The (Long-Dormant) Relational-Adaptation Theory of the Firm

We revisit (and seek to resuscitate) one of the earliest arguments in the theory of the firm: Williamson’s (1975) use of Simon’s (1951) conception of an employment relationship to argue that integration is the efficient governance structure for transactions that require adaptive, sequential decision-making. We contend that this relational-adaptation argument is distinct from later work by Klein, Crawford, and Alchian (1978), Williamson (1979, 1985), and Grossman and Hart (1986) that emphasizes hold-up motivated by specific investments. Furthermore, we argue that, while the relational-adaptation argument was informal and incomplete in 1975, it can now be formalized and completed, and that doing so delivers both theoretical and empirical insights. * We are grateful for research support from Harvard Business School (Baker and Gibbons), MIT’s Sloan School and Program for Research on Innovation in Markets and Organizations (Gibbons), and USC’s Marshall School (Murphy). s Happens: The (Long-Dormant) Relational-Adaptation Theory of the Firm by George Baker, Robert Gibbons, and Kevin J. Murphy We revisit one of the earliest arguments in the theory of the firm: Williamson’s (1975) use of Simon’s (1951) conception of an employment relationship to argue that integration is the efficient governance structure for transactions that require “adaptive, sequential decision-making” (Williamson 1975: 40