Kelembagaan DAS

Louis M. Imbeau


by: Louis M. Imbeau
Département de science politique, Université Laval, Québec, G1K-7P4, Canada;

Paper presented at the annual conference of the European Public Choice Society Aarhus, 26-28 April 2003


In this paper, I propose a theory of budgeting based on the principal-agent theoretical framework. First, I provide an institutional description of the delegation structure and information asymmetry in a public university, as an example with which to substantiate the following theoretical discussion. In the section that follows this description, I explore the dynamics of a simple principal-agent relationship where both actors can play either transparency or opacity with regard to the budget information each one controls. I show that the outcome of this relationship varies according to the presence or absence of built-in incentives aimed at modifying the agent’s preferences. In the third section, I develop a more complex model where the principal delegates part of her responsibilities to a supervisor. I show that the budget outcome depends on the supervisor’s collusion behaviour which in turn varies with the incentive structure in place. In the last section I discuss several implications of the model, both normative and positive.


Transparency has to do with the flow of information among the participants to a decisionmaking process. As suggested by Webster’s definition, a transparent process is frank (i.e. «free from pretense or deceit»), obvious (i.e. «easily detected or seen through»), and clear (i.e. «readily understood»). In the budget process, transparency is most often looked at from that normative standpoint. For the International Monetary Fund (IMF), for example, «transparency in government operations is widely regarded as an important precondition for macroeconomic fiscal sustainability, good governance, and overall fiscal rectitude» (Kopits & Craig 1998: 1). In other words, budgeting is more efficient in a transparent setting. This is what the IMF argues when it pressures public administrations in developing countries to adopt explicit budgetary rules. This is the assumption behind much of the empirical research that tends to show that public deficits are lower when budget rules are more stringent (Alesina & Bayoumi 1996; Haan & Sturm 1994; Hagen & Harden 1994). But things are not that simple. Obviously, there are situations where frankness, obviousness, or clearness may have unwanted effects. In a parliamentary system, for example, budgetary secrecy is often seen as necessary if the corruption of the whole process is to be avoided. In a bureaucratic organisation, a superior must often hide information from subordinates to protect the integrity of her plans. A more moderate normative position, then, would be that complete transparency in a budget process is generally impossible to reach and, sometimes, undesirable, as bureaucratic organisations are characterized by uncertainty.

Indeed, adopting a positive perspective, many observers of decision-making processes have looked at the role of information and have concluded that uncertainty or information asymmetry and delegation of responsibility are fundamental characteristics of bureaucratic organisations, if not of any organisation. For Max Weber, politicians confront uncertainty as bureaucrats derive power from their control over information: «Under normal conditions, the power position of a fully developed bureaucracy is always over towering. The ‘political master’ finds himself in the position of the ‘dilettante’ who stands opposite the ‘expert’, facing the trained official who stands within the management  of administration. This holds whether the ‘master’ whom the bureaucracy serves is a ‘people’, equipped with the weapons of ‘legislative initiative’, the ‘referendum’, and the right to remove officials, or a parliament, elected on a more aristocratic or more ‘democratic’ basis and equipped with the right to vote a lack of confidence» …. (Weber, in Gerth & Mills 1946: 232). In other words, we can observe a tendency of politicians to use authority-based budget control and of bureaucrats to use expertise-based budget manipulation (Bendor et al. 1987). But as Bendor, Taylor and Van Gaalen (1987: 797) argued, «the Weberian emphasis on the uncertainty confronting politicians must be complemented by an analysis of the uncertainty confronting bureaucrats». Indeed, they showed that a legislature may find it advantageous deliberately to increase the uncertainty that a bureau faces: «if the legislature makes it harder for a risk-averse bureau chief to predict demand or penalty, the bureau will restrain its deception» (Bendor, Taylor and Van Gaalen 1985: 1041). In short, information asymmetry is pervasive in bureaucratic organisations. It appears when some participants to a process have control over information that others lack. Subordinates have access to information that is unknown to their superiors concerning the characteristics of the goods and services they produce or the needs of the population they serve. Lacking that information, a superior may find it difficult to make efficient budgetary allocations.

One way to compensate this information asymmetry is through the delegation of responsibility. Indeed, budgeting decisions, or decisions regarding spending allocations, may be delegated or not. Where there is no delegation the budget is entirely decided by the superior for every level of her organisation. Subordinates have no budgeting decision to make. They are mere executants. This situation is typical of small or highly centralized public administrations where a few persons make all budgetary decisions. In larger or more decentralized organisations, the superior delegates part of her budgeting responsibilities to subordinates thus creating two sets of spending allocation decisions, or two budgets: 1- the superior’s budget in which the superior decides the share of her total financial resources each subordinate will receive; 2- the subordinate’s budget in which he allocates funds within his sub-division to executants. The superior delegates part of her responsibility to subordinates in large part because of uncertainty. But this method does  not go without risks. As Robert Dahl (1967: 21) observed «there are decisions that require me to delegate authority to others … but if I delegate, may I not, in practice, end up with a kind of aristocracy of experts, or even false experts?» Roberto Michels (1915) had expressed the same idea concerning political parties in his famous Iron law of oligarchy: «Organization gives birth to the domination of the elected over the electors, of the mandataries over the mandators, of the delegates over the delegators. Who says organization says oligarchy».

When information asymmetry and delegation of responsibility combine, a unique situation is created, coined as the «agency relationship» in the literature of neoinstitutional economics and political science. Indeed, at the most general level, Stiglitz (1987: 967) suggested that a principal-agent situation arises «whenever one individual’s action have an effect on another individual». More specifically, Gilardi (2001: 3) defines the principal-agent relationship as «a social transaction, or interaction, in which one actor, the agent, carries out actions that are intended to fulfil the interests of another actor, the principal». In the specific situation where one considers the process of allocating financial resources to programs and activities in a bureaucratic organisation, I argue that a principal-agent relationship appears when both information asymmetry and delegation of responsibility are present.

In this paper, I propose a theory of budgeting based on the principal-agent theoretical framework. First, I provide an institutional description of the delegation structure and information asymmetry in a public university, as an example with which to substantiate the following theoretical discussion. In the section that follows this description, I explore the dynamics of a simple principal-agent relationship where both actors can play either transparency or opacity with regard to the budget information each one controls. I show that the outcome of this relationship varies according to the presence or absence of builtin incentives aimed at modifying the agent’s preferences. In the third section, I develop a more complex model where the principal delegates part of her responsibilities to a supervisor. I show that the budget outcome depends on the supervisor’s collusion behaviour which in turn varies with the incentive structure in place. In the last section I discuss several implications of the model, both normative and positive.

Delegation and information asymmetry in a public university: an illustration

To illustrate the workings of delegation and information asymmetry in budgeting, let us look at the budget process in a public university as an example. After Bergman, Muller & Strom (2000), let us conceive of the budgetary process in a democratic society as a «chain of delegation» going from the electorate to budget executants. For example, the most important delegation nodes in the budget process in the higher education system in the province of Quebec could be depicted as follows:

Electorate – Provincial Legislature

President of the Treasury Board – Departments

Minister of Education – Universities and Colleges

Finance Vice-rector – Faculties

Dean – Departments

The electorate delegated to the provincial Legislature the responsibility to spend or otherwise allocate the equivalent of 24% of provincial GDP, a total of 43 480 M$ for the 1999-2000 budget. The provincial Treasury Board then allocated this amount among the various government programs and services, of which 9 676 M$ were allocated to the ministry of education which in turn allocated 278 M$ to the university I use as an example here. At each one of these levels, a subordinate receives an amount of money from his superior and reallocates it among his own subordinates. The same process continues within the university. At this level, the central budgeting question is: how are monetary resources allocated to university activity? I limit the discussion to this part of the university budget that is used to staff and operate faculties and departments (mainly salaries for teaching, administrative, and support personnel, as well as material resource expenditures). In the university annual financial statement, these expenditures are included in the Operating fund under the heading «Teaching and free research». Expenditures in «Teaching and free research» for fiscal year 1999-2000 totalled 217.7 M$. There is no need to say that the largest part of this amount was already committed when the University prepared its 1999-2000 budget. Permanent employees (professors, secretaries, technicians, and professionals) directly involved in teaching and free research functions consumed around 195 M$1 which left about 23 M$ for annual allocations. In this description, I ask two questions regarding the annual allocations for «teaching and free research» expenditures: Is there a delegation of responsibility in the budgeting process? Is there information asymmetry?

Delegation of responsibility

There is delegation of budgeting responsibilities for both the auxiliary teaching personnel budget (14.9 M$) and the material resource budget (4.9 M$). Using a complex formula based on a weighted volume of activities (number of students, courses, diplomas, etc.), the vice-rector allocates these amounts among the 16 faculties. In single department faculties (Dentistry, Law, Music, Theology, Philosophy, Pharmacy, and Nursing) the delegation stops there. In the other 9 faculties, the dean allocates the amount received among the departments after having extracted the amount necessary for faculty operations. In those cases, the delegation stops at the department level, the chair being responsible for allocating the money received among various spending areas under his responsibility. Therefore, we have a two-level structure of delegation within the university: Vice-rector to Dean and Dean to department Chair. This process has been working with only minor modifications for at least the last 20 years.

Things are similar with the decisions to hire new permanent (or likely to become permanent) personnel (tenure-track professors, secretaries, technicians, and professionals) but they use to be quite different. Indeed, up until 1997, the delegation structure in this area was a one-level delegation. The decision to «open a position» was taken by the vicerector after a request by the department chair channelled through the dean. Once the decision was made, it was left to the department to choose in which specific sector the hiring would take place and what would the hiring requirements be (completed PhD or not, postdoctoral experience or not, etc.). The dean only made formal interventions in those decisions in multiple-department faculties. However, in 1997, as part of a scheme aimed at reducing the University budget deficit. hiring decisions were decentralised to the faculty level. Though the vice-rector kept the authority to authorize each position, he delegated to the dean the responsibility to allocate teaching positions among departments while leaving to the department its traditional role in choosing the candidate. In short, after 1997, the delegation structure for hiring decisions was also a two-level structure. This change in the delegation structure raises the issue of its impact on overall spending. Is it possible that the decision to delegate, made to foster the objective of eliminating the university deficit, actually increased costs? In order to answer that question, one also has to consider the role of information in the budget process.


1 This is an approximation.

Information asymmetry

There is quite an important amount of information that is available to the participants in the university budgetary process: information concerning revenue and spending in previous years (in particular, detailed information concerning each transaction recorded by the university Finance service), and information concerning the level of activity of each unit in previous years (number of students taught, number of diplomas, number of courses, etc.). This information is controlled by the vice-rector. Therefore, there is information asymmetry in favour of the vice-rector. There is, however, information asymmetry in favour of the dean against the vice-rector and of the department chair against the dean when actual needs are considered. Indeed, each faculty is a «black box» to the vice-rector (except maybe his or her own faculty) and each department to the dean. The culture of a faculty or a department, as well as the quality standards for research and the training requirements for students, are, to a certain extent, unique to each unit. A vice-rector (dean) may observe the inputs and the outputs of a faculty (department) but may find it difficult to understand what goes on in-between. Therefore, when it comes to making allocation decisions, the vice-rector (dean) is somewhat dependent on the information provided by the dean (department chair). The information asymmetry here is in favour of subordinates over superiors. If the superior relies on the information she has (information about past activity) she runs the risk of making the wrong choices in establishing institutional priorities, thus over funding faculties or departments that have less development potential and under funding those that have more.

One way to solve this problem is to get involved in a long and costly process of consultation to gather appropriate information. The other way is to let subordinates (deans and department chairs) make the decision and bear the consequences. This problem is particularly acute when it comes to cutting expenditures, as was the case in this university in 1997. How does one go about choosing which department or which faculty should have less personnel? On the one hand, one may want to spread spending cuts equally across the board, a potentially inefficient decision since some units may be already overstaffed while others are in need of additional personnel. One may also want to scatter the resistance to budget cuts, or to deviate toward others the criticism one’s budget policy may cause. In this case, the solution is to delegate the responsibility to make cuts to lower tier decision-makers in the organisation. This is what the university did in 1997: delegating the responsibility to make hiring decisions actually meant delegating the responsibility to decrease the number of teaching, administrative, and support personnel positions. As long as faculties and departments were to compete against each other, the university spared the resistance its budget policy could have provoked.

Therefore delegation is a way for superiors to solve the problem of uncertainty caused by information asymmetry in favour of subordinates. It may also be a solution to the opposition or resistance of subordinates to difficult budgetary decisions. But what is the effect of delegation on spending level if it has any? Let us see what a principal-agent theory treatment of this issue can tell us.

A theory of budgeting in a bureaucratic organisation

I assume that actors are rational, i.e. they are utility maximizers. Their utility function
includes three arguments:

  • money, a short cut for «things of value» like trips abroad, cars, comfortable houses, good food, nice offices, big spending accounts, etc.;
  • power, i.e. influence over others or the capacity to make others do what they would not otherwise; this capacity is generally linked to authority positions in the organisation;
  • esteem, the assurance of being recognized as knowledgeable, efficient, intelligent, etc. by others, i.e. actors value the fact of having their opinion taken into account in the allocation of resources.

The maximization of one argument often coincides with the maximization of others. Positions of authority, for example, generally come with higher wages, higher influence over others and higher possibility of seeing one’s ideas incorporated into the content of decision. However, it may happen that the search for one of these objectives be incompatible with another. For example, one may resign from office on a question of principle. This amounts to giving up on money and power to protect one’s (self) esteem. Choosing a career in the private sector may be seen as giving more importance to money than to power since wages are higher for private than public servants, yet authority may be seen as lower for the former. Hence, actors in the budget process want a budget that provides them with money, power, and esteem. They want a budget that includes monetary advantages for themselves (big spending account, high salary, etc.), leverage over their subordinates (high discretionary budget to distribute among subordinates), and characteristics that correspond to their idea of a good budget and that help them implement the ideas they have about a good policy.

Actors are further assumed to be involved in a relationship characterized by delegation of authority and information asymmetry, a principal-agent relationship. The principal delegates part of her authority to an agent who controls an information that is not available to her. To decrease her level of risks, the principal may make a formal contract with the agent in which is imbedded a set of incentives for the agent to divulge the information he has. The principal and the agent may also indulge in informal transactions (Breton & Wintrobe 1982) in which efficiency behaviour is exchanged for trust (to be immediately consumed as «esteem» or to be capitalized in view of a future use). Here I concentrate on this part of the informal transactions between the principal and the agent that concerns information. Two types of information are particularly relevant: uncompressible expenditures and total revenue.

Agents control the information concerning uncompressible spending. Indeed they have a better knowledge of the amount that is necessary to maintain at their actual level the outputs of their units: for a university department chair, for example, how many lecturers are needed to keep a given program running or how many secretaries are necessary to maintain efficiency. From the principal’s point of view, the difference between the total budget allocated to an agent’s unit and this unit’s uncompressible spending may be seen as superfluous expenditures –the «discretionary surplus» (Migué and Bélanger 1974) that the agent tries to maximize. Information about the level of uncompressible spending is necessary for the principal to make an efficient allocation of resources (i.e. an efficient budget) among the services she is heading. Having that information, the principal can better distribute her budget among agents and avoid the risk of free riding (one agent getting a higher proportion of the budget than necessary) and of adverse selection (over funding one service while under funding another). This is why the principal tries to convince the agent to be transparent, that is to relay the information he controls to her so as to increase her own efficiency. This efficient behaviour on the part of the agent is what I call «upward transparency» (TU), which I define as the behaviour of the agent that provides the principal with the information she needs to make efficient budget allocations.

Total revenue is an important information controlled by the principal. Revenue sources at each budgeting level are numerous and complex enough for a superior to have the capacity to hide part of that information from her subordinates. In the university budget process, for example, the subsidy given by the ministry of Education and voted on by the provincial legislature represented only 55% of total university budget in 1999-2000, the rest coming from various other sources. Likewise, revenue given to departments through the general allocation scheme is supplemented by other sources — like compensations for professors assigned to administrative duties or funds for graduate student support — and these supplementary revenue sources could make up a substantial part of a department revenue. The multiplicity and the complexity of revenue sources thus create an asymmetry in favour of the principal who knows her total revenue whereas the agent does not. From his point of view, the difference between his superior’s total revenue and her [the superior’s] uncompressible spending may be viewed as her room to manoeuvre of which the agent wants to get a part. The agent needs the information concerning the principal’s real room to manoeuvre in order to set his budget requests at the most appropriate level. If his requests are too far below the principal’s capacity, he gets a budgetary allocation that is inferior to the one he would have gotten had he asked for more. If his requests are too far above the principal’s capacity, they could be seen as unrealistic and his evaluation of costs and needs as non credible, thus eroding his capital of trust with the principal. The agent can save the cost of finding out about the principal’s real manoeuvre room and, most importantly, the cost of cheating (that is of artificially inflating service costs and clients’ needs) if the principal adopts a «downward transparency» (TD) behaviour, the behaviour of the principal that provides the agent with a clear picture of her real room to manoeuvre.

Thus, using a game theory presentation in normal form, we have two actors, the principal (P) and the agent (A), with two strategies, transparency (T) and opacity (O) (or, equivalently, more or less transparency), the combination of which yields four budget outcomes (figure 1):

  1. A Delegation-Success Budget (DSB) appears when the agent is more transparent and the principal less. Here the principal gets to know everything she needs in order to make an efficient budget while preserving her freedom of action by not revealing her real room to manoeuvre. This budget is called a delegation-success because it corresponds to the situation depicted by Arthur Lupia and Mathiew D. McCubbins where the delegation of authority yields a budget that is better than it would have been without delegation2.
  2. Blind budget (BB): when none of the actors cooperate. This type of budget corresponds to the outcome the principal would have realised had she not delegated to her agent. Indeed, since all the information controlled by the agent is kept from the principal, this budget corresponds to the budget the principal could have achieved by herself.
  3. Cooperative budget (CB) is the result of both actors adopting a cooperative strategy of «more transparency». This outcome also improves the principal’s welfare relative to BB but less so than DSB.
  4. Delegation-failure budget (DFB): when the agent is less transparent and the principal is more. In this case, the agent keeps for himself the advantages of being the one informed while drawing profits from the principal’s transparency. The information provided by the principal is used by the agent to increase his share of the budget.


2 «We say that delegation succeeds when an agent’s actions improve a principal’s welfare relative to the status quo, where the status quo is the outcome that the principal would have realized had she not delegated to her agent. We say that delegation fails when an agent’s actions reduce a principal’s welfare relative to the status quo» (Lupia & McCubbins 2000: 296).


It is reasonable to assume that the actors’ preference scales are the following ones, given their utility maximizing propensity:

Principal : DSB > CB > BB > DFB
Agent       :  dFB > CB > BB > DSB.

Hence the game depicted in figure 1.

The budget relationship depicted in figure 1 is a typical prisoner dilemma game (game #12 in Anatol Rapoport’s and Melvin Guyer’s (1966) taxonomy) with Blind Budget as the natural solution and Nash equilibrium. Assuming a one-shot game, the likely outcome of the interaction between a principal and an agent maximizing their utility is a blind budget. Allowing for long term learning through experience, the theory of meta-games shows that Compromise Budget is an equilibrium if both actors adopt a strategy of reciprocal response: I will cooperate if you do, I won’t if you don’t (Brams 1975). Therefore the Blind Budget outcome may be avoided through personnel stability. If both the principal and the agent stay in office long enough for them to learn from repeated interactions, they may end up at  a Compromise Budget.


Another way of avoiding the blind budget outcome is through the modification of the preference scale of one actor (or of both). This is possible through a structure of incentives that increases the costs and the benefits of a given strategy for an actor. This solution is typical of normative applications of the principal-agent theory. Imagine that the principal links specific benefits to upward transparency (a financial bonus, a symbolic recognition (esteem) or a promotion for the transparent agent) and specific costs to upward opacity (a financial penalty, or indifference, or the demotion, for an opaque agent) so as to change the agent’s preference scale to this one: CB>DFB>DSB>BB. We would then have a new game, depicted in figure 2.

The game depicted in figure 2 corresponds to game #21 in Rapoport and Guyer’s taxonomy, a «class I game with a single threat-vulnerable equilibrium» (Rapoport & Guyer 1966: 207). Both players have a dominant strategy: Opacity for the principal and Transparency for the agent. Assuming that both players follow their dominant strategy, the natural outcome is DSB which is also a Nash equilibrium.

What is the implication of this section? In the absence of a long term relationship with the agent, the principal may modify the outcome of the basic principal-agent budget model by building incentives into the relationship in order to convince the agent to be more transparent. This outcome is an improvement over the non cooperative solution of the basic model without incentives and over the status quo the principal would have realised without delegation.

Now, how can we take into account the fact that budgetary processes often involve multilayer principal-agent relationships? There are two ways of theoretically dealing with a chain of principal-agent relationships. One may assume that the effect of each relationship adds up into a combined effect of the process on the budgetary outcome. In such a case, there is no need to go any further in terms of theoretical development than a two-actor relationship. However, following Tirole (1986), I argue that there is an interaction between two overlapping principal-agent relationships because the same actor plays the agent in the upper relationship while playing the principal in the lower one. Therefore we need to go a step further in the theoretical construction to take this interaction into account. I suggest to use Tirole’s concept of «supervisor» to represent the actor who plays the role of agent in a relationship while playing the principal in another one.

Let P stand for the principal and A for the agent, in a principal-agent relationship. Let the following expression stand for a two-level nested relationship:

P1 → A1

P2 → A2

where A1 and P2 represent a single actor playing different roles in a two level chain of agency relationships. The actor playing the roles A1 and P2 in the two-level nested agency relationships becomes ST (the «supervisor» in Tirole’s agency relationship):

PT → ST → AT

The strategies AT has at his disposal is to increase or reduce the distortion in the bottomup flow of information (upward transparency or opacity) whereas ST has two sets of strategies at his disposal: downward transparency or opacity in his relationship with AT and upward transparency or opacity in his relationship with PT. PT is assumed to be the first principal in a series of principal-agent relationships. As such, she is assumed not be involved as an agent in a higher level relationship.

In general, a principal faces several supervisors and a supervisor faces several agents. Therefore a more accurate representation of a Tirole’s agency relationship is the following one:


Tirole has shown the importance of collusion in such a setting. There may be «horizontal
collusion» among supervisors against the principal or among agents of a given branch against the branch supervisor. This possibility of collusion may be theoretically treated within the single principal-agent relationship presented in the previous section. Supervisors (agents) may collude to modify the principal’s (supervisor’s) preference scale thus making the principal (supervisor) adopt a more transparent behaviour. To avoid horizontal collusion, the principal (supervisor) has two strategies: downward opacity (i.e. confer with each supervisor (agent) separately and share with each one a different information along the old principle «divide and conquer») or downward transparency (i.e. building trust through the sharing of information).

A vertical collusion takes place between a superior and his immediate subordinates in order to deceive the third party in the three-tier relationship. There may be a collusion between a principal and his supervisors in order to increase the quality of the information coming from the field in spite of the resistance of agents (principal-supervisor collusion) and there may be a collusion between a supervisor and her agents to deceive the principal (supervisor-agent collusion). The principal has two types of instruments to avoid the building of a supervisor-agent collusion or to counter the effects of such a collusion when it exists: monitoring (v.g. close control of spending down the line in order to know what is actually going on, appointment of trusted individuals as supervisors, regular switching of supervisors in order to prevent the building of trust within a division) and incentives (v.g. bonus or promotion linked to upward transparency on the part of the supervisor). To see the impact of collusion on the dynamics of the basic principal-agent game, let consider two nested principal-agent relationships where the supervisor plays the role of agent in the upper relationship and the role of principal in the lower one (Figure 3).

By assumption, a principal-supervisor (P-S) collusion implies that both the principal and the supervisor adopt a cooperative strategy in the upper budget and that both the supervisor and the agent adopt a maximizing strategy in the lower budget. A supervisoragent (S-A) collusion implies cooperative strategies in the lower budget and maximizing strategies in the upper budget. Collaboration (Co) stands for an agreed cooperative behaviour on the part of all three players.

Table 1 gives the outcomes and the payoffs of four collusion scenarios (No: no collusion; P-S: principal-supervisor collusion; S-A: supervisor-agent collusion; Co: collaboration) under four incentive structures (IS1: no incentives in either budget; IS2: incentives in both budgets; IS3: incentives in lower budget only; IS4: incentives in upper budget only). The collusion scenario preferred by the supervisor and the ensuing outcome is determined by the addition of the payoffs to the supervisor. For example, consider the payoffs when there is no built in incentive (IS1). Each collusion scenario yields a unique outcome in each budget. No collusion (No) implies that the outcomes of the games correspond to the natural solution, each player adopting a maximizing strategy: BB in both budgets. A P-S collusion implies that players adopt cooperative strategies in the upper budget (outcome: CB) and a maximizing strategy in the lower budget (outcome: BB). Conversely, in a S-A collusion, players adopt maximizing strategies in the upper game (outcome: BB) and cooperative strategies in the lower game (outcome: CB). Finally, a collaboration scenario implies that all players adopt the cooperative strategy of more transparency, yielding the outcome CB in both budgets. Given these results, the supervisor prefers a collaboration scenario to a collusion scenario since his total payoff is 6 (3+3) under this scenario as opposed to 5 (3+2 and 2+3) 4 (2+2) under the two collusion scenarios P-S and S-A. Applying this reasoning to all combinations of incentive structures and collusion scenarios, we find that total payoffs to the supervisor are higher  when she colludes with the principal, except under IS-1, where no incentive exists, and IS-4, where there are incentives only in the upper game. In those cases, the supervisor prefers to collaborate with both the principal and the agent thus maximizing her gains. But what about costs? Don’t these choices have an impact on agency costs and overall spending?


Three types of costs can be identified in the budget relationship: information-search costs, efficiency costs, and incentive costs. Three assumptions about cost distribution can help assess them:

  1. A player’s opacity results in information-search costs for the other player. In other words, resources – money, time, and energy – are devoted to looking for the information obfuscated by the other player. Therefore, information-search costs vary from one budget outcome to another; they are higher when both players are opaque (outcome: BB) and they are lower when both players are transparent (outcome: CB). Hence an ordinal scale that, can be represented by an interval scale for computing purposes:
    BB > DFB = DSB > CB
    4         2         2         0
  2. Some budget outcomes are more efficient (i.e. closer to a situation that cannot be improved for one individual without deteriorating the situation of another) than others. Efficiency is improved when the principal is better informed or, equivalently, when the agent is transparent. There is also a gain in efficiency when the principal can keep a certain latitude by being opaque. Considering  a loss in efficiency as a cost, we have the following ordinal scale of efficiency costs with their associated cardinal values:
    DFB > BB > CB > DSB
    4         3       2          1
  3. The implementation of incentives and controls aimed at modifying an agent’s preferences is costly. Assuming that the «no incentive» structure (IS1) bears no cost, assuming equal costs for incentives in the upper (IS4) or the lower (IS3) budgets, and assuming that these costs add up when incentives are implemented at both levels (IS2), we have the following distribution of incentive costs:
    IS2 > IS4 = IS3 > IS1
    4         2         2         0

Using the interval scales for each cost, it is easy to compute the relative agency costs of each outcome. Outcome BB, for example, which is the outcome of the simple game reported above in figure 1, bears an information cost equal to 4, an efficiency cost equal to 3 and an incentive cost equal to 0, for a total agency cost of 7. The computation of agency costs resulting from complex structures (i.e. agency relationship with a supervisor), though a little more cumbersome, follows the same logic with the additional rule that information and incentive costs for upper and lower budgets are added whereas efficiency costs are averaged. This is to take into account the idea that information and incentive costs increase as additional layers of agency relationships are added, but that efficiency gains may be reached by additional delegations. Thus, for a type 1 incentive structure (IS1) and a «no collusion» scenario, we have the cost structure given in table 2.


Applying these rules, one can compute the agency costs related to each outcome resulting from each collusion scenario and each incentive structure given in table 1. This distribution of agency costs is displayed in table 3, together with the agency costs related to the outcomes of the simple games analysed above. Two situations stand out as involving a cost level that is far below the others: simple agency relationship with long term interaction (Prisoner dilemma in a metagame), and Collaboration scenario without incentive. The lowest cost situation in simple games does not correspond to the outcome preferred by the principal on the basis of her payoffs. Indeed, the Nash equilibrium of game #21 is DSB, the principal’s most preferred outcome, whereas the cost level is markedly higher than the lowest cost (5, as compared to 2). In complex games, the lowest cost outcome corresponds to the supervisor’s most preferred outcome.

Discussion and conclusion

The budget model described above has several normative and positive implications. From a normative standpoint, the model helps answer at least three questions.

  1. When is it rational for a principal to delegate responsibilities to a supervisor? Or, equivalently, can a principal improve her payoffs by moving from a simple to a complex budget relationship? The answer depends on the status quo position. It is clear from table 4 that any move from RG#12 (simple relationship without incentive) pays: a principal can improve her situation from BB to at least CB by delegating to a supervisor. However if the status quo position is RG#21 (simple relationship with incentives) delegation deteriorates the principal’s payoff. Therefore a rational principal should not delegate and should rather resort to incentives as a budget instrument. This is all the more true when we consider that delegation strips the principal from some of her power as it gives the supervisor a dominant role in determining the collusion scenario that is to prevail. Why, then, did the authorities in the university we analysed choose to delegate part of their budgeting responsibilities to the deans as supervisors? Because of an exogenous constraint. Indeed, the potential for resistance or outright opposition on the part of faculty members, support personnel, and students, was a serious threat to the political feasibility of the restrictive budgetary policy that the university wanted to implement in order to eliminate its budgetary deficit. One way to deal with this opposition was to scatter the resistance points through delegation. This way, pressures against the university budgetary policy could be buffered through 60 departments and 16 faculties. In other words, the choice of a policy instrument is not always motivated by a single policy objective. Decision-makers simultaneously pursue several objectives. When facing a potentially strong political opposition, the principal may be willing to give up her most preferred budgetary outcome in order to secure the political benefits generated by delegation.


  2. When, and with whom, is it rational for the supervisor to collude? The supervisor is in the unique position of choosing whether or not to collude and with whom. The analysis provided in table 1 tells us that it is never to the supervisor’s interest to let things go their own way nor to collude with agents against the principal3. Rather, the supervisor will prefer either a principal-supervisor collusion scenario or Collaboration, depending on the incentive structure in place. When there is no incentive in the lower budget, the supervisor prefers a scenario of collaboration. Otherwise, her preference goes for a scenario of collusion with the principal. My empirical observation of the university budget process is not detailed enough to provide a convincing confirmation of this hypothesis. However what I have observed so far seems to lead to that conclusion: there is no incentive structure in the university budget relationship and the collusion behaviour I have observed so far resembles a collaboration scenario.
  3. Given the collusion behaviour of the supervisor, when is it rational for the principal to change the incentive structure? Let us start with the situation where there is no incentive (IS-1) and where the supervisor chooses a scenario of collaboration. The principal cannot improve her payoff in the upper budget by modifying the incentive structure. But assuming that she shares the supervisor’s preferences in the lower budget, she will get a better result in that budget if she implements incentives aimed at modifying the agent’s preferences (IS-3). However, in doing that, the principal increases agency costs.  Therefore, a principal who is insensitive to agency costs would try to implement incentives at least in the lower budget. A similar supervisor would pressure the principal for incentives in the upper budget as well. The result is the most costly situation, IS-2, with agency costs reaching 11,5 points. If we assume that agency costs enter the utility function of both the principal and the supervisor, an assumption that is coherent with our previous discussion about the importance of esteem in the utility function of budget actors, the best situation results from a scenario of collaboration without incentive neither in the upper nor in the lower budgets. In other words, the use of incentives in the budget process involving a supervisor is never the choice of a rational, well informed, principal. Incentive structure in the budget process may allow savings in agency costs but only in a simple structure process. With complex structures involving supervisors, it is cheaper to rely on individual’s preferences and not to try to change the preference ordering of supervisors and agents through incentives. Higher agency costs are unavoidable for the most efficient budget outcome (DSB) to be possible.


3 Letting things go means allowing the budget outcome to be set only by the maximizing rationality of actors (i.e. a No collusion scenario).

From a positive standpoint, the model presented here also has important implications for overall spending and actors’ behaviour. Indeed, agency costs end up in actual financial costs for the organisation. The time and energy invested in information search and in the implementation of incentives aimed at convincing supervisors and agents to be transparent, are resources that are deviated from the production of goods and services. Therefore, for a given output level, higher costs translate into higher spending, everything else being equal. Hence the following hypotheses concerning the level of aggregate spending:

H: Ceteris paribus, the level of aggregate spending is lower when an efficient incentive structure is at play in a simple organisation (without supervisor).
H: Ceteris paribus, the level of aggregate spending is higher when an efficient incentive structure is at play in a complex organisation (with supervisor).

In addition to these two hypotheses that can be falsified through a quantitative analysis predicting the level of aggregate spending, several hypotheses predicting actors’ choices can also be deduced in order to guide a qualitative analysis of the budgetary process in a given sector of government intervention.

H: A supervisor’s collusion behaviour varies with the existing incentive structure: in the absence of incentives in the lower budget, she will tend to collaborate with both her principal and her agents; when there are incentives in the lower budget, the supervisor will collude with her principal.
H: A principal will resort to delegation as a budget instrument only if exogenous constraints makes it desirable. As soon as this constraints disappear, the process will be centralized again.

Transparency and opacity are two strategies in the hand of superiors and subordinates in the budgetary process of a bureaucratic organisation. The ideal world for the superior is one where she can be opaque while her subordinates are transparent. This is how she can get the most out of her interaction with them. However, their rationality tells them not to reveal the information they have, yielding a result that she could improve if she correctly used two policy instruments: incentives and delegation. Through incentives, a superior can alter her subordinates’ preference scales thus making it more advantageous for them to be transparent. Through delegation, she can free herself from the dependence on her subordinates by making them responsible for the decision they make. For delegation to lead to her best outcome, incentives must be added, thus increasing costs.

There is an alternative instrument a superior may use to improve the outcome for both her and her subordinates while maintaining costs at their lowest level : trust. Acknowledging her personal interests and those of her subordinates, a superior may induce them to be more transparent through a long term relationship based on reciprocity: I will cooperate if you do, I will not if you do not. For this method to work, three conditions are essential: first, the superior must be willing to risk the uncooperative outcome where everybody is opaque; second, terms of office for both superiors and subordinates must be long enough to allow repeated interactions; third, the superior must give up her attempt to get her most preferred outcome and be content with her second best. Under these conditions, a cooperative solution may be stable.

When one looks at the new public management agenda for government bureaucracy, one can see that the preferred instruments are incentives and delegation rather than trust. The consequence may well be a stabilisation or a decrease in spending accompanied by a much sharper decrease in the quantity and quality of government services as most of the savings will have been absorbed by agency costs. And opponents to «big governments» will comfort themselves, unaware that they created the real problem of diminishing efficiency by trying to solve the false problem of a benevolent superior facing malevolent subordinates.


Bendor, J., S. Taylor, et al. (1987). “Politicians, Bureaucrats, and Asymmetric Information.” American Journal of Political Science 31(4): 796-828.

Bendor, J., S. Taylor, et al. (1985). “Bureaucratic Expertise vs. Legislative Authority: A Model of Deception and Monitoring in Budgeting.” American Political Science Review 79(4): 1041-1060.

Bergman, T., W. C. Muller, et al. (2000). “Introduction: Parliamentary democracy and the chain of delegation.” European Journal of Political Research 37: 255- 260.

Brams, S. J. (1975). Game Theory and Politics. New York, The Free Press.

Breton, A. and R. Wintrobe (1982). The logical of bureaucratic conduct: an economic analysis of competition, exchange and efficiency in private and public organizations, Cambridge University Press.

Dahl, R. A. (1967). Pluralist democracy in the United States: conflict and consent. Chicago, Rand McNally.

Gerth, H. H. and C. W. Mills, Eds. (1946). From Max Weber: Essays in sociology. New York, Oxford University Press.

Gilardi, F. (2001). Principal-Agent Models Go to Europe: Independent Regulatory Agencies as Ultimate Step of Delegation. ECPR General Conference, Canterbury, UK.

Imbeau, L. M. and K. Chenard (2002). The Political-Economy of Public Deficits: A Review Essay. European Public Choice Society, Belgirate, .

Kopits, G. and J. Craig (1998). Transparency in Government Operations. Washington DC, International Monetray Funds: 42p.

Lupia, A. and M. D. McCubbins (2000). “Representation or abdication? How citizens use institutions to help delegation succeed.” European Journal of Political Research 37: 291-307.

Michels, R. (1915). Political Parties: A sociological study of the oligarchical tendencies of modern democracy. Glencoe, IL, Free Press.

Migué, J.-L. and G. Bélanger (1974). “Towards a General Theory of Managerial Discretion.” Public Choice 17.

Rapoport, A. and M. Guyer (1966). A Taxonomy of 2×2 Games. General Systems: Yearbook of the Society for General Systems Research. L. Von

Bertalanffy and A. Rapoport. Ann Arbor, MI, Society for General Systems
Research: 203-214.

Stiglitz, J. (1987). Principal and Agent. The New Palgrave: A Dictionary of Economics. J. Eatwell, M. Milgate and P. Newman. London, Macmillan. 3: 966- 972.

Tirole, J. (1986). “Hierarchies and bureaucraties.” Journal of Law, Economics and Organization 2: 181-214.

Leave a Comment »

No comments yet.

RSS feed for comments on this post. TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Blog at

%d bloggers like this: