INTERNATIONAL ECONOMIC REVIE} Vol. 29, No. 2, May 1988 OPTIMAL FINANCIAL STRUCTURE IN EXCHANGE ECONOMIES* By JosepH G. HAUBRICH* 1. nTRODUCTION The current upheaval in the financial system has forced economists to examine financial institutions and structures in a new light. Discovering why some funding occurs through banks rather than asset markets poses a challenge not only to academics, but also to policy makers who must determine their role in the financial system. The new approach starts from the premise that financial struc- tures arise to solve informational difficulties inherent in the underlying economy. Though recent work has often shown how specific contracts or institutions can mitigate an information problem, it has rarely investigated the optimal arrange- ment for that particular economy. Such a distinction is particularly important if we consider free-entry and competition among the highly motivated “rocket scientists” and entrepreneurs who design and set up institutions. Even work that has examined optimal arrangements has artificially restricted the options and opportunities of investors within the model (Diamond and Dybvig 1983; Jacklin 1987). These restrictions are felt particularly acutely because of the sophistication of modern financial markets. In part, such restrictions were imposed by the state of optimal contract and allocation theory; similarly, recent advances in that theory provide a richer framework for research. This paper works toward a positive theory of financial structure. Using results and techniques from the modern theory of economic mechanisms (Hammond 1987; Prescott and Townsend 1984) and contract theory (Holmstrom and Weiss 1983; Stiglitz 1982), it carefully specifies the basic structure of the economy and derives the optimal financial arrangements in a simple general equilibrium set- ting. In contrast to earlier work, this paper explicitly considers coalitions and syndicates and permits the introduction of contrived uncertainty into contracts. These general arbitrage arguments and contractual possibilities hold promise of more fruitfully representing an increasingly deregulated financial world, where bankers and investment groups put together innovative financing and create new instruments and accounts. The new techniques also re-introduce some important elements of finance that had been missing from contract theory. These previously missing elements are arbitrage and contrived uncertainty; for example, adding randomness to your actions by keeping the size and timing of your offering secret. + Manuscript received October 1986; revised February 1987, * An earlier version of this paper was presented at the Econometric Society meetings, Winter, 1984, I wish to thank Robert King, Mark Gertler, Peter Hammond, Richard Kihlstrom and two anony- ‘mous referees, as well as seminar participants at the Board of Governors and Pena for valuable help. Any errors remain my responsibilty 217218 JOSEPH G, HAUBRICH Clarifying the assumptions on information, observability, and enforcement that lead to banks, insurance companies, or asset markets may shed special light on bank regulation. Most justifications of bank regulation have viewed it not as a fine tuning exercise, but as essential to the viability of the entire industry. We can clarify the advantages that banks and other intermediaries bring, and also exam- ine when, and what sort of, regulation is necessary to preserve those advantages. While providing a basis for addressing questions about the existence, regula- tion and change of financial institutions and markets, this approach has its costs. First, it provides no immediate policy recommendations. In that sense this work is the polar extreme to such works as Tobin and Brainard (1963) and Smith (1984), which simply posit demands for liabilities and loans in order to develop the richer models needed to obtain substantive policy implications. The work of Diamond and Dybvig (1983) on banking panics falls somewhere in between. Secondly, this simple model misses some important attributes. The analysis ab- stracts from the costly monitoring of agents (Diamond 1984; Campbell and Kracaw 1980) as well as the possible economies of scope that lead to the bund- ling of separate financial services (Gilligan, Smirlock, and Marshall 1984). This Paper avoids those admittedly important factors to focus on other issues. Both these disadvantages occur because for now, tractability limits this methodology to simple versions of the models currently used in banking research. Analyzing financial arrangements also provides a simple context for the expo- sition and clarification of several new concepts in mechanism theory, along with a chance to examine their usefulness in a particular setting. Using an endowment economy means that the main theorems and proofs can be illustrated with Edge- worth boxes. This paper employs a one good, two period endowment model with individual uncertainty. This simplification (in the tradition of Lucas’s “Asset Prices in an Exchange Economy” (1978) and Townsend's “Economic Organization with Lim- ited Communication” (1987)) permits calculation of optimal contracts yet still provides time and uncertainty, the two essential elements of a financial economy. ‘The work starts by formulating a maximization problem, solving it, and finding the market or institutional support for the optimal allocation determined by the problem. The main results of the paper lie in showing how the informational assump- tions lead to constraints upon the planning problem which then determine the resulting institutions. The most striking features are that recognizable, though simple, financial institutions emerge from this procedure and that multilateral incentive compatibility significantly alters the resulting structures. Perfect infor- mation corresponds to imposing only resource constraints, This results in a standard insurance contract, with perfect risk sharing. Private information in- duces a need for incentive compatibility (self-selection) constraints which result in a financial intermediary of the sort considered by Diamond and Dybvig (1983), although some payoffs may be random. When there is private information and when individual trading is also unobservable, multilateral incentive compatibility constraints become relevant. These restrict the allocation to be the ex post com- petitive equilibrium: the equilibrium of a consumption bond market whereOPTIMAL