Printer Friendly

Incentive issues in R&D consortia: insights from applied game theory.


A. Aims and Premise

How can applied game theory help analysts understand Research and Development (R&D) consortia? Our basic premise is as follows. We view institutions such as R&D consortia as mechanisms for dealing with incentive problems that are obstacles to realizing the benefits of cooperation. These incentive problems involve strategic interaction. Hence, we need game theory to understand them. Yet, a full-blown game model the incorporates all relevant incentive issues is not feasible--the mathematics is too complex. Thus, we review a series of relevant insights--drawing upon several strands of more rigorous literature on incentive problems--and synthesize these insights into a single narrative or "logic." We use a case study of an R&D consortium, the Latin American Fund for Irrigated Rice (Spanish acronym: FLAR), to illustrate how the applied game-theoretic insights fit together and how they are relevant to a real-world institution like FLAR. (1) We introduce our proposed logic and FLAR in the present section.

Our principal aims are (1) to set out this logic step by step, (2) to persuade the reader of its relevance to R&D consortia, and (3) to suggest how insights from several strands of applied game theory can be applied to R&D consortia.

In the course of our argument, we introduce several concepts that are building blocks for our logic. These include in particular the notion that complexification should precede simplification (crucial to bridging the gap between applied game theory and real-world institutional analysis), the multidimensionality of subsidies, the coexistence of crowding in and reverse crowding out, a novel concept of leadership, and a "big-picture game" (i.e., one that abstracts from specific incentive issues) that is a repeated prisoners' dilemma (PD) with unknown horizon.

B. Approach and Overview

Olson (1965) elaborated a key insight from the literature on public goods (e.g., Samuelson, 1954, 1955)--namely that the existence of a group of players with a common interest does not automatically give rise to collective action or may result in collective action at a suboptimal level. At the core of Olson's logic is a simple mathematical model that yields several propositions, for example, those relating group size to the severity of the collective action problem. The decades since Olson (1965) have witnessed an explosion of applied game-theoretic models of a number of incentive problems that are relevant to collective action. Hence, we propose an updated, more conceptual, logic that can accommodate much of this subsequent literature and does more justice to the complexity of real-world institutions (Figure 1). We think of participants in an R&D consortium as players in a game, who have certain reasons for joining the consortium (Box A in Figure 1) but face incentive issues (Box B) that are obstacles to optimal cooperation. We analyze the resulting arrangement as a solution to those incentive issues, emphasizing the role of leadership (Box F).


Who are the players--individuals, or households, or organizations, or governments, etc.? This is a tricky and important question. We take a flexible approach. In the next subsection, we sketch FLAR and argue that its "representative organizations" (one for each country) are the relevant collective action players. However, a few individuals were instrumental in founding and designing FLAR (see bellow).

Why should a group of players form an R&D consortium? There are three key technical reasons for doing so (Box A). (2) First, R&D outputs are typically (imperfect) public goods: they are characterized by imperfect rivalry and imperfect (or costly) appropriability. This is the reason for collective action that Olson (1965) and the collective action literature generally focus on. A second key reason is the existence of complementary or synergistic assets--such as intellectual property (IP), genetic materials, and employees' expertise--that serve as inputs into the R&D process. The third rationale for collective action consists of economies of scale and scope. We discuss each of these rationales in Section II. They share the feature of mutual positive externalities. In other words, players' interests in cooperating converge to some extent.

Suppose that due to some combination of these three key reasons, it is optimal to form an R&D consortium. Incentive problems (Box B)--which may be internal or external to the consortium--constitute obstacles to optimal cooperation. The fundamental reason underlying these incentive problems is that players' interests diverge to some extent too. This tension between convergence and divergence of interests is the core theme of the economics-of-contracts literature. (3)

The formal literature on R&D consortia commonly assumes that players are for-profit firms. How, then, to characterize players, objectives in hybrid public-private R&D consortia? While organizations' objectives are typically complex, we believe own-institution bias to be a common phenomenon. That is, even nonprofit organizations that have prosocial missions will often have their own institutional interests: while they seek to promote their missions, they are keen on receiving the resources and credit for doing so. Thus, if other organizations pursue the same mission, tensions may result.

We aim to do justice to complexity by sketching the full picture of relevant incentive problems (Box C). We show in Section III--and illustrate using the FLAR case--that many incentive problems are simultaneously relevant. A single game incorporating all these problems would be far too complex to fully model mathematically. Moreover, players are not perfectly rational or self-interested. Thus, we need to take into account insights from the social and cognitive sciences that may increase complexity even further (horizontal arrows above Box C). However, social and cognitive factors can also simplify the game (horizontal arrows below Box C). For example, players reduce complex games to manageable structures through pattern recognition; moral values limit opportunistic behavior.

We devote Section IV to Boxes D, E, and F. A central tenet of our approach is that complexification should precede simplification. Only after having sketched the issues in their full complexity do we simplify the game to a manageable structure (Box D). Due to the core feature of mutual positive externalities, this will typically be a repeated PD with unknown horizon. In such games, the Folk Theorem applies: various gradations and combinations of noncooperation and cooperation are incentive compatible (Box E). To narrow down this range to a unique solution--an explanation for the institutional arrangement--one needs to impose equilibrium refinements. We emphasize one particular kind of refinement, interpreting leadership as the ability to choose one scenario from the incentive-compatible set. Social and cognitive factors help explain this ability (Box F and adjacent boxes).

We conclude by reviewing the article's main insights (Section V).

C. Sketch of FLAR

Before the mid-1990s, practically all developing world agricultural R&D was publicly funded. (4) FLAR is among a small number of pioneering attempts to catalyze the local private sector to play a more active role in agricultural R&D (Binenbaum, Pardey, and Sanint, 2003). It was founded in 1995 to conduct a range of activities--primarily plant breeding--related to irrigated rice. Recently, the first FLAR rice varieties were released. Additionally, FLAR has recently become involved in an agronomy project. While FLAR is relatively small--its annual de facto budget, which is larger than its formal budget (including in-kind contributions such as facilities and delegates' time and travel but excluding the agronomy project), is about $1 million--it does illustrate key elements of our proposed logic.

