Why do some policies in international organizations (IOs) reflect member states’ interests, while others echo IO staff’s bureaucratic interests? IOs, especially international financial institutions (IFIs), established since the 1940s, have long been criticized as a tool of US hegemony. The International Bank for Reconstruction and Development (IBRD), now a major organization within the World Bank, and the International Monetary Fund (IMF), were founded as a result of the Bretton Woods Agreement led by the United States. These two institutions, along with the General Agreement on Tariffs and Trade (GATT)/World Trade Organization (WTO), shaped the post-World War II economic world order under the leadership of the United States. During the Cold War, the United States could afford to reconstruct Western Europe and take a superior position over its rival, the Soviet Union, thanks, in part, to these IOs (Vreeland 2006). From this perspective, characterizing IFIs as “servants” to a superpower seems justifiable. This characterization of IFIs aligns well with the realist theoretical tradition in international relations that states are the main actors in international politics and that IOs are only a reflection of the global distribution of power (Waltz 1979; Mearsheimer 1994/1995). According to realism, IOs are “epiphenomenal” and subordinate to a superpower, serving its interests (Mearsheimer 1994/1995).

International public administration (IPA) scholars have recently raised doubts about the realist or member state-centric view of IOs. They argue that IO staff are also deserving autonomous actors in world politics when they pursue their own bureaucratic interests, as they alone have the ability to significantly influence decision making in IOs (Vreeland 2006; Vaubel 1996; Vaubel et al. 2007). Since the end of the Cold War, the role of IOs has become more salient concerning a wide variety of global issues, such as poverty and inequality, environmental problems, communicable disease pandemics, and human rights concerns, which cannot be solved through the efforts of a single country. With this growing importance of IOs in addressing diverse global issues, states have delegated some authority to IO staff members, especially concerning the management of day-to-day operations.

A potential problem with this authority delegation may arise when the preferences of IO staff conflict with those of member states (Hawkins et al. 2006; Barnett and Finnemore 1999). It is natural to assume that member states would seek to pursue their national interests in IOs, while IO staff members would seek to provide global public goods and/or maximize organizational and bureaucratic benefits. Equipped with greater knowledge, expertise, and inside information, staff members can shape design and implementation of policies, and their performance of day-to-day international administrative procedures is a way to realize their own preferences. As states are unable to closely monitor staff behavior, the asymmetric information and authority over daily tasks are likely to cause some “agency slippage (Hawkins et al. 2006).” This is in sharp contrast to the state-centric realist characterization of IOs, as IO decisions are likely to reflect staff preferences.

To bridge the theoretical discord between realist and IPA scholarship, this study attempts to explain the dynamics of decision making in IOs based on the principal-agent framework (Bernheim and Whinston 1986; Copelovitch 2010b; Hawkins et al. 2006). We begin with An IO consisted of collective principals (multiple member states) and a common agent (the IO bureaucracy). Member states generally influence decision making by casting a vote or reaching a consensus, depending on the specified procedures, while IO staff exert influence by setting the agenda in meetings or providing advice based on their professional knowledge and superior information. We contend that when member states and staff have conflicting preferences yet major members maintain converging preferences, the staff’s attempt to influence decision making will be less likely to produce a noticeable difference. This is because an unfavorable proposal prepared by the staff is not likely to be passed by the member states that reserve the authority to make a final decision. Conversely, if member preferences are divergent, the staff are more likely to succeed in manipulating the policy in their favor. Therefore, staff will attempt to engage in and steer member states’ deliberations in decision making by setting the agenda and preparing favorable policy proposals. In sum, the direction of IO policy and the extent to which IO staff preferences materialize are conditional on whether members’ preferences are divergent or convergent.

We conducted two comparative case studies of World Bank reform cases in 2010 and 2018 to examine the proposed argument. By comparing the explanatory power of our main thesis to that of possible alternative hypotheses, we were able to confidently rule out the rival hypotheses. Therefore, the empirical findings of this study provide support for the theoretical argument. In 2010, the convergence of G6 preferences led to the World Bank’s policy of increasing support for low-income countries (LICs) to prevent further deterioration of the global economy after the 2008 financial crisis. The content of the 2010 reform was inconsistent with the World Bank staff preference, as it significantly diminished the World Bank’s expected profits acquired by lending to middle-income countries (MICs). By contrast, in 2018, G6 preferences diverged. The World Bank staff stressed the importance of a capital increase and turned members’ attention away from the issue of more tightly regulating lending to MICs. As a result, the largest capital increase in World Bank history was agreed to, while significant reform to reduce agency slippage did not take place. These two cases clearly demonstrate the merits of the theoretical argument.

We proceed with the research in four steps. In the following section, we review studies of decision making in IOs and highlight the discrepancies between the two main theoretical traditions. In the theoretical section, we elaborate on the principal-agent framework and how it applies to the World Bank. In the fourth section, we describe the case selection and variable operationalization. We draw supporting evidence from the two selected cases for the principal-agent argument. In the conclusion, we discuss the main academic and policy contributions and implications.

What are the factors that contribute to IO decision making including critical actors in the decision-making process? We recap here three possible factors from existing literature: the institutional setting, the influence of IO staff, and the influence of an hegemon and other major state members.

IO staff and member states interact to translate societal demands into policy outputs in the form of organizational decisions (Eckhard and Ege 2016, 965; Easton 1957; Luhmann 1968; Scott 2003). It varies by specific institutional settings, but generally speaking, delegates from member countries have the right to participate in decision making by casting votes or reaching a consensus, while the staff indirectly steers decisions by preparing meetings, setting agendas, and providing advice through policy reports. These decisions range from ordinary tasks, such as project and program approvals and evaluations in IFIs, to major decisions, such as organizational reforms.

2.1. The Institutional Setting of IOs

The institutional setting of IOs may affect various decisions. As IOs are created by agreements among multiple countries that seek to benefit from them, the setting that prescribes the decision-making rule is an important feature. If an IO had no formal rules and procedures, a powerful country could hold sway over most decisions (Koremenos et al. 2001; Lall 2017; Haftel and Thompson 2006; Hawkins et al. 2006). To prevent this, most IOs have formal decision-making procedures. One example of such a safeguard is a voting system. Voting power can lessen or encourage the unruly behavior of a few powerful countries. In an egalitarian voting system (one-nation-one-vote), all countries have equal voting power, which makes a developing country’s ballot more meaningful. Conversely, in the one-dollar-one-vote system, voting power is determined by the subscription shares a country contributes to the organization. The United States has more than 15% of the total votes at the World Bank, while Afghanistan has only 0.05%, which means Afghanistan’s impact on decision results is hardly significant (Bayer et al. 2015). In addition to voting power allocations, the voting requirement for approval may affect whether a proposal passes or not. Proposals are passed more easily through a simple majority voting than a supermajority voting rule. In the case of approvals that need a majority vote at the Asian Infrastructure Investment Bank (AIIB), an alliance of a few powerful shareholders theoretically can pass them.

While the institutional setting, such as the voting system, can stymie or facilitate decision making, it does not indicate why some decisions are made, and others are not. To explain the dynamics, it is necessary to consider the preferences and strategic choices of the involved actors. Some studies have demonstrated that the voting patterns of developing countries in the United Nations (UN) tend to be influenced by the vote-buying strategy of powerful countries (Woo and Chung 2017; Kuziemko and Werker 2006). Considering that the actor’s strategic choices can make influence independent of the institutional setting, we briefly summarize the literature on the influence of the two main actors involved in IO decision-making.

2.2. The Influence of IO Bureaucrats

Recently, IPA scholars have argued that the IO itself reserves the capability and incentives to influence IO decisions. IOs have long been regarded as tools of superpowers or as mediators without independent preferences (Waltz 1979; Mearsheimer 1994; Keohane and Martin 1995; Keohane 1984). However, a number of studies have indicated that not only do IO staff play an important role in encouraging states to internalize global norms (Barnett and Finnemore 2004; Finnemore and Sikkink 1998), but they also have their own preferences over policies that might significantly depart from the collective will of member states (Vetterlein 2012; Trondal et al 2015).

The independent preferences of IO staff can lean either toward service for the global public good or toward self-interested profit maximizing. In the first case, staff are devoted to fighting common problems faced by humanity (Eckhard and Ege 2016, 971; Xu and Weller 2008, 49; Biermann et al. 2009; Juncos and Pomorska 2013). For example, a main goal of the World Bank since the 1970s has been poverty reduction (Vetterlein 2012). This preference is often consistent with the national interests of member states but not always. In the second case, staff unhesitatingly pursue organizational interests and seek to maximize their profits using delegated authority (Bauer 2012). For instance, IMF and International Development Association (IDA) staff tend to rush to extravagant lending right before quota reviews and replenishment for the sake of budget maximization (Vaubel 1996). Moreover, IO staff sometimes inexplicably increase the total number of staff to share the workload (Vaubel et al. 2007). Such self-interest contradicts member states’ expectations of staff.

Even if staff have no voting power in decision making and must rely on revenues acquired from member states, they can push their preferences via administrative procedures and the information they monopolize. Information about day-to-day operations is not open to the public, and this information monopoly allows staff to select which knowledge will be disclosed (Benner et al. 2009) and thereby influence the country’s deliberations on decisions. In addition, IO staff can manipulate decisions through their delegated authority: they “propose new programs, projects, regulations, and other policies; [have] the ability to draft the annual budget […] and […] to prepare the work program for IO governing bodies” (Lall 2017, 254). Thus, staff can nudge member countries in certain directions.

However, this explanation of IO staff influence overlooks the formal voting procedures in decision making. Despite attempts by staff to manipulate decisions, proposals must gain the approval of member countries. If a staff member submits a proposal against the member countries’ collective will, they can simply reject it. Thus we turn to member countries’ influence, especially one by the more powerful state members of IOs.

2.3. The Influence of Hegemon and Other Powerful States

In IOs, member states are the most powerful players in decision making, casting votes, or reaching consensus. This is consistent with the long-lived claim by realists that states are the sole main actors in international politics (Waltz 1979) and that IOs are only a reflection of the global distribution of power (Mearsheimer 1994). According to this tenet, IOs’ policy decisions should echo hegemonic preferences, as attested to by numerous empirical studies. For example, affinity with the United States in UN voting brings about favorable conditions for a country in IO resource allocation. Voting patterns that follow those of the United States in the UN increase the probability of loan approval for the country by the World Bank where the United States holds overwhelming influence (Kilby 2009), minimize the gap between commitment and disbursement (Kilby 2013), and raise the amount of IDA loans to the country (Andersen et al. 2006). Moreover, a country with a temporary seat on the UN Security Council, whose resolution is salient for the justification of military actions, attracts more lending from both the World Bank and the IMF because the United States wants to win a favor from temporary members of the UNSC (Dreher et al. 2009a; 2009b). Furthermore, US commercial interests seem to correlate with the World Bank’s lending decisions (Fleck and Kilby 2006). These studies indicate that IOs are exploited by the hegemon wanting to pursue its national interests through IOs.

However, the mechanism by which the hegemon makes arbitrary determinations, even when opposed by other stakeholders, and why policy sometimes reflects staff preferences remain unclear. Given the institutional setting, even the hegemon cannot unilaterally determine policies in structured IOs, as there are other stakeholder countries pursuing their national interests. Yet, the literature often ignores their influence. Moreover, this argument does not offer an explanation for why some IO decisions reflect staff preferences when they are at odds with those of the hegemon. Considering the limitations of the three mainstream theories, we introduce an alternative theory in the following section.

The principal-agent theory provides a useful framework for understanding IO decision making while factoring both member states and IO staff into the policy making process. The principal-agent problem arises when an agent carrying out a delegated task goes against the wishes of principals to pursue their own interests (Nielson and Tierney 2003, 246; Copelovitch 2010b, 55; Kiewiet and McCubbins 1991). For instance, a fund manager (agent) is less wary of investment risks, while an investor (principal) desires to at least secure their initial investment. The fundamental problem is that by entrusting its agent with daily tasks, the principal has limited information and incomplete monitoring mechanisms. The principal can minimize agency slippage through screening prior to hiring the agent, or by devising either direct “police patrol oversight” or indirect “fire alarm oversight” through third parties (Nielson and Tierney 2003, 246).

While managing agency slippage in a one-on-one relationship is a relatively simple task, a situation in which collective principals and a common agent exist becomes complicated due to the principals’ diverse preferences. For example, an elected politician is a common agent of voters and should represent their interests, in principle. If they put self-interest before the voters’ collective will, the voters’ converging preferences can punish the politician in the coming election (Ferejohn 1986). However, if voters’ preferences are divergent, or polarized, the agent’s self-interest may correspond with some voters’ preferences. Then, the probability of the politician’s survival increases, as the politician equipped with better information and policy instruments can attempt to exploit the split among voters.

We contend that the same logic can be applied to IO decision making. An IO is a common agent of multiple state members. In principle, the staff should sacrifice for the principal’s prosperity. Yet, the staff is a rational actor seeking self-interest. By controlling information disclosure and performing administrative procedures, staff members have room to steer decisions and manifest their preferences should member states’ preferences diverge (Vreeland 2006, 37–41).

Meanwhile, it is the member states that pass staff proposals. Staff cannot change policy themselves. As in the politician–voter relationship, each member state agrees or disagrees with a suggested proposal. For example, some states pursue their commercial interests or security, while others value altruism or environmental concerns (Alesina and Dollar 2000; Berthelemy 2006). Given the multifaceted preferences associated with an issue, whether staff efforts make a noticeable difference depends on the degree of convergence of the principals’ preferences (Nielson and Tierney 2003; Copelovitch 2010a; 2010b; Breßlein and Schmaljohann 2013).

When IO staff has their own policy preferences that significantly depart from those of member states, and member states maintain converging preferences, the policy direction is more likely to reflect the preferences of the member states. In this case, approval of decisions reflecting the staff’s self-interest is not likely because the member states collectively would reject it, because final decisions are made through voting or consensus of the delegates from member states. Facing an objection by the members or expecting an objection, IO staff is likely to prepare proposals that echo member states’ preferences when the member states have converging preferences.

H1. When member states have converging preferences and IO staff does not agree, policies reflecting the member states’ preferences are likely to prevail.

By contrast, if member states’ preferences diverge, we anticipate that policy outcomes will reflect the staff’s preferences. In particular, the polarization of member states’ preferences allows staff to maximize their influence by exploiting their professional knowledge and administrative authority. Diverging preferences over an issue usually lead to long-term discussions to produce a consensus. The staff can manipulate members’ deliberations in subsequent discussions by selecting agendas that are favorable to them. In addition, staff can publish policy documents that implicitly or explicitly include their preferences as essential information sources for states when making decisions. Therefore, divergent preferences among member states are more likely to result in policy that reflects the staff’s preferences.

H2. Member states’ divergent preferences are more likely to lead to policy decisions that are reflective of IO staff.

4.1. Case Justification

We attempt to test the hypotheses with two reform cases within the World Bank. The World Bank is a great case to test our hypotheses for several reasons. First, it is a structured IO. Its membership consists of 189 countries whose delegates make decisions following formal procedures. At the same time, more than 10,000 staff members perform daily tasks such as agenda setting for meetings, preparation of documents, and implementation of projects. Second, unlike the clandestine decision-making procedures in most IOs, the World Bank discloses extensive information by allowing access to documents, which offer abundant credible resources for researchers tracing decision-making procedures. Third, the World Bank’s financial health allows the staff to have some degree of autonomy. While many IOs are highly dependent on members’ subscriptions, the World Bank makes a profit and holds enormous, retained earnings, even after paying out dividends to shareholders. This financial stability allows the staff to pursue their self-interest, as they are less wary of a sudden organizational collapse.

Among the myriad reform cases, we selected two that highlight the conflict of interests among states and between states and staff. Organizational reforms generally entail massive changes to organizational rules and structure. Among all possible reforms, one of the most contentious one is capital increase, as it raises member states’ financial contributions to the IO, which can trigger domestic oppositions. Hence, if governments agree with the capital increase, despite the potential domestic political risk, they will attempt to maximize their interests within the World Bank. Therefore, instead of examining all reform cases, we narrow our focus on reforms undertaken concurrently with capital increase. There have been only five capital increase decisions in the World Bank’s history. While three were endorsed in 1959, 1980, and 1988 during the Cold War period, two other cases were agreed to after the end of the Cold War. Given that the role and status of IOs were transformed after the collapse of the Communist Bloc (Dreher et al. 2013), the selection of two post-Cold War cases allows us to compare two different observations in the most similar settings.

4.2. Variables

The outcome variable policy direction in the hypotheses is operationalized by the amount of capital increase and the resource allocation policy. First, staff seek to increase capital as much as possible, while member countries are usually unwilling to support this unless it is in their national interest. As previously mentioned, the staff pursue either global common prosperity or their self-interest, including organizational influence and job security. One necessary condition for realizing these goals is the procurement of available, maximum-possible resources. Increasing the regular subscription amount through capital increase is an easy way to strengthen financial capability of the World Bank. However, the capital increase must be endorsed by a “three-fourths majority of the total voting power” (IBRD 2012) in the Board of Governors of the World Bank, the collection of all delegates of member states. Governors would not favor capital increase unless it is beneficial for their countries. Therefore, we assume that a high capital increase reflects the staff’s preference, while disapproval or the proper amount of capital increase denotes the member states’ preferences. In sum, staff wants larger capital increase while state members are reluctant to it.

Second, staff prefer to use funds in such a way as to maximize returns. For example, they expect more benefits from lending to MICs, as LICs are more vulnerable to economic and political turbulence. However, the Board of Directors, consisting of 25 executive directors from larger member states that oversee day-to-day decisions of the Bank, has the authority to have a final say in lending decisions (World Bank 2014). Among the 25 executive directors, six are appointed by G6 countries, while 19 are elected. Each executive director might attempt to adjust lending policy to pursue its national interests and create regulations to ameliorate agency slippage. Given this institutional setting, the creation of new regulations and tighter oversight mechanisms for a lending is consistent with member states’ preferences, while their repeal or the status quo reflects the staff’s interest. In sum, staff seeks to be able to lends more liberally especially to MICs while member states wants to maintain tighter regulations over lending decisions to minimize agent slippage.

The independent variable convergence of member states’ preferences is measured by G6 preferences. The G6 consists of the United States, Japan, China, Germany, France, and United Kingdom, the six largest members of the World Bank in the order of voting power as of 2021. Focusing on the G6 allowed us to avoid the arduous task of investigating all the member states’s preferences. First, the G6 has a large share of the total voting power, and thus they appoint their own executive directors. For example, the United States possesses more than 15% of the total voting power, and the sum of the G6’s voting power reaches over 40% in the IBRD. This large voting share of the G6 allows them to collectively exercise a veto over various important issues. Second, in contrast, none of them can unilaterally decide a policy, despite their strong voices. For instance, a three-fourths majority vote is necessary for the approval of capital increase, but even the United States possesses only slightly more than 15% of the total voting power. Thus, the United States alone is neither a veto power nor a unilateral decision-maker in the World Bank. Thus it is the collective preferences of these six members that are important.

The G6 countries have shown tendencies to actively promote their own policy preferences through IOs. Several studies have documented US vote-buying in IOs (Koziemko and Werker 2006; Woo and Chung 2017; Dreher et al. 2009a) and the influence of the UK and France on former colonies (Brysk et al. 2002; Nagtegaal and De Bruin 1994). In addition, due to their powerful economies, Germany and Japan are actively engaged in global affairs. While these five countries have usually been considered as the main actors in extant studies (Nielson and Tierney 2003; Copelovitch 2010a; 2010b; Breßlein and Schmaljohann 2013), the distinguishing point of this study is that we consider China, whose influence in IOs has often been discounted despite its growing importance in international relations. Not only does China represent the lending developing countries as a member of BRICS but it has been extending its influence globally, especially on the African continent (Carmody and Owusu 2007; Brookes and Shin 2006; Rotberg 2009). And China’s voting power ranks third among 189 members and trails only that of the United States and Japan. Thus, China deserves to be considered as a stakeholder in the World Bank.

To measure the degree of convergence, we categorize G6 preferences concerning capital increase and resource allocation policy into approval, qualified approval, and disapproval. These classification criteria for the two issues decrease the problem arising from the vague nature of member preferences. G6 preferences converge when they lean toward one side, while divergence occurs when their preferences are polarized. We classify the preferences declared in the meeting statements of the World Bank Group Development Committee Meeting (DC), a group of 25 representative governors who meet twice a year. After each meeting, the statements and related staff reports are disclosed to the public on the DC website. Therefore, analyzing these documents allows us to observe the preferences of each member state and the staff concerning the issues.

4.3. Research Strategy

To evaluate the validity of the hypotheses derived from the multiple principal and a common agent theoretic framework, we compare their explanatory power to rival explanations: the technocratic view , influence of IO staff, and influence of a hegemon. If some observed evidence cannot be better or equally well explained by the rival explanations, while the observation is consistent with our theory’s predictions, we can confidently rule out the rival hypotheses (Musgrave and Nexon 2018, 605; Bennett and Checkel 2014; Humphreys and Jacobs 2015; Zaks 2017). The main arguments and predictions of each theory are summarized in Table 1. In the next section, we analyze two World Bank reform cases to evaluate the competing explanations.

5.1. World Bank Reform in 2010

Reform talks in 2010 were precipitated by the 2008 financial crisis, which severely damaged the least developed countries (LDCs). LDCs were especially vulnerable to this economic shock originated in the United States, as they were highly dependent on trade and foreign aid from foreign countries (Presbitero and Zazzaro 2012, 1944; Allen and Giovannetti 2011). The steep increase in the price of food, petroleum, and necessities could have propelled citizens to participate in massive protests and riots in LDCs (Winters 2011, 57). In this situation, IFIs were assigned to prevent political instability as well as economic one in these countries. Figure 1 shows that the total amount of IBRD and IDA loans soared rapidly during the 2007–2010 period.

5.1.1. G6 Converging Preferences and Staff’s Preferences

Commensurate with the World Bank’s crucial role, G6 preferences converged on the extension of World Bank lending to LICs to ameliorate the negative consequences of the global financial crisis. In documents on the Post-Crisis Direction, member countries suggested several goals. The first objective involved “targeting the poor and vulnerable countries” (DC 2010a). In line with this goal, the G6 argued that the World Bank should switch its focus to fragile, conflict-affected states by increasing IDA loans and grants for LICs. Moreover, the G6 urged the World Bank to help LICs manage the risks and prepare for future crises (DC 2010b).

The G6 governors’ statements reveal the converging preferences in detail. They unanimously endorsed the World Bank’s new role in providing more grants and concessional loans to LICs that had poor access to private capital markets. The American governor Timothy Geithner claimed in the Development Committee that “a top priority for all of the Multinational Development Banks (MDBs) must be support for the poorest” (DC 2009a), not just in quantity but in quality of loans, to lift up the lives of the poorest, ameliorate climate change, and foster collective security (DC 2010c). The Japanese governor Naoki Minezaki chimed in that “differently from the way we support MICs, we need to secure appropriate concessional resources, such as IDA, in assisting LICs” (DC 2009b). The German governor Heidenmarie Wieczorek-Zeul took the same position, saying, “It is […] all the more important that there are no funding gaps, particularly for low-income countries” (DC 2009c). In addition, he urged staff to be cautious, saying that “economic stimulus packages in developed countries must not be implemented at the expense of the developing countries” (DC 2009d). The French governor Christine Lagarde encouraged the World Bank “to continue to accord the highest priority to poverty reduction, and thus deploy […] its interventions for the benefit of the poorest” (DC 2009e). The British governor Douglas Alexander called on “governors to ensure that the poorest countries are not left behind in these difficult times” (DC 2009f). Lastly, the Chinese governor, Xuren Xie, urged the World Bank to stand at the developing countries’ side, saying that it “should always be innovative in providing developing countries flexible and user-oriented services” (DC 2009g). In summary, all the G6 governors agreed that the World Bank should concentrate economic recovery efforts to LICs.

All welcomed the capital increase, despite the contracting economy, but Japan and France added a condition. After the United States had reviewed the financial capacity of all the MDBs (DC 2009h), the American governor expressed a positive opinion of the capital increase, claiming that “the Bank will be more indispensable and effective […] with a stronger capital base” (DC 2010c). The German governor emphasized the World Bank’s role in making “sure that the long-term impacts of the crisis can be cushioned” (DC 2009c) and the need for the capital increase, saying that “it can continue to provide developing countries with the necessary funds” (DC 2009c). The British governor claimed that “new capital for IBRD also benefits the poorest countries through secure and increased transfers to IDA” (DC 2010d). China, one of the beneficiaries of World Bank loans, welcomed the capital increase, calling “for timely implementation of the […] general capital increase proposals” (DC 2009g).

In contrast to the unwavering endorsement by these four countries for capital increase, Japan and France supported the idea with reservations. Japan was wary of the possible shrinkage of its voting power after Voice Reform, saying that the World Bank “should take into full consideration the cumulative contributions to IDA of current donors (because Japanese contributions to IDA is one of the largest)” (DC 2009b) and “that will provide a supportive basis for Japan’s contributions to a general capital increase in IBRD” (DC 2009b). Japan’s main concern was to remain the second-largest voting power. The French governor urged the World Bank to assure that enhancing capital inflows “effectively shore up the financial capacity of the World Bank […] before contributing additional resources” (DC 2009i). Table 2 summarizes the G6 positions on the two main issues during the 2010 reform talks.

In contrast with the G6’s converging preferences to heavily focus on low income countries, the World Bank staff had lent a massive amount of money to middle income countries. Figure 1 shows a steeper increase in commitment to MICs than to LICs during the 2007–2010 period. The staff preferred MICs to LICs as borrowers for a number of reasons. First, this was because the risk of default was much lower among MICs. Second, lending to MICs was more profitable to the World Bank (Birdsall 2017; Lerrick 2006; Saldinger 2019), despite its relatively low interest rates, while IDA grants or concessional loans to LICs were unprofitable. In addition, if IDA resources dried up, the retained IBRD (and IFC) earnings could be transferred to the IDA (Clemens and Kremer 2016). Therefore, more IDA grants meant lower profits for the World Bank. In sum, the staff’s preferences were at odds with the G6’s converging preferences and their active lending to MICs possibly indicated agency slippage.

5.1.2. Results of the 2010 Reform

The converging preferences of the G6 brought about the result that was favorable to them, leaving little room for the staff to steer their deliberations. Noticeable changes included capital increase, Voice Reform, Access to Information (AI), and modernization of lending procedures.

In accordance with the G6 preferences, capital increase was agreed to (DC 2010e). Had the G6 demanded that the World Bank increase IDA commitments without financial reinforcements, while reducing IBRD lending, it would have severely shrunk the IBRD’s profits and made transfers from IBRD to IDA impossible. Therefore, the capital increase was inevitable in 2010. Without intense opposition, the staff expedited the resources review to calculate the deficit (DC 2010e). As a result, the member countries promised 5.1 billion dollars in paid-in capital increase for IBRD and 200 million dollars for IFC.

The second notable change was Voice Reform. The voting power of some developing countries had leaped so dramatically that their preferences could be more reflected in the World Bank decisions. For example, the voting power of developing and transitional countries grew by 3.1% to 4.7%, while the voting power of the G6 countries was cut slightly (DC 2010f). China experienced the largest voting power increase from 2.78% to 4.42%, and Vietnam’s voting power more than doubled from 0.08% to 0.20%. In contrast, US voting power declined from 16.36% to 15.85%, and Japan’s voting power fell by 1.01% to 6.84%. This change was consistent with the G6’s converging, albeit reluctant, preference to listen to developing countries’ voices more.

In addition, a revolutionary information disclosure policy called AI was devised. As agency slippage is enabled by asymmetric information between member states and the staff, member countries had long sought to mitigate this asymmetry by disclosing hidden information the staff had monopolized. Although several previous reforms, such as “Project Information Document” and “Public Information Center” had improved organizational transparency (Nielson and Tierney 2003, 263), the guidelines on information disclosure remained vague (OPCS 2009). This ambiguity left room for the staff not to disclose information. The newly introduced AI policy allowed access to all information with clearly defined exceptions. In addition, the third-party “Access to Information Committee” was established to resolve information disclosure conflicts. If member countries considered staff refusal of information disclosure to be unreasonable, they could challenge it through the committee. Furthermore, “Open Data,” “Open Finances,” and the “Open Knowledge Repository” maximized AI and significantly alleviated asymmetric information (World Bank 2015).

Finally, a modernized, integrated lending framework was introduced. For example, the World Bank carried out a revision of the “Investment Lending” procedure, which covered most of its loans. The purpose was to improve loan efficiency by establishing an “Operational Risk Assessment Framework” (ORAF) (DC 2011). With this framework, the probability of approval of high-risk loans was minimized. Moreover, “Program-for-Result” (PforR) financing was designed. The step-by-step PforR disbursement depended on “the achievement and verified disbursement-linked results specified as” disbursement-linked indicators (World Bank 2017). Thus, the staff and recipient country had to carefully design and prepare the project to achieve the defined goals.

The AI and the newly introduced lending framework reduced the staff’s discretion and strengthened the oversight by member states. This is obviously something that the staff would not desire. These strengthened oversight by member states and the reduced staff discretion allowed the member states’ preferences prevail and resulted in more projects being approved for LICs, as the G6 preferred, than for MICs in the immediate aftermath of the 2010 reforms.

To sum up, the 2010 reform resulted in capital increase to a reasonable extent, strengthening of developing countries’ voting power, maximization of AI, and modernization of lending procedures. These were closely related to the G6’s collective will to aid LICs and minimize agency slippage. Figure 1 provides evidence that the total commitment to MICs decreased following the reform as of 2011, while Figure 2 shows that IDA commitments soared. In other words, the G6’s converging preference to carry out reforms remedied agency slippage.

Additionally, the staff’s attempt to manipulate decisions did not make a noticeable difference in the face of the G6’s converging preferences. In a report, the staff contended that the role of the World Bank in building infrastructure in MICs was a priority, especially in response to dwindling international investments (World Bank 2010). If the staff had pushed for the reform focused on MIC needs, it would not have succeeded, due to the G6’s focus on assisting LICs. Therefore, the staff did not have any other option but to conform to the G6’s preferences and publish various documents favorable to them concerning capital increase (DC 2009j), the reform of the World Bank’s internal governance (DC 2009k), and the LIC debt sustainability framework (DC 2009l).

5.2. World Bank Reform in 2018

Even after the massive reforms in 2010, a large amount of money was still flowing into upper-middle-income countries (UMICs), such as China, Brazil, Mexico, and Turkey. Figure 3 depicts IBRD and IDA commitments from 2011 to 2017. China is the second-largest recipient in the graph. Two other UMICs are also listed among the top 10. This would suggest that despite the G6’s efforts to provide more loans and grants to LICs, the World Bank as a whole has provided more financial assistance to MICs, indicating possible agency slippage, despite the increase oversight.

Against this backdrop, reform talks began again in 2015. The staff claimed there was a need for another capital increase to achieve the Sustainable Development Goals, as they lacked the necessary resources. In other words, the two same concerns resurfaced: capital increase and resource allocation policy.

5.2.1. G6 Diverging Preferences and Staff Preferences

Unlike the previous reform talks, G6 preferences concerning the two issues diverged in 2018. The United States was dissatisfied with the existing lending pattern and disagreed with the capital increase. In the 2016 DC meeting, Jacob J. Lew, the American governor under the Obama administration, said that the IBRD should provide “knowledge and technical advice” (DC 2016a) to higher-income countries, rather than money. This dissatisfaction turned into fury under the Trump administration. In 2017, Steve Mnuchin argued that “continued high lending to countries above the World Bank’s own graduation threshold is problematic since it diverts limited World Bank resources to countries with substantial access to other sources of finance” (DC 2017a). In addition, he said that “more can be done to optimize the World Bank’s balance sheet” and did not deem “the original schedule for considering the World Bank’s capital position as necessary or realistic” (DC 2017a). This rhetoric specifically targeted China. During the presidential election campaign in Fort Wayne, Donald Trump said that “we can’t continue to allow China to rape our country” (Stracqualursi 2017).

China was annoyed with the US statements. Despite its standing as the world’s second largest country, China preferred to continue receiving IBRD loans due to the favorable interest rate and, more importantly, the staff’s development knowledge. The statement by Governor Jiwei Lou reflected China’s preference that “with the help of the WBG [World Bank Group], the positive spillovers of MICs’ economic development will bring benefits to LICs” (DC 2016b). Furthermore, the Chinese governor Jie Xiao stated in the 2017 DC meeting that “the WBG needs to increase its financial strength and lending capacity” and “render stronger support to the Belt and Road and other initiatives” (DC 2017b). Therefore, China wished to remain an IBRD beneficiary and to harness the World Bank’s knowledge for China-led projects.

The others’ preferences were also diverged. Japan took a negative position on the capital increase, while it showed qualified approval for continuous lending to UMICs. The Japanese Governor Taro Aso said, “Proposals on capital strengthening measures other than capital increase […] are also needed” (DC 2017c). Japan spurned the capital increase to avoid the subsequent voting power adjustment, which could result in a net loss in voting power. Furthermore, he said that “with regard to the assistance towards the” UMICs, “IBRD should focus its scarce public resources on targeted interventions in […] Global Public Goods” (DC 2017c). Germany supported the capital increase, as it would help fulfill the implementation of the Climate Change Action Plan (CCAP) (DC 2016c; 2017d), and conditionally agreed to lending to MICs (DC 2016c). France blamed the staff for not focusing on LICs, while it expressed qualified approval for the capital increase. The French Governor Michel Sapin stated in the 2017 DC meeting that “we need to continue those discussions [of the capital increase] in a pragmatic frame of mind, examining all the options available so that […] we can agree on a consistent set of measures that will lead to greater financial sustainability” (DC 2017e). Finally, the British governor Priti Patel conditionally supported the capital increase and lending to UMICs, saying that “the UK expects consideration to be given to all ways to increase the financial sustainability and capital adequacy […], including pricing and efficiency measures, further balance sheet optimization as well as a possible capital increase” (DC 2016d). She further claimed that UMICs “should focus more on partnering with the Bank’s knowledge […] with a particular emphasis on global and regional public goods” (DC 2017f). To sum up, the G6 preferences could be divided into approval, qualified approval, and disapproval. Table 3 summarizes the G6’s divergent preferences.

With these divergent preferences, the staff actively attempted to steer decisions by including their preferences in policy reports. In contrast to the various issues covered by the reports published during the previous reform talks, the need for the capital increase was the main subject of the documents (DC 2017g; 2017h; 2017i; 2017j). In these reports, the staff stressed that the World Bank would lack resources, even after implementing other measures to improve the staff’s financial efficiency. In contrast to the ample justification of the capital increase, there were limited reports on the regulation of IBRD lending. Instead, the staff claimed that many poor people exist in MICs and attempted to highlight the needs for loans flowing into poorer regions in MICs by arguing that they were crucial for poverty reduction (Clemens and Kremer 2016). In addition, Jim Yong Kim, the president of the World Bank, also stated, “the lessons we learn in China […] are very helpful to our work in other developing countries” (Allen-Ebrahimian 2017).

5.2.2. Results of the 2018 Reform

With the divergence of the G6’s preferences, the staff’s attempts worked well, and the largest capital increase ever in World Bank history was attained. This unprecedented scale was idiosyncratic, given that there was no financial crisis and the United States, as the largest shareholder in the World Bank, expressed opposition in the early stages of the talks. Taking advantage of this divergence, the staff set capital increase as the main agenda item in consecutive meetings and emphasized its importance in reports. In addition, Jim Yong Kim attempted to orchestrate opposition through his diplomatic efforts. He established a relationship with the Trump administration by launching the Women’s Empowerment Fund and aligning with his daughter, Ivanka Trump (Donnan and Fleming 2018). As a result, member countries agreed to an increase in paid-in capital to 7.5 billion dollars for IBRD and 5.5 billion dollars for IFC (World Bank 2018). This massive capital increase in 2018 resulted from the staff’s enthusiastic engagement when faced with the G6’s diverging preferences.

The staff was reticent concerning the lending restriction to UMICs, and the reform resulted in no substantive change other than loan price differentiation based on the recipient’s income. For example, Graduate Discussion Income countries that were encouraged to graduate from IBRD or IDA were to pay higher maturity premiums (Bank Directive 2018). However, this would not fundamentally solve agency slippage. Rather, this new policy prompted staff to seek self-interest by providing profitable loans to UMICs. Moreover, the increased loan price did not impede UMICs from borrowing loans from the World Bank. First, despite the soaring price, the loan conditions remained more favorable than the interest rates of private capital. Second, some UMICs, such as China, focused on development knowledge transfers. For these reasons, the loan price differentiation did not ameliorate agency slippage. Therefore, staff continued to offer a large number of loans to rich areas in China for mysterious capacity building, even after the reform. As this was not aligned with the intended regulation, it indicates ongoing agency slippage (Saldinger 2019). To briefly recapitulate, the 2018 reform drastically enlarged the World Bank’s financial capacity and increased the expected profitability of interest payments from UMICs instead of constraining lending to them. Both of these reflect the staff’s preferences.

5.3. Alternative Explanations

5.3.1. Technocratic View

According to the technocratic view, the World Bank is an apolitical institution, and its policies are based on macro-economic principles (Knight Santaella 1997; Bird and Rowlands 2003; Joyce 2004; Copelovitch 2010a, 5). This theory provides a plausible explanation of the 2010 reform case. After the global financial crisis, the expected role of the World Bank was to back vulnerable LICs whose domestic instability could have damaged the global economy. To accomplish this, a capital increase and proper lending management system were implemented. However, the largest capital increase in 2018 cannot be adequately explained by this view. Until 2018, the total amount of the World Bank’s commitment has been in decline, and there has been no severe economic crisis. Therefore, this apolitical, technocratic perspective is limited in its ability to explain the two reform cases.

5.3.2. Influence of IO Staff

While the result reflecting the staff’s preference in the second case can be well explained by this theory, the tightened restrictions on themselves in the first case remain unaccounted for. In the second case, the staff’s ability to intervene in the decision-making process led to successful manipulation of the decisions. By contrast, many proposals inconsistent with the staff’s preferences were reported during the 2010 reform talks. In this case, the staff were passive servants of the member states, which implemented a strict lending framework and vexing AI to maximize the principals’ access to information and oversight. This alternative thus does not help us in understanding the outcome of the 2010 reform.

5.3.3. Influence of a Hegemon (and Other Powerful States)

Finally, the traditional realist explanation anticipates that IO policy reflects the hegemon’s preference because IOs are only a tool to realize its global strategy. In 2010, the reform direction was consistent with the US preference, yet the contents of the reform in 2018 contradicted the US aversion to the capital increase and lending to UMICs. Despite the hegemon’s initial opposition, the World Bank’s financial capacity was significantly strengthened, and the staff could expect more profits from giving loans to UMICs due to the loan price differentiation. Thus, the inexplicable second case is at odds with the realist explanation.

This research attempted to compare the validity of principal-agent theory with rival theories in explaining decision-making patterns in IOs using the World Bank reforms as cases. While the three competing theories provide limited explanations, the observed evidence of the two examined cases is consistent with the expectations of our main theory. In the first case, the G6’s preferences converged on support for the capital increase and LICs. At that time, the staff rarely demonstrated their power through agenda setting and document publication. As a result, a capital increase, an adjusted lending framework,and maximized disclosure of hidden information were agreed to. The latter two changes reduced the staff’s discretion and in effect ameliorated agency slippage.

By contrast, in the case of 2018 World Bank reform, the G6’s preferences were polarized, and staff took the lead in DC meetings. They emphasized the capital increase, while remaining laconic on the regulation of lending to UMICs. This attempt of the staff under the divergence of G6’s preferences resulted in the largest capital increase in the World Bank’s history and loan price differentiation that improved profitability. This evidence supports the principal-agent theory, which hypothesized that the direction of decision making would be determined by the degree of convergence of the member countries’ preferences.

One of the main contributions of this study is our attempt to broaden the theoretical understanding of decision making in IOs by filling the gap between realist and IPA scholarship. Unlike other theories, the principal-agent theory considers the interaction between states and IO staff and excludes no one. Using a simple variable, this parsimonious theory provides a more plausible explanation of IO decision making and constructs a bridge between the two very different disciplines. Another contribution is the study’s focus on the World Bank, which scholars have paid little attention to, despite its importance in the international political economy. As the largest multilateral development bank in the world, the World Bank is often recognized as an apolitical economic institution, while political interaction is more apparent in the IMF, the international lender of last resort. Nevertheless, the existing literature shows that World Bank decisions reflect the dynamics of international politics (Kilby 2009; 2013; Andersen et al. 2006; Dreher et al. 2009a; 2009b; Fleck and Kilby 2006). In this study, we elaborate on the idea by defining the main actors and their preferences. Finally, the point that distinguishes our research the most from existing studies is the inclusion of China. To date, studies of IOs have tended to exclude China, despite its growing importance. This regional power has attempted to exert its influence globally through intense engagement in IOs. We therefore conclude that China should be considered a main actor commensurate with its global position.

This study also has policy implications for the ongoing “IO crisis.” Nationalistic sentiment has been rising in recent years around the world, and the existence of some IOs is threatened. The North Atlantic Treaty Organization (NATO); United Nations Educational, Scientific and Cultural Organization (UNESCO), from which Donald Trump unilaterally declared withdrawal; and World Trade Organization (WTO) face such a threat. However, the principal-agent theory counterintuitively suggests that this crisis could be an opportunity for IO staff to exert their influence when member countries’ opinions are polarized.