13 where an agent evaluates his own project,and then issues securities to investors that promise specified returns.Or,a coalition of agents can offer investors a claim on group returns.Financial intermediaries are coalitions of agents that evaluate projects,invest in those determined to be high-value projects,and share the returns from the portfolio of projects. More specifically,the outline of the model is as follows.Agents live for two periods.Each agent is endowed with a project of unknown type (good or bad).Agents know their own type,so there is no opportunity to enter into contracts before knowing their types.Each project type can have a high or low return (good projects are more likely to realize the high return).An agent can expend his endowment either on producing information about a single project's type or as an investment in a single project,his own if he has not evaluated it or another agent's project.If a project is evaluated,then a noisy signal of true project type is received.Project evaluation and investment are publicly observable and verifiable,as are project returns,evaluation results,consumption outcomes,and contract terms. An efficient outcome invests in as many good projects as possible.But,the difficulty in accomplishing this is that bad-type agents will want to mimic good-type agents,claiming that they are good,promising the same high return to investors as the good-type agents,and then hoping that their project realizes the high return.Indeed,there is such a securities market equilibrium,but it is one in which some bad-type projects are evaluated,by mimicking agents.This is inefficient. The alternative is the financial intermediary coalition.The model is one of mechanism design. One interpretation of how to implement the equilibrium with the coalition(given by Boyd and Prescott)is as follows.Coalition members deliver their endowments to the coalition prior to investment.These endowments are used for project evaluation.Depositors are other agents who turn over their endowments to the coalition in exchange for a promised amount of consumption.The depositors give the coalition the right to invest in their project and to receive the entire project output,if the coalition desires.Project owners are promised very high returns if evaluation reveals a good project and if the realized return is high.Otherwise,depositors are promised an amount of consumption which is more than a bad-type agent could achieve on his own,but less than the promised amount for projects with a good evaluation and high realized returns.Members of the coalition are residual claimants and share profits equally. The coalition's sharing rules induce truthful revelation of agent type.The coalition then evaluates good-type projects and funds each of these projects with a good evaluation.It uses the remaining proceeds to fund bad-type projects without evaluation.This is the critical point.The promised returns separate types,and since good types are relatively scarce,the coalition ends up funding some bad- type projects,but it does not waste resources evaluating those projects.This is why it dominates the securities market
13 where an agent evaluates his own project, and then issues securities to investors that promise specified returns. Or, a coalition of agents can offer investors a claim on group returns. Financial intermediaries are coalitions of agents that evaluate projects, invest in those determined to be high-value projects, and share the returns from the portfolio of projects. More specifically, the outline of the model is as follows. Agents live for two periods. Each agent is endowed with a project of unknown type (good or bad). Agents know their own type, so there is no opportunity to enter into contracts before knowing their types. Each project type can have a high or low return (good projects are more likely to realize the high return). An agent can expend his endowment either on producing information about a single project’s type or as an investment in a single project, his own if he has not evaluated it or another agent’s project. If a project is evaluated, then a noisy signal of true project type is received. Project evaluation and investment are publicly observable and verifiable, as are project returns, evaluation results, consumption outcomes, and contract terms. An efficient outcome invests in as many good projects as possible. But, the difficulty in accomplishing this is that bad-type agents will want to mimic good-type agents, claiming that they are good, promising the same high return to investors as the good-type agents, and then hoping that their project realizes the high return. Indeed, there is such a securities market equilibrium, but it is one in which some bad-type projects are evaluated, by mimicking agents. This is inefficient. The alternative is the financial intermediary coalition. The model is one of mechanism design. One interpretation of how to implement the equilibrium with the coalition (given by Boyd and Prescott) is as follows. Coalition members deliver their endowments to the coalition prior to investment. These endowments are used for project evaluation. Depositors are other agents who turn over their endowments to the coalition in exchange for a promised amount of consumption. The depositors give the coalition the right to invest in their project and to receive the entire project output, if the coalition desires. Project owners are promised very high returns if evaluation reveals a good project and if the realized return is high. Otherwise, depositors are promised an amount of consumption which is more than a bad-type agent could achieve on his own, but less than the promised amount for projects with a good evaluation and high realized returns. Members of the coalition are residual claimants and share profits equally. The coalition’s sharing rules induce truthful revelation of agent type. The coalition then evaluates good-type projects and funds each of these projects with a good evaluation. It uses the remaining proceeds to fund bad-type projects without evaluation. This is the critical point. The promised returns separate types, and since good types are relatively scarce, the coalition ends up funding some badtype projects, but it does not waste resources evaluating those projects. This is why it dominates the securities market
14 The intermediary dominates the securities market because the intermediary coalition can induce agents to truthfully reveal their type and this cannot be achieved in the securities market.Truthful revelation allows the coalition to avoid inefficiently evaluating some bad-type projects.The reason is that,by conditioning returns on the coalition's portfolio returns,rather than on the returns of a single project,the coalition can offer higher returns to bad-type agents,so they will participate in the coalition. The relative proportions of good-types and bad-types are also important.In particular,good-type agents must be scarce.Note also that it is important that a coalition be large because a small coalition may end up with so many good-type projects that they cannot all be funded.In the population,good-type projects are relatively scarce and this must be reflected in coalition membership.Thus,as in Diamond(1984),size of the coalition is critical for the argument. The equilibrium concept in Boyd and Prescott is based on the core of an economy.That is,an allocation is an equilibrium if no large coalition of agents,with specified fractions of agent types,can achieve a different allocation,satisfying resource,consumption,incentive and other constraints,and make at least some agent type better off without reducing any other type's utility.Deviating coalitions are not allowed to attract higher than population proportions of type-i agents unless it makes them strictly better off.Although the solution of the model is standard in that it relies on the revelation principal,the equilibrium concept is less common in the finance and financial contracting literature.This may account for why this paper has not led to a successor literature in banking per se;instead,it has been more influential in macroeconomics,where the equilibrium concept has been taken up,though see the discussion of Williamson(1988),below. Boyd and Prescott's intermediary has the characteristics of bank-like intermediaries identified in the introduction.Other researchers have pursued solutions to the problems of reliability and appropriability of valuable private information,but these other solutions do not involve bank-like intermediation.Two settings in particular have been examined.The first considers delegated portfolio management,i.e.,a setting where a fund manager may claim to have superior information or superior ability and offers to invest on behalf of investors.The second considers the sale of valuable information about investments when the information producer does not invest on behalf of investors.A theory of intermediation must distinguish between firms that sell information,like rating agencies,firms that are delegated portfolio managers,like mutual funds or hedge funds,and bank-like financial intermediaries. At the level of casual empiricism there are identifiable differences between these types of arrangements.A bank-like intermediary does not sell information that it produces.Rather,as in Boyd and Prescott,it uses the information internally to improve the returns to coalition members.This is very different from the case of a firm that sells information to investors,like a rating agency.Firms selling information face problems of reliability and appropriability,but they do not lend money.Purchasers of
14 The intermediary dominates the securities market because the intermediary coalition can induce agents to truthfully reveal their type and this cannot be achieved in the securities market. Truthful revelation allows the coalition to avoid inefficiently evaluating some bad-type projects. The reason is that, by conditioning returns on the coalition’s portfolio returns, rather than on the returns of a single project, the coalition can offer higher returns to bad-type agents, so they will participate in the coalition. The relative proportions of good-types and bad-types are also important. In particular, good-type agents must be scarce. Note also that it is important that a coalition be large because a small coalition may end up with so many good-type projects that they cannot all be funded. In the population, good-type projects are relatively scarce and this must be reflected in coalition membership. Thus, as in Diamond (1984), size of the coalition is critical for the argument. The equilibrium concept in Boyd and Prescott is based on the core of an economy. That is, an allocation is an equilibrium if no large coalition of agents, with specified fractions of agent types, can achieve a different allocation, satisfying resource, consumption, incentive and other constraints, and make at least some agent type better off without reducing any other type’s utility. Deviating coalitions are not allowed to attract higher than population proportions of type-i agents unless it makes them strictly better off. Although the solution of the model is standard in that it relies on the revelation principal, the equilibrium concept is less common in the finance and financial contracting literature. This may account for why this paper has not led to a successor literature in banking per se; instead, it has been more influential in macroeconomics, where the equilibrium concept has been taken up, though see the discussion of Williamson (1988), below. Boyd and Prescott’s intermediary has the characteristics of bank-like intermediaries identified in the introduction. Other researchers have pursued solutions to the problems of reliability and appropriability of valuable private information, but these other solutions do not involve bank-like intermediation. Two settings in particular have been examined. The first considers delegated portfolio management, i.e., a setting where a fund manager may claim to have superior information or superior ability and offers to invest on behalf of investors. The second considers the sale of valuable information about investments when the information producer does not invest on behalf of investors. A theory of intermediation must distinguish between firms that sell information, like rating agencies, firms that are delegated portfolio managers, like mutual funds or hedge funds, and bank-like financial intermediaries. At the level of casual empiricism there are identifiable differences between these types of arrangements. A bank-like intermediary does not sell information that it produces. Rather, as in Boyd and Prescott, it uses the information internally to improve the returns to coalition members. This is very different from the case of a firm that sells information to investors, like a rating agency. Firms selling information face problems of reliability and appropriability, but they do not lend money. Purchasers of
15 the information may lend in reliance on the information purchased,but they are then directly lending,not via an intermediary.A portfolio manager,claiming to have superior information,accepts investments from one set of agents and then uses the proceeds to invest in securities.This seems very similar to a bank-like intermediary.One difference is that the claims held by the investors do not have different state contingent payoffs than the payoff on the portfolio of claims chosen by the portfolio manager,essentially, the investors and the portfolio manager all hold equity claims in the portfolio. In Bhattacharya and Pfleiderer's(1985)model,investors want to hire portfolio managers,but there are two sources of private information that make this difficult.First,investors must hire a manager from pool of managers with heterogeneous abilities.A manager or agent has the ability to receive an informative signal about the risky asset(there is also a riskless asset).Once a manager has been hired,he must be induced to truthfully reveal the signal he has received.However,once the principal has designed the contract and hired a manager,the manager/agent's only role is to transmit the information to the principal.There is no portfolio management by the manager/agent since the principal can directly invest using the information supplied by the manager/agent.There is no intermediary(nor do Bhattacharya and Pfleiderer claim that there is;the purposes of their paper are different). Allen (1990)presents a model that distinguishes conditions under which information is sold to agents who then use the information to make investments from the case where the buyers of the information then act as intermediaries and resell the information.Essentially,reselling the information allows more of the value of the information to be captured.Because the initial information seller must distinguish himself from potential uniformed mimics,he faces a number of constraints.These constraints limit the amount of profit he can take in from selling the information.This is the basis for information resellers to enter the market;they find it profitable to resell the information rather than use it as a basis for their own investments because they can capture more of the value of the information.Here there is a type of intermediation:there are agents who buy information and then resell it.But,these agents do not invest on behalf of others. Ramakrishnan and Thakor(1984)consider a setting in which firms issuing new shares to the public can hire an agent to produce information about their quality.Information production requires a costly,and unobservable,effort,so the information producer would like to avoid this cost if he can do so without being detected.There is an ex post noisy indicator of the information producer's effort choice,so compensation for information production can be linked to this indicator.Because information producers are risk averse,they would prefer to avoid the risk that the noise in the indicator prevents them from obtaining compensation for their efforts.The main point of Ramakrishnan and Thakor is that this risk is mitigated if one infinitely large intermediary is formed since this diversifies the risk associated with the effort indicator.The large intermediary is formed when information producers can costlessly monitor
15 the information may lend in reliance on the information purchased, but they are then directly lending, not via an intermediary. A portfolio manager, claiming to have superior information, accepts investments from one set of agents and then uses the proceeds to invest in securities. This seems very similar to a bank-like intermediary. One difference is that the claims held by the investors do not have different state contingent payoffs than the payoff on the portfolio of claims chosen by the portfolio manager; essentially, the investors and the portfolio manager all hold equity claims in the portfolio. In Bhattacharya and Pfleiderer’s (1985) model, investors want to hire portfolio managers, but there are two sources of private information that make this difficult. First, investors must hire a manager from pool of managers with heterogeneous abilities. A manager or agent has the ability to receive an informative signal about the risky asset (there is also a riskless asset). Once a manager has been hired, he must be induced to truthfully reveal the signal he has received. However, once the principal has designed the contract and hired a manager, the manager/agent’s only role is to transmit the information to the principal. There is no portfolio management by the manager/agent since the principal can directly invest using the information supplied by the manager/agent. There is no intermediary (nor do Bhattacharya and Pfleiderer claim that there is; the purposes of their paper are different). Allen (1990) presents a model that distinguishes conditions under which information is sold to agents who then use the information to make investments from the case where the buyers of the information then act as intermediaries and resell the information. Essentially, reselling the information allows more of the value of the information to be captured. Because the initial information seller must distinguish himself from potential uniformed mimics, he faces a number of constraints. These constraints limit the amount of profit he can take in from selling the information. This is the basis for information resellers to enter the market; they find it profitable to resell the information rather than use it as a basis for their own investments because they can capture more of the value of the information. Here there is a type of intermediation: there are agents who buy information and then resell it. But, these agents do not invest on behalf of others. Ramakrishnan and Thakor (1984) consider a setting in which firms issuing new shares to the public can hire an agent to produce information about their quality. Information production requires a costly, and unobservable, effort, so the information producer would like to avoid this cost if he can do so without being detected. There is an ex post noisy indicator of the information producer’s effort choice, so compensation for information production can be linked to this indicator. Because information producers are risk averse, they would prefer to avoid the risk that the noise in the indicator prevents them from obtaining compensation for their efforts. The main point of Ramakrishnan and Thakor is that this risk is mitigated if one infinitely large intermediary is formed since this diversifies the risk associated with the effort indicator. The large intermediary is formed when information producers can costlessly monitor
16 each other's efforts.(Millon and Thakor (1985)extend the analysis to the case where the internal monitoring is costly.)Ramakrishnan and Thakor's intermediary,however,does not accept funds for investment.Rather,it is a pure information seller.In this regard,also see Lizzeri(1999) In general,the differences in settings where some agents would like valuable,but costly, information produced for investment purposes are subtle.In many models there is no need for the information seller to actually accept the funds that will be invested on the basis of the superior information.In Bhattacharya and Pfleiderer,Allen,and Ramakrishnan and Thakor,the information producer sells the information to investors,but does not need to actually invest the funds of the investors. In Boyd and Prescott the intermediary accepts deposits,produces information,and invests in projects based on the information produced.Only by conditioning the returns on the portfolio that is produced by the coalition can truthful revelation be induced. A potentially important aspect of information production by banks concerns whether the information is produced upon first contact with the borrower or is instead learned through repeated interaction with the borrower over time.Another strand of the literature on banks as information producers argues that banks acquire (private)information over time through repeatedly lending to a borrower.The acquisition of this private information over time is known as a"customer relationship" and is discussed in Section III below. D.Banks as Consumption Smoothers Bryant (1980)and Diamond and Dybvig (1983)develop a role for bank liabilities,without stressing any particular features of bank assets.Bank liabilities do not function as a transactions medium. Rather,banks are vehicles for consumption smoothing;they offer insurance against shocks to a consumer's consumption path. The Diamond and Dybvig model assumes that the payoffs from the available investment opportunities are inconsistent with the possible consumption paths desired by consumers.In particular, consumers have random consumption needs,and satisfying these needs may require them to prematurely end investments unless they save via intermediation so that they can to some extent diversify these consumption shocks.The model offers a view of the liability side of banking;the right to withdraw from the bank,prematurely ending investment in order to satisfy sudden consumption needs,corresponds with notions of how demand deposits actually work.The model also focuses on banking panics,a separate topic that we discuss in Section IV below. The outlines of the Diamond and Dybvig model are as follows.There are three dates 0,1,and 2 and a single good.The available technology allows one unit of investment to be transformed over two periods into R>I units at the final date.If this investment is interrupted at the interim date,then it just
16 each other’s efforts. (Millon and Thakor (1985) extend the analysis to the case where the internal monitoring is costly.) Ramakrishnan and Thakor’s intermediary, however, does not accept funds for investment. Rather, it is a pure information seller. In this regard, also see Lizzeri (1999). In general, the differences in settings where some agents would like valuable, but costly, information produced for investment purposes are subtle. In many models there is no need for the information seller to actually accept the funds that will be invested on the basis of the superior information. In Bhattacharya and Pfleiderer, Allen, and Ramakrishnan and Thakor, the information producer sells the information to investors, but does not need to actually invest the funds of the investors. In Boyd and Prescott the intermediary accepts deposits, produces information, and invests in projects based on the information produced. Only by conditioning the returns on the portfolio that is produced by the coalition can truthful revelation be induced. A potentially important aspect of information production by banks concerns whether the information is produced upon first contact with the borrower or is instead learned through repeated interaction with the borrower over time. Another strand of the literature on banks as information producers argues that banks acquire (private) information over time through repeatedly lending to a borrower. The acquisition of this private information over time is known as a “customer relationship” and is discussed in Section III below. D. Banks as Consumption Smoothers Bryant (1980) and Diamond and Dybvig (1983) develop a role for bank liabilities, without stressing any particular features of bank assets. Bank liabilities do not function as a transactions medium. Rather, banks are vehicles for consumption smoothing; they offer insurance against shocks to a consumer’s consumption path. The Diamond and Dybvig model assumes that the payoffs from the available investment opportunities are inconsistent with the possible consumption paths desired by consumers. In particular, consumers have random consumption needs, and satisfying these needs may require them to prematurely end investments unless they save via intermediation so that they can to some extent diversify these consumption shocks. The model offers a view of the liability side of banking; the right to withdraw from the bank, prematurely ending investment in order to satisfy sudden consumption needs, corresponds with notions of how demand deposits actually work. The model also focuses on banking panics, a separate topic that we discuss in Section IV below. The outlines of the Diamond and Dybvig model are as follows. There are three dates 0, 1, and 2 and a single good. The available technology allows one unit of investment to be transformed over two periods into R>1 units at the final date. If this investment is interrupted at the interim date, then it just
17 returns the initial one unit.Importantly,the long-term investment only realizes a return over the initial investment if it reaches fruition at date 2.All consumers are identical initially,at date 0,but each faces a privately observable,uninsurable risk with regard to their preferences.At date 1,each consumer learns whether he cares only about consumption at date 1,an "early consumer,"or only about consumption at date 2,a "late consumer."The problem is evident:consumers would like to insure themselves against the bad luck of being an early consumer.Without being able to write such insurance contracts,because consumer type is not observable,early consumers can do no better than consuming their single unit of endowment,which was invested in the investment technology but which is liquidated early.The lucky late consumers consume R>1. Diamond and Dybvig(1983)argue that a bank can provide insurance against the risk of being an early consumer.Basically,a bank works as follows.At date 0 the bank opens and accepts"deposits"of endowment.The bank promises a fixed claim of r per unit deposited will be paid out to consumers who withdraw at date 1.The return on a deposit that is not withdrawn at date 1,but is withdrawn at date 2, depends on how much was withdrawn at date 1.Suppose the fraction of consumers who will turn out to be early consumers is fixed and known.Then Diamond and Dybvig show that the return of r can be set to the amount that an early consumer would achieve if there were complete insurance markets.So the bank can support the full-information risk-sharing equilibrium. The Diamond and Dybvig model has important features of intermediaries and the real world environment.First,it incorporates the idea that consumers have uncertain preferences for expenditure streams,producing a demand for liquid assets.Furthermore,the modeling representation of this uncertainty,the technique of early and late consumers,has been very influential in its own right. Uncertainty about preferences for expenditure streams leads to the bank offering claims that look like demand deposits.This is combined with a second important feature,namely,real investment projects are irreversible,or at least costly to restart once stopped.A third important feature of the model is the idea that individual consumers have private information about the realization of their type,the realization of their preferred consumption stream.There is no credible way to truthfully reveal this information. We now turn to some details about why insurance or securities markets cannot provide consumption smoothing or insurance against the risk of uncertain preferences for expenditure streams.In Diamond and Dybvig,an intermediary that issues demand deposits allows greater risk sharing than autarky.Diamond and Dybvig assume that demand deposits cannot be traded and do not consider other securities markets.Their model assumes a sequential service constraint,that is,a first-come-first-served ruled under which at date 1 the bank honors claims to withdraw in the order in which they are received until the bank runs out of resources to honor the claims.The remaining consumers seeking to withdraw receive nothing and the bank fails.We discuss models that motivate the sequential service constraint in
17 returns the initial one unit. Importantly, the long-term investment only realizes a return over the initial investment if it reaches fruition at date 2. All consumers are identical initially, at date 0, but each faces a privately observable, uninsurable risk with regard to their preferences. At date 1, each consumer learns whether he cares only about consumption at date 1, an “early consumer,” or only about consumption at date 2, a “late consumer.” The problem is evident: consumers would like to insure themselves against the bad luck of being an early consumer. Without being able to write such insurance contracts, because consumer type is not observable, early consumers can do no better than consuming their single unit of endowment, which was invested in the investment technology but which is liquidated early. The lucky late consumers consume R>1. Diamond and Dybvig (1983) argue that a bank can provide insurance against the risk of being an early consumer. Basically, a bank works as follows. At date 0 the bank opens and accepts “deposits” of endowment. The bank promises a fixed claim of r1 per unit deposited will be paid out to consumers who withdraw at date 1. The return on a deposit that is not withdrawn at date 1, but is withdrawn at date 2, depends on how much was withdrawn at date 1. Suppose the fraction of consumers who will turn out to be early consumers is fixed and known. Then Diamond and Dybvig show that the return of r1 can be set to the amount that an early consumer would achieve if there were complete insurance markets. So the bank can support the full-information risk-sharing equilibrium. The Diamond and Dybvig model has important features of intermediaries and the real world environment. First, it incorporates the idea that consumers have uncertain preferences for expenditure streams, producing a demand for liquid assets. Furthermore, the modeling representation of this uncertainty, the technique of early and late consumers, has been very influential in its own right. Uncertainty about preferences for expenditure streams leads to the bank offering claims that look like demand deposits. This is combined with a second important feature, namely, real investment projects are irreversible, or at least costly to restart once stopped. A third important feature of the model is the idea that individual consumers have private information about the realization of their type, the realization of their preferred consumption stream. There is no credible way to truthfully reveal this information. We now turn to some details about why insurance or securities markets cannot provide consumption smoothing or insurance against the risk of uncertain preferences for expenditure streams. In Diamond and Dybvig, an intermediary that issues demand deposits allows greater risk sharing than autarky. Diamond and Dybvig assume that demand deposits cannot be traded and do not consider other securities markets. Their model assumes a sequential service constraint, that is, a first-come-first-served ruled under which at date 1 the bank honors claims to withdraw in the order in which they are received until the bank runs out of resources to honor the claims. The remaining consumers seeking to withdraw receive nothing and the bank fails. We discuss models that motivate the sequential service constraint in