Economics of Sovereign Wealth Funds

Chapter 8 Regulating a Sovereign Wealth Fund Through an External Fund Manager

Udaibir Das, Adnan Mazarei, and Han Hoorn
Published Date:
December 2010
  • ShareShare
Show Summary Details
André de Palma, Luc Leruth and Adnan Mazarei1 

Only a short while ago, sovereign wealth funds (SWFs) were seen by many as the newest malignant instrument of state-sponsored capitalism on the global financial stage. SWFs, with large amounts of assets projected to rise rapidly on the backs of rising commodity prices and widening global imbalances, were expected to invade and occupy financial markets and hold them hostage for the unfriendly objectives of their home governments. Little empirical evidence supports these concerns and, in fact, SWFs have generally behaved noncontroversially. In international forums where these issues are debated, many observers were not surprised by SWFs’ benign behavior and expressed serious doubts when others feared that SWFs would pursue objective functions vastly at odds with the objectives of other economic actors operating in the host country, and in particular, that profit maximization would not be at the core of their strategies.2 Yet, the concerns were sufficiently strong to have led to calls from many quarters for more regulation of SWF operations through a number of means (e.g., Kimmit, 2008).

Fortunately, with improved information, persistent communication efforts by SWFs explaining their activities, the preparation of a voluntary set of best practices (known as the Santiago Principles; IWG, 2008), and efforts by various international organizations (principally the IMF and the Organisation for Economic Co-operation and Development) to clarify the rules and regulations governing SWFs’ investments, tempers have cooled. Perhaps more important, with the global financial crisis unfolding, other issues have taken center stage. Also, SWFs have lost a portion of their portfolio value as a result of the global decline in asset prices, and many have turned inward in an attempt to save their domestic economies through increased support for local banks or purchases of domestic assets.

So, was the debate about SWFs all about nothing? The authors of this chapter do not think so. Although many of the fears about sovereign funds have been exaggerated, SWFs do pose important issues for regulation, as foreign investors and as key players in the economies in which they operate, and also, more simply and irrespective of the location of their activities, as players whose characteristics are not fully known to other economic actors. That many SWFs turned inward during the crisis is certainly not an argument against looking into these issues more deeply. In parallel, the recent crisis has heightened global concerns about risk management of all financial institutions, especially systemically important ones. Some SWFs have already started to make their comebacks and, as the crisis recedes, their operations in foreign markets are likely to accelerate quickly, which will tend to revive concerns and misgivings in recipient countries.

Before moving further, this section details some of the key issues surrounding SWFs’ operations.3 Concerns in recipient countries about SWFs have generally involved three broad issues. First is ownership of SWFs by foreign sovereigns, which may have noncommercial, perhaps strategic, motives for investment. Sovereign ownership generally raises the role of governments in economic activity, with a concomitant distortionary impact on business decisions. Second is their size, which may unduly affect asset prices and financial stability in host countries. Third is the transparency of their operations. Notwithstanding the decline in SWFs’ foreign-based operations during the recent global financial crisis, clearly the issue of governance by the SWF of its operations in countries receiving SWF investments will remain relevant for a long time, if only because some key SWF operations are locked in.

SWFs have also raised important issues in their home countries although these concerns differ substantially from those in recipient countries. Irrespective of its location, an SWF can be of one or a combination of the following types: stabilization fund, savings fund, reserve management corporation, development fund, or contingent pension reserve fund (IMF, 2008). Accordingly, even when they operate at home, SWFs raise a number of concerns for their sovereign owners.4 Unfortunately, the principal-agent (P-A) model is not well suited to analysis of an SWF’s operations at home and, while there may be some lessons to be drawn, they are not the focus of this chapter.

Rather, this chapter concentrates on the concern that SWFs’ operations may distort the decision-making processes and governance of the foreign firms they invest in (though not unique to SWFs), thereby harming the national interest of the recipient country—even leaving security-related concerns aside. At the root of this concern is the lack of trust among all parties in the objectives (the utility functions) pursued by the others. This concern implies that perceptions and the difficulty in interpreting the actions of any party will affect the interactions between the SWF and the recipient country, which may lead to a lack of trust that is not necessarily justified. This chapter thus looks in detail at the possibility that SWFs invest in recipient countries with legitimate (i.e., not related to strategic or “unfriendly” motives) objectives that are hard to interpret because they are not necessarily associated with short-term profit maximization. For example, an SWF could invest to learn more about a certain business. In that case, it is difficult for authorities in the recipient country to accurately interpret the signals sent by the SWF through its actions.

This chapter examines the relationships between SWFs and their recipient countries, concentrating on the dependence of that relationship on the nature of the objectives pursued by the SWFs, using a P-A framework. In particular, when an SWF has multiple objectives, signals can be misinterpreted, leading to misguided reactions by authorities in the recipient country. Thus, hard-to-interpret signals do not provide a sufficient case for the imposition of constraints on the SWF. However, the chapter shows and discusses later that requiring the SWF to invest through intermediary asset managers may foster cooperation, especially when the objectives of the SWF and of the authorities are closely aligned. An SWF may also alleviate concerns in a recipient country by acting as an investor for (and accepting the funds of) other SWF and non-SWF investors.

The next section of this chapter discusses the foreign operations of SWFs that may create the need for regulation through an external fund manager. The subsequent section provides the outlines of an analytical framework, and is followed by a section discussing the scope for regulation under different assumptions about the information set of the agents involved. The penultimate section complicates the basic model set out earlier by introducing agents with multiple objectives, thus increasing the risks that signals are misinterpreted by recipient countries. The final section concludes.


For a variety of reasons, countries regulate financial domestic companies. Upon becoming active in a foreign country, some or all of the activities of an SWF also become subject to regulations imposed by the recipient government. Although generally encouraging foreign investment, many recipient countries tend to be more suspicious of SWF activities than they are of domestic firms, because they fear that the SWF may somehow find a way to circumvent the regulatory process, even if not actively trying to avoid scrutiny.

Hence, various propositions have been made to regulate SWFs, including imposing upper limits on ownership or voting rights, prohibiting SWFs’ investments in sensitive areas, and subjecting SWF operations to a specific set of procedures. The recipient country experiences a clear trade-off, however. If an SWF brings economic benefits to the recipient country (as most have argued), limiting its room for maneuver through more regulations than typically applied to domestic firms engaged in similar activities may not be wise—in addition to the extra costs to the regulatory authority (and the compliance costs to the SWF), the SWF will face additional constraints (some arguably not necessary) and, if there are too many hurdles, it could even forego the opportunity of operating in that country altogether.

A more subtle idea that has received support is to ask an SWF to invest through a fund manager in the recipient country (Gibson and Milhaupt, 2008). By introducing an additional layer between the SWF and the companies in which the SWF’s money is to be invested, the hope is that the probability of detrimental activities (be it to the companies or, more generally, to the country) would be reduced for two reasons: the scope for collusion between the SWF and the target company, or undue influence by the SWF on that company, would be more limited; and investment funds tend to be more tightly regulated than other commercial activities. Another approach for allaying the concerns of recipient countries about the possible noncommercial motivations of SWFs has been initiated by Temasek of Singapore. Temasek has taken steps to establish a more general fund that would also take investments from other SWFs and non-SWF investors. Through this approach, Temasek and a few of its non-SWF partners could signal to recipient countries that their investments involve resources from diverse investors with different political interests, thus tamping down concerns about any noncommercial objectives of their investments.5

The next subsections look at the proposal to use an external manager by considering the case of an intermediary fund in the recipient country. P-A theory is used for this analysis. Broadly speaking, both the regulatory authority (or a trusted domestic investor) and the SWF become the principals of the recipient-country fund manager (the agent). As is standard in contract theory, the strategy of each principal is to induce the agent to act in such a way that the resulting output maximizes the principal’s utility. In doing so, the principals interact with and learn more about each other. Thus, the principals compete in trying to influence the actions of the agent in a way that could protect the interests of the recipient country (the principal represented by the regulatory authority or a domestic investor) and of the sovereign owner (the principal represented by the SWF). The analysis demonstrates that a critical consideration in determining whether the approach is likely to have the desired results is the extent to which the objectives of the domestic regulator and an SWF can be accurately interpreted and, if they can, whether these objectives are complementary or antagonistic.


Theoretical Considerations

The discussion now turns to a simple description of the model used to develop the ideas and the policy arguments. The model, inspired by Martimort (2006), will not be solved here. The framework is a simple way to sketch the case of one agent being supervised by several principals (the model is limited here to two principals) and is consistent with the typical model used in the field.

Let us assume that two principals (Pi, i = 1,2) have an objective function Si(q), where q = q(q1,q2) is the output of the agent. In return for the agent’s output, each principal makes a transfer ti to the agent. The utility Vi of each principal is given by

The utility U of the agent is given by the transfers received from both principals minus the effort it makes to produce q. It is further assumed that the agent is efficient with an intensity θ, so that the cost C(q) of producing q affects the agent’s utility as follows:

where the term θC(q) can be interpreted as the opportunity cost to the agent of producing q—or in managing a company F on behalf of its principals in such a way that F produces q. The term θ can be interpreted as the agent’s type (e.g., efficient or not) and this parameter may or may not be known to the principals. As detailed below, the analysis can then refer to cases of perfect or imperfect information, respectively. The model appears in Figure 8.1 (which does not include company F). The cost function C(q) is supposed to be known to the principals.

Figure 8.1An SWF Invests through a Recipient-Country Fund Manager

Source: Authors.

This formulation of the utility function of the agent is very general. In fact, the analysis allows for the possibility that each principal Pi enters into a contract with the agent on a specific component of the output that matters to that principal. This quantity is called qi (i = 1,2) with q = q(q1,q2). A further assumption that principal P1 (resp. P2) can only observe q1 (resp. q2), and is not able to observe q2 (resp. q1) (the output going to the other principal), yields the case of private agency. If the contract is on the entire output q = q(q1, q2), it is public agency. In both cases, as discussed in the next subsection, the principals interact with each other via the contracts they offer to their common agent. The analysis shows that the ability of the regulatory body (or the recipient-country fund manager) to constrain the SWF depends on whether the agency is private or public. For the cost function, this distinction is not necessary because the agent is assumed to have perfect information about it. Yet, C(q) can play a role in the results, as discussed later.

Considerations on the Basic Assumptions

The main elements of the model relevant for the purpose of this chapter’s thesis follow:

  • First, who are the principals? Because the P-A model with two principals assumes a transfer to the agent, the regulatory authority cannot technically be associated to a principal. Instead, it should be assumed that the “domestic” principal is a recipient-country financial institution in which the authorities have full trust while the other principal is the SWF. This simplification could affect the results, because a P-A relationship would also exist between the regulatory authority (which would become an additional, main principal as shown in Figure 8.2) and the recipient-country fund manager (the domestic principal); the recipient-country fund manager is, in fact, an agent of the regulatory authority. Similarly, there would be a P-A relationship between the recipient-country fund manager (the agent in Figure 8.1) and the domestic company F in which money is ultimately invested. Just as in the other case, the recipient-country fund manager—agent of principals 1 and 2—would itself become one of the principals of company F. Yet another realistic complication would be to take into account the P-A relationship that necessarily exists between the SWF and its sovereign owner (not included in Figure 8.2). Hence, the domestic principal is associated to the “government,” or to the “recipient-country fund manager,” as the case may be. The simple situation presented in Figure 8.1 would then become as in Figure 8.2.6 Unfortunately, these complex features cannot be included in the model because cascade P-A relationships have not been studied (although Mookherjee [2006] provides a qualitative discussion of such models). The analysis also neglects firm F, the domestic company, as a player.
  • Second, who knows what? With complete information (see “Perfect Information” section later in this chapter), the parameter θ is known to both principals (see Figures 8.1 and 8.2). Both principals knowing θ corresponds to a situation in which the regulatory authority feels confident that, through the recipient-country financial institution acting as domestic principal, it knows the characteristics of the agent or is in a position to monitor it closely enough to be well informed about its behavior. This may not always be the case because the relationship between the regulatory authority and the domestic principal is not as straightforward as has been assumed (as discussed in the previous bullet) or because there is less than perfect information (as discussed in the next bullet). More important, and contrary to the simple assumption of perfect information, the SWF is unlikely to know an agent unilaterally selected by the regulatory authority as well as the economic actors of the recipient country know the agent, implying an asymmetry between principals, which, to the authors’ knowledge, has not been studied in the literature. However, it could be realistic to assume that the SWF is allowed to choose an agent from a list of candidates deemed acceptable by the regulatory authority of the recipient country (perhaps a fund manager with ties to the SWF’s country), which would restore some balance of information between the two principals. If the agent is sufficiently well known, the assumption of perfect information holds. In addition, there is an aspect of information that involves sending messages to and properly reading the intentions of the other players (which can be interpreted correctly or incorrectly). This situation is discussed in detail in the section titled “P-A Players with Multiple Objectives.”
  • Third, situations of imperfect information present additional issues.7 With imperfect information, the agent is not well known to either principal and therefore will retain some advantage in its ability to extract surplus from both principals as they compete for the agent’s services. The regulatory authority will also face uncertainty about the characteristics of the SWF. This situation appears to be a more realistic assumption than q being known. For example, because q is an aggregate measure, it is reasonable to assume that a principal may be fully aware of some characteristics of the agent and ignorant of others, making the principal’s aggregate perception imperfect. In addition, as will be shown, this model allows a related issue to be explored, specifically, the extent to which antagonistic or complementary utility functions belonging to the principals can affect the outcome. It is important because an agent could be very good at certain tasks and not as good at others while the global output q = q(q1,q2) received by each principal is a combination of these tasks. This is also important because a key consideration in the debates about SWFs is precisely the extent to which their objectives conflict with those of domestic actors.
  • Fourth, the interpretation of q should not be restricted to the notion of quantity. In fact, the output q, in the context of SWFs’ operations, corresponds to a portfolio that the principals ask the agent to manage for them. Martimort (2006), for example, considers the agent’s output to be a portfolio. An ambitious approach (not used in this chapter) would be to consider that the sovereign breaks up its SWF into a number of agents, each with a comparative advantage in dealing with various aspects of portfolio management, to maximize the sovereign’s overall utility with the combination of their outputs (just as an investor strikes a balance between risks and returns). Such a model would provide useful insight into the optimal architecture of any organization; one obvious example would be to determine the share of foreign reserves that the SWF would be allowed to manage while the central bank retains responsibility for the rest. This topic is discussed briefly later in this chapter, but the issue is not modeled.
  • Fifth, the model is here restricted such that the agent has an obligation to work for both principals simultaneously. In the P-A literature, this is a special case because most models allow the agent to reject one contract and accept the other. However, doing so would not make sense in this context. Indeed, this analysis rules out the possibility that the SWF could be operating independently in the recipient country. Thus, if the agent rejects the possibility of working with the SWF (or the other way around), the SWF would simply not be allowed to operate.
  • Finally, the welfare aspects of multiprincipal models are generally not well defined because the objectives of the two principals may overlap with each other. This overlap means that utilities cannot easily be added up and the theoretical literature has therefore neglected welfare analysis to instead focus on various notions of economic efficiency, but these notions are not discussed in the context of this chapter.

Figure 8.2Additional P-A Relationships

Source: Authors.


Perfect Information

With perfect information, both principals in this model know the characteristics θ of the common agent. Let us first consider the case in which the common agent operates as a private agency (i.e., deals with each principal Pi individually and negotiates on the specific output qi that only the two parties involved can observe). Because the principals know θ, each principal is able to reduce its own transfer ti to the point at which the agent is indifferent between undertaking the task and rejecting the offer (although the rejection option is not allowed here, as discussed earlier). Each principal behaves in the same way, and an equilibrium is reached when none of the parties wants to deviate.

Even in this simple case, both principals interact with each other when they negotiate with the agent. This interaction occurs during the negotiations, as each principal offers a contract in the form of a set of options to the agent. Most of these options will not materialize, but they are nevertheless important because they act as strategic signals to the other principal.8 With both principals acting through the agent, the result may be inefficient outcomes, a somewhat unexpected result since there is otherwise perfect information in the game (better information being often associated with more efficiency).

In a private agency, and from the point of view of the regulatory authority, the ability to monitor the activities of the SWF is limited by the fact that each principal negotiates directly with the agent and that the interactions between the principals are limited to sending signals to each other through a set of contracting options. Therefore, even with perfect information—and private agency—it is difficult to see how forcing an SWF to go through an agent with which it would be able to contract on its specific output would limit the scope for undesirable behavior. In fact, although it is possible that the signals sent by the other principal somehow affect (in a socially positive manner) the SWF, it is also equally likely that the signals sent by the SWF to the domestic principal have a less desirable impact (if the SWF has undesirable objectives). This chapter argues below and in the subsection devoted to the case of imperfect information that this is, indeed, a risk for the recipient country.

When the agency is public (i.e., both principals can negotiate with the agent on the entire output), there is clearly more scope for interaction between the two principals (through the agent). However, although these interactions can lead to more coordination (and that is presumably what the regulatory authority would like to see), there is also potentially more room for failing to coordinate properly (a situation the regulatory authority would like to avoid). The latter is especially likely to happen if the SWF’s objectives diverge substantially from those of the recipient country. This is precisely the situation in which the recipient country would want to increase its control over the activities of the SWF, and the model suggests that the ensuing lack of coordination leads to inefficient equilibriums. If, however, the SWF has objectives that are broadly compatible with those of the recipient country, then using a common fund manager as a public agency is indeed likely to strengthen collaboration and therefore yield efficient equilibriums. Overall, if the room for coordination is limited, having an agent in common is unlikely to help. If scope for coordination exists, having an agent in common may help reinforce that coordination.

Although this issue is explored in more detail in the section “P-A Players with Multiple Objectives,” it is worth noting that, in this game in which signaling plays an important role, the signals sent by one principal (e.g., the SWF) could be misread by the other principal (e.g., the regulatory authority). A misreading is likely to happen when two intrinsically different principals use their investments through their common agent as tools to learn about each other. As discussed in the next section, it seems that if signals are misread, the outcome is likely to be inefficient. Once again, an agent in common may help, but will be a more efficient instrument once some common ground in the form of compatible objectives has been established.

Imperfect Information

The previous subsection assumed that the principals operate with perfect information about the agent’s type (θ), which means that the recipient country feels confident enough about the behavior of the recipient-country fund manager to put it in charge of handling the SWF’s activities. That the SWF also knows the agent perfectly is clearly a restriction, but the theoretical models typically do not consider the case of asymmetry of information between principals. Despite its limitations, however, the concept of perfect information helps to provide insight into the communication channels between the two principals, even when the agent is “transparent” (i.e., when the agent’s type θ is known).

However, the analysis must also consider the case in which the agent retains an informational advantage. It may be reasonable to assume that the authority in the recipient country does not know the agent well. For example, the 2008–09 global financial crisis revealed that regulatory authorities, in spite of the time and resources spent, were taken by surprise by the types of financial products developed (and their valuation) by the most trusted firms they were supervising, even those operating domestically.

There is also another situation leading to imperfect information: the risk for an existing principal that the agent changes its type (θ) when a new and unknown principal—the SWF—enters. Such an argument is often made in clubs: the new member of a club could alter the nature of the club. A similar argument has also been used by some observers to argue that the inclusion of China in the World Trade Organization would not only affect China’s approach to trade, but could also fundamentally change the way the World Trade Organization operates.

One of the results obtained in multiprincipal models with imperfect information is that the agent can strategically use its informational advantage to play one principal against the other, especially if the principals have conflicting objectives (see Laffont and Pouyet [2004] for a formal approach). This is not surprising because this result was already obtained in perfect information, under both public and private agency, but the effect is amplified when the agent retains private information. Because the principals do not have perfect information, they have an incentive to monitor the agent to minimize its rent-seeking behavior. However, by monitoring the agent, each principal will interact, although involuntarily, with the other principal and will therefore send strategic messages. Khalil and Lawarée (2006) study that framework for principals asking a common fund manager (the agent) to manage their investments. As with perfect information, the results depend on whether the agency is private or public, although, even when the agency is private, the principals still interact with each other by performing some monitoring. That monitoring may reveal to the other principal some information about the characteristics of the agent. When the agency is public, each principal may want to use monitoring to convince the other principal that the agent is not really good for the other principal’s purpose. In that sense, although θ is a one-dimensional parameter not designed to capture the multidimensional aspects of the agent’s skills, it also serves as a blurring device that prevents principals from correctly assessing these skills (assumed given) in the different areas that each cares about. If each principal does the same, too much monitoring, from a societal perspective, may be occurring. Too little monitoring may occur when the agency is private.

This is an interesting conclusion for SWFs’ operations. The idea that there can be excessive monitoring is not always accepted (much like the idea that there can be excessive availability of information), and a recipient-country fund manager may resent becoming the object of increased scrutiny by the domestic regulators and possibly by the SWF. Therefore, although some monitoring is beneficial, the regulatory authority may fall into the trap of excessive monitoring. This may also occur with public agency in which the principals have more opportunities to collaborate—and compete—with each other, suggesting that forcing a public-agency approach would not be without pitfalls. Excessive monitoring could also be thought of by SWF home countries as a protectionist step by the recipient country, creating demands in the SWF’s home country for retaliatory steps against all foreign investment.

After having explored the way in which imperfect information can affect the results obtained with perfect information, the analysis uses the model to look briefly into other areas that could be relevant to the activities of SWFs. To keep the argument simple, the discussion assumes a private-agency model in which the principals are concerned with two different aspects of the activities (the output) of the common agent.

From the social point of view, a crucial consideration is the extent to which the outputs q1 and q2 are complementary or are substitutes (or antagonistic, as termed earlier in this chapter). Let us look at the private-agency model again. The agent is tasked by both principals to produce the output that is relevant to each of them (q1 for P1 and q2 for P2). For example, the common agent could be asked by one principal to invest in a stock with certain return and variance characteristics, while the other principal favors another stock with different characteristics. In that situation, the principals requesting different stocks compete with each other (when their respective desired outputs are substitutes) and they support each other (when their desired outputs are complementary) for the agent’s time and to capture the output. Technically, the agent’s outputs (activities) are complements when C12 < 0, and substitutes when C12 > 0, where Cij denotes the cross derivatives of the cost function introduced in the section titled “A Simple Model.” It is easy to show that an increase in output q1 triggers a decline in output q2 when activities are substitutes, while the reverse holds when they are complements. Complementarity could be interpreted as a situation in which the SWF has “friendly” objectives that do not conflict with those of the recipient country.

One general lesson can be drawn from such imperfect information models. Principal P1 always tries to induce the agent to do more for it (of q1) and less for the other principal (P2) by offering to the agent signals and incentives aimed at convincing P2 that the agent is not good at performing the task that P2 cares about (q2). When outputs are substitutes (or compete with each other), this signaling behavior leads to output distortions as both principals play the same game and successfully manage to reduce the output for the other. When outputs are complementary, the game does not lead to a substantial loss in overall output.

Therefore, the extent to which the objectives of the regulatory authority and the SWF conflict with (substitutes) or reinforce (complements) each other is an important consideration, as noted in the introduction to this chapter. Not surprisingly, fewer distortions will occur with complements and the recipient country is correct to feel more comfortable with the arrival on its territory of an SWF with objectives that do not conflict with domestic objectives. Once again, however, all these results depend on how each player interprets (correctly or incorrectly) the signals sent by the others. The chapter now shows, using one example, how even good intentions can be misinterpreted, and lead to a loss of efficiency.


While formally assuming that players maximize some general utility function, P-A models are often limited to profit maximization, as already indicated. Yet, some of the key issues emerging from the debates on SWF activities suggest that there are suspicions (on the parts of authorities in recipient countries, think tanks, and even home-country authorities) about SWFs’ motives, even though achieving profits must clearly be an important component of a more complex utility function. The discussion in this section will primarily be relevant to the SWF operating in a recipient country, although some of the issues would also be relevant to the SWF operating domestically.

A review of the various objectives that an SWF could pursue indicates that some of these objectives could be characterized as “learning by investing.” By investing in a line of business, the SWF will learn about the business’s activities and can, in turn, provide its sovereign with useful information (irrespective of whether such learning is aimed at strategic, commercial, security, or simply development purposes). Learning by investing contrasts with the well-researched field of “learning by doing,” in which costs continue to decrease with the cumulative quantity of goods produced (e.g., Dasgupta and Stiglitz, 1980; Tirole, 1988). If one of the players (in general, it could be either the principal or the agent but this chapter focuses on the principal) is pursuing multiple objectives in the long run, these objectives would probably not be compatible with short-term financial objectives, but could well be compatible, in the long run, with standard (although enlarged) financial objectives. Thus, the SWF is “investing” in the short run for a long-run payoff that the recipient country may find difficult to define or even understand at the time of the investment. In such a framework, the signals received by the other players (i.e., information about the investor’s action) will be hard to interpret. When signals are not read correctly, any move could lead to suboptimal reactions by other players and therefore to suboptimal outcomes from the point of view of the other players (including the agent and the principal). Such situations could clearly occur when an SWF operates in a recipient country, but could also arise for an SWF operating domestically. This issue has been touched on in the context of a model with two principals and one agent. For tractability, the model in this section is restricted to one principal and one agent in the first instance, before briefly moving to the case of several principals.

One Principal

A simple example illustrates how learning by investing can take place. Assume that an SWF acts as a principal and wishes to invest in each period (time is discrete) in one of two funds, denoted by j, with j = A, B. The financial returns from investing in either fund are stochastic and denoted by Rj, j = A, B. It can also be assumed that the principal can acquire some private additional benefit Tj from investing in either of these funds (e.g., control or acquisition of information). Again, this information can be directly related to the profitability of the fund or to more general benefits that the principal wishes to acquire. It is further assumed that the principal is facing uncertainty; therefore, the benefit Γj is modeled as a latent variable (not observable directly by the principal) that can take two values normalized to 0 or 1, without loss of generality. The possible realizations of Γj are denoted by Γj = 0 (referred to as “bad outcome”) or Γj = 1 (referred to as “good outcome”). The words “good” and “bad” are arbitrary. It is assumed that the underlying process generating the good and bad outcomes is stationary and unknown.

The principal is assumed to be facing uncertainty because it does not know the probabilities of occurrence of the good and bad outcomes. The principal can reduce this uncertainty, but at a cost. The analysis denotes by pj, the probability that the outcome is unfavorable Prob{Γj = 0} = pj, with fP(pj), the prior belief, in other words the probability distribution for parameter pj. Without any extra prior information, it is reasonable to assume the simplest possible prior, that is, that the distribution is uniform (fP(pm) = 1). One extreme case occurs when the random variables Rj and Γj are perfectly correlated; the other extreme case occurs when the random variables Rj and Γj are independent; at these extremes, neither a large nor a low return on investment j tells anything about the value of Γj.

If there is perfect correlation (between the financial benefits and the additional benefits), acquisition of information by the principal is required to maximize the principal’s long-term objectives. If the two benefits are independent, acquisition of information and profit maximization correspond to two different objectives held by the principal in a multicriteria problem.

The analysis has assumed that the investment occurs in a discrete time frame and, in each period, the principal decides where to allocate its assets (the choice is discrete, i.e., either fund A or fund B is chosen). The proposed model is dynamic and captures the time lag associated with the concept of learning by investing. It works as follows: after each period, the principal observes a realization of the random variable Γj, and adjusts its previous knowledge accordingly. Note that the observation of fund j is costly because the principal has to invest in this fund to be allowed to have access to a realization of Γj. After n observations of the realization of Γj, the distribution of prior knowledge has changed and is determined by the observed number of good and bad outcomes; this provides sufficient statistics for making the next decision. This problem is difficult and has no closed-form solution, but rigorous numerical solutions exist. Some of the computations related to this problem are provided in the Annex to this chapter.

Investing in a fund can therefore play a dual role for the SWF because it gets both some financial return and some private benefit. Some simple cases are easy to solve.

Assume, for example, that returns and private benefits are perfectly correlated. To fix ideas, assume that the value of pA for fund A is known; we then have a “one-armed bandit” problem.9 It can be shown that in this case, the best strategy is one of three: (1) the principal always invests in fund A and never changes; (2) the principal invests in fund B, and if there are too many bad outcomes, it stops and switches to fund A forever; or (3) the principal invests in fund B and never changes (note that it would be irrational to invest in A and then switch to B). Case 2 is the most interesting, and illustrates what is meant by learning by investing. The choice of fund B can be interpreted as follows: the principal is ready to possibly lose some money to find out if fund B will be a good source of private information. After exploring that possibility for some time, the principal decides (in case 2) that fund B is probably not a good source of private information and switches to fund A. Stopping rules must be designed and computed on the basis of past observations so that the principal knows when to switch; although they are complex, such rules (known as the Gittins Indices after Gittins [1979, 1989]) can be computed analytically.10 In more complex cases, the quality of the private information provided by both funds is unknown. In that situation, shifts by the SWF from fund A to fund B (and vice versa) can always occur, that is, the SWF can switch back and forth several times because a sufficiently long series of bad outcomes can discourage the principal from continuing to invest in the fund originally selected and induce it to switch.

Several Principals

When several principals are willing to invest in the same funds, the situation is far more complex. The principals will observe each other, and the rational SWF behaving as just described sends a signal to the other principals when it decides to switch from one fund to another. Although these signals are hard to interpret, they cannot be ignored. Given that all these signals convey information, they should and will be interpreted by the other principals. However, the solution is not easy to compute—a point made earlier—because the acquisition of some share of a fund could be explained by several considerations that are hard to disentangle: a principal could select a fund in the short-run investment stage to acquire potential private information conveyed by this fund. Or the principal may just be interested in the financial return on this investment. Moreover, the correlation between the private information that two principals could extract also matters in the discussion.

The analysis thus far has assumed that the recipient-country fund manager (the agent) is not acting strategically. The situation becomes more complicated when the agent ceases to be a passive player that ignores that its choices convey information to the principals. As seen in the “Perfect Information” section, even with perfect information, the agent may be in a position to play one principal against the other. The assumption of a passive agent may be realistic in situations in which the number of principals is large, but will generally not be a good assumption if there are only a few. In addition, if the principals are few, they will be aware that their actions are observed and, in turn, may benefit from acting strategically. This is usually true (Martimort, 2006), and the discussion has already shown that signals outside equilibrium can play a critical role in the game with single objectives. This analysis suggests that, with multiple objectives, the signals sent by any player (one of the principals or the agent) can be significantly blurred, reducing the ability of the other players to interpret them.

The fees charged by the agent will depend on the value of the private information it can provide, called θ earlier in the chapter. As is usually assumed, this parameter will remain unknown to the principals, but while the value of the information (T m) gathered by the principals is unknown to the agent, the agent can learn from the principals’ investment behavior (via a market study, for example). This problem is complex and has not been addressed in the literature. However, in simple cases, it can be shown that when principals have multicriteria objectives, the agent may be able to extract a positive rent (even under perfect information) that it would not be able to extract otherwise (a result similar to that obtained in Martimort [2006] with a single objective). This is worrisome from the policy standpoint because the outcome of forcing an SWF to operate through a recipient-country fund manager (when it is suspected that the objectives of the SWF depart strongly from those of the regulatory authority), could be to increase the profits of the common agent and make that common agent move to an inefficient outcome. Once again, channeling the investment activities of the SWF through a recipient-country agent may not always increase efficiency.


Some observers have suggested that SWFs could partly allay the concerns about possible political motivations behind their activities by investing through fund managers located in the recipient countries. This chapter examines the usefulness of this proposal by using agency theory. The results show that, under reasonable assumptions, the use of fund managers may not necessarily address these concerns. This result holds when an SWF pursues only its profit-maximization motives, but even more so when it pursues multiple objectives, including learning by investing. These results indicate that recipient countries may try to address their concerns through more direct regulation, which may add to the protectionist trends observed in many countries. To avoid protectionism, SWFs and recipient countries need to work toward greater organizational and operational transparency. The recent creation by Temasek of a new investment division that hopes to seek backing from institutional investors (and possibly the general public in the future) may prove to be a useful example of an innovation that would help dispel the concerns of recipient countries.


    Chancelier, J.-P., M.De Lara, and Palma,2009, “Risk Aversion in Expected Intertemporal Discounted Utilities Bandit Problems,”Theory and Decision, Vol. 67, No. 4, pp. 433–40.

    • Search Google Scholar
    • Export Citation

    Das, Udaibir S., AdnanMazarei, and AlisonStuart,2010, “Sovereign Wealth Funds and the Santiago Principles,” Chapter 5, this volume.

    Dasgupta, P., and J.Stiglitz,1998, “Learning by Doing, Market Structure and Industrial and Trade Policies,”Oxford Economic Papers, Vol. 40, No. 2, pp. 246–68.

    • Search Google Scholar
    • Export Citation

    Gittins, J. C.,1979, “Bandit Processes and Dynamic Allocation Indices,”Journal of the Royal Statistical Society, Series B, Vol. 41, No. 2, pp. 148–77.

    • Search Google Scholar
    • Export Citation

    Gittins, J. C.,1989, Multi-Armed Bandit Allocation Indices (New York: John Wiley & Sons).

    Gibson, Ronald J., and Curtis J.Milhaupt,2008, “Sovereign Wealth Funds and Corporate Governance: A Minimalist Response to the New Mercantilism,”Stanford Law Review, Vol. 60, No. 5, pp. 1345–69.

    • Search Google Scholar
    • Export Citation

    Hammer, C., P.Kunzel, and I.Petrova,2008, “Sovereign Wealth Funds: Current Institutional and Operational Practices,”IMF Working Paper WP/08/254 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    International Monetary Fund, 2008, “Sovereign Wealth Funds—A Work Agenda” (Washington). Available via the Internet:

    • Search Google Scholar
    • Export Citation

    IWG (International Working Group of Sovereign Wealth Funds), 2008, Sovereign Wealth Funds: Generally Accepted Principles and Practices—“Santiago Principles” (Washington). Available via the Internet: (Reprinted as Appendix 1 to this volume.)

    • Search Google Scholar
    • Export Citation

    Khalil, F., and J.Lawarée,2006, “Incentives for Corruptible Auditors in the Absence of Commitment,”Journal of Industrial Economics, Vol. 54, No. 2, pp. 269–91.

    • Search Google Scholar
    • Export Citation

    Laffont, J.-J., and J.Pouyet,2004, “The Subsidiarity Bias in Regulation,”Journal of Public Economics, Vol. 88, No. 1–2, pp. 255–83.

    • Search Google Scholar
    • Export Citation

    Martimort, D.,2006, “Multi-Contracting Mechanism Design,” in Advances in Economics and Econometrics, ed. by R.Blundell, W.Newey, and T.Persson (Cambridge, UK: Cambridge University Press).

    • Search Google Scholar
    • Export Citation

    Mookherjee, D.,2006, “Decentralization, Hierarchies, and Incentives: A Mechanism Design Perspective,”Journal of Economic Literature, Vol. 44, No. 2, pp. 367–90.

    • Search Google Scholar
    • Export Citation

    Parlour, C. A., and U.Rajan,2001, “Competition in Loan Contracts,”American Economic Review, Vol. 91, No. 5, pp. 1311–28.

    Tirole, J.,1988, The Theory of Industrial Organization (Cambridge, Massachusetts: MIT Press).


Denote by Nb the number of periods a bad realization (tj = 0) is observed. Then, the a priori distribution of t, conditional on the observation of k = Nb bad outcomes for j over n observations of j, can be computed according to Bayes’ rule. Some standard computations show that

The expectation of the probability Γj conditional on Nb = k is

Note that, as expected, this converges to k/n, as k and n tend to infinity.

Denote the benefit (conditional on the realization tj) of the principal, during one period by Tjj) = U(rj + θ j; Γj), where U(.;.), the utility function of the principal, is based on the return of fund j (j = A or B) and on the private information provided by this fund. For example, if the principal is risk neutral, and if there is a constant trade-off between the two criteria (profitability and learning), then its utility is linear, and Tjj) = rj + θ Γj. If the principal is risk averse, the utility function will have a nonlinear specification, such as constant relative risk aversion or constant absolute risk aversion. With risk-neutral principals, the unconditional benefit of the principal, after n investment periods (observations) is then

This expression is linear in the probability. It can therefore be simplified as follows:

The principal should then maximize the discounted value of the flows of benefits over the total investment period.

Note that if the principal has some perception biases of the unknown probabilities, the expression will be nonlinear in the probability:

where v(·) and w(·) denote the probability weighting functions. With no perception biases, v(·) and w(·) reduce to the identity. When v(·) and w(·) are nonlinear, a closed-form solution for T(n) does not necessarily exist. However, numerical computations could still determine the switching points in the investment strategies.


The authors have benefited from comments from E. Barot, M. El-Gamal, A. Ferrière, and J. Pihlman.


Others have also argued that, far from being sophisticated, SWF managers were actually not aware of their potential impact, or were reluctant, for reasons of “image,” to activate all the tools of leverage at their disposal.


Das, Mazarei, and Stuart (Chapter 5, this volume) discuss this and many other concerns raised by SWFs’ operations. They also elaborate on the rationale for the Santiago Principles and how the principles seek to allay those concerns.


First, as public investors, SWFs must properly invest public funds and secure adequate risk-adjusted returns. Second, SWFs need to make sure their investments are immune from undue domestic political interference. Third, because of their size, their activities need to be well coordinated with the country’s overall macroeconomic policies, particularly with the activities of other actors tasked with the implementation of macroeconomic objectives (e.g., the central bank). And fourth, as mentioned in the case of an SWF operating overseas, there is the possibility that some of the multiple objectives pursued by the manager of the SWF may not be compatible with those of the other domestic economic actors. For example, the finance minister of a centralized economy may have misgivings about the profit-maximizing instincts of the SWF manager. As mentioned, these issues became more relevant during the 2008–09 financial crisis as SWFs were asked by their governments to shift some of their focus to their domestic economies.


“Temasek to Launch $4bn Investment Division,” Financial Times, February 10, 2010.


Several other P-A relationships are ignored in this brief description, some of which can be very important, such as the relationship already mentioned between the foreign authority as a principal and the SWF as its agent.


In the model considered here, imperfect information corresponds to adverse selection. Other variations of the model could also have been investigated but, in the context of this chapter, the analysis would not yield substantially different conclusions.


Parlour and Rajan (2001) provide a model of private agency where the principals lend money to a common agent who may default. In that case, however, the signaling between the principals is made through that possibility of default. This type of signaling may also lead to inefficient equilibriums.


A one-armed bandit is a process akin to a jackpot. The probabilities of success are unknown but could be determined more accurately as the machine is used, but using the machine is costly (see Gittins, 1979, 1989).


Note that this problem, known as the “bandit problem,” evoked at the beginning of this section, has been treated in the literature only for risk-neutral players, with the exception of Chancelier, De Lara, and de Palma (2009), who consider the case of players who are risk averse.

    Other Resources Citing This Publication