, times the number of sets in her chosen partitioning. A second measure is θ times the partitioning's “Shannon content,” which may be interpreted as average search time when search is efficient. We ask whether improved IT, i.e., a drop in θ, indeed lowers the decentralization penalty. We obtain a strongly affirmative answer to this question for both cost measures in a class of examples and a mixed answer when we generalize so as to preserve some of the key properties of those examples. In a parallel manner we explore another conjecture suggested in the empirical literature, namely that better IT raises the coordination benefit, which we define as the increase in net profit when the firm bases its actions on pooled information, rather than letting each action variable depend on the information gathered by just one manager.     

Mechanisms that efficiently verify the optimality of a proposed action

The best known achievement of the literature on resource-allocating mechanisms and their message spaces is the first rigorous proof of the competitive mechanism's informational efficiency. In an exchange economy withN persons andK+1 commodities (including a numeraire), that mechanism announcesK prices as well as aK-compenent trade vector for each ofN−1 persons, making a total ofNK message variables. Trial messages are successively announced and after each announcement each personprivately determines, usingprivate information, whether she finds the proposed trades acceptable at the announced prices. When a message is reached with which all are content, then the trades specified in that message take place, and they satisfy Pareto optimality and individual rationality. The literature shows that no (suitably regular) mechanism can achieve the same thing with fewer thanNK message variables. In the classic proof, all the candidate mechanisms have the privacy property, and the proof uses that property in a crucial way. ‘Non-private’ mechanisms are, however, well-defined. We present a proof that forN>K,NK remains a lower bound even when we permit ‘non-private’ mechanisms. Our new proof does not use privacy at all. But in a non-private mechanism, minimality of the number of message variables can hardly be defended as the hallmark of informational efficiency, since a non-private mechanism requires some persons to know something about the private information of othersin addition to the information contained in the messages. The new proof of the lower boundNK invites a new interpretation of the competitive mechanism's informational efficiency. We provide a new concept of efficiency which the competitive mechanism exhibits and which does rest on privacy even whenN>K. To do so, we first define a class ofprojection mechanisms, wherein some of the message variables are proposed values of the action to be taken, and the rest are auxiliary variables. The competitive mechanism has the projection property, with a trade vector as its action and prices as the auxiliary variables. A projection mechanism proposes an action; for each proposal, the agents then use the auxiliary variables, together with their private information, to verify that the proposed action meets the mechanism's goal (Pareto optimality and individual rationality for the competitive mechanism) if, indeed, it does meet that goal. For a given goal, we seek projection mechanisms for which theverification effort (suitably measured) is not greater than that of any other projection mechanism that achieves the goal. We show the competitive mechanism to be verification-minimal within the class of private projection mechanisms that achieve Pareto optimality and individual rationality; that proofdoes use the privacy of the candidate mechanisms. We also show, under certain conditions, that a verification-minimal projection mechanism achieving a given goal has smallest ‘total communication effort’ (which is locally equivalent to the classic ‘message-space size’) among all private mechanisms that achieve the goal, whether or not they have the projection property.     

Inefficiency and complementarity in sharing games

A sharing game is a very simple device for partially reconciling an organization’s goal with the interests of its members. Each member chooses an action, bears its cost, and receives a share of the revenue which the members’ actions generate. A (pure-strategy) equilibrium of the game may be inefficient: surplus (revenue minus the sum of costs) may be less than maximal. In a previous paper, we found that for a wide class of reward functions, no one squanders at an inefficient equilibrium (spends more than at an efficient profile) if the revenue function has a complementarity property. In the present paper, we examine the “opposite” of the complementarity property (Substitutes) and we study a class of finite games where squandering equilibria indeed occur if Substitutes holds strongly enough. Squandering equilibria play a key role when one traces the effect of technological improvement on a sharing game’s surplus shortfall. We then turn to the question of choice among reward functions in a principal/agents setting. We find that if we again assume complementarity then strong conclusions can be reached about the reward functions preferred by “society”, by the players (agents), and by the principal.     

Comparing finite mechanisms

Summary. This paper obtains finite analogues to propositions that a previous literature obtained about the informational efficiency of mechanisms whose possible messages form a continuum. Upon reaching an equilibrium message, to which all persons “agree”, a mechanism obtains an action appropriate to the organization's environment. Each person's privately observed characteristic (a part of the organization's environment) enters her agreement rule. An example is the Walrasian mechanism in an exchange economy. There a message specifies a proposed trade vector for each trader as well as a price for each non-numeraire commodity. A trader agrees if the price of each non-numeraire commodity equals her marginal utility for that commodity (at the proposed trades) divided by her marginal utility for the numeraire. At an equilibrium message, the mechanism's action consists of the trades specified in that message, and (for classic economies) those trades are Pareto-optimal and individually rational. Even though the space of environments (characteristics) is a continuum, mechanisms with a continuum of possible messages are unrealistic, since transmitting every point of a continuum is impossible. In reality, messages have to be rounded off and the number of possible messages has to be finite. Moreover, reaching a continuum mechanism's equilibrium message typically requires infinite time and that difficulty is absent if the number of possible messages is finite. The question therefore arises whether results about continuum mechanisms have finite counterparts. If we measure a continuum mechanism's communication cost by its message-space dimension, then our corresponding cost measure for a finite mechanism is the (finite) number of possible equilibrium messages. We find that if two continuum mechanisms yield the same action but the first has higher message-space dimension, then a sufficiently fine finite approximation of the first has larger error than an approximation of the second if the cost of the first approximation is no higher than the cost of the second approximation. An approximation's “error” is the largest distance between the continuum mechanism's action and the approximation's action. We obtain bounds on error. We also study the performance of Direct Revelation (DR) mechanisms relative to “indirect” mechanisms, both yielding the same action, when the environment set grows. We find that as the environment-set dimension goes to infinity, so does the extra cost of the DR approximation, if the error of the DR approximation is at least as small as the error of the indirect approximation. While the paper deals with information-processing costs and not incentives, it is related to the incentive literature, since the Revelation Principle is central to much of that literature and one of our main results is the informational inefficiency of finite Direct Revelation mechanisms. Received: May 21, 2001; revised version: December 14, 2001 RID="*" ID="*" Earlier versions of this paper were presented at the Decentralization Conference, Washington University, St Louis, April 2000 and at the Eighth World Congress of the Econometric Society, August 2000. We are grateful for comments received on those occasions. The second author gratefully acknowledges support from National Science Foundation grant #IIS9712131. Correspondence to: T. Marschak