Current and past FLAR member countries are listed in Table 1. Among members, the International Center for Tropical Agriculture (Spanish acronym: CIAT) plays a central role. CIAT is one of the (research) centers of the Consultative Group on International Agricultural Research (CGIAR). The CGIAR and its centers are essentially a conduit for development assistance (sponsored by rich countries and private philanthropic foundations) and have played--and are still playing--a big role in the introduction and adaptation of modern agricultural technology to the developing world. CIAT currently is the only noncountry FLAR member and hosts FLAR at its headquarters and main research campus near Cali, Colombia.
FLAR Member Countries/Industries April 2006

Current Members Past Members

Argentina Chile
Bolivia CIRAD
Brazil (a) Cuba
Colombia (a) Paraguay
Costa Rica
Venezuela (a)

Notes: CIAT = Centro Internacional de Agricultura Tropical;
CIRAD = Centre de Cooperation International en Recherche Agronomique
pour le Developpement; IRRI = International Rice Research Institute.

(a) Founding members.

Two salient features of FLAR affect the proper characterization of its members as players. First, FLAR members vary widely in size as measured by irrigated rice output (Figure 2). Their R&D capabilities are similarly asymmetric. Second, an important principle of FLAR is single representation: each member country is represented by one organization, called a representative. A national government has veto power over which organization will act as its representative but is only indirectly involved itself. FLAR representatives are closely connected to the rice sector and/or agricultural R&D. The key players in FLAR are not member countries' governments, but their respective representative organizations, whose role derives from domestic dominance in rice production, processing, and/or breeding. The representatives' objective is primarily to bolster the productivity and competitiveness of their domestic rice sector, in particular of their members or clients--rice farmers and/or rice-processing industries. However, there may be additional objectives, namely, to reinforce the organization's domestic leadership position and, relatedly, to receive the credit for benefiting their constituents. We have found own-institution bias to play a significant role in FLAR (Binenbaum, Pardey, and Sanint, 2003).


FLAR representatives occupy a wide spectrum along the public-private axis. For example, Cuba's representative--when Cuba was a member--was a 100% public-sector organization, while Costa Rica's representative is a 100% private sector consortium. Most representatives fall somewhere between these two extremes. Sometimes, a public-sector representative is privately funded. Representatives' constituents include governments, rice farmers, seed firms, and rice-processing firms. They typically play a latent role: they are mostly not active players but might become so under certain conditions. We rarely observed this in our case study (Binenbaum, Pardey, and Sanint, 2003).


A. A Mixed Public Goods Characterization
 To understand an R&D consortium, one needs to properly characterize
 its outputs. This is not a straightforward exercise--R&D outputs
 possess in varying degrees the public goods characteristics of
 nonrivalry (also known as nondiminishability) and nonexcludability.
 Thus, different strands of literature are-sometimes

When both excludability and rivalry are low, Samuelson's (1954, 1955) model of pure public goods may be appropriate. With low excludability but high rivalry, the common-access-resource literature (which builds on Hardin, 1968) is relevant. When excludability is high but rivalry low, a club-goods model (as pioneered by Buchanan, 1965) is nearer the mark.

Intellectual property rights (IPR) confer some excludability. Perfect excludability would require costless and universal IP protection and perfect enforcement. Clearly, this is not the case in the real world. Application costs are substantial; the requisite IP expertise is expensive; legal systems are overburdened; enforcement is imperfect and requires costly and time-consuming litigation. All these factors are compounded by the existence of multiple jurisdictions. Many types of R&D output are not patentable in most jurisdictions. For example, Latin American IP systems do not allow patent protection for plant varieties; instead, they use a weaker form of IPR, plant variety protection. On the other hand, a technology's excludability can be strengthened by complementary assets such as tacit knowledge codified but confidential information, and objects embodying the technology.

R&D outputs may not be completely non-rivalrous either. It is true that one player's possession of knowledge does not diminish the amount of knowledge available to other players. But if the players are competitors in product markets, then one player's possession of information may enable her to capture another player's market share. Clearly, this introduces an aspect of R&D output rivalry into the game. In analyzing the likely (in)efficiency properties of consortium arrangements, it is important to consider the degree of competition (or rivalry in public goods terms) in output markets. This is a key component of R&D consortium models such as Katz (1986). Members' ex ante commitments in the consortium agreement, as well as their level of cooperative behavior once the consortium is in place, depend on it. In Katz (1986), the consortium output vector consists of members' scalar cost parameters of a single product. They then compete in the output market with their reduced cost parameters. Competitors' prospective cost reductions are a disincentive for cooperative behavior of consortium members.

How to characterize FLAR's outputs? FLAR's breeding outputs are excludable to a significant extent. Plant variety protection is reinforced by the need to physically possess the plant genetic resources (PGR) required for breeding. As to nonrivalry, most FLAR members sell their rice output almost exclusively domestically. An important exception to this pattern of insignificance of rice exports of FLAR members is Uruguay, which exports large amounts of rice (worth between US$100 and US$300 million annually) to Brazil in direct competition with Brazilian rice producers. This may have had a negative effect on Uruguayan and Brazilian cooperativeness (such as the extent of sharing of PGR) within FLAR (Binenbaum, Pardey, and Sanint, 2003).

FLAR's agronomy services are near-pure public goods, generating quick benefits across FLAR member countries and probably spilling over to farmers elsewhere (L.R. Sanint, pers. comm.).

In summary, the FLAR case combines elements of club games, pure public goods, and--to a much lesser extent--common-access resources. Prior to the agronomy project, FLAR clearly came closest to the club model.

B. Economies of Scale and Scope
 Economies of scale are ubiquitous in modern R&D, particularly because
 some indivisible infrastructure--a minimum scale--is needed. The
 significance of scale economies in a consortium game depends on their
 range and magnitude relative to each player's size.

Additionally, joint R&D projects are often, but not always, characterized by economies of scope. Even highly focused projects will often spawn valuable spin-off activities. Economies of scale are, however, a more universal feature and are more suitable for a simple and general mathematical characterization.

