A principal-agent model of corruption

One of the new avenues in the study of political corruption is that of neo-institutional economics, of which the principal-agent theory is a part. In this article a principal-agent model of corruption is presented, in which there are two principals (one of which is corrupting), and one agent (who is corrupted). The behaviour of these principals and agent is analysed in terms of the costs and benefits associated with different actions. The model is applied to political corruption in representative democracies, showing that, contrary to common belief, the use of principal-agent models is not limited to bureaucratic corruption. Introduction: Neo-institutional economics and corruption Although corruption is generally analysed as a social phenomenon, especially in comparative research, it is from the decisions of individuals that corruption stems. In this article a principal-agent model of corruption is presented, one which focuses on these decisions. Our choice of individualism is a methodological one, and must not be understood as a denial of the significance of the social structures of which individuals are a part. Indeed, neo-institutional economics, of which the principal-agent literature is a part, deals explicitly with the institutional factors that affect the decisions of individuals. What can neo-institutional economics contribute to our understanding of corruption? To answer that question let us first give a brief overview of neo-institutional economics in general.1 Institutions can be defined as collectively binding working rules.2 A contract between two persons is such a set of rules, as are school regulations. Proportional representation is an institution, just as the stock market and marriage are. Institutions are distinguished from the preferences of individuals; they structure and constrain the decisions made by them. These institutional constraints are emphasised in various ways, leading to three different, but related branches within the field of neo-institutional economics: the property rights branch, the transaction costs branch, and the principal-agent branch.3 In the property rights literature decisions by individuals concerning goods are assumed to be affected not so much by the goods themselves, but by the kind of property rights individuals can exercise. These property rights can be user rights, usufructuary rights and transfer rights. An individual who owns 208 NICO GROENENDIJK a house will make different decisions about maintenance than an individual who is a tenant. In the property rights literature institutions constitute property rights of individuals, as well as the means for these individuals to defend themselves against infringements of these rights. Stemming from the work of Coase, the transaction costs branch emphasises the importance of transaction costs for the allocation of resources and the structure of economic organization.4 Prior to Coase, the role of transaction costs was minimal since such costs are for the greater part information costs, and most economic theories and models assumed full information. Following Williamson, transaction costs can be divided into ex ante and ex post costs. Ex ante costs are the costs of drafting, negotiating and safeguarding an agreement.5 The costs of monitoring and controlling the execution of an agreement are ex post costs, as are possible revision costs. Transaction costs will differ depending on the incidence of the transactions, the degree of uncertainty that the individuals face, and the “asset specificity”, e.g. the extent to which the good and the transaction concerned are geared to one another. It should be mentioned that what Williamson calls an “agreement” must be understood to include not only contracts, but other more extensive institutions as well. In the principal-agent branch of neo-institutional economics (or: agency theory) elements of both the property rights theory and the transaction costs theory are integrated and used to model relationships between principals and agents. Principal-agent models assume that the interests of principal and agent diverge, that there is informational asymmetry to the advantage of the agent, but that the principal can prescribe the pay-off rules in their relationship. Furthermore, an important feature of the agency literature is that decisions are considered to be trade-offs between different kinds of costs (monitoring costs, bonding costs and residual loss). What contribution has neo-institutional economics made, to date, to the study of corruption? The most comprehensive economic models of corruption are those of Rose-Ackerman, Lui, Andvig and Moene, Klitgaard, and Alam.6 Rose-Ackerman focuses on a traditional form of corruption, namely, bribery of bureaucrats by companies in the government contracting process.7 The incidence of corruption is found to differ according to the degree of competition and the extent to which government has well-defined preferences for the good in question. Her analysis pinpoints the areas in which resources for anti-corruption measures can best be used. In later work, she extends her analysis to other types of corruption, namely, high-level political corruption as well as low-level bureaucratic corruption.8 Lui assumes that corruption is more difficult to detect and deter when a greater proportion of the officials is corrupt.9 Using a two-period model with overlapping generations of officials, A PRINCIPAL-AGENT MODEL OF CORRUPTION 209 Lui shows that there can be several stationary equilibrium levels of corruption in an economy. Andvig and Moene offer a fine model of bureaucratic corruption, focusing on the demand as well as the supply side.10 Their model also has multiple equilibria, which sheds light on the varying incidence of corruption across similar societies and activities. Klitgaard offers a largely heuristic principal-agent model, which encompasses a principal, a corrupt agent, and a client.11 He offers an overview of various measures to fight corruption as well. Alam explains variations in corruption across agencies and societies by the ability of the victims of corruption to engage in countervailing actions.12 These countervailing actions can be evasive, direct or illicit. The incentives to engage in countervailing activities depend on two kinds of factors, global factors and specific factors. Global factors like property rights explain variation across societies; specific factors such as the type of corruption explain variation across agencies. Elements of neo-institutional economics have been used by Bakker and Schulte Nordholt, to elucidate the mechanisms by which corruption can take root in societies, especially when societies are going through institutional turnovers.13 Similar findings have been put forward by Johnson and Huntington, with regard to institutional discrepancies within a society.14 To the possibility of such a horizontal institutional gap, Bakker and Schulte Nordholt add the possibility of a vertical institutional gap, which arises when national economic, political and legal institutions are not equipped to deal with internationalization. Other authors have combined neo-institutional economics with game theory. Manion analyses bribery in Chinese enterprise licencing, assuming that those who offer the bribes are fully entitled to their licenses and do not know whether the bribes are overpayments or fees for valued illicit services.15 Bribery can be shown to be an equilibrium solution in such an asymmetric information game. A similar analysis is that of Cadot who focuses on the situation in which an applicant has to pass a test in order to get a permit.16 Different solutions exist for this game, depending on the information sets of both players. Basu, Bhattacharya and Mishra, using game theory as well as the economic theory of crime, based on the work of Becker, offer a simple model of recursive corruption.17 Recursive corruption occurs when a corrupt official gets caught and bribes the official that caught him. Related work is that of Mookherjee and Png who analyse the delegation of enforcement authority in environmental policy, and the opportunities such delegation brings about for corruption.18 Focusing on political corruption, Rasmusen and Ramseyer have combined game theory with the theory of “rent-seeking”, which is related to neo-institutional economics.19 They note that the size of bribes in political 210 NICO GROENENDIJK corruption is small, and put that down to the imperfect coordination between legislators. When a legislator expects other legislators to sell their votes, and there is no coordination, he will take whatever bribe is offered. Because this is hardly satisfactory, legislators may opt for a ban on bribery. Myerson uses game theory to analyse the effectiveness of different electoral systems in reducing corruption in a multiparty system.20 Galeotti and Merlo, who also investigate the relationship between electoral rules and government corruption, have challenged Myerson’s conclusion that proportional representation guarantees the exclusion from government of corrupt parties.21 Looking at the contributions from neo-institutional economics we find that, apart from Klitgaard, and to some extent Rose-Ackerman, the principal-agent theory has not been fully used to enhance our understanding of corruption. This article aims to fill that gap, and to do so in two ways. First, unlike Klitgaard and Rose-Ackerman, who assume only one principal-agent relation namely the relation between a corrupt official and his superior, we will look upon the relation between the “client” and the corrupt official as a principalagent relation as well. Second, the notion of trading-off different costs is taken into account more explicitly. Johnston remarks that while the principal-agent approach is well-suited to the analysis of bureaucratic corruption, other more broadly-based forms of corruption may fit the framework awkwardly or not at all.22 Moreover, he seems to suggest that the application of the principal-agent approach to corruption, because of its supposed bias towards bureaucratic corruption, takes politics out of our notion of corruption. In this article we will put this proposition to the test by applying the principal-agent framework to political corruption in representative democracies. In the next section, a basic principal-agent model will be introduced. This basic model demonstrates how non-co-operative behaviour by an agent can occur, and which factors, in terms of different costs, condition such behaviour. In the third section the basic model will be modified. First of all, an appropriate definition of corruption is selected. In this article corruption is defined as any unauthorized transaction between agents and a third party. In the case of bureaucratic corruption the agents are bureaucrats, and the principals are elected or appointed officials. In the case of political corruption the latter are the agents, and the principals are the voters. The third party is considered to be a second principal to the agent. Given our definition of corruption the basic principal-agent model is amended and turned into a principals-agent model of corruption. In the fourth section this model is applied in a tentative way to political corruption.The fifth section, finally, contains some conclusions. A PRINCIPAL-AGENT MODEL OF CORRUPTION 211 Modelling principal-agent relations A basic principal-agent model Let P be a principal, and A an agent. Assume that A can choose between two alternative actions x and y, and that these actions contribute to the realization of two different outcomes, X and Y. In choosing between different actions A is supposed to represent P’s interests in return for payment of some kind. Assume that P is unable – costlessly – to observe A’s actions, so that P has only the outcomes by which to assess and to reward A’s behaviour. If these outcomes were fully determined by A’s actions, P could base his rewarding scheme on these outcomes, and the lack of information about A’s actions would not be relevant. The outcomes, however, are not determined by A’s actions, they are only affected by them. Assume that P’s interests do not match A’s interests, as can be shown by their utility functions, which have different arguments: UP = U(a; b; c) (1.1) UA = U(d; e; f) (1.2) Assume also that P wants to maximise his own welfare, and that the utilities associated with the two possible outcomes X and Y are UP and UP Y . Suppose that UP > UP Y , so that P will prefer outcome X. If UA and UA are the utilities for A that are involved with the possible actions x and y, and if we assume welfare-maximisation by A, A will choose action x if UA >UA , and action y if UA UA , in which case A will choose x+. Of course, P could also 212 NICO GROENENDIJK turn y into a less attractive alternative, y-, hoping that UA > UA . This use of incentives or disincentives can be seen as a refinement of the original rewarding scheme. Note that the use of these positive and negative incentives does not change A’s utility function in any way. Note also that this use of incentives, and the intended revision of A’s choice, does not guarantee that the actual outcome will be X. Second, P can use persuasion. The difference between this measure and the use of incentives, is that A’s utility function is meant to converge, from UA to UA, in the direction of UP . A possible new utility function for A is then: UA = U(a; b; e; f) (1.3) which resembles (1.1) more than did (1.2). If UA > UA, A will choose action x, which makes outcome X, preferred by P, more likely. Third, assuming an hierarchical relationship between P and A, P, as A’s superior, can use directives, thus reducing the discretion A has. Supposing there are only two alternative actions, x and y, which A can take, the use of directives would mean that one action is prescribed (x) and the other (y) is not allowed.23 This increases the probability that the appropriate action (i.e. x) is chosen, but P can never be sure of that. More in general, the use of directives is aimed at reducing the set of alternative actions for A, whereas the use of incentives can be seen as changing the characteristics of the alternatives, and the use of persuasion as changing the criteria by which the alternatives are assessed. As mentioned before, without any monitoring, P has only the outcomes X and Y to evaluate A’s behaviour. The use of measures controlling that behaviour requires more information on P’s part. Thus, there are two different costs for P associated with monitoring and controlling A’s behaviour: inspection costs (or: appraisal costs) and prevention costs.24 What if the actual outcome is X? Can P rest for a while? No, there is still the possibility that A has been non-co-operative by choosing action y (the fourth possibility), although he has been paid by P to act in P’s interests. Even though P does not suffer a direct welfare loss as a result of a non-preferred outcome, there is the possibility that he has been wasting money. To prevent that state of affairs P will use the same measures as when Y is the actual outcome. So either way P will incur inspection and prevention costs. Is there any reason for A to react to all this monitoring and controlling by P? First, suppose A’s original choice was y. When P uses a negative incentive, the choice A had between x and y becomes a choice between x and y-. If UA > UA > UA the welfare loss for A associated with this negative incentive is UA-UA. However, when P uses a positive incentive, the choice A had between x and y becomes a choice A PRINCIPAL-AGENT MODEL OF CORRUPTION 213 between x+ and y, and, supposing UA > UA > UA, A’s welfare will increase from UA to UA. When P uses persuasion it is not clear whether A is worse off or not. First A chose y because UA > UA; now he chooses x because UA > UA . A’s utility level UA cannot meaningfully be compared with his utility level UA, because utility functions UA and UA differ. When P uses directives, A’s welfare loss is the same as when P uses a negative incentive: UA-UA. Now, suppose A was already co-operative; his original action was x. He will welcome the use of positive incentives, because UA>UA. He will be indifferent to the use of negative incentives, because they affect an alternative he did not choose in the first place. The use of persuasion could mean a welfare improvement if UA > UA, but then we would be comparing apples and oranges. He will be indifferent to the use of directives, finally, unless his discretion as such represents a utility to him, in which case his welfare will decrease. On the whole, the best thing for A to do is to conceal his actions in order to frustrate P’s monitoring, and at the same time to try to avoid negative incentives and directives as well as to try to lure P into the use of positive incentives. That strategy is recommendable irrespective of whether A is cooperative or not. However, if A is already co-operative, he can use that fact to conciliate P. Opposed to P’s inspection costs and prevention costs, there are A’s concealment costs and diversion costs. Comparison with the prevailing principal-agent model The principal-agent model that is presented above differs fundamentally from the model that prevails in the agency literature, although the assumptions are the same: there is divergence of interests between principal and agent, the actions of the agent do not fully determine the outcomes, and the principal cannot costlessly monitor these actions. In the prevailing model the principal is thought to engage in “monitoring”, which covers a variety of activities,25 and the agent is sometimes expected to “bond”, e.g. the agent is expected to expend resources to guarantee his principal that he will not take actions that will harm him.26 The agencyproblem is then to minimise the agency costs, which are the sum of: – the monitoring expenditures incurred by the principal; – the bonding expenditures incurred by the agent; – the residual loss, e.g. the resulting welfare loss for the principal, given a certain amount of monitoring and bonding. The main weakness of this model is that it lumps together disutilities of different individuals. The monitoring costs and the residual loss are disutilities 214 NICO GROENENDIJK of the principal. The bonding costs, however, are borne by the agent. It is sometimes argued that the bonding costs are disutilities of the principal because it is likely that the agent meets the bonding expenditures using the payment he gets from the principal in return for looking after the principal’s interests. However, whether the agent uses these resources, or he uses other resources to pay for these bonding expenditures, is not relevant: they are a disutility of the agent. A related criticism is that it is unclear whom the “agency problem” concerns: the principal or the agent? Much of the earlier agency literature focused on the normative aspects of the principal-agent relation, from the angle of the principal. That literature showed how to structure the principal-agent relation in order for the agent to make choices that will maximize the principal’s welfare given the existence of uncertainty and imperfect monitoring.27 The subsequent positive agency literature has not lived down that original imbalance, which emerges when we want to explain the agent’s behaviour. Whereas the principal is supposed to minimise the sum of his monitoring costs and his residual loss, the model does not offer such a trade-off for the agent. Is the agent only to minimise his bonding costs? Why would he bond in the first place? Bonding can only be understood as a means, among other things, of diversion by an agent who is facing the possibility of a welfare loss as a consequence of the controlling activities of his principal. That is why we have, in our basic principal-agent model, two well-matched opponents, who are frustrating each other’s plans, by inspecting and camouflaging, by controlling and diverting. One of these individuals, the principal, has the advantage that he can prescribe the pay-off rules, and that (supposing he is the agent’s superior) he can use directives to control the agent. For his part, the agent can take advantage of the fact that the principal is faced with a lack of information about the agent’s actions and with uncertainty about the link between these actions and the outcomes. That in our model principal and agent are equally matched is no repudiation of the possible hierarchical character of the relationship between principal and agent, as is shown by the possibility that the principal, as the agent’s superior, uses directives to curb the agent’s behaviour. Neither are we of the opinion that a hierarchy has to exist between principal and agent, in order to label their relationship a principal-agent relation. The question whether a hierarchy is essential for an agency relationship or not, is often confused with the question whether there is any delegation of decision-making authority.28 In our view delegation refers to alleviating one’s duties by letting others make choices for you and take actions on your behalf. Delegation refers to tasks that could well have been performed by the principal, were it not for lack of time. Just as hierarchy, delegation is not essential for agency relationships. If a car A PRINCIPAL-AGENT MODEL OF CORRUPTION 215 Figure 1. Inspection costs, prevention costs and failure costs. is brought to a garage for an overhaul, the car-owner becomes a principal and the garagist an agent, even though neither are part of some hierarchical organization, and even though there is no delegation of authority. What is essential for an agency relationship is discretion of the agent. This discretion follows from the informational asymmetry in the relation between principal and agent,29 and not from the delegation of authority.30 Trading-off different costs Turning back to our basic model, P has to bear three different costs: – inspection costs, IP ; – prevention costs, PP ; – failure costs, FP . These are the costs associated with the welfare-loss P eventually suffers (his “residual loss”). The inspection and prevention costs will vary with the amount of inspecting and preventing that takes place, as will the failure costs, as is shown in Figure 1. The total costs are the sum of inspection and prevention costs on the one hand, and failure costs on the other.31 P aims at minimizing his total costs, TCP . A also has to bear three different costs: – concealment costs, CA; – diversion costs, DA; 216 NICO GROENENDIJK – failure costs, FA. This is the “residual loss” A eventually suffers, compared to the situation in which A could have it his own way, without any monitoring and controlling by P. Just like P, A will want to minimise his total costs (TCA), which are the sum of his concealment, diversion, and failure costs.32 The course of the FP -curve in Figure 1 is affected by the amount of concealment and diversion by A. If there is much concealment and diversion, P’s inspection and prevention will be less effective than if there is only a little concealment and diversion. The effectiveness of P’s activities is reflected in his failure costs. The less effective his activities are, the higher his failure costs. Suppose A increases his concealment and diversion activities, resulting in a shift upwards of the FP -curve and thus of the TCP -curve. With increasing concealment and diversion, new optima for P are brought about, at higher total costs, and at a higher level of inspection and prevention. Of course, the increased level of inspection and prevention makes A less successful in his activities to reduce his residual loss, shifting up the FAand TCA-curves, with optima at increased levels of concealment and diversion. These increased levels of concealment and diversion make P’s activities less effective, shift up the FP and TCP -curves, and increase the level of inspection and prevention, thus making A’s concealment and diversion less successful etc. What is striking is that P and A, as a result of each others’ activities, are pushed towards these high levels of inspection, prevention, concealment and diversion. This problem is similar to that of a duopoly, and whether it is solved or not depends on additional assumptions being made about the behavior of P and A. If P is passive he chooses an as his optimal amount of inspection and prevention, regardless of the indirect effect that level of inspection and prevention has on his own failure costs. If A is also passive (the Cournot-type of duopoly), he will also make an independent choice concerning his optimal amount of concealment and diversion. If A is not passive (the Stackelbergtype of duopoly) a solution exists as long as P remains passive. If neither are passive, i.e. both P and A react to each other’s decisions (the indeterminate type of duopoly), then they are on the road to high levels of their activities, and high total costs. Just as in the case of a collective monopoly, it is imaginable that some kind of collusion takes place, in which P and A bury the hatchet, and arrangements are made that minimize the sum of P’s total costs and A’s total costs. Apart from choosing the optimal amount of inspection and prevention given the FP and IP+PP -curves, and given the level of A’s activities, P can try to shift these curves downwards. Shifting downwards the IP+PP -curve is equivalent to raising the cost-efficiency of his monitoring and controlling activities. To get the FP -curve downward, P can choose more effective monA PRINCIPAL-AGENT MODEL OF CORRUPTION 217 itoring and controlling activities. But even if P can persuade A to prefer the same outcome as he does himself, even if there is no divergence of interests between P and A whatsoever, there will still be failure costs for P, because the outcomes are not fully determined by A’s actions. And even if there is no monitoring and controlling problem at all, that is to say if P can costlessly monitor and control A’s actions, these failure costs will arise, for the same reason. Ultimately then, P will have to turn his mind to the problem of indeterminacy of the outcomes. The more vaguely and abstractly P has defined these outcomes, the less likely it is that A’s actions will determine them. And the more agents are involved with the same outcomes, the less likely it is that one agent’s actions will be determining. Corruption as a principal-agent problem A definition of corruption There are almost as many different definitions of corruption as there are contributions to the research on the phenomenon. Surveys and discussions can be found in Johnston, Holmes, Hoetjes, and Heidenheimer.33 The aim of this article is definitely not to conjure up a new definition of corruption. However, in order to present an adequate principal-agent model of corruption, we need an adequate definition of corruption. A concise definition, and therefore a good starting-point, is that of Bayley, who defines corruption as the misuse of authority, as a result of considerations of personal gain.34 There are several elements of this definition that are worth discussing. First, there is the element of considerations of personal gain. In economics the behaviour of an individual is assumed to be brought about by his looking after his own interests. What these interests are is for the individual to decide; he can be a materialist or an idealist, an egoist or an altruist, a crook or a saint, or (like most people) a little bit of all of these. From the point of view of (neo-institutional) economics the element of personal gain in a definition of corruption is a superfluous one. This is more important than it may seem at first glance. If for example, a bureaucrat accepts a bribe from a company, and in return grants a license to that company, we unmistakably have a case of corruption. Now what if the company does not offer a bribe but offers a job to the bureaucrat’s brother? And what if the bureaucrat turns down a bribe, but instead wants the company to donate to the campaign of his favourite politician, or to Greenpeace? If any of these favours in return are able to bring about a misuse of authority, they apparently signify an increase in welfare for the corrupt bureaucrat. There is no need for us to go into the nature of 218 NICO GROENENDIJK these favours, as we have to if we confine ourselves to “personal gain” as the bureaucrat’s motive. Second, there is the element of misuse of authority. As was discussed above, it is best to swap “authority” for “discretion”, which leaves us with “misuse of discretion”. What differentiates non-cooperative behaviour in general, as pictured in our basic principal-agent model, from corruption? Not the fact that the corrupt agent acts contrary to his principal’s interests, in choosing an action that the principal would not have chosen. Is it the nature of the action itself then? If an agent chooses an action which not only his principal would not give preference to, but which is unauthorized or illegal as well, then we have to do with fraud. Corruption differs from fraud, however, in two ways. In the first place, corruption in itself does not necessarily involve an unauthorized or illegal action by the agent. There is nothing illegal in granting a license to a company, nor is it unauthorized to do so for the bureaucrat involved. Fraud, however, is always about illicit actions. In the second place, corruption always involves a third party, whereas fraud is committed by a single person. It is this involvement with an accomplice that is essential for corruption. In other words, it is the unauthorized dealing with a third party that distinguishes corruption from non-co-operative behaviour in general, as well as from simple fraud. The granting of a licence (the agent’s action) may be authorized, taking a bribe in return (the transaction) is not. In our view then corruption can be defined as any unauthorized transaction between agents and a third party. This third party acts as a second principal to the corrupt agent, in which respect we differ from Klitgaard, who thinks of the third party as a client to the agent.35 Which transactions are permitted and which are not is for the first principal to decide, on moral grounds. This first principal can be an elected politician or an appointed official, in which case it is a matter of bureaucratic corruption. Political corruption refers to the situation in which the electorate acts as a principal to a politician-agent. Below we will turn our basic principal-agent model into a principals-agent model of corruption. A principals-agent model of corruption What does the possibility of corruption mean to P, A? Who will we label C? Let us concentrate on newcomer C first. C faces two different problems. First of all he has to consider whether corruption in itself pays or not. The benefits of corruption for C consist of the service rendered or the good provided by A, and/or not having to bear any deadweight costs. Against these benefits, A PRINCIPAL-AGENT MODEL OF CORRUPTION 219 there are the costs associated with corruption, such as the costs of finding a corrupt agent, of negotiating the size of the bribe, and of course of the bribe itself. Besides that, there are the costs of keeping the transaction secret, the expected value of the penalty in case C gets caught, and moral costs. Supposing that a decision is made in favour of corruption, C is confronted with a second problem: he becomes a principal to A. Suppose C, facing the possibility of a long wait at a customs post, bribes a customs officer in order to get a swift passage through.36 Just as in the previous section, there are four different possibilities: 1. C is promptly allowed to pass, partly as a result of A’s action in return for C’s bribe. This is the only possibility that would fully please C. 2. C is promptly allowed to pass, although A merely collects the bribe and does nothing to C’s advantage. 3. C has to wait like everyone else, despite A’s efforts. 4. C has to wait like everyone else, with A just pocketing the bribe. Assume that C cannot costlessly monitor A’s actions, and that these actions affect but do not determine the outcomes. What steps can we expect C to take? If we are dealing with one-off corruption, and C is allowed a swift passage through, he will probably congratulate himself on his foresight. If he has to wait, he will look upon the ineffectual bribe as a one-time item of loss. If we are dealing with recurrent corruption, however, the monitoring and controlling activities that are part of the basic principal-agent model enter the picture. Suppose C is confronted with a constant wait, despite his bribes. C is not in a position to issue directives that narrow the set of alternative actions A can choose from, so he has to rely on either changing the characteristics of the alternative actions themselves (through incentives), or changing the way the agent assesses these alternative actions (by persuasion). The use of these instruments requires monitoring by C. His inspection and prevention costs (IC and PC) are to be traded-off against his failure costs (FC). These failure costs are either the costs of a non-preferred outcome (a wait) or the cost of a redundant bribe. Of course, C’s decisions are linked. His total costs as a principal will be part of his deliberations about engaging in, and proceeding with, corruption. C’s costs and benefits are stated in Table 1. Second, let us turn to the corrupt agent, A. He faces three different problems. First, there is his principal-agent relation with P, and the minimisation of CA P , DA P and FA P .37 In the second place, there is his second principal C. In this second principal-agent relation A will use the same concealment and diversion tactics as in the first, and he will be concerned with a similar minimisation of his total costs as agent (CA C , DA C and FA C). Third, there is the decision to be corrupt or not. The costs associated with corruption 220 NICO GROENENDIJK Table 1. Costs and benefits for the corrupting principal.