FINANCIAL STRUCTURES 219 people trade claims to income after the resolution of the uncertainty. These same constraints also rule out any randomness in the final allocations. We further show that imposing multilateral incentive compatibility corresponds to imposing no-arbitrage conditions. The results from this general setting should also serve as a warning. When intuitively plausible economic environments are carefully specified, it turns out that many types of financial institutions cannot arise. This is not to argue that banks do not exist but rather that a particular, popular, form of economic analy- sis is fraught with pitfalls. One goal of this paper is to clarify the assumptions that lead to different institutions. Seemingly innocuous specifications—such as the observability of consumption—can have major consequences for the resulting equilibrium, The remainder of the paper is as follows. Section 2 describes the economy and sets up the mathematical program to be solved. Section 3 examines the solution to the basic problem with deterministic allocations and individual incentive com- patibility constraints, and discusses the financial system that supports the solu- tion. The next two sections address more elaborate versions of the problem. Section 4 allows random allocations while Section 5 brings in multilateral incen- tive compatibility constraints and discusses support for the new allocations. Sec- tion 6 presents a summary and conclusion. 2. DESCRIPTION OF THE ECONOMY To examine the basic issues in as simple a manner as possible, I use a two period endowment model with uncertainty in the first period. This provides a motivation for banks along the lines discussed in Diamond and Dybvig (1983) and Haubrich and King (1988). This captures the liquidity and insurance factors emphasized in those papers, but also answers questions involving more sophisti- cated behavior. This exchange economy has one good x consumed in each of two periods T = 1,2 with no storage between periods. At T = 2, all agents receive the identi- cal endowment y,. At T= 1, endowments differ. A lucky fraction ¢ of the continuum? of agents will receive a first period endowment of y, + 6, > ys, While an unlucky fraction (1 ~ 1) receives y, +8, < y,, where for consistency 10, + (1 — 1), = 0. The shock (6,, 8,) is private information: it is impossible to verify if, an agent has become “rich” or “poor.” Notice that this endowment uncertainty is perfectly diversifiable but that perfect diversification may not occur because of information (adverse selection) problems. All agents have identical? preferences represented by a utility function U(c,, ¢2) defined over consumption in both periods. The utility function is strictly increas- ing in c,, continuous, twice differentiable and strictly concave. In addition it exhibits normality in consumption in both periods. * Judd (1985) demonstrates the mathematical consistency of this modelling technique, which in- solves a continuum of independent draws whose realization matches the true density /(,). * One may also represent the uncertainty by an additive shift in preferences as well, in which case endowments would be identical and preferences random, The two cases are more than just formally equivalent. Both model uncertainty arising from a household production function.220 JOSEPH G, HAUBRICH Since this model incorporates both time and uncertainty, an optimal contract specifies both intertemporal trade (borrowing and lending) and risk-pooling (in- surance). Finding the optimal contract/institution involves first solving a maximi- zation problem to determine the optimal allocation of the good across states and time, and then finding a contract that supports such allocations. Before discussing the nature of the optimization problem, though, it may be useful to consider the ex post competitive equilibrium of this economy. This simple borrowing and lending market provides a reference point for comparison with the allocations and institutions discussed later. It may also serve to clarify the nature of the economy. Define an (ex post) competitive equilibrium to be a set of prices p, and quantities ¢(0,) such that (j) cach agent is maximizing utility given Prices and his or her endowment, and (ii) supply equals demand for each good. ‘That is, ¢{(9,) solves max U(c,, ¢2) subject to Pils + P22 = Pi +O) + Pr Po and in turn satisfies the aggregate consistency condition L lO /8) =¥ yl@)F0) for al In other words, after the resolution of the uncertainty, this is the standard two- period exchange economy. Individuals engage in borrowing and lending to smooth consumption over the two periods. Since no storage is possible, economy-wide net borrowing will be zero in each period, Moving beyond the ex post competitive equilibrium ¢o the optimal financial contract involves two steps. The first derives the optimal contract using a pro- gramming problem. The next interprets that solution as a contract, exhibiting an institution that implements the solution. Formally, the Revelation principle and the work on mechanism design (Harris and Townsend 1981; Meyerson 1979) justify this approach, but the basic idea just extends the second welfare theorem: in a classical economy, any Pareto optimal allocation can be supported by a decentralized market. This paper also examines when a decentralized system upholds the contract and when a centralized authority (such as a government) must intervene to support the social optimum. Choosing the optimal contract requires solving the mathematical program- ming problem. Since all individuals are alike ex ante, the program must maximize the expected utility of a representative agent who has yet to learn his or her type. This chooses the best allocation (which may be random) subject to the resource constraints, the incentive compatibility constraints, and the multilateral incentive compatibility constraints. These constraints have straightforward interpretations. First, a distribution of the good among agents must be feasible. Second, since types are unverifiable, output must be distributed so that no agent gains by misrepresenting his or her type. Finally, no group or coalition can improve upon an allocation by selective misrepresentation combined with trading among itself.OPTIMAL FINANCIAL STRUCTURES 221 With this in mind, we now proceed to the formal statement of the problem and its solution. 3. THE PROBLEM AND ITS SOLUTION This first stage of the analysis will characterize optimal allocations, where optimizing involves maximizing expected utility subject to various constraints. Since a major task of this paper involves determining the effects of allowing random allocations and imposing group incentive compatibility constraints, we solve the problem in steps. The first step chooses non-random allocations subject only to the individual incentive compatibility constraints. The next step allows random allocations (lotteries), and the final step involves imposing the multilat- eral incentive compatibility constraints. These constraints imposed upon the problem correspond to differing levels of the social planner’s (and thus ultimately the contracting parties’) ability to moni- tor the actions of the private agents. Fully observable income permits allocations contingent on agent type. Only the resource constraints would need to be con- sidered; the incentive compatibility constraints would be superfluous. However, when the social planner observes only transfers, incentive compatibility con- straints are required, since the planner cannot identify agents by type beforehand during the distribution of allocations. Still, there is no need to impose group or multilateral constraints since agents cannot trade or make side payments among themselves: the social planner can make sure they consume what they are given. Without observable transfers, however, group incentive constraints come into play, The allocated goods can now be traded and bargained with. The sequence of constraints we consider forms a natural progression following the technology ‘of communication and monitoring present in the economy. To solve the problem, then, one looks for allocations x,, which maximize the expected utility, EU(x,, + 0,, x2,), of agents before they know their type, where i = 1, 2 indexes period and j = 1, 2 indexes agent type. Recall that t denotes the probability that agents become type I and receive @,, and that y, denotes the per capita income of period i, These allocations are the solutions to o MAX 1U(x 41 + O45 X21) + UL = DUC 2 + O25 22) subject to thy FU OX = Me (RESOURCE) tay + (l= X22 = V2 and Uy + Oy, X24) = UOy, + Oy, X22) re) Ub 2 +02, X29) = Ulys + On, X20) which incorporates individual incentive compatibility conditions,* or (II), which * Allen (1985) shows that conditioning allocations on the announced distribution of characteristics results in superior sorting. Using his procedure changes the allocations considered in this paper, but the propositions (i. the full information solution is unattainable, etc.) still hold2 JOSEPH G. HAUBRICH is (I) plus coalition incentive compatibility conditions Uw: (micy U(Bx1, +L BiXy2 + 2, BXz1 + (1 BX32) S Uy: + O25 X22) where 0 < y, B < 1 and denote the fraction of a coalition’s members representing themselves as type 1's. That is, after people have been to the planner they meet in groups and trade. This ex post coalition, then, cannot force, and indeed may not Know, what agents report to the planner. However, for interpreting the MIC constraints in problem Il it is easiest to think of the coalition knowing member's reports, and thus knowing the fractions of reported types and (1 — 7), and then. giving each member an equal share of the coalition’s goods. The 4's, being pri vate, are excluded from the redistribution. The form of this constraint appears to restrict the bargaining process since coalition members receive payment exactly in proportion to the number of people claiming to be each type. In fact, this is no constraint because each individual is free to choose which coalition to enter (or form), and by varying the proportions of agents, can achieve any proportion possible in a bargaining situation.’ Notice that, in line with other contract models, the solution to this optimal contracting problem specifies quantities, not prices. Since goods produce utility, we directly choose the amounts going to individuals. In some cases, there may exist prices (interest rates) such that an agent would choose the appropriate consumption pattern. In many cases, though, the person faces a discrete choice and must choose between two distinct patterns of consumption. The budget set facing an individual need not be a straight line. In fact, it may consist of several distinct points. That difference distinguishes institutions—such as insurance com- panies that give you a choice between two non-combinable contracts—and markets—where you can borrow and lend any amount at the prevailing interest rate. ‘The Lagrangian, and hence the first order conditions for problem (I) will involve four multipliers: two from the resource constraint, 4,, 22, and two for the incentive compatibility constraints, j1,, 1. A constraint is said to bind if the solution satisfies it with equality, ie., except when it just binds, the associated lagrange multiplier is non-zero. We do not show the first order conditions for problems (I) and (II) for two reasons. First, the more interesting facts about the solution do not easily follow from the first order conditions, and secondly the constraints need not be convex. Other techniques, along the lines of Holmstrom (1982) and Holstrom and Weiss (1983) will prove more helpful. +L = PX a2 + Or, P21 + — X32) S Ul + Oy. X21) * Note that the form of the MIC constraints allows coalitions of infinite size and positive measure. Because of this, the problem also (implicitly) assumes that the social planner knows the distribution ‘of characteristics (1) beforehand, Otherwise, a coalition of positive measure could manipulate the apparent distribution and render the allocation infeasible, An alternative approach (following Ham- mond 1983, or Kaneko and Wooders 1988), which would matter only for the first proof of proposi- tion 6 below, would be to consider only finite coalitions. The constraints labelled MIC would remain “appropriate, but with y and f rational numbers. Continuity, though, would then imply the same MIC ‘inequalities would continue to hold.OPTIMAL FINANCIAL STRUCTURES 23 The following three propositions exhibit the basic properties of the solution. This section will concentrate on the intuition behind these results for the model at hand, as these results are particular cases of well-known general results (Harris and Townsend 1981; Guesnerie and Laffont 1984; Prescott and Townsend 19846). First, however, we establish a useful benchmark: the full information or unconstrained optimum, the allocation attainable if 0, were public information. With agent type observable, allocations can be made contingent on that type. Each agent then receives (y, —@;. y2) and thus consumes (),, y3). The public information allows perfect insurance of the ex ante identical agents since the risk is diversifiable: agents consume the expected value of income. An insurance company that took in premiums from lucky agents and made payments to the unlucky could work here. Proposition 1. The full information social optimum does not solve (I). Poor. The full information optimum removes all uncertainty. This implies equal consumption for all, or XO, = yy =x +O, and xy, =, but then x,, +6, > x, +0, since 6, <0, showing Uler2 + 91, X22) > Ulery + 91, X24) so this allocation is not incentive compatible. QED. Proposition 1 should not be surprising and as general result is well known. Imperfect information creates the need for the IC constraints, which preclude reaching the unconstrained social optimum. In relation to the other benchmark, Proposition 3 (below) shows that the optimum allocation improves upon the ex post competitive equilibrium of this exchange economy. The allocation that emerges as the solution to (1) need not lie on the contract curve of this economy (a simple example of this occurs with a utility function ule, ¢3) = cf*cl* and equal numbers of agents). That is, the allocation that maximizes ex ante expected utility does not equalize, ex post, the marginal rates of substitution of the two types of agents. This is an application of the general theory of the second best: satisfying the necessary conditions for an unreachable first-best optimum does not lead to the second-best solution (Lipsey and Lan- caster 1956). This often occurs in areas involving sorting and screening. For example, in Grossman and Hart (1981), private information and the consequent demand for insurance produces labor contracts that exhibit ex post inefficiency. Unemployment exists and the marginal value product of labor does not equal the value of the marginal disutility of labor. Removing that inefficiency would de- stroy the ex ante (prior to the resolution of the uncertainty) Pareto efficiency of the contract; everyone would lose if the unemployment could not occur. A simi- lar situation occurs in problem (1). Because of the incentive compatibility con- straints, moving off the contract curve allows greater insurance. The IC constraints, coupled with the desire to smooth consumption over time, and the assumption that all goods are normal, imply a time pattern in the224 JOSEPH G. HAUBRICH ct Ficure 1 ONLY ONE IC CONSTRAINT BINDS: AN AGENT INDIFFERENT BETWEEN (x, + Og, X34) AND (¥3 + Oz. a2) WILL NOT BE INDIFFERENT TO (x + ,.€33) AND (Xp + 8. %33) allocations: x,, x2. People with a positive draw in T =1 desire to save and consume some of the windfall tomorrow. Those with a nega- tive draw wish to borrow and spread the loss over both periods. This enables the planner to separate the two groups, but care must be taken not to give either group more in both periods. This background leads to PROPOSITION 2. Only one IC constraint binds at the optimum of (I). PRoor. Without loss of generality, consider the case where the poor IC con- straint binds. That implies U(xy2 + 82, X23) = Uli, + 82, X2,) (ie. on the same indifference curve). If the rich constraint also binds, then U(x,, + 61, X21) = U(x42 + 84, X22). The indifference curve through (x,, + 0,, x2.) must be flatter than the curve through (x,, +, X21) since consumption of both goods is normal. Similarly, the slope at (x,; + @,, X22) is flatter than at (x,, + 0, X22) (see Figure 1). Normality and convexity preclude mutual satisfaction of these two conditions. A convex indifference curve cannot pass through (x,, + 6,, x,,) and (42 + 04, Xaq) and be flatter at both points. The rich IC constraint either does not bind or is not satisfied. QED.‘OPTIMAL FINANCIAL STRUCTURES 225 One and only one Lagrangian multiplier associated with the incentive compati- bility constraints can have a non-zero value. Intuitively, it seems unlikely that both groups will envy each other at an optimum allocation. Since reducing risk involves redistributing income from the lucky (@,) to the unlucky (8,), one expects that it is the IC constraint of the lucky rich that precludes the full information optimum. The allocation mechanism functions as a partial insurance program. Lucky agents pay a premium while unlucky agents receive a payment. Too large a transfer, though, and the lucky agents would pretend to be poor. PRoposttion 3. The IC constraint of the rich binds at the optimum of (I) Proor. From Proposition 2, one IC constraint must bind since Proposition 1 rules out the unconstrained solution. Now assume the constraint binds the poor, not the rich. Then @ Uy, + Or, X21) > U2 + O15 X22) and Gi) UG, + O25 X22) = Ulxas + 82, X21) but recall x,, <.x,3 and x,, > xz,. This implies we can take dx,, from the rich and give dx, to the poor, increasing x,, and decreasing x, , provided tdx,, + (1 — t)dx,, = 0. Notice (i) still holds for small transfers while (ii) becomes U(x, + 03, X22) > Ule1 + 92, X21). The IC and resource constraints remain satisfied, but the redistribution reduces uncertainty by moving weight from the tails of the distribution to the center, increasing expected utility for the risk averse.° Put another way, relaxing a binding constraint increases the maximum, QED. The optimal allocation characterized above dominates that supported by a simple ex post borrowing market where securities are claims (not state contin- gent) on consumption in a given period.” Fully state contingent securities are impossible, of course, because of the private information. Still, the desire of agents “As an example, consider separable utility, Uley, ¢:)= ule.) + Brea). The reallocation then changes expected utility from tue, + 0,) + Bubxa)) + (1 — HMM, + 2) + Brea) 10 tabeyy + 0,) + Bvbea, — dxz,)) + (I~ Oulxy, + 8) + Bibra ~ dxg)) Note that duc to separability, first period utility does not change. Second period dispersion of consumption decreases, and since v is concave, expected utility increases, a direct application of Rothschild and Stiglitz Theorem 2 (1970) * Since this result, that the allocation resulting from a non-contingent market is not Pareto ‘optimal, is well known both in a general context (Harris and Townsend 1981) and in the banking literature cited below, presenting a proposition would be superfivous. Sul tis straightforward to sketch a proof of the claim. At a competitive equilibrium, indifference curves of both types will be tangent to the budget line through the endowment. The “cheat point” or allocation a person gets when he misrepresents his type can be shown to lie on the same budget line, but on the opposite side of the endowment (see Figure 4a). Since the indifference curves are tangent at the equilibrium (point ‘Ain the figure) both types strictly prefer the competitive allocation to the “cheat point™ (B). By Proposition 1, however, we know an incentive compatibility constraint binds at the optimum, s0 the ‘© post competitive allocation is not second-best. A more intuitive approach would be to mimic the technique used to prove Proposition 3, and show how a redistribution which reduces uncertainty (in Figure 4a, a south-west movement) is feasible.226 JOSEPH G. HAUBRICH to smooth consumption allows a contract that reduces income risk. The allo- cation requires some contract, some institution such as an intermediary which can transform those ‘primitive’ assets of agents’ endowments into more sophisti- cated securities. Following recent work stressing the insurance and liquidity role of the banking sector (Diamond and Dybvig 1983; Smith 1984; Bhattacharya and Gale 1987; Jacklin 1987; Haubrich and King 1988), we may consider the institution a bank. As Diamond and Dybvig (1983, p. 402) point out, “banks issuing demand deposits can improve on a competitive market by providing better risk sharing among people.” Imagine the following scenario. Agents agree to deposit y, + 8, (the maximum verifiable amount) in the bank in period 1 and also agree to deposit y, in period 2. They are then given a choice between two withdrawal streams. (d,,, d3,) and (d,,, dy) where dyy — 02 = X44, day = Xay, dy = xyz and dy, = x33. Like other theoretical banks posited in the literature, this one sets an interest rate below the no-bank level. This serves to redistribute income from the rich, who tend to lend, to the poor, who wish to borrow, thus providing insurance because ex ante, agents do not know their shock. In this simple set up, of course, the interest rate is only implicit. The agents’ opportunity set is not a line, but rather two points. This depository institution therefore displays two important features of banks: intertemporal trade and insurance. This simple model thus captures the essence of many recent banking models, which have proved useful in analyzing many questions, including panies, deposit insurance, and usury laws. Later, in Section 5, we shall see how this bank exhibits yet another quality of recent banking models, namely, the need for exclusivity. The bank must be able to guarantee that agents consume their allocation, either by directly observing consumption or by exclud- ing markets for assets that are claims to income. As the reason for such exclusi- vity involves coalitions, we postpone further discussion of this until Section 5. 4, LOTTERIES AS ALLOCATIONS In an important recent contribution, Prescott and Townsend (1984a) have argued for the consideration of random allocations, which they term lotteries. In the real world, such contrived uncertainty may take the form of secrecy in the size and timing of offerings, deliberately vague rules and procedures (subject to random enforcement by a committee or court), or contracts tied to uncertain but irrelevant events. One possible benefit of such randomness, as Prescott and Townsend have pointed out (in a slightly different context) is to convexity Prob- Jem (D) and use risk aversion to sort people. Figure 2 shows that the incentive compatibility constraints of (I) are not convex. Let point A denote the consump- tion level of the rich agent when the rich agent tells the truth. The IC constraints imply that the utility of this “truth-telling” bundle must be at least as high as the bundle received when the rich agent claims to be poor. Thus a rich person’s “cheat point” (the poor allocation of relatively more today, plus the income shock), for example B, must lie on or below the indifference curve through A. Point B satisfies the constraint, but convex combinations of A and B do not. Define a lottery L, as a probability distribution over all possible allocations x,,‘OPTIMAL FINANCIAL STRUCTURES 227 4 Ficure 2 IC CONSTRAINTS NOF CONVEX in the commodity space. The utility of a lottery L, will be EIU(L)) = J Ula + js Xaelajs X29) Mey, dea where 2(x1;, X2,) is the probability of a bundle (x,;, x2,). The incentive compat bility constraints become (LIC) for, + O jy, Xaj)2l% aj, X)) dx > fom +O p, Xapel%ias Xap dx, Aj. Likewise the resource constraints become j 4208150 Xa) dey With a continuum of agents the lottery may be independent across agents. Other- wise, the scheme must introduce some dependence to satisfy the resource con- straints, which would break if too many winners arose. Any allocation can be specified by the probability weights z, since z is from a distribution over the entire commodity space. Thus, by the expected utility theorem, utility is linear in z and, hence, the LIC are linear in z, and constitute convex constraints. A deposit228 JOSEPH G. HAUBRICH contract now gives you a choice of lotteries—you withdraw a randomized amount. For example, your place in line during a bank run (the Diamond- Dybvig sequential service constraint) provides a lottery. You might get your money out, or you might not. By curing non-convexities lotteries allow many problems involving uncertainty to be studied with the tools of modern general equilibrium theory.* However, lotteries change the qualitative nature of the optimal allocation less than might initially be expected. Notice the tension inherent in contrived randomness: the purpose of a bank is to reduce uncertainty, but a lottery adds uncertainty. PROPOSITION 4. If agents are risk averse, lotteries cannot achieve the full infor- ‘mation social optimum, PRoor, Assume using lotteries increases expected utility, that is, assume the optimal lottery is not degenerate. Hence, there must be randomness in the allo- cation. The full information solution to (1), however, is a non-random amount, so with risk aversion the optimal lottery allocation is strictly inferior to the un- constrained optimum. QED. Prescott and Townsend in their Section 6 present an example where the use of lotteries leads to the full information solution. A necessary part of that example, however, is that one party be risk neutral and thus able to bear a lottery with no loss of utility. Proposition 4 may serve as a counterweight to that example by showing that in general such extreme results are not possible, It seems intuitive that lotteries cannot totally remove the effect of imperfect, incomplete infor- mation, Often, though, they do considerably less. Stiglitz (1982) provides conditions for beneficial randomization in the related context of the optimal income tax. There, the government cannot directly ascer- tain whether an individual is skilled or unskilled. Thus, “the objective of rando- mization is to increase the effectiveness of screening (or, to put it another way, to reduce the welfare loss associated with the self-selection constraints).” Stiglitz’s results apply directly to our model, as he also uses two goods, consumption and leisure, While randomization frequently proves desirable, there exist broad cases where it does not. With additively separable utility, for example, decreasing abso- lute risk aversion precludes randomization, In general the usefulness of lotteries depends on the exact specification of endowments, uncertainty and preferences. As the proof of Proposition 4 points out, randomness cuts both ways. It can relax self-selection constraints, but may harm risk averse agents by adding uncertainty. 5. MULTILATERAL INCENTIVE COMPATIBILITY If the social planner cannot explicitly specify consumption nor exclude the possibility of exchanges between individuals, individual incentive compatibility constraints alone are not enough to guarantee that people actually consume their * For more details, and many applications, see Prescott and Townsend (1984a, blOPTIMAL FINANCIAL STRUCTURES 229 assigned allocations. A simple example is when the IC constraints put an allo- cation off the contract curve. Agents then trade from this allocation to reach the contract curve (CC). The resulting point is not incentive compatible: if it were, the allocation could have been there, since it is Pareto superior to the original point. The trade makes both individuals better off, so E(U) rises. This makes the original allocation not incentive compatible, given the possibility of trade, More formally, an allocation is multilateral incentive compatible (MIC) if no coalition can improve upon the allocation. The strategies open to the coalition involve the types claimed by members and the exchanges between members. Of course, the coalition cannot verify agents’ income shocks any more than the planner can. Still, as the above example shows, that is not necessary for coalition formation to restrict feasible allocations. After the allocation, and after the income shock, there can be incentives (arbitrage possibilities) for groups, In the above case, everyone truthfully reports his or her type, receives the incentive compatible allocation, the private idiosyncratic shock, and being off the contract curve, trades. Two agents making a mutually beneficial exchange do not need full information about endowments: an enforceable trade can be three units today for five tomorrow. All they need to know is that someone is willing to trade with them. ‘One noticeable restriction concerns lotteries.” Proposition 5. If lottery results are public information then lotteries cannot improve upon the deterministic solution to (I). PRoor. Without loss of generality, consider a lottery given to the poor, (L;) Then consider the coalition consisting of all agents claiming to be poor (and thus receiving the poor allocation). The public nature of the lottery enables the co- alition to collect each agent's realization and dispense E(L,) to each. This re- moves the uncertainty and makes the allocation deterministic. Since agents are risk averse, they will join the coalition QED. Notice the important place of observability in the proof of Proposition 5. If lottery results were private information, coalitions could not diversify away the contrived risk. The following proposition proves that the bank allocation is not MIC, showing that additional asset markets enable agents to redistribute the coalition’s pro- ceeds. With such coalitions possible, some groups can do better than the bank allocation with a strategy of misrepresentation and redistribution. This is the reason that we have stressed the exclusive nature of the bank, and the need for the absence of other asset markets. Propostnion 6. The solution to problem (1) does not solve problem (IX). Proor. Since the fractions ¢ and (1 — 1) will cancel out, consider the case where t = $, We showed above how points off the contract curve do not solve (II) * 1 wish to thank Robert King and Robert Townsend for pointing this out to me.230 JOSEPH G. HAUBRICH cat Unlucky (Poor) Lucky (Rich) 4 Figune3 POINT A IS NOT MULTILATERAL INCENTIVE COMPATIBLE So we concentrate on solutions to (I) which lie on the contract curve. For any such point A (see Figure 3) corresponding to (x,, +0}, x31) find the “cheat point” B of (x,, + 0;, X,3)- These points lie on the same indifference curve by Proposition 3. By the MIC constraints in (II) agents can consume anywhere along the chord AB by having different proportions of the coalition membership misrepresent their type. Convexity of preferences implies this is preferred to A. QED. In fact, the multilateral incentive constraints put rather stringent limits on the insurance possibilities of problem (II). Proposition 7. An allocation solves problem (Il) if and only if it is an ex post competitive equilibrium for the economy. Proor. As in the previous proposition, it suffices to examine 1 = 4. The proof shows that any solution to (1) and the corresponding “cheat point” lie along a price line through the original endowment point, (y, + 0, vai 1 +92. ¥2). This implies (i) any allocation 4 that is a competitive equilibrium is MIC, and (ii) any other allocation is not MIC (by a variant of the proof of Proposition 7). Let E be the initial endowment, E =(y, + 0, y,). The allocation point is A= (x1, +0 X12) While the “cheat point” B= (x12 +0, X22)= (2), — X14 +8 22 — X24OPTIMAL FINANCIAL STRUCTURES 231 cc. Fioune 44 (See Figure 4, a, b.) the slope from E to A is ar — yeVer +O = 01 + 4) while the slope from B to E is Ya ~ v2 Xa WO + 8 = Ay, = X11 + ON both of which equal Dea — y2V/Dea1 — ide Now if A is a competitive equilibrium for E, it must also be for B. Hence A is incentive compatible (also MIC). Should A not be an equilibrium, some section of cc Fioure 42 [AN ALLOCATION 1s MIC IF AND ONLY IFFT IS 4 COMPETITIVE EQUILIBRIUM232 JOSEPH G. HAUBRICH the line segment AB will be above an indifference curve through A. Convexity implies one coalition can improve upon A (though it may be a coalition of the poor) QED. This result, that the only multilaterally incentive compatible financial structure is the competitive equilibrium, bears a close relation to other work. In particular, Hammond (1983, 1987) has shown under quite general conditions that MIC precludes non-linear pricing of exchangeable goods. In the economy considered here, exchangeability depends upon observability. If the central planner or Bank cannot observe consumption or transfers directly, agents can trade income. In effect, agents have access to a loan market. Income becomes an exchangeable commodity. Furthermore, the bank allocation (solution to ()), based on discrete deposits and differential payments, is decidedly nonlinear. In that case, Ham- mond’s results suggest that the non-linear solution to (1) (without MIC) will not survive coalition formation. From an ex ante perspective, everyone is better off with a bank (I) than with only a loan market (I). However, the extra benefit of insurance comes from nonlinearities that get arbitraged away by groups. Ham- mond also relates multilateral incentive compatibility to the literature that uses the more normative concept of fairness. Proposition 7 also differs from Hammond’s work in several ways. Hammond concentrates on preference shocks while this paper uses random endowments. This leads to different proof techniques, and somewhat different theorems. Per- haps the most important divergence is that this paper shows below how bilateral trading can duplicate the full effects of multilateral incentive compatibility even though the outcome of the random shock does not form a connected set (as it does not, taking on only + or ~6). From a purely theoretical standpoint, then, while Hammond’s results are much more general, they do not encompass the results of this paper. A corollary to Proposition 7 extends Proposition 5 to the cases of private lottery results and finite coalitions. If the final allocation, after the lottery draw- ing, is not an ex post competitive equilibrium, it is not multilateral incentive compatible. There will exist gains from trade that agents will exploit, disrupting the allocation. Conversely, when the multilateral constraints are not applicable, Proposition 5 does not hold and banks have some scope to induce randomness. Economically, there is some reason to fear results that depend too heavily on the formation of coalitions. We intuitively expect that organization, agency, and transaction costs severely restrict coalition formation. However, a non- cooperative approach to the problem reveals that much weaker conditions yield the same results. The proof of Proposition 7 does not require complicated co- alitions. All that is really needed is for one individual to claim an incorrect type, move to the “cheat point” B and offer to trade with other agents. This results in an allocation along AB. If everyone else has told the truth and received the non-MIC allocation, it remains in the interest of the remaining player! to misre- °° Mote pre a set of positive measure must follow this strategy. It remains true that each ‘member of the set acts individually to take advantage of the arbitrage opportunity. For more general economies, Hammond (1987) shows that a sufficient condition for this “bilateral” incentive compati- bility to force a competitive solution is forthe 6's to form a connected set.OPTIMAL FINANCIAL STRUCTURES 233 present his type and trade with others. The real force behind multilateral incen- tive compatibility is the ability of agents to exchange goods, not the complex machinations of coalitions. As Hammond (1983, p. 19) puts it, “consumers need rely on each other only to make mutually advantageous exchanges. The proof of Proposition 7, then, ultimately relies only on arbitrage consider- ations. The time pattern of consumption, which sorts agents in order to provide insurance, imposes different rates of return on implicit “consumption bonds” for the two groups. Income being exchangeable allows agents to arbitrage between these bonds. Agents do not need to know the private information that lies behind why some will trade at that price, only that they will. A planner (or contract enforcing authority such as a court system) with enough information, however, can prohibit this activity. In effect, this closes down the market and thus elimi- nates the possibility of arbitrage. Since the solution to problem (I1) must be a competitive equilibrium one type of support is simply to let agents trade claims to income. Of course, other equivalent mechanisms are possible. For instance, a bank taking in deposits and allowing withdrawals can exist side by side with a loan market. The rate of return on bank deposits would then equal the competitive return on other assets. Em- pirically this is false, suggesting that banks are probably more than disguised mutual funds. ‘The lower expected utility when income is exchangeable creates an incentive to somehow make income non-tradeable, and thus allow a bank to redistribute income,"! Indeed, much government regulation of the banking system may rest on this incentive. For example, some assets, such as IRA's, not only cannot be redeemed early, but, according to regulations cannot be used as collateral for loans. This seems an attempt to restrict the exchangeability of income, Similarly, interest rate ceilings may exist to prevent arbitrage of the low rates set by the bank. This corresponds to much of the rhetoric justifying bank regulation.'? Without a mechanism to generate exclusivity and prevent ex post trading, the banks providing insurance and liquidity services could not exist. Banks provide a real service to the economy, but “excessive competition” and “speculation” preclude a viable banking system. This creates a clearly defined role for the government, which promulgates laws preventing the exchange of income and thus protects the banking system. There remains the problem of why the government has superior knowledge or enforcement ability. In the model as presented the government has no such advantage. Observable transfers allow private parties to prevent the destabilizing arbitrage; but unobservable transfers cannot be observed, and thus stopped, by the government any better than by anyone else. However, perhaps we call an enforceable contract among all agents the government. 6. SUMMARY AND Cor IONS. Rather than coming to any definitive conclusions on the nature of financial structure, this paper has been an analytical first step in building a model that 1 Jacklin (1987) also makes this point. "2 For a summary of recent works on this topic, see the “Symposium on Bank Regulation,” Journal of Business (1986).234 JOSEPH G, HAUBRICH captures the important features of financial systems. This paper has shown (I hope) that such an exercise is possible, and has raised several important issues along the way. Despite the extreme simplicity of the economy and the stark formalism of the method, recognizable institutions emerged as the outcome of an explicit optimization process. Much more work needs to be done to clarify many theoretical issues and to produce empirically testable structures that can provide guidance in matters of policy. This study may also serve as a warning, Justifying a financial institution as the outcome of a mathematical programming problem requires careful consideration of the contracts and constraints. Traditional assumptions, such as unobserved income, are usually not enough to produce traditional solutions. This paper, however, attempts to go beyond that negative result by clarifying when the new possibilities and restrictions will apply. For example, contrived uncertainty has a rather limited scope in models of this, ‘general class, especially when the primary good, in this case income, is exchange- able. In some fields, however, it will be most plausible to posit non- exchangeability; for example, in labor contracts. Prescott and Townsend (1984a) and Stiglitz (1982) have had success in this area. More work can help delineate the circumstances under which lotteries will form a significant part of the institu- tional make-up Perhaps more importantly, we have seen that multilateral incentive compati- bility constraints deserve serious consideration. These constraints dramatically limit potential allocations and institutions. Equally important, their strength comes from the basic notion of arbitrage. It seems important to proceed with this, concept on at least two fronts. A richer model may be needed to explain banks and other financial institutions, perhaps containing transactions costs, infor- mation production, or government regulation. Secondly, current and future models of contractual arrangements should be subject to the full discipline of arbitrage arguments. University of Pennsylvania, U.S.A. REFERENCES, ALLEN, F. “AL 255-260, BHATTACHARYA, S. AND D. Gate, “Preference Shocks, Liquidity, and Central Banks Policy,” in New Approaches to Monetary Economics, W. Barnett and K. Singleton, eds. (New York: Cambridge Press, 1987) CaWPRELL, T. S, AND W. A. KRacAW, “Information Production, Market Signaling and the Theory of Financial Intermediation.” Journal of Finance 35 (September 1980), 863-882. Diawonp, D. W,, "Financial Intermediation and Delegated Monitoring,” Review of Economic Studies St (July 1984), 393-414. AnD P. H. DyBviG, “Bank Runs, Deposit Insurance, and Liquidity,” Journal of Political Economy 91 (lune 1983), 401-419, GILLIGAK, T, M, SMIRLOCK AND W. MARSHALL, “Seale and Scope Economies in the Multi-Produet Banking Firm,” Journal of Monetary Economics 13 (March 1984), 393-406. GRossMAN, S.J. ANDO. Hakr, “Implicit Contracts, Moral Hazard and Unemployment,” American Economic Review papers and proceedings 71 (May 1981), 301-307. ieving the First Best in Small Economi Journal of Public Economics 27 (July 1985),OPTIMAL FINANCIAL STRUCTURES 235 Gueswenie, R. AND J. LaFFONT, "A Complete Solution to a Class of Principal-Agent Problems with fan Application to the Control of a Self-Managed Firms,” Journal of Public Economics 25 (December 1984), 329-369. HamaonD, P. J.,"Multilateral Incentive Compatibility in Continuum Economies,” IMSSS Technical Report No. 435, Stanford University (November 1983), “Markets as Constraints: Multilateral Incentive Compatibility in Continuum Econom Review of Economie Studies 54 (July 1987), 399-412. Hargis, M. AND R, M. TowNSEND, "Resource Allocat ‘metrica 49 January 1981), 33-64 Havanici, J. G., aND RG, Kin, “Banking and Insurance, coming). Houssteom, B., “Moral Hazard in Teams,” Bell Journal of Economics 13 (Autumn 1982), 324-340, ———, anp L. Weiss, “Managerial Incentives, Investment and Aggregate Implications: Scale Ef- fects,” Review of Economic Studies 52 (Suly 1985), 403-426, IACKLIN, C. J, "Demand Deposits, Trading Restrictions, and Risk Sharing,” in Contractual Arrange- ‘ments for Intertemporal Trade E. Prescott and N. Wallace, eds. (Minneapolis: University of Minnesota Press, 1987). Journal of Business, “Symposium on Bank Regulation,” $9 (January 1986), 1-110. Jupp, K. L, “The Law of Large Numbers with a Continuum of IID Random Variables,” Journal of Economic Theory 35 (February 1985), 19-25, Kaneko, M, AND M. H. Wovens, “The Core of a Game with a Continuum of Players and Finite Coalitions: The Model and Some Results," University of Toronto (August 1984) Lursey, R. G. Ano K. Lancaster, “The General Theory of the Second Best,” Review of Economic Studies 24 (1956-1957), 11-32. Lucas, R. E. JR, “Asset Prices in an Exchange Economy,” Econometrica 16 (November 1978), 1429-1446. Myerson, R. B, “Incentive Compatibility and the Bargaining Problem,” Econometrica 47 (January 1979}, 61-74 Prescorr, E.C. AND R. M. Townsenp, “General Competitive Analysis in an Economy with Private Information,” International Economic Review 25 (February 1984), 1-20. ——~ anv ———, “Pareto Optima and Competitive Equilibria with Adverse Selection and Moral Hazard,” Econometrica 32 (January 1984), 21-46, Roruscuito, M. axb J. E, SriGuitz, “Increasing Risk: |. A Definition,” Journal of Economic Theory 2 (March 1970), 225-243, ‘Suivi, Bruce D., “Private Information, Deposit Interest Rates, and the ‘Stability’ of the Banking System,” Journal of Monetary Economics 13 (November 1984), 203-317. Snauitz, J. £, "SelfSelection and Pareto Efficient Taxation,” Journal of Public Economics 17 (March 1982), 213-240. Tost, J. aND W. BRAIvano, “Financial Intermediaries and the Effectiveness of Monetary Controls,” American Economic Review 53 (May 1963), 384-386. Townstnp, R. M., “Economic Organization with Limited Communication,” American Economic Review 77 (December 1987), 954-971 n-under Asymmetric Information,” Econo- Journal of Monetary Economics (forth-