The threshold idea can be modeled using a standard U-shaped average curve derived from an R&D production function. Output consists of technological innovations, for example, increases in a performance variable such as crop yield improvement Y. The minimum infrastructure entails a fixed cost F. Thus, there is a threshold effect: for any budget B < F, Y = 0. Assuming that B is spent efficiently, it is given by the cost function: C(Y) = B. The average cost function C(Y)/Y is initially decreasing because it starts at infinity and, subsequently, F is divided over more and more R&D output units. Additionally, marginal cost C'(Y) is typically initially decreasing as well. Adding critical inputs to the minimum infrastructure will drive average cost down further; specialization ("pin factory") effects occur. Thus, the threshold effect is augmented by additional scale effects; the efficient scale Y* = argmin C(Y)/Y exceeds the minimum scale [Y.sup.min] = [C.sup.inv] (F). Eventually, as B increases, scale diseconomies due to planning complexity, bureaucratization, and reduced communication of innovative ideas set in. In summary, we expect U-shaped average and marginal cost curves to be characacteristic of R&D production processes generally. Olson (1965) takes these same features into account in his model of collective action.

Often, the efficient scale will exceed the investment capacities of separate players, but they can approach or reach it by pooling resources. Due to incentive problems, their investments tend to be suboptimal (as argued below) and may be close to the minimum scale.

FLAR may be not far above its minimum critical mass but substantially below its efficient size. One indication that overall funding for FLAR is at a suboptimal level is the approximately US$10 million budget of its African counterpart, West African Rice Development Association (WARDA, also known as the Africa Rice Center)--an order of magnitude larger than FLAR. WARDA's activities are sim ilar to FLAR's, but it is a traditional CGIAR center, 100% funded as development assistance. While Latin American national agricultural research systems (NARS) have superior resources compared to their African counterparts--and are thus able to carry out a larger share of R&D activity--this difference is unlikely to fully account for the discrepancy between FLAR and WARDA (Binenbaum, Pardey, and Sanint, 2003). (5)

Initially, economies of scope appeared to be irrelevant to FLAR: it was set up with a strong focus on irrigated rice breeding (and supporting other players' efforts in this area). However, the success of the agronomy project may have benefited from FLAR's primary focus. Crop performance is central to the success of agricultural systems; thus, FLAR--supported by CIAT's agronomic capabilities--may have been particularly well placed to undertake the project. Agronomy may now become a permanent feature of FLAR. FLAR may, on closer inspection, turn out to exhibit economies of scope after all.

C. Asset Complementarities
 Due to asset complementarities, a grand coalition with full sharing
 of assets would be optimal. This important stylized feature translates
 into a simple superadditivity game--the starting point for our
 discussion of incentive problems.

The collective-action literature has focused on the public goods and scale economies rationales for collective action. Less attention has been given to another key reason for collective action: the need to combine complementary (synergistic) assets. The assets may be tacit knowledge, codified information, IPR, or technology-embodying objects such as genetic resources. They comprise inputs and rights to inputs in R&D. The literature on cooperative R&D emphasizes asset complementarities (Hagedoorn, Link, and Vonortas, 2000) but may have underused collective-action concepts. On the other hand, it is important that asset complementarities be incorporated into a collective-action narrative.

Graff, Rausser, and Small (2003) and Heller and Eisenberg (1998) point to the critical problem of combining complementary IP and related assets in the life sciences. Due to genetic combinatorics, genetic resources are a prime example of complementary assets. For this reason, cooperative plant breeding games where each coalition pools its members' PGR tend to satisfy the strict super-additivity property: for any two disjoint coalitions S and S', the value of the combination exceeds the sum of the separate values: V(SUS')>V(S) + V(S'). A simple superadditivity game with costless bargaining and transferable value has a unique and Pareto-efficient equilibrium: all PGR owners will join a grand coalition, share their genetic resources for a collective breeding program, and share the resulting benefits according to their respective bargaining positions. This result, while trivial in its own right, is our starting point for analyzing R&D consortia such as FLAR.

This model of coalition formation can be applied to the other collective-action rationales--public goods and economies of scale and scope--as well. We will refer to it as the "baseline model."


A. Creation of a Leviathan as a Solution to the Holdup Problem
 A variation on the baseline model involving an extreme "holdup"
 yields a drastically different result: zero cooperation. Players
 can mitigate this incentive problem by initially delegating power to
 another--possibly newly created--player.

In holdup games, players have an incentive to wait to improve their bargaining positions during later negotiation rounds (Cohen, 1998; Klein, 1998). Suppose that coalitions can grow in successive stages by allowing new members on board. Is it possible for the grand coalition to emerge gradually by having players join one by one, negotiating with those already in the coalition? Suppose that the last player to join has the strongest bargaining position, the next-to-last player the second strongest, etc. We can construct such a game so as to obtain an impossibility result: even in the presence of superadditivity, there exists no equilibrium with any coalition formation, for example, there is no PGR sharing whatsoever.

Now consider another model variation. Consider again the extreme holdup game, but now suppose that prior to the formation of PGR sharing pools, the players can jointly create another player with the power of principal and with the objective to maximize joint value and satisfy some sort of fairness crtierion (say, Shapley values or the Nash bargaining solution). Of course, we will get the (trivial) result that the new player will be created if the cost of doing so is low enough. This may be called the Leviathan result/game, as it is a variation on the theme--developed by Hobbes and subsequent enlightenment and "contractarian" thinkers--of a social contract whereby players voluntarily delegate power to a government so as to avoid conflict among themselves (Kraus, 1998).

We may also construct less extreme versions of holdup and Leviathan games. Thus, we can construct holdup games that result in partial coalition formation and Leviathan games that result in partial delegation of power.

This is one possible rough explanation for the existence of FLAR. Its members agreed to yield to it some of their control over PGR. In fact, the FLAR agreement does not compel members to share all their genetic resources, but it does provide a mechanism whereby members voluntarily place PGR in the FLAR breeding program.

B. Coexistence of Reverse Crowding Out and Crowding In
 We differentiate between crowding in and reverse crowding out and
 show that these features can coexist even in a univariate (cash)
 subsidy model. We also point out that they may coexist due to the
 multivariate nature of subsidies.

There often exist outside players (i.e., external to the consortium group) willing to subsidize the consortium. The optimal design of such subsidies is not a straightforward exercise--it requires an understanding of the potentially confusing concept of crowding in/out. "Crowding in" and "crowding out" both refer to the effect of public sector spending on private sector spending; in the case of crowding in (out), this effect is positive (negative). In other words, the terms mean that public and private expenditures are strategic complements or substitutes, respectively. If a reduction in public expenditure results in increased private expenditure, then we should speak of "reverse crowding out.''

Reverse crowding out is of particular interest as it allows benevolent sponsors to reallocate funds from one worthy cause to another at a reduced opportunity cost. A model of reverse crowding out is easily constructed. An R&D consortium Ss value V(S) is a function of--among other things--its budget, which is the sum of its members' contributions: B = [[SIGMA].sub.k[member of]s] [b.sub.k]. We make the following two "natural" (in terms of modeling technique) assumptions: [partial derivative]V/[partial derivative]B > 0; [[partial derivative].sup.2]V/[[partial derivative].sup.2]B < 0. Now ignore all other complicating factors and assume the baseline model, with one additional feature: the presence of an exogenous subsidy, which is some fraction of the optimal funding amount B*. The budget contributions [b.sub.k] are the Nash bargaining solution. Now consider a simple comparative statics exercise. Suppose that the subsidy is exogenously reduced, then under the assumptions made, all [b.sub.k] will increase. This is a reverse crowding out result.

The trouble with this model is that it assumes marginal diseconomies of scale ([[partial derivative].sup.2]V/[[partial derivative].sup.2]B < 0). Here, we run into a fundamental problem of neoclassical economics, namely, its reliance on convex technology sets. Under the opposite assumption of marginal economies of scale ([[partial derivative].sup.2]V/[[partial derivative].sup.2] B > 0)--which as implied above is more appropriate--it is far more difficult to obtain a general result, but reverse crowding out is still possible.

Matching funding arrangements are characteristic of some consortia, such as the Australian Research and Development Corporations (RDCs) (Alston, Freebairn, and James, 2004; Alston et al., 1999). In such arrangements, the subsidy is a function of member contributions. Suppose that the subsidy rate s is constrained to be a fixed proportion (which can assume any nonnegative value) of member contributions and is announced in advance. Then, member contributions are a (response) function of s, which will usually feature positive (crowding in) as well as negative (crowding out) sections for different intervals of s. This is one way in which crowding in and crowding out can coexist in a single game.

Furthermore, outside support can be multivariate. For instance, sponsors can provide infrastructure, consulting services, etc., to an R&D consortium, in addition to cash subsidies. Taking into account the multivariate nature of subsidies, sponsors may find a combination of crowding out (for some variables) and reverse crowding in (for other variables) optimal. This is another type of coexistence of crowding in and crowding out.

Reverse crowding out is central to FLAR's genesis. FLAR emerged as a funding arrangement to enable the continuation of an existing breeding program, namely, CIAT's irrigated rice program. Due to a budgetary setback unrelated to the program, it had to be discontinued; it was at this point that CIAT's Director General, Grant Scobie, and its rice economist, Luis Sanint, decided to try and mobilize funds from the Latin American rice sector (Binenbaum, Pardey, and Sanint, 2003). Inspired by the Australian RDC arrangement, they designed and founded FLAR. Thus, the breeding program was essentially continued, but its funding source was shifted from development assistance to local industry and public sector support. This is a clear example of reverse crowding out.

FLAR receives in-kind support through the usage of CIAT's facilities and its access to the CGIAR's network of personnel and germplasm exchanges. These advantages doubtlessly enhance FLAR's value to its members and stimulate their cash and in-kind investments in FLAR. Thus, FLAR demonstrates both the multivariate nature of outside support and the possibility of an incentive structure that combines reverse crowding out with crowding in.

C. Underfunding the to Differences in Willingness to Contribute
 In contrast to partnerships with smaller number of members, R&D
 consortia tend to feature standardized funding arrangements, which
 cannot fully exploit members' willingness to contribute. This is
 another cause of underfunding.

An important theme in the public goods literature is that private information on willingness to pay constiutes a major obstacle to efficient public goods provision. The public goods literature suggests that nifty revelation mechanisms such as the one devised by Groves and Ledyard (1977) achieve second-best results in the presence of such informational asymmetries. Such mechanisms are complex even in relatively simple games. They would thus be hard to design for R&D consortia due to the latter's excessive technological and strategic complexity. Thus, we rarely, if ever, observe sophisticated revelation mechanisms in R&D consortia.

Partnerships with fewer members offer more scope for fine-tuning contractual details. Transaction costs--for example, those associated with negotiations--increases rapidly with the number of players. Hence, multimember consortium funding arrangements tend to be based on simple formulas where member contributions are a function of some measure of member size.

Such simplicity and transparency come at a cost. Funding arrangements in R&D consortia will generally not be able to achieve second-best solutions a la Groves-Ledyard. Funding shares are based on an imperfect correlate of willingness to pay and will thus vary in relationship to players' true and undisclosed willingness to pay. The "marginal member"--the one whose funding share is the highest relative to her willingness to pay and just about agrees to join the consortium--may exert a disproportionate and downward influence on the consortium's overall budget. Thus, under-funding is likely to be worse than in the Groves-Ledyard case.

In FLAR, funding is tied to a country's rice output--a likely correlate of willingness (or ability) to pay. In this way, FLAR has saved itself cumbersome negotiations; this was a provision that all prosepective members could easily agree on. It is an open question whether one of the sophisticated revelation mechanisms could have been put in place with the consent of all members--at the very least, this would have entialed costlier negotiations and a more complex and less transparent consortium agreement.

The actual functioning of FLAR features a revelation mechanism of sorts: members can get more out of FLAR through supplementary investments, such as sending additional experts to meetings in excess of the amount of delegates required. The members with less developed rice R&D programs are the ones who tend to make such supplementary investments, presumably as a way to catch up. Such members would likely agree to a larger FLAR, if all others agreed to increase contributions.

D. The Ambiguous Effect of Member Size on Relative Contribution
 In most R&D consortia, the players' sizes matter. There are two
 opposing effects of size on relative contribution: the "greater
 stakes" and "bargaining power" effects.

Most formal models of R&D consortia do not take into account their members' sizes. These models often conceive of the typical consortium member as a firm without a budget constraint. Even with players who do not differ exante, a game may yield asymmetric equilibria in terms of strategies and/or payoffs. Models of R&D consortia with more than two members, such as that of Katz (1986), use ex post symmetry as a modeling device that enables them to prove many of their most interesting propositions. To our knowledge, only one recent article, Amir, Evstigneev, and Wooders (2003), manages to model an R&D consortium while disposing of ex post firm symmetry.

How to think of size asymmetries in a collective-action framework? In Olson's (1965) model of collective action, there is a tendency for exploitation of large group members by small ones, in the sense of the former bearing a disproportionate share of the costs. This is because larger members tend to have grater stakes in the joint project, creating an opportunity for smaller ones to free ride on the former's contributions. Let us call this the "stakes effect." This phenomenon can be observed, for example, in the United States bearing a disproportionate share of the North Atlantic Treaty Organization burden. However, in R&D consortia, this tendency is often contravened by the larger members' greater bargaining power. Large R&D players tend to have more bargaining chips and "outside options" due to their larger and more diverse pool of resources and partners. For this reason, they can be expected to extract favorable conditions, including lower funding shares, from other players. This is the "bargaining power effect." In politicized contexts, this effect is mitigated by legitimacy constraints: players do not wish (or cannot afford) to be perceived by their constituents as "suckers." Being "exploited" by other players does not look good.

As illustrated in Figure 2, FLAR's members vary widely in size. The largest member country, Brazil, has an output of irrigated rice which is about 80 times that of the smallest one, Guatemala, and a factor 10 or more larger than any of four other member countries. FLAR members differ vastly not only in terms of output but also in terms of R&D capabilities. Output is correlated with R&D capability. Brazil's NARS has a large budget and is a global player in both qualitative and quantitative terms. It has much to offer to FLAR or to other players in terms of rice varieties, expertise, facilities, etc. Brazil would be an interesting agricultural R&D partner for multinationals such as Monsanto. These factors contribute to Brazil's bargaining strength within FLAR. Similar considerations apply to other relatively large FLAR member countries such as Colombia. One might expect to observe an effect opposite to Olson's, namely, "exploitation" of the small by the great. However, that would probably contravene political legitimacy. FLAR definitely has political aspects, and it might well be unacceptable in domestic politics for the organization representing a small rice-producing country to be seen as securing an unfairly small share of FLAR's net benefits.

E. Ex ante IPR Allocation as an Incomplete Contracting Problem
 Relationship-specific investments and contractual incompleteness are
 pervasive features of R&D consortia. This helps us understand their
 IPR arrangements.

The success of any consortium whose primary mission is to conduct R&D depends on ongoing member contributions long after the initial consortium agreement has been signed. Such contributions can take many forms, such as personnel exchanges and the sharing of ideas, data, and physical resources. These contributions are generally relationship specific and impossible to fully specify contractually ex ante. Thus, insights from the incomplete-contracting literature (pioneered in Grossman and Hart, 1986, and Hart and Moore, 1988), which features games with incomplete ex ante contracts combined with relationship-specific investments and renegotiation, are applicable to R&D consortia. (6)

Due to its incompleteness, the ex ante contract is too blunt an incentive instrument to induce the optimum level of specific investment. Hence, some underinvestment relative to the first-best outcome is inevitable. (7) Gross-man and Hart (1986) show that the second-best ex ante property rights allocation is the one that gives the relevant property rights to the player whose specific investment adds the most value to the jointly developed product. Thus, property rights are allocated so as to minimize the efficiency loss due to underinvestment.

The key relationship-specific investment in FLAR is the sharing of PGR. Such sharing is problematic, even among consortium members. In cases such as FLAR, players would never have consented ex ante to complete PGR sharing. Under fundamental uncertainty concerning rice markets and the future value of any PGR, and under conditions of great strategic complexity, players are extremely reluctant to give up all control over their key assets--PGR. Therefore, it is left up to FLAR members to share, or not share, their PGR with FLAR. Thus, a consortium like FLAR needs an incentive mechanism to encourage PGR sharing. Part of this involves rules to protect against leakage of FLAR varieties to third parties. Another part of this mechanism involves ex ante rules on property rights.

It is clearly not feasible to stipulate ex ante a property rights allocation for each and every specific variety developed by FLAR--uncertainty and complexity are too great for this. For example, it is not even clear ex ante which successful varieties will emerge from FLAR's breeding efforts. Thus, we have a clear case of incomplete contracting, where most future contingencies cannot be covered by an ex ante contract (the consortium agreement). Instead, the best the ex ante contract can do is to stipulate a few simple, general rules on property rights allocation. By Grossman and Hart's (1986) logic, this allocation should correspond to the relative importance of members' specific investments. FLAR has an arrangement in place, which roughly--but only roughly--follows this principle. Each member gets the domestic plant variety protection rights to plant varieties developed by FLAR. The size of a country's domestic market is correlated with its plant breeding capabilities and its potential to contribute to FLAR's breeding program.

F. External Toughness Fosters Internal Cohesion
 The consortium arrangement impinges on relations with third parties.
 Internal resource sharing in R&D consortia can be encouraged by
 restrictions on external sharing. The anticompetitive aspect of R&D
 consortia can be welfare enhancing if it counterbalances third
 parties' market power.

To complete our description of elements of the R&D consortium game, we consider provisions regarding nonmembers. The two main categories of nonmembers are those eligible for membership and those not eligible. The consortium must find ways to restrict non-member access to its outputs. Thus, non-members cannot free ride on consortium investments and eligible ones have an incentive to join. Moreover, cooperative behavior--especially resource sharing--by consortium members is encouraged by provisions preventing "leakage" to third parties.

Rules on the allocation of IPR and associated licensing provisions play a key role in the design of such incentives. The consortium may hold internally generated IPR and exclusively license it to members. Furthermore, the consortium agreement may stipulate reach-through provisions in such licenses restricting sublicensing or making it conditional on consortium approval.

The consortium itself may be a means for competitors to reduce competition in the output market. It may even be used to reduce the members' overall (costs of) innovative activity by eliminating competitive R&D. Thus, there is a fundamental tension in R&D consortia: the rationales for cooperation must be weighed against antitrust considerations (Katz, 1986; Katz and Shapiro, 1987; Shapiro and Willig, 1990). However, collusion can be welfare enhancing if it counterbalances the market power of third parties.

In summary, external toughness--including restrictions on insider-outsider interactions--tends to foster internal cohesion.

FLAR illustrates all these points. While it is not incorporated and thus cannot itself hold IPR, FLAR members get to hold the rights to FLAR rice varieties in their respective home jurisdictions. Agreements between a FLAR member and a third party (e.g., a multinational) that involve the transfer of FLAR materials or IPR generated by FLAR require the approval of the other members. An interesting FLAR provision that has a collusive aspect is its rule for royalties. The FLAR agreement stipulates specific formulas for royalties percentages, should FLAR members license their plant breeders' rights for FLAR varieties to third parties. This provision is included in the agreement to prevent FLAR members from being played off against each other when they negotiate licensing contracts with third parties. This collusive provision may well be justifiable in social welfare terms because it enhances members' incentives to share their PGR with FLAR and strengthens their bargaining positions with multinationals, thus counterbalancing the latter's market power. Private sector plant breeding and related technological tools and IPR are highly concentrated in the hands of a few powerful multinationals. Concerns about leakage of FLAR resources and products to multinationals played a prominent role in early FLAR discussions. Relevant FLAR provisions were designed to allay such concerns (Binenbaum, Pardey, and Sanint, 2003).


A. Characterization of the Big-Picture Game as a Repeated PD
 Consortium games can be simplified to a multistage PD with multiple
 levels of (non) cooperation and unknown horizon. Thus, we can apply
 the Folk Theorem, that is, identify incentive-compatible scenarios.

The amalgam of incentive problems discussed in the previous section might seem to be exceedingly complex and unwieldy, impossible to analyze as a single game. Not so: the overall picture can be drastically simplified. Collective action games, including R&D consortia, are often PDs. PDs have two key features: mutual positive externalities of cooperation (or, equivalently, negative externalities of noncooperation) and Pareto superiority of symmetric cooperation over symmetric non-cooperation. Both features are generally implied by the three technical rationales of Section 2. However, the incentive problems of Section 3 diminish both the positive externalities and the Pareto superiority of cooperation. If, weighed against the rationales, the incentive problems are not too severe, a PD characterization of the game is appropriate. If they are too severe, then nothing happens (i.e., the outcome is noncooperation) and the game is not interesting.

The kind of PD game that is relevant to R&D consortia is characterized by repetition and by an uncertain horizon. Players will generally interact repeatedly and they do not know when their interactions will end. The Folk The-orem says that a wide range of subgame-perfect Nash equilibria (SPNE) are sustainable in an infinitely repeated PD, including perpetual noncooperation and perpetual cooperation. Neyman (1999) extends this result to the case of finitely repeated PD where the common knowledge assumption regarding the number of stages is slightly perturbed. Neyman does not explicitly model uncertain payoffs that vary from one period to the next--as in FLAR (8) and most real-world games--but his result should be generalizable to this case.

For our purposes, it is further useful to allow for multiple levels of cooperation and for restrictions on the strategy space. A key simplifying element here is that the level of cooperation summarizes multiple underlying variables, such as funding, input sharing, information sharing, IPR sharing, etc. The extension of the PD to multiple levels of cooperation is straightforward: if there are, say, four levels of cooperation, then for each pair of levels considered separately, we have PD payoffs. An example of a restriction on the strategy space is that differences of level of cooperation can be only 0 or 1. Thus, if one player plays Level II, then Level IV is ruled out--only Levels I, II, and III can be played by the other players. If one player plays Level I, then the others can play only I or II.

How to construct the big-picture game in the FLAR case? Consider the game played by the primary players--the representatives:

1. A representative's set of actions can be reduced to four levels of participation: (I) none; (II) support FLAR at level close to its minimal critical mass, with limited sharing of PGR, data, and expertise; (III) support common FLAR budget at level close to its minimal critical mass but with supplementary investments to enhance technology transfer and with more extensive sharing; and (IV) support FLAR at high level, maximizing joint payoff, with extensive sharing.

2. PD payoffs to a country's rice sector are associated with an increase in participation

3. Adjustments to the PD payoffs must be made due to size asymmetrics, output market competition, and own-institution bias. For example, own-institution bias implies a divergence between the interests of a country's rice sector and its FLAR representative: while the former receives PD payoffs, the latter (the actual player) perceives it would be made worse off by increases in overall participation to Levels III and IV, as these would strengthen FLAR too much.

4. Legitimacy constraint and consensual decision making: an arrangement based on willingness to pay is ruled out, and instead, all players must support FLAR's budget at a proportionately equal level--as in the observed budget rule. Thus, players contributing at Levels II and III can coexist, but it anyone plays II or III, then nobody will play IV.

This game features multiple stages. Every year, players may leave or join FLAR (implying another adjustment in payoffs: a change from or to Level I is costly) and make decisions on the sharing of PGR and other resources; members can renegotiate the FLAR agreement, increasing or decreasing their support.

Due to the output competition and own institution bias, Levels III and IV are not incentive compatible for some players. Combined with legitimacy constraints, this implies that nobody will play IV. This still leaves a range of incentive-compatible scenarios where some players play I or II and the remaining players play I, II, or III. Which of these prevails?

B. Leadership as an Equilibrium Refinement
 We introduce a novel leadership concept, namely, the ability to
 choose one of the game's incentive-compatible scenarios in order to
 maximize a social welfare function. This function will often
 incorporate the interests of players not represented in the

Generally, when a game-theoretic solution concept yields multiple solutions, refinements are called for. In this case, we have a repeated PD yielding multiple SPNE. How to narrow down this range to a unique solution? Here, the focus is on leadership.

In the context of our logic, leadership can be defined in two ways: (1) the ability to alter the payoffs (e.g., through persuasion) or structure of the game such that players reach a dynamic equilibrium socially superior to the equilibrium in the absence of the leader's intervention or (2) the ability to choose one of the equilibria on the basis of some social welfare criterion. (9) Let us call these Type I and Type II leadership, respectively.

The personal leadership of Scobie and Sanint may have been crucial for the formation of FLAR. Although it is impossible to prove this counterfactual, one senses that, absent their initiative, the joint Latin American program of irrigated rice breeding would have died prematurely following CIAT's budget cut.

In our logic, the factors conducive to leadership (Figure 1) determine the extent to which both types of leadership can be exercised. Were these factors present in the formation of FLAR? Scobie and Sanint did not have formal power over the future FLAR members. They did not control significant financial incentives, but CIAT did possess breeding expertise and facilities as well as connectivity to global nonprofit rice research and PGR flows. They understood the system, that is, they properly anticipated the FLAR game--for example, they sensed a significant willingness to pay in the Latin American rice sector for some sort of continuation of the breeding program. They had professional standing but not to the extent that they could affect the other players' institutional structure and objectives. Importantly, they were perceived to be impartial--not biased in favor or against any players. In summary, conditions were in place for Type II leadership, but by and large not for Type I leadership.

The social welfare function depends on the leader's views. For example, Scobie and Sanint were probably interested in promoting not just the rice industry's but consumers' interests as well. But while the FLAR arrangement addresses to some extent mutual positive externalities among members, it does not address externalities that are external to their group. Rice consumers derive a surplus from FLAR activities but are not represented in it. Advances in rice breeding and agronomy inspire future agricultural innovations. Such "external externalities" can only be internalized through public or philanthropic funding.

FLAR's history can be interpreted as follows: Scobie and Sanint exercised leadership. They offered core players--leading institutions in the continent's rice sector--a win-win solution whereby the latter moved from Level I participation to Level II or III, depending on the institutional payoffs for each country's representative. The players might not have achieved this on their own as none of them sufficiently satisfied the conditions for leadership. However, given the limitations to Scobie's and Sanint's leadership role, they could not possibly have achieved the first-best outcome, namely, Level IV participation by all primary players. Further, Pareto improvements (not counting narrow institutional interests) can only happen if latent players become active. Consider two examples: first, farmers or an activist government in a country which is currently a Level II player could force the country's representative to move to Level III. If this happened to all Level II players, an overall move to Level IV--a significant FLAR expansion--might become feasible. Second, FLAR could be supplemented by philanthropic funding to bring it closer to Level IV operations, say, comparable to WARDA. Such funding could include matching arrangements (as in the RDCs) to incentivize current Level II players to increase their support of FLAR.


Real-world R&D consortia appear to be an extremely complex and diverse phenomenon that does not lend itself for game-theoretic modeling. Nevertheless, on closer inspection, there are many common features that most R&D consortia share.

First, they are motivated by a mix of technical rationales: their outputs are (pure or impure) public goods; their R&D technologies exhibit economies of scale (and often scope as well); their R&D inputs are complementary. These rationales can be summarized as "mutual positive externalities."

Second, a range of incentive problems hamper the players' efforts to realize the mutual positive externalities of cooperation. Due to the relationship-specific nature of consortium investments and the incomplete nature of contracts, holdups and underinvestment--that is, suboptimal contributions to the consortium of various kinds--occur. Players' willingness to contribute to the consortium in various ways cannot be observed but only (imperfectly) be inferred from their actions. "Leakage" of consortium resources and outputs to third parties is costly to prevent. Size asymmetries cause additional transaction costs. All these factors exacerbate underinvestment.

Third, the consortium contract is a response to these incentive issues. Relationship-specific investments such as the sharing of inputs, people, and information can be encouraged by a carefully tailored ex ante allocation of intellectual and other property rights. Membership contributions are based on (albeit imperfect) correlates of willingness to pay. Provisions against leakage of consortium resources and outputs to third parties encourage internal sharing. Collusive elements in the consortium agreement increase member payoffs and strengthen their bargaining position vis-a-vis third parties.

These incentive issues are fruitfully analyuzed in the respective strands of literature that address them (except for the issue of external leakage, which appears to have barely been addressed in applied game theory so far). While combining them explicity in a single game model may not be feasible. games that focus on each issue separately are useful for understanding real-world consortia.

Fourth, it is still possible and useful to construct an overall game that abstracts from specific incentive issues. As noted in the collective-action literature, there is a striking generality to the PD. We interpret the R&D consortium game as a multistage PD with multiple levels of (non) cooperation and unknown horizon. Thus, we can apply the Folk Theorem, that is, identify incentive-compatible scenarios. This is a fruitful way of thinking about real-world consortia: a range of alternative scenarios would have been compatiable with players' incentives.

Another key real-world element of the consortium game which can be modeled as a PD game is the existence of multiple levels of (non) cooperation.

Fifth, we can explain (or even predict) which of the incentive-compatible scenarios emerges by reference to leadership. The consortium's leaders help determine the level at which the consortium is funded and the extent of members' cooperativeness. The extent or strength of leadership can, in turn, be explained (or predicted) by a range of socioeconomic and cognitive conditions.

Jointly, these five elements comprise a logic of collective action in R&D which, we hope, will help bridge the gap between applied game theory and real-world consortia (and other institutions).

Benevolent designers of R&D consortia can combine this scheme with incentives to address external externalities. Almost all the internal and external incentive issues cause underfunding. In order to properly address both the internal incentive issues and the external externalities, a benevolent catalyst should recognize the multidimensional nature of subsidies and the coexistence of crowding in and negative crowding out. That is, (1) in addition to cash subsidies, various kinds of in-kind support are important in incentivizing (prospective) consortium members; (2) it makes sense to increase some of these choice variables in order to induce greater contributions by consortium members; while (3) other choice variables should be decreased to encourage cooperation.

As to follow-up work, much remains to be done in further developing our conceptual logic and using it to generate and test hypotheses. Furthermore, our article contains a wealth of largely undeveloped modeling ideas. The reader is invited to pursue these.


Alston, J. M., J. W. Freebairn, and J. S. James. "Levy-Funded Research Choices by Producers and Society." Australian Journal of Agricultural and Resource Economics, 48, 2004, 33-64.

Alston, J. M., M. S. Harris, J. D. Mullen, and P. G. Pardey. "Agricultural R&D Policy in Australia, " in Paying for Agricultural Productivity, edited by J. M. Alston, P. G. Pardey, and V. H. Smith. Baltimore: Johns Hopkins University Press, 1999, 118-71.

Amir, R., I. Evstigneev, and J. Wooders. "Noncooperative versus Cooperative R&D with Endogenous Spillover Rates." Games and Economic Behavior. 42, 2003, 183-207.

Binenbaum, E., P. G. Pardey, and L. R. Sanint. "Intellectual Property Arrangements in a Public-Private International R&D Consortium: The Latin American Fund for Irrigated Rice." School of Economics, University of Adelaide: Unpublished Paper, 2003.

Buchanan, J. M. "An Economic Theory of Clubs. " Economica, 32, 1965, 1-14.

Cohen, L. "Holdouts," edited by P. Newman, New York: MacMillan, 1998.

Graff, G. D., G. C. Rausser, and A. A. Small. "Agricultural Biotechnology's Complementary Assets." Review of Economics and Statistics, 85, 2003, 349-63.

Grossman, S. J., and O. D. Hart. "The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration." Journal of Political Economy, 94, 1986, 691-719.

Groves, T., and J. Ledyard. "Optimal Allocation of Public Goods: A Solution to the 'Free Rider' Problem." Econometrica, 45, 1977, 783-809.

Hagedoorn, J., A. N. Link, and N. S. Vonortas. "Research Partnerships." Research Policy, 29, 2000, 467-86.

Hardin, G. "The Tragedy of the Commons." Science, 162, 1968, 1243-8.

Hart, O., and J. Moore. "Incomplete Contracts and Renegotiation." Econometrica, 56, 1988, 755-86.

Heller, M. A., and R. S. Eisenberg. "Can Patents Deter Innovation? The Anticommons in Biomedical Research." Science, 280, 1998, 698-701.

Hermalin, B. E. "Toward an Economic Theory of Leadership: Leading by Example." American Economic Review, 88, 1998, 1188-206.

Katz, M. L. "An Analysis of Cooperative Research and Development." RAND Journal of Economics, 17, 1986, 527-43.

Katz, M. L., and C. Shapiro. "Research-and-Development Rivalry with Licensing or Imitation." American Economic Review, 77, 1987, 402-20.

Klein, B. "Hold-Up Problem, " in The New Palgrave Dictionary of Economics and the Law, edited by P. New man, New York: Macmillan, 1998.

Kraus, J. "Hobbes and Contractarianism," in The New Palgrave Dictionary of Economics and the Law, edited by P. Newman, New York: Macmillan, 1998.

Maskin, E., and J. Tirole. "Unforeseen Contingencies and Incomplete Contracts." Review of Economic Studies, 75, 1999, 597-626.

Neyman, A. "Cooperation in Repeated Games When the Number of Stages in Not Commonly Known." Econometrica, 67, 1999, 45-64.

Olson, M. A. The Logic of Collective Action, Cambridge, MA: Harvard University Press, 1965.

Pardey, P. G., and N. M. Beintema. Slow Magic: Agricultural R&D a Century After Mendel. Washington, DC: International Food Policy Research Institute, 2001.

Rosenkranz, S., and P. W. Schmitz. "Optimal Allocation of Ownership Rights in Dynamic R&D Alliances." Games and Economic Behavior, 43, 2003, 153-73.

Salanie, B. The Economics of Contracts. 2nd ed, Cambridge, MA: MIT Press, 2005.

Samuelson, P. A. "The Pure Theory of Public Expenditure." Review of Economics and Statistics, 36, 1954, 387-9.

--. "A Diagrammatic Exposition of a Theory of Public Expenditure." Review of Economics and Statistics, 37, 1955, 350-6.

Shapiro, C., and Willig, R. D. "On the Antitrust Treatment of Production Joint Ventures." Journal of Economic Perspectives, 4(3), 1990, 113-30.


AI: Artificial Intelligence

CGIAR: Consultative Group on International Agricultural Research

CIAT (Spanish acronym): International Center for Tropical Agriculture

FLAR: (Spanish acronym): Latin American Fund for Irrigated Rice

IP: Intellectual Property

IPR: Intellectual Property Rights

NARS: National Agricultural Research Systems

PD: Prisoners' Dilemma

PGR: Plant Genetic Resources

R&D: Research and Development

RDCs: Research and Development Corporations (Australia)

SPNE: Subgame-Perfect Nash Equilibria

WARDA: West African Rice Development Association (also known as Africa Rice Center)


* Thanks are due to John Mathews, Phil Pardey, Ahmad Rahman, and Perry Shapiro; to attendants of 2005 conference sessions of the WEAI, AAEA, and ASE; to two anonymous referees; and to seminar participants at the University of Sydney, School of Economics and Political Science, for helpful comments. Additionally, I am grateful to Pardey for helping to produce Figure 1 and to Luis Sanint (formerly of FLAR) for some very useful ideas and expert insider insights. The usual disclaimer applies.

Binenbaum: School of Economics, University of Adelaide, SA 5005, Australia. Phone +61-8-84311405, Fax +61-8-82231460, E-mail

(1.) A companion article (Binenbaum, Pardey, and Sanint, 2003), focuses on the details of the case study and draws more specific conclusions on agricultural research policy.

(2.) Additional rationales for R&D consortia include mutual learning, risk reduction, duplication avoidance, and network positioning (Binenbaum, Pardey, and Sanint, 2003).

(3.) For a handy summary of this literature, see Salanie (2005).

(4.) We are referring here to modern R&D not to the important, ubiquitous, and age-old stream of agricultural innovations by successive generations of farmers. For data on public versus private funding of R&D worldwide, see Pardey and Beintema (2001).

(5.) WARDA serves similar numbers of producers and consumers, but due to organizational and funding constraints, an arrangement similar to FLAR may not be feasible for West African rice R&D.

(6.) Rosenkranz and Schmitz (2003) suggest a way of modeling this aspect of R&D consortia. Their model combines incomplete contracting with mutual know-how disclosure in a two-firm partnership.

(7.) Maskin and Tirole (1999) dispute this conclusion by showing that sophisticated contracts can in fact achieve first-best outcomes in an incomplete contracting setting. However, such contracts are generally not feasible in R&D consortia due to the latter's complexity.

(8.) Key uncertainties for FLAR include (1) the future performance of plant varieties yet to be bred and (2) rice price fluctuations in global markets. from one level to the next. That is, if player X moves up one level while the other's do not, this move makes X worse off while making the others better off, but if a sufficiently large subset of players move up on level, then all benefit in net terms.

(9.) Many different leadership concepts have been proposed in the management literature. Hermalin (1998) appears to be the only extant game-theoretic model of leadership. His definition of leadership--the ability to lead by example (in a moral-hazard-in-teams setting)--is narrower than ours. However, Hermalin's leadership concept, ours, and those advanced in the management literature are not mutually incompatible. They highlight different aspects of leadership, which is a complex phenomenon.
COPYRIGHT 2008 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2008 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:research and development
Author:Binenbaum, Eran
Publication:Contemporary Economic Policy
Article Type:Report
Geographic Code:1USA
Date:Oct 1, 2008
Previous Article:Are financial sector policies effective in deepening the Malaysian financial system?
Next Article:University cost structure in Taiwan.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters