Company influence on foreign aid disbursement: is conditionality credible when donors have mixed motives?

When donors enforce conditionality upon recipients who do not implement the conditions, companies can suffer from cancellation of their contracts with the recipient when aid dries up. A strategic recipient may avoid implementing controversial conditions by only granting a contract to a company that puts pressure on the donor to keep aid flowing. In our model, each of these three agents takes account of each of the two other agents' actions. We show that this triadic structure can be crucial when explaining recipients' use of companies to influence donors to give aid unconditionally, and we offer a time-consistent explanation for the failure of conditionality.

JEL Classification: E61, F35

1. Introduction

Empirical evidence indicates that poor countries frequently abstain from implementing the conditions that donors have set as a requirement for granting foreign aid. Still, it is found that the aid is disbursed irrespective of the recipient's implementation record (World Bank 1992; Mosley, Harrigan, and Toye 1995; Collier 1997; Svensson 2000a). The World Bank (1992) concluded that even though the rate of compliance with World Bank conditions was only 50%, the release rate of loans was nearly 100%. In other words, despite the donors' intention to induce the recipients to undertake what are perceived by the donors to be "good policies," usually regarding fiscal, monetary, and trade policies designed to increase economic growth, it is found that aid does not induce these policies (Burnside and Dollar 2000). Even more uncontroversial conditions, like setting a certain minimum level of expenditure on health care and education, seem to fail (World Bank 1992; Mosley, Harrigan, and Toye 1995; Oxfam 1995).

The malfunctioning of conditionality is a serious problem for the donor community and the multilaterals because this instrument is viewed as a necessity for achieving the goals of aid (Kanbur 2000). At the same time, receiving aid is a very important income source for poor countries. On average, aid accounted for more than half of central government expenditure for 50 of the most aid-dependent countries from 1975 to 1995 (World Bank 1998b), and a typical low-income country now receives around 7% to 8 % of GNP in foreign aid (World Bank 1998a). This gives rise to a puzzle: If it is vital for the recipients to get aid, and also essential for the donor to have the conditions implemented, why cannot the seemingly powerful donors force the seemingly weak recipients to implement the conditions before aid is disbursed?

Ravi Kanbur's (2000) observations as a World Bank representative in Ghana in 1992 illustrate one potential explanation for the failure of conditionality. At that time, the Ghanaian government had refused to implement the conditions set by the World Bank for granting a loan, and the bank had to decide whether or not to disburse this loan. In this situation, private companies that had contracts with the Ghanaian government put pressure on the World Bank to release the loan because they were afraid of not getting paid. Eventually, the loan was disbursed without the implementation of the conditions, and Kanbur concludes that the pressure surrounding conditionality is important in explaining its failure. Thus, strategic recipients may refuse to implement the conditions and then threaten to cancel contracts with companies in order to put pressure on the donor to disburse aid. (2)

Building upon the triadic (3) modeling structure of Basu (1986), we consider the interdependence that sometimes arises between donors, recipients, and large companies with interests in both countries. (4) In this model, the donor takes into account both its own and the company's relationship with the recipient when deciding on aid disbursement, and, more generally, each of the agents always takes account of the triadic structure when making their decisions. If the recipient can influence the company to put pressure on the donor to disburse the aid, we show that this could make the donor provide the aid even when the recipient has not implemented the conditions. We show that the recipient is not able to influence the company in a situation where the parties interact only once. However, donors usually give aid to the same recipients over time, and allowing for repeated interactions between the parties gives the result that the recipient is able to influence the company to generate pressure against conditionality. Hence, the recipient is not necessarily as weak as it may seem, because the recipient might be able to use the company's influence over the donor.

This article is related to the literature on foreign aid in general and to the work on the failure of aid conditionality in particular. Despite its importance, conditionality has received little attention with regard to theoretical modeling (Drazen 2000). The main contribution to the understanding of the failure of aid conditionality is Svensson's (2000c) principal-agent model where he illustrates how altruistic donors' time inconsistency problem gives rise to continued aid even when conditions are not implemented. (5) In Svensson's framework, the donor and the recipient share the perception that implementing the conditions improves the well-being of the recipient, so the recipient would implement some of the conditions even if the donor did not exist. The failure of conditionality arises because the level of poverty determines the amount of aid, and this gives incentives for the recipient not to implement costly poverty-reducing policy conditions. Another contribution that illuminates the problem is Mosley, Harrigan, and Toye (1995), who model the interaction between the donor and recipient as a dyadic bargaining game. They find that there will always be some slippage on the conditions that the recipient has agreed to implement.

In our model, we incorporate the empirical finding that, frequently, there are disagreements between the donor and the recipient on what constitute "good policies" (Mosley, Harrigan, and Toye 1995; Dollar and Svensson 2000; Kanbur 2000). One reason for the unwillingness of the recipients to implement the conditions is that this change in policies will often harm politically important groups in the recipient country (Summers and Pritchett 1993). (6) Another reason could be that it is not in the recipient country's interest to implement these policies, as could be the case when the poor country maintains import barriers to protect "infant industries." The donors, on the other hand, would profit from having access to another foreign market and would put pressure on the recipient country to liberalize trade. Thus, whether some of the conditions are imposed for the donor's own benefit, or the benefit of its allies, or whether they are imposed for the good of the recipient, is an open question. In our model, we assume that the recipient would prefer not to implement the conditions, ceteris paribus.

To illustrate conditionality, we assume that the donor follows the World Bank (1998a) in adhering to the empirical findings of Burnside and Dollar (2000) that aid only increases economic growth if the tight macroeconomic policies are sustained. Thus, the donor makes aid conditional upon these policies, believing that implementation of the policy conditions is essential in order to achieve the intended effects from aid. So, even if aid has some positive effects if the conditions are not implemented, the donor's assessment is that granting aid is wasted in this situation. With this divergence of opinion, it is not necessary to restrict our focus to Samaritan donors, and this differs substantially from the existing work on foreign aid policy, in which time inconsistency is crucial (see Mosley, Harrigan, and Toye 1995; Pedersen 1996, 2001; Svensson 2000c). Hence, our results are not related to the Samaritan's dilemma.

Obviously, as Svensson (2000c) and World Bank (1998a) also note, donors' rationale for giving aid may, in reality, be guided both by altruistic and self-interested motives. Thus, both motives will naturally be reflected in the conditions attached to aid. The empirical studies of Burnside and Dollar (2000), Trumbull and Wall (1994), and Alesina and Dollar (2000) all suggest that both motives are present among donors. Thus, we incorporate a self-interest motive and an altruistic motive in the donor's preferences. So, if a donor has self-interests with regard to its own domestic industries, and an altruistic motive for maintaining aid conditionality, we show how recipients can grant contracts strategically to companies with origins in the donor country and that this may cause a time-consistent failure of conditionality.

In general, the literature on foreign aid uses traditional dyadic models to explore the donor-recipient relationship, and we show why the triadic framework may be important in explaining the failure when companies are able to influence the donor's disbursement decision. In our model, restricting the donor, the recipient, and the company's interaction to being dyadic yields the opposite results in comparison to allowing for triadic relations: Assuming dyadic interactions implies that the recipient is unable to influence the company's eventual pressure on the donor; this causes the recipient to implement the conditions, and conditionality becomes successful. Our framework should thus be regarded as complementary to Mosley, Harrigan, and Toye (1995) and Svensson (2000c). One novel policy implication that may be important for the players on the foreign aid scene is that this setup provides a rationale for donor guarantees to companies (from the donor countries) that operate in the recipient country.

This article is organized as follows. A simple introduction to the triadic structure of the problem is given in section 2, and a formal game-theoretical framework that models the interdependency between donors, recipient, and companies is proposed and analyzed in section 3. Section 4 discusses suggestions for improving the record of conditionality, and section 5 provides some concluding remarks.

2. Triadic Structure

Assume first that there is a traditional bilateral relationship between a donor and the recipient, where the donor adheres to conditionality. From the donor's point of view. maintaining conditionality implies that aid is only disbursed if the recipient has implemented the conditions. In line with the empirical findings, we also assume that the recipient would prefer not to implement these conditions, but submits to the donor's demand because implementation results in aid being disbursed. Hence, in this pair-wise (dyadic) relationship, conditionality is successful.

In our setting, the recipient is going to build a road. Let a construction company be the potential constructor, and hence, a third party to the donor-recipient relationship. Assume also that the company is large and influential, so it has some leverage on the donor. We could treat the company's influence on the donor as a black box, but for illustrative purposes we assume that the company is about to locate a new project, and that the donor is interested in having this project located in its own country. Note also that the donor is indifferent as to whether or not the road is built. Assuming a dyadic structure implies that each pair of agents interacts in isolation, and hence, aid conditionality is not influenced by the introduction of the company as illustrated in Figure 1.

[FIGURE 1 OMITTED]

In a general triadic structure with three agents, however, each agent's optimization problem not only takes into account its own interaction with the two other agents. In addition, each agent also takes account of the other agents' interaction with each other, as is illustrated in Figure 2.

[FIGURE 2 OMITTED]

The interesting cases in the triadic structure arise when one agent takes actions it would not have taken in traditional dyadic relations. In our triadic setting, the recipient may be able to make the company influence the donor's conditionality decision by making the contract to the company contingent upon the company's pressure for aid disbursement. Assume that the recipient announces that it will give the road contract to the company only if the company manages to influence the donor to disburse the aid even if the conditions are not implemented. Thus, the company may be forced to use the location decision to influence the donor's disbursement decision in order to secure the contract: The company can let the donor know that it will only locate in the donor country if the donor disburses the aid. If the location is more important to the donor than enforcing aid conditionality, this pressure may cause the donor to disburse aid even if the conditions are not implemented, which is opposite to the result in dyadic relations.

3. The Model

We use a game-theoretical framework to model the interaction between a recipient, a donor, and a company, where the company has business in both the donor's and the recipient's country. In the section "The General Structure of the Payoff Functions" below, we develop the utility functions for the agents, and the section "'Specific Payoffs" describes the payoffs necessary to illustrate the important features of the game. The outcome of the game when assuming dyadic relations shows that conditionality is successful under traditional assumptions of pair-wise interaction. This result is described in the section "The Dyadic Outcome." In the section "The Failure of Conditionality in the Infinitely Repeated Triadic Game" is discussed a triadic solution: Even though the recipient's threat of not giving the road contract to the company is not credible in the stage game, it becomes credible under certain conditions when we allow for infinitely repeated interactions. The subgame perfect equilibrium of the infinitely repeated game shows that it is possible for the recipient to lock the donor into repeatedly granting aid even if the conditions are not implemented.

In this game, it is common knowledge that the recipient will link the donor's disbursement decision to the road contract. No threat is explicitly stated, but the agents know that if the donor does not disburse the aid, then the recipient's intention is to refuse to give the contract to the company unless it locates the new activity abroad. This implicit triadic threat is meant to create pressure on the donor to give aid even if the conditions are not satisfied, and we will return to the credibility issue.

The timing of the game is as follows. At stage 1, the recipient decides whether or not to implement some conditions defined by the donor ex ante, and at stage 2 the donor chooses whether or not to give aid to the recipient. (7) At stage 3, the company decides on whether to locate a new activity in the donor's country, termed "at home," or some other country, which is denoted as "abroad." Finally, at stage 4, the recipient decides on whether or not to pay the company for building a road in the recipient's country. The game tree in Figure 3 illustrates the feasible actions. All we need in order to analyze the game and determine the equilibria is a specification of the payoffs of all three players at each terminal mode in the game tree below. This is the topic of the next two sections.

[FIGURE 3 OMITTED]

The General Structure of the Payoff Functions

All three agents are assumed to be rational, forward-looking, and acting in a utility- or profit-maximizing manner. The profit of the company can be represented as

(1) C = C(m, s),

where s denotes the level of activity the company has in the donor country and m is the recipient's payment for the road. Assume that the company gets an assignment from an agent other than the donor and the recipient. The company can choose between two levels of activity: either [s.sup.*] if it locates the new activities to the donor's country or [s.sup.u] if it locates in some third country different from the recipient's country, where [s.sup.u] < [s.sup.*] and [s.sup.*] are the optimal choice. Hence, if the new activities are located outside the donor's country, this would represent a net cost for the company. Let [s.sup.*] - [s.sup.u] be interpreted as that cost.

The recipient's payment to the company, m, takes only two values: m = [m.sub.0] = 0 if the recipient does not buy the road or m = [m.sub.1] > 0 if the recipient does buy the road. The price [m.sub.1] for the road should be understood as the result of negotiations between the recipient and the company, and hence, we can assume that both parties are better off if the road is built for the price of [m.sub.1] compared to not having the road built.

The recipient, in addition to being better-off paying [m.sub.1] and having the road built, also experiences an increase in utility R if the donor grants bilateral aid a > 0. However, the donor specifies a set of conditions, for instance, some "sound" macroeconomic policies, that the recipient must implement in order to receive the aid. Let c be a dummy variable that indicates whether or not these conditions are implemented, let [c.sub.1] = 1 denote that the conditions are implemented, and [c.sub.0] = 0 that they not.

We incorporate the empirical finding that recipients frequently oppose these conditions (see Summers and Pritchett 1993; Mosley, Harrigan, and Toye 1995; Kanbur 2000), and this is typically the case if the new policies redistribute resources away from groups that are favored by the governments. Thus, the impact on the recipient's utility of implementation is assumed to be negative, ceteris paribus, and the utility function can be expressed as

(2) R = R(m, a, c).

We assume that the donor's utility increases in the amount of aid granted if the recipient complies with the conditions. So, if the conditions are not implemented, the donor believes it is a waste giving aid because macroeconomic disruption would cause the aid to have no effect. In those circumstances, the donor prefers not to give the aid. However, if aid is still given in this situation, the donor has a decreasing utility in the amount disbursed because giving more aid means losing more funds. Consistent with the idea of conditionality, this means that, ceteris paribus: (i) It is rational for a donor to grant aid to the recipient if it implements the conditions, and (ii) it is rational not to grant aid if the conditions are not implemented.

The idea of conditionality can then be captured in the donor's utility function, D, which, for the sake of simplicity, is assumed to be differentiable: Define the aid conditionality function h(a, c) and let D depend on h in that [partial derivative]D/[partial derivative]h > 0, where h is increasing in a if c = [c.sub.1], and decreasing if c = [c.sub.0]. Because the donor believes that implementation of the conditions yields the best policies for the recipient, the utility of the donor is increased if the recipient were to implement the conditions even in the absence of aid [i.e., h(a, [c.sub.0]) < h(a, [c.sub.1])]. The donor's preferences with regard to conditionality can be illustrated in Figure 4.

[FIGURE 4 OMITTED]

Take any given amount of aid, for example, [a.sub.1]. Note that according to the partial aid conditionality function, the donor would be worse off if it must disburse aid when the conditions are not implemented ([h.sub.2]), compared to not giving aid in that situation ([h.sub.1]). In addition, the donor is interested in having as much of the company's activity located in its own country as possible. Hence, the company's scale of business in the donor's country, s, is also included in the donor's utility function, D, which we can state as

(3) D = D[h(a,c),s].

The donor's preferences in this model, as stated in Equation 3, then reflect the empirical findings that both self-interest and altruistic motives play a role for the donors (Trumbull and Wall 1994; Alesina and Dollar 2000).

Specific' Payoffs

To keep the analysis as simple as possible, and to highlight the endogeneity of the implementation of the conditions, assume that the parameter values [a.sub.0], [a.sub.1], [m.sub.0, [m.sub.1], [s.sup.u] and [s.sup.*] are exogenously determined. (8,9) Moreover, assume that the donor and the recipient have signed a contract ex ante that guarantees a certain amount of aid a1 to the recipient if the recipient implements some conditions [c.sub.1]. The contract also states that if the recipient does not implement the conditions, then there will be no aid disbursement, and let "no aid" be denoted [a.sub.0] = 0.

Initially, we assume that the recipient prefers to fulfill the conditions and get the aid, compared to refraining from the aid/conditionality package: R(m, [a.sub.1], [c.sub.1]) > R(m, [a.sub.0], [c.sub.0]). The recipient's utility is determined by whether or not the road is built, whether or not aid is granted, and whether or not conditions are implemented:

(4) [R.sub.a] = R([m.sub.1], [a.sub.1], [c.sub.0]) buy road, get aid, not implement conditions

(5) [R.sub.b] = R([m.sub.1], [a.sub.1], [c.sub.1]) buy road, get aid, implement conditions

(6) [R.sub.c] = R([a.sub.1], ]a.sub.0], [c.sub.0]) buy road, no aid, not implement conditions

(7) [R.sub.g] = R([m.sub.0], [a.sub.0], [c.sub.0]) no road, no aid, not implement conditions

From the assumptions of the recipient's utility function, note that [R.sub.a] > [R.sub.b] > [R.sub.c] > [R.sub.g]. (10)

The company's payoff depends on whether or not it is granted the road contract and whether or not it locates its new activity in the donor country:

(8) [C.sub.a] = C([m.sub.1], [s.sup.*]) road is built and location in the donor's country

(9) [C.sub.b] = C([m.sub.1], [s.sub.u]) road is built and location in another country (abroad)

(10) [C.sub.c] = C([m.sub.0], [s.sup.*]) road is not built and location in the donor's country.

Recall the assumption that if the road contract is not tied to the company's location decision, the company would always choose to locate at home. Note also that, usually, infrastructure projects must be of a substantial dimension to attract multinational companies. Thus, we assume that the road project to be built is large and that the contract is of great value to the company. In this setting, then, it seems most realistic to assume that the value to the company of the road contract is larger than the cost of locating the new activity abroad. Hence, let the company always choose to "build the road and locate abroad" instead of "not build the road and locate the new activity in the donor's country" if it must choose between the two:

(11) [C.sub.b] > [C.sub.c].

By assumption, then, [C.sub.a] > [C.sub.b.] > [C.sub.c].

As stated above, the recipient intends to withhold the contract from the company unless the company locates the new activities abroad in case no aid is granted. Thus, the donor's choice of giving aid or not in a situation where the recipient has not implemented the conditions may be determined by which of the states yield the highest utility for the donor: Either to achieve

(12) [D.sub.1] = D[h([a.sub.0], [c.sub.0), [s.sub.u]

by not granting the aid (and hence, maintaining conditionality) and losing the company's new activities to another country, or to achieve

(13) [D.sub.c] = D[h([a.sub.1], [c.sub.0]), [s.sup.*]]

by giving aid and having the new activity located in the donor country.

We assume that the donor is more concerned about the activities in its own country than the potential waste that could occur in the aid conditionality scheme. So, losing the new activity to another country is considered a substantial loss for the donor, whereas the waste of granting the aid if the conditions are not implemented is considered to be less severe. Hence, we assume

(14) [D.sub.c] > [D.sub.t].

Note also that the donor achieves its highest feasible payoff in this game if aid is disbursed when the recipient has implemented the conditions and the company locates in the donor country. This case is defined by

(15) [D.sub.a] = D[h([a.sub.1], [c.sub.1]), [s.sup.*]].

Before turning to the triadic game, we explore the outcome of the game if the players interact in a paired manner.

The Dyadic Outcome

Assume for now that the three players are restricted to only interact pair-wise, that is, we impose traditional dyadic relations in the game. Because this assumes that it is not possible for the donor to take account of the recipient's relationship with the company when interacting with either the company or the recipient, the dyadic donor-recipient game is restricted to the two first stages of the stage game in Figure 2. Then, we can see from the backward induction solution in the game tree in Figure 2 that conditionality is successful because the recipient will always implement the conditions to get the aid. Hence, the outcome will be (implement conditions, give aid).

In the interaction between the company and the recipient regarding the road contract, we assume that they have found a price that is acceptable to both. Hence, the outcome will be (buy road). Similarly, for the interaction between the company and the donor, we have assumed that the optimal location for the company is in the donor country, and this is also the best result for the donor. Therefore, the outcome will be (locate at home). From these actions, it is evident from Equations 5, 8, and 15 that the payoffs to the recipient, the company, and the donor will be [R.sub.b], [C.sub.a], and [D.sub.a], respectively.

It is essential to notice that the donor achieves its best feasible payoff if we have dyadic interactions. However, it is evident that there is a gain for the recipient if it is able to get aid without implementing the conditions, although this would induce a loss for the donor. So in order to capture this gain, the recipient can go into an alliance with the company. This three-party structure requires a triadic framework to be fully analyzed.

The Failure of Conditionality in the Infinitely Repeated Triadic Game

In this section, we will show that the pattern of repeated interactions in these relationships is crucial for this explanation of the failure of conditionality. From the game tree in Figure 3, the backward induction outcome of the triadic stage game reveals that the recipient's threat is not credible because it is in the recipient's own interest to grant the road contract to the company at stage 4. So the subgame perfect equilibrium path of the stage game follows the sequence that the recipient implements the conditions, the donor disburses the aid, the company locates in the donor country, and the recipient grants the road contract: (implement, aid, home, [m.sub.1]). (11) Thus, compared to the payoff in subgame perfect equilibrium of the stage game, there would be a gain to the recipient if it could manage to get aid without implementing conditions.

Because the company is sure to have the road contract in this game, the recipient needs another "carrot" to make the company willing to punish the donor, because locating abroad imposes a cost on the company. (12) In other words, if the donor does not disburse aid, then the recipient must compensate the company for taking these costs. The repeated nature of the failure of aid conditionality leads us to investigate the game in infinitely repeated interactions when such a compensation is offered.

We show here that by repeating the stage game infinitely, the recipient's threat may become credible if it shares some of the gain by offering a road contract on better terms than the original contract [m.sub.1]. Hence, a subgame perfect equilibrium where the donor must grant the aid without the conditions being implemented can be sustained. (13) Two differences need to be highlighted. First, we assume now that the company is able to switch parts of its tax base between its branch in the donor's country and a branch in another country, and this decision replaces the company's decision on location at stage 3. To follow the specified payoff function of the company (see "'The General Structure of the Payoff Functions," above), assume that the tax system in the donor's country is slightly more preferable to the company than the tax system in the other country. So, an optimal tax decision in isolation from other concerns would be to render the full tax base to the donor country's tax authorities.

The second modification of the game is more important. Assume that the recipient may offer the company a strictly more favorable contract, [m.sub.f], compared to the contract described in "The General Structure of the Payoff Functions", so [m.sub.f] > [m.sub.1]. (14) Then let the game tree in Figure 2 describe the players' feasible actions each year. Together with the payoff functions in "The General Structure of the Payoff Functions," above and the restrictions in "Specific Payoffs," this game tree describes the stage game that is repeated infinitely. Here we provide the intuitive explanation for the existence of a subgame perfect equilibrium where aid is disbursed when conditions are not implemented and describe its four necessary conditions. For the mathematical representation of the conditions and the proof of the equilibrium, see the Appendix.

Assume that the recipient applies a trigger strategy where it announces that it will only offer the favorable contract each year as long as the company complies with the recipient and influences the donor to keep the aid flowing even if the conditions are not implemented. Then, from the recipient's perspective, the company complies as long as it locates abroad if aid is not disbursed, but also if it locates at home when aid is granted. If the company does not comply in a particular year, the recipient will play according to the subgame perfect equilibrium of the stage game, and hence, offer the regular contract, [m.sub.1], forever after. In that case, the company loses the opportunity to earn the extra profit from the favorable contract in the future, and that serves as the recipient's punishment of the company for not complying.

Assume further that the company also applies a trigger strategy, under which it will only punish the donor for not giving aid as long as the recipient has offered the favorable contract in the previous years. Should the ordinary contract be offered, then this triggers the company into playing according to the subgame perfect equilibrium of the stage game forever. Thus, offering an ordinary contract implies that the company will never punish the donor in the future, which, in tam, removes the recipient's opportunity to get aid without implementing the conditions forever after.

The recipient and the company's trigger strategies are crucial in this game, but in order to explain the necessary conditions for the equilibrium of interest, we need to specify the full set of the players' strategies. So let one element in the players' strategy profile be to play according to the sequence (not implement, aid, home, [m.sub.f]), the equilibrium path, as long as no player deviates from this path. (15) Assume further that the other part of the players' strategy profile is that any deviation from the equilibrium path by any player leads to one of the two following paths. The first is (not implement, no aid, abroad, [m.sub.f]). This path can be termed "the donor-specific punishment path" because it is followed once only in the situation where the donor does not give aid and the company complies with the recipient and punishes the donor by locating abroad. In case this deviation from the equilibrium path occurs, the players revert to the equilibrium path in the next stage game. The second is the path that may be triggered by the recipient or the company from the trigger strategy specified above. If the company does not comply, that is, if it locates at home in a situation where no aid is given, or if the recipient does not offer a favorable contract, then this triggers the subgame perfect equilibrium of the stage game forever: (implement, aid, home, [m.sub.i]). This can be termed "the company-specific punishment path.'" Moreover, this path is triggered by any other deviation from the equilibrium path than the donor-specific punishment path, and hence, both the recipient's and the company's trigger strategies described above are contained in this strategy profile.

Now turn to the necessary conditions for these strategies to yield a subgame perfect equilibrium where the recipient succeeds in having aid without implementing the conditions. It is straightforward to see that if maintaining conditionality is very important for the donor while the company's location is not, then aid may not be disbursed if the conditions are not implemented even if the company should punish the donor for not giving aid. In this case, there is no scope for the recipient to use the company to influence the donor. Hence, the first necessary condition is that if the conditions are not implemented, then the donor must be better off in disbursing aid and having the location in its own country, compared to not giving aid and losing the location abroad. Thus, the first condition is

(16) Condition 1: D[h([a.sub.1], [c.sub.0]), [s.sup.*]] > D[h([a.sub.0], [c.sub.0]),[s.sup.u]].

The assumption that donor self-interest may overshadow other goals in foreign aid policy seems to be well documented for several large contributors of foreign assistance, as noted earlier.

The second condition is that the company must be willing to take the cost of rendering its tax base to another country in case the donor does not disburse aid. If the company is not willing, then the recipient has no means of influencing the donor. So, assume that the recipient has not implemented the conditions and the donor has rejected disbursing the aid. Then two important scenarios can be depicted.

In the first scenario, the company chooses not to comply with the recipient, and hence, continues to render its tax base to the donor's country. Then the company can secure itself the payoff from the subgame perfect equilibrium of the stage game each year, because non-cooperation triggers this outcome forever.

In the second scenario, the company complies with the recipient and punishes the donor. Hence, the company follows the donor-specific punishment path and incurs the cost that particular year when locating abroad. This secures the company [m.sub.f] that stage game, as the recipient's strategy is to offer a favorable contract as long as the company complies. Then, according to the strategy profile, all will play according to the equilibrium path forever after, which results in the highest feasible yearly payoff to the company.

Thus, if the recipient is to get aid without implementing the conditions, then the company must be better off in the latter scenario. If this is the case, then the company would actually carry out the punishment of the donor in case no aid is given. Hence, the second condition implies that the company must be sufficiently patient to await the future reward of cooperating: The discounted payoff to the company from following the subgame perfect equilibrium of the stage game must be lower than the total discounted payoff from punishing the donor once for not giving aid and then "'being rewarded" with the favorable contract (and locate at home) forever after:

(17a) Condition 2: [1/1 - [delta.sup.C]]C([m.sub.1], [s.sup.*]) < C([m.sub.f], [s.sup.u]) + [[delta.sup.C]/1 - [delta.sup.C]] C([m.sub.f], [s.sup.*]),

where [delta.sup.C] is the company's discount factor. Then the contract must be of such a value that [delta.sup.C] > [??.sup.C] where

(17b) [??.sup.C] = [[m.sub.f] - [m.sup.1] - [s.sup.*] + [s.sup.u]/[s.sup.*] - [s.sup.u]]

is the critical discount factor from Inequality 17a. In other words, the second necessary condition implies that the company's extra profit from the favorable contract must be sufficiently large to allow for punishment once, given the company's discount factor. The intuition is that the company will maintain cooperation with the recipient and punish the donor once, if warranted, in order to continue to get a share of the recipient's gain from having aid without implementation.

Now turn to the third necessary condition. Assume that all behave according to the recipient's intentions and follow the equilibrium path, which implies that the recipient has not implemented the conditions, the donor has given aid in fear of being punished, and the company has located at home. In this case, the recipient can grab a rent by offering an ordinary contract, [m.sub.1]. Hence, by walking away from the stated promise of offering a favorable contract, the recipient can achieve both the gains from having aid without implementing the conditions and save itself the costs of offering the company a favorable contract. However, this triggers the subgame perfect equilibrium of the stage game forever after, because the company requires the favorable contract to cooperate. Hence, the third condition is that the favorable contract must not be so costly for the recipient that it is worthwhile grabbing the rent by offering an ordinary contract when the other players follow the equilibrium path and then having the outcome of the subgame perfect equilibrium forever after, compared to not deviating from the equilibrium path every year:

(18) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

which implies that [delta.sup.R] must satisfy [delta.sup.R] > [??.sup.R]. Note that a "costly contract" in this setting could imply a high value of the favorable contract in terms of money, but also that the recipient is impatient.

The same reasoning can be used to see that the recipient must also be better off to follow the donor-specific punishment path in case the conditions are not implemented, aid is not given, and the company complies and locates abroad. This fourth condition arises from a similar situation to that described in the previous paragraph: The recipient can grab a one-period surplus by offering [m.sub.1] instead of following the donor-specific punishment path and offering [m.sub.f]. The former offer triggers the subgame perfect equilibrium of the stage game forever, whereas the latter results in the equilibrium path being played forever. Hence, the final condition is that offering [m.sub.f] must yield a higher total discounted payoff to the recipient than offering [m.sub.1] in a situation where the company has punished the donor for not giving aid:

(19) Condition 4: R([m.sub.f], [a.sub.0], [c.sub.0]) + [[delta.sup.R]/1 - [delta.sup.R]] R([m.sub.f, [a.sub.1], [c.sub.0]) > R([m.sub.1, [a.sub.1], [c.sub.0]) + [[delta.sup.R]/1 - [delta.sup.R]] R(m.sub.1], [a.sub.1], [c.sub.1]),

where [delta.sup.R] is the recipient's discount factor, [delta.sup.R] [member of] <0,1>, which implies that [deltasup.R] must satisfy [delta.sup.R] > [??.sup.R], where

(19b) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

is the critical discount factor for Condition 4. This may not be very different from the third condition, and we show in the Appendix that these conditions become identical if the recipient's utility loss of offering a favorable contract compared to an ordinary contract is constant.

So, if these four conditions are satisfied, the recipient starts out by declining to implement the conditions. Then the donor gives the aid, the company does not alter its optimal location of tax base. and the company is granted the favorable contract. Hence, in our subgame perfect equilibrium, the yearly payoff to the recipient, the donor, and the company will be R([m.sub.f], [a.sub.1], [c.sub.0]), D[h([a.sub.1], [c.sub.0])[s.sup.*]), C[m.sub.f], [s.sup.*]), respectively. If the donor were to deviate from the equilibrium path and refuse to disburse the aid in a given year. the company would punish the donor that year by shifting its tax base to another country. Then the players, in accordance with their equilibrium strategies, revert to the equilibrium path the following year.

This result sheds light on what may happen in these relations over time. The recipient ensures that the company puts pressure on the donor by offering a share of the gain attained by getting aid without implementing conditions. Note also that the utility of the donor if it did not involve itself in giving aid would have been

(20) [D.sub.j] = D[h([a.sub.0], [c.sub.0]), [s.sup.*]] > D[h([a.sub.1], [c.sub.0]), [s.sup.*]].

Hence, the donor would, in fact, be better off if it had never entered the arena of aid at all. This is due to the fact that the donor believes it is a waste of money to give aid when the conditions are not implemented but is forced to disburse the aid because of the fear of losing some tax income abroad.

4. Aid Institutions for Preventing Strategic Behavior

Despite the abstraction from a number of issues relevant for the interaction between donors, recipients, and private firms at the development scene, our model offers new insight into the failure of aid conditionality. When building aid institutions, an important question to the donor is how to make the recipient implement controversial conditions when there is scope for strategic behavior. Three suggestions emerge from the above analysis and are only briefly noted here: delegation of the disbursement decision, governmental guarantee schemes, and cooperation between the donor and the company.

Delegation to an agent with different objectives is often used as a means of avoiding commitment problems. Svensson (2000c) shows that delegating responsibility to a donor agency with less aversion to poverty than the donor can be optimal when the donor is unable to commit itself. In our model, however, it is only required that the agency to which the responsibility is delegated does not have interests competing with the concern for the aid/conditionality decision. It follows from the structure of the infinitely repeated game that, as long as the agency is indifferent with respect to the company's tax-base decision, there is no scope for the recipient's strategic behavior.

Another opportunity for the donor to enforce conditionality is to offer a governmental guarantee that the company shall not become worse off from engaging in the recipient's country than the equilibrium payoff. (16) Take the infinitely repeated game first. Once the company has been chosen as the constructor of the road, the donor can guarantee compensation equal to the company's loss if it is denied the favorable contract. Such a scheme is credible without a contract as long as the donor is better off redeeming the guarantee than being forced to give aid without the conditions implemented. On the margin, the donor would set the monetary value of the guarantee, denoted G. equal to the monetary loss for the company if the recipient carries out its threat and punishes the company. Hence, in the infinitely repeated game, the company must at least be guaranteed compensation for the favorable contract, so G [greater than or equal to] [m.sub.f] - [m.sub.1] must be satisfied every year. Then, if paying G to the company is better for the donor than taking the loss of giving aid without conditions being implemented, which is [D.sub.a] - [D.sub.c], the guarantee scheme is credible. In addition, if the donor and the company are from the same country, this could be set up in an enforceable contract, and hence, the guarantee would be credible without the donor's condition for a credible scheme being satisfied. Either using a guarantee or a contract ensures that the recipient's threat to influence the company would not have a bite.

The last suggestion on mitigating the problem of a strategic recipient could be for the company and the donor to agree on speaking with one voice on issues involving the recipient. It is straightforward to show that there will be no scope for strategic behavior in such a setting as long as a joint body between the donor and the company attaches sufficiently low weight to the company interests. Keeping in mind the close ties between many of the large private firms with interests in developing countries and their governments as exemplified by tied aid, such a constellation does not seem too unrealistic.

5. Concluding Remarks

Though earlier work has pointed toward time inconsistencies of altruistic donors as an explanation for the failure of aid conditionality, this article sheds light on failure arising from company influence on the disbursement decision. Supported by the institutional setting in which the donor--recipient relationship is embedded, our model suggests that recipients can influence the decision on granting aid through introducing commercial interests into the game. Such strategic behavior can enable a recipient to avoid implementing the conditions attached to the aid.

It is also found that a donor subjected to credible threats could, in fact, be better off by never involving itself in the aid relationship. When donors perceive that giving aid when the conditions are not implemented is equivalent to wasting the money, they would rather withdraw over time when forced by private companies to give aid when the conditions are not implemented. Even if players' preferences are constant in game-theoretical models, we believe that donors might be less eager to give aid over time if they are subjected to the type of strategic behavior modeled here. If correct, the trend toward reducing the level of aid (World Bank 1998a) during the past decade is not surprising in light of our model.

These problems of strategic behavior arise because of the recipient taking advantage of the donor's many competing interests, and the model offers some theoretical solutions. One possible solution to this problem is to give governmental guarantees to companies that involve themselves in the recipient country. Another way to deal with this particular strategic behavior would be to delegate the disbursement decisions to an agent that is confined to maintaining the prevailing disbursement policy. Finally, by creating a joint body for donor and company interests toward the recipient, conditionality could also be maintained.

Appendix: The Subgame Perfect Equilibrium in Infinitely Repeated Games and the Proof

The Strategy Profile

Suppose that Equation 16 is satisfied and that [delta.sup.R] > [??.sup.R], [??.sup.R] and [delta.sup.C] > [??.sup.C] as defined in Equations 17 through 19. Then, the following complete strategy profile constitutes a subgame perfect Nash equilibrium:

Equilibrium Path. The players' sequence of actions is, at stages 1 to 4, respectively: not implement conditions, give aid. locate in donor country, offer favorable contract; ([c.sub.0],[a.sub.1], [s.sub.*], [m.sub.f]).

Any deviation from the equilibrium path implies that one of the following punishment paths is followed:

* If the donor deviates from the equilibrium path and plays "no aid" after the recipient has played "not implement conditions": Play ([s.sup.u], [m.sub.f]), hence the donor-specific punishment path ([c.sub.0], [a.sub.0], [s.sup.u],[m.sub.f]) is played once. Then return to the equilibrium path.

* Any other deviation, by any player, from the equilibrium path or the donor-specific punishment path: All agents play according to the stage game outcome ([c.sub.1], [a.sub.1], [s.sup.*], [m.sub.1]) forever.

The Proof

No player will deviate from the subgame perfect equilibrium of the stage game once triggered. Thus, neither the donor nor the company will deviate from the company-specific punishment path if triggered.

The donor will never deviate from the equilibrium outcome if the payoff when being punished once for not giving aid when conditions are not implemented, D[h([a.sub.0], [c.sub.0]), [s.sup.*]], leaves the donor worse off than giving aid without the conditions being implemented and thereby avoiding punishment. D[h([a.sub.1], [c.sub.0]), [s.sup.*]]. This follows directly from the assumption in Equation 16.

The company will never deviate from the:

* donor-specific punishment path as long as [delta.sup.C] > [??.sup.C] is assumed. Assume that the donor plays (no aid), which, according to the strategy profile, implies that the donor-specific punishment path is played once. If the company deviates from this path and locates at home, it will achieve a total profit of (1/1 - [delta.sup.C])C([m.sub.1], [s.sup.*]) because this triggers the subgame perfect equilibrium of the stage game forever (according to the strategy profile). However, to follow the donor-specific punishment path after (no aid) is observed yields C([m.sub.f], [s.sup.u]) + ([delta.sup.C/1 - [delta.sup.C])C([m.sub.f], [s.sup.*]). Because [delta.sup.C] > [??.sup.C] implies that Equation 17 (1/1 - [delta])C([m.sub.1], [s.sup.*]) < C([m.sub.f], [s.sup.u]) + ([delta/1 - [delta])C([m.sub.f], [s.sup.*] is satisfied, the company will never deviate from the donor-specific punishment path.

* equilibrium path, both because it is strictly better off when having [m.sub.f] when the equilibrium path is followed, compared to [m.sub.1] in the subgame perfect equilibrium of the stage game, and because there is a cost for the company in locating abroad. Moreover, the company achieves its highest feasible profit in this game when the equilibrium path is played, so there exists no gain to the company from any deviation from the equilibrium path.

The recipient will never deviate from the:

* equilibrium path at stage 4 as long as the total payoff from following this sequence infinitely, (1/1 - [delta.sup.R])R([m.sub.f], [a.sub.1], [c.sub.0]), is larger than the payoff from deviating by offering the company an ordinary contract. This deviation from the equilibrium path yields R([m.sub.1], [a.sub.1], [c.sub.0]) in the deviation year and R([m.sub.1],[a.sub.1, [c.sub.0]) forever after because the strategy profile implies that the subgame perfect equilibrium of the stage game is triggered. Because [delta.sup.R] > [??.sup.R] satisfies Equation 18, (1/1 - [delta.sup.R])R([m.sub.f], [a.sub.1], [c.sub.0]) > R([m.sub.1], [a.sub.1], [c.sub.0]) + ([delta.sup.R/1 - [delta.sup.R])R([m.sub.1], [a.sub.1], [c.sub.1]), the recipient will never deviate from the equilibrium path at stage 4.

* equilibrium path at stage 1 by implementing the conditions because, according to the strategy profile, this triggers the stage game forever. To deviate from the equilibrium path at stage one is worse for the recipient than not to deviate: R([m.sub.f], [a.sub.1], [c.sub.0]) > R(m.sub.1], [a.sub.1], [c.sub.c]), from Equation 18.

* donor-specific punishment path as long as [delta.sup.R] > [??.sup.R]. Assume that (not implement, no aid, abroad) has been played, which implies that the recipient should offer [m.sub.f] according to the donor-specific punishment path. However, assume that the recipient deviates from this path and offers an ordinary contract, [m.sub.1]a According to the strategy profile, this triggers the stage game equilibrium forever, and the total pay off to the recipient is thus R([m.sub.1], [a.sub.0], [c.sub.0]) +([delta.sup.R]/1 - [delta.sup.R])R([m.sub.1], [a.sub.1], [c.sub.1]). If the recipient does not deviate from the donor--specific punishment path, it would have R([m.sub.f], [a.sub.0], [c.sub.0]) + [delta.sup.R]/[delta.sup.R])R([m.sub.f], [a.sub.1] [c.sub.0])instead because offering a favorable contract implies that the company and the recipient continue to cooperate (according to the strategy profile). Because [delta.sup.R] > [??.sup.R] satisfies 19, R([m.sub.f], [a.sub.0], [c.sub.0]) + ([delta.sup.R]/1 - [delta.sup.R])R([m.sub.f], [a.sub.1], [c.sub.0]) > R([m.sub.1], [a.sub.0], [c.sub.0] + ([delta.sup.R]/1 - [delta.sup.R])R([m.sub.1], [a.sub.1], [c.sub.1]), the recipient will never deviate from the donor-specific punishment path. Note that the donor-specific punishment path is followed once (no aid) is observed at stage 2, so it is not a feasible action for the recipient to deviate from the donor-specific punishment path at stage one. QED.

On the Recipient's Discount Factor and the Cooperation

Because both Equations 18 and 19 specify a lower bound for the recipient's discount factor, we know that the recipient's time preferences must satisfy the largest of the two factors in order for the recipient to be willing to carry out the specified actions in the strategies. Thus, for the equilibrium to exist where the recipient gets aid without implementing the conditions, the recipient's discount factor must be larger than Equations 18 and 19. Unfortunately, we are unable to find an analytical solution to which factor is the largest. However, a straightforward solution is within reach if we assume that the cost of offering the favorable contract compared to the ordinary contact has the same impact on the recipient's utility irrespective of the value of the other variables.

So assume that, ceteris paribus, R([m.sub.1], a, c) - R([m.sub.f], a, c) = x. In that case, we can see that Equations 18 and 19 represent the same condition:

(A1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(A2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

However, as shown in the proof, it is not necessary to restrict the recipient's utility function for our equilibrium to exist.

Let us contrast the proof with Hatlebakk (2002). There are two important differences. First, we let the third party keep the favorable contract when cooperating on the donor-specific punishment path, which seems more realistic than giving the third party an ordinary contract along this path. Second, we specify the condition that the first mover must not be better off by offering an ordinary contract once the equilibrium path is started (see Eqn. 19). A parallel to this condition is also necessary for the solution in Hatlebakk (2002) but not stated explicitly. As we have shown in the section "The Failure of Conditionality in the Infinitely Repeated Triadic Game," it is not sufficient to assume that the player that offers the favorable contract does not offer so favorable terms that this player loses the whole surplus from making the threat credible (i.e., making the third party comply). The equilibrium also requires the restriction that this player does not grab the rent from offering an ordinary contract once the equilibrium path is started.

I am grateful for valuable comments by Kaushik Basu, Rune Jansen Hagen, Magnus Hatlebakk, Sanjay Jain, Jens Josephson Ravi Kanbur, Erik Sorensen, and Bertil Tungodden. This article was developed during my stay with Kaushik Basu and the Department of Economics. Cornell University. and I thank them for their hospitality. Thanks also go to participants at NEUDC 2001 for useful comments on an earlier version of this article, titled "Credibility. conditionality and strategic behavior: A new explanation for the failure of aid conditionality."

(1) There is a discussion in the literature about whether or not "good" macroeconomic policies are necessary for aid to induce economic growth. In addition to Burnside and Dollar (2000), see Hansen and Tarp (2000, 2001) and Dalgaard and Hansen (2001).

(2) The literature on international relations offers anecdotal evidence of a number of situations where commercial third parties are drawn into the game between two agents. See Gwynne (1983) for an account of how a large American company put pressure on a bank to grant a loan to a Philippine company so that the American company would get a contract with the Philippine company. This three-party relationship is formalized in Deshpande (1999). For other examples, see Cohen (1986), Darity and Horn (1988), Jain (1986), and Wellons (1986).

(3) Most models of economics are dyadic, which means that all the agents interact pairwise. In a triadic model, an agent i not only takes account of his relationship with agent j. but also takes account of his own and agent j's relationship with a third agent k. See Basu (2000) for a discussion on dyads and triads.

(4) The main difference between our approach and Basu (1986) is that Basu alms at explaining how a landlord in a rural economy may be able to use a third party (a merchant) to extract a larger surplus from a laborer than would be possible without the third party's influence. Our focus is on foreign aid and, as opposed to Basu's model, we do not operate explicitly with markets. However, the important triadic structure is the same as the rural setting in Basu (1986).

(5) For other principal-agent models that discuss foreign aid, see Pedersen (1996) on how it is crucial for the donor to have the first mover advantage if aid is to increase investment, or Pedersen (2001) for an illustration of how the adverse incentive effects of aid may cause poverty to increase due to a Samaritan's dilemma problem. See Drazen (2000) for a survey of the political economy of foreign aid. A related problem to the failure of conditionality is that aid can be fungible, see. for example, Hagen (2001).

(6) Policy conditions can yield a change in relative prices, reduced government spending, and reduced absorption to achieve external balance, which, in turn. may hurt or benefit different groups in a country (Swensson 2000b; Summers and Pritchett 1993).

(7) For other models with recipients as first movers, see Pedersen (1996), Svensson (2000c), and Hagen (2001),

(8) It could be more confusing than clarifying to display all possible payoffs for all players in these games, so only the payoffs necessary to illustrate the important equilibria of the games are described.

(9) The amount of bilateral aid that is to be granted to a particular recipient can also be determined outside the donor agency, for instance, in a parliamentary committee. This is only a simplifying assumption that does not influence the results.

(10) Note also that we let the subscripts of the utilities of each player follow in alphabetical order to indicate the ranking of that player's utility.

(11) The bold lines in Figure 3 show the optimal choice for each agent at each stage, so the subgame perfect equilibrium path of the stage game is easily traced.

(12) The threat is never credible if the company's actions at stage 3 do not matter for the decision of whether or not to build the road at stage 4. Note the interesting case where we assume that the price for the road can be interpreted as the opportunity cost of having the road built, where this cost decreases the resources at the disposal of the government, and thus, decreases aid. The opportunity cost is then a function of aid, a, denoted m(a) and where [partial derivative]m/[partial derivative]a < 0, Assume that [a.sup.*] is the amount of aid that makes the recipient indifferent between building the road or not, ceteris paribus. Now, the interesting scenario arises if the aid promised to the recipient is larger than the critical value, a > [a.sup.*], and if [a.sup.*] > 0. Thus. if the recipient does not get aid, this will imply that the recipient will not build the road, irrespective of the company's actions (as this is the optimal choice for the recipient at stage 4 in the game). Then the company knows that it will not get the road contract no matter what actions it takes at stage 3. Hence, it will always choose to locate at home, as this is the optimal choice in this situation. Then the donor knows that the recipient is not able to influence the company, so the donor will not be punished for not giving aid in a situation where the recipient has abstained from implementing the conditions. As long as a > [a.sup.*] and [a.sup.*] > 0 (note that [a.sup.*] increases when the favorable contract is offered in the infinitely repeated game), then the results from the infinitely repeated game will be sustained, thus, the threat is credible. Note also that if a < [a.sup.*], then there will be no scope for strategic behavior, as the recipient will not under any circumstances build the road. I am grateful to an anonymous referee for pointing this situation out for me.

(13) Such a result might also arise in more short-term relationships. One example of a credible threat in a one-shot game arises if we allow for a simultaneous coordination game with one good and one bad equilibrium to be played between the recipient and the company after the fourth stage in the original stage game. The details are available from the author upon request.

(14) See Hatlebakk 2002 for the idea of offering a favorable contract in the triadic structure. See the appendix for the differences between the two theoretical solutions.

(15) See Abreu (1988) for the specification of strategies in terms of paths

(16) One empirical example of governmental guarantees for national companies that engage in poor countries can be found in the Norwegian Guarantee Institute.

References

Abreu, Dillip. 1988. On the theory of infinitely repeated games with discounting. Econometrica 56:383-96.

Alesina, Alberto, and David Dollar. 2000. Who gives foreign aid to whom and why? Journal of Economic Growth 5:33-63.

Basu, Kaushik. 1986. One kind of power. Oxford Economic Papers 38:259-82.

Basu, Kaushik. 2000. Prelude to political economy: A study of the social and political foundations of economic. New York: Oxford University Press.

Burnside, Craig, and David Dollar. 2000. Aid, policies, and growth. American Economic Review 90:847-68.

Cohen, Benjamin. 1986. In whose interest? International banking and American foreign policy. New Haven: Yale University Press.

Collier, Paul. 1997. The failure of conditionality. In Perspectives on aid and development, edited by Catherine Gwin and J. Nelson. Washington, DC: Overseas Development Council, pp, 51-77.

Dalgaard, Carl-Johan, and Henrik Hansen. 2001. On aid, growth and good policies. Journal of Development Studies 37:17-41.

Darity, William, and Bobbie L. Horn. 1988. The loan pushers: The role of commercial banks in the international debt crisis. Cambridge, MA: Ballinger Publishing Company.

Deshpande, A. 1999. Loan pushing and triadic relations. Southern Economic Journal 65:914-26.

Dollar, David. and Jacob Svensson. 2000. What explains the success or failure of structural adjustment programmes? Economic Journal 110:894-917.

Drazen, Allan. 2000. Political economy in macroeconomics. Princeton: Princeton University Press.

Gwynne, S.C. 1983. Adventures in the loan trade. Harper's Magazine 267:22-6.

Hagen, Rune. 2001. Aspects of the political economy of foreign aid. Mimeo. Foundation for Research in Economics and Business Administration, Norway.

Hansen, Henrik, and Finn Tarp. 2000. The aid effectiveness disputed. Journal of International Development 12:375-98.

Hansen, Henrik, and Finn Tarp. 2001. Aid and growth regressions. Journal of Development Economics 64:547-70.

Haflebakk, Magnus. 2002. A new and robust subgame perfect equilibrium in a model of triadic power relations. Journal of Development Economics 68:225-32.

Jain, Arvind K. 1986. International lending patterns of U.S. commercial banks. Journal of International Business Studies 17: 73-88.

Kanbur, Ravi, 2000. Aid, conditionality and debt in Africa. In Foreign aid and development: Lessons learnt and directions for the future, edited by Finn Tarp. London: Routledge, pp. 409-23.

Mosley, Paul, Jane Harrigan, and John Toye. 1995. Aid and power: The World Bank and policy based lending. London: Routledge.

Oxfam, 1995. The Oxfam poverty report. London, England: Oxfam Publications.

Pedersen, Karl Rolf. 1996. Aid, investment and incentives. Scandinavian Journal of Economics 98:423-38.

Pedersen, Karl Roll 2001. The Samaritan's dilemma and the effectiveness of foreign aid. Journal of International Tax and Public Finance 8:693-703.

Summers, Lawrence H., and Lant H. Pritchett. 1993. The structural-adjustment debate. American Economic Review 83: 383-9.

Svensson, Jacob. 2000a. Why conditional aid doesn't work and what can be done about it? Mimeo. Institute for International Economic Studies, Stockholm University.

Svensson, Jacob. 2000b. Foreign aid and rent-seeking. Journal of International Economics 51:437-61.

Svensson, Jacob. 2000c. When is foreign aid policy credible? Aid dependence and conditionality. Journal of Development Economics 61:61-84.

Trumbull, William N., and Howard J. Wall. 1994. Estimating aid-allocating criteria with panel data. Economic Journal 104: 876-82.

Wellons, Philip A. 1986. Multinational institutions and the debt crisis: National interests and long term consequences. In World debt crisis: International lending on trial, edited by Michael P. Claudon. Cambridge, MA: Ballinger Publishing, pp. 147-69.

World Bank. 1992. World Bank structural and sectoral adjustment operations: The second OED review. Washington, DC: Operations Evaluation Department Report No. 10870.

World Bank. 1998a. Assessing aid: What works, what doesn't, and why. New York: Oxford University Press.

World Bank. 1998b. World development indicators. Washington D.C.: World Bank.

Espen Villanger, Chr. Michelsen Institute, N-5892 Bergen, Norway; E-mail [e-mail].

Received April 2003; accepted March 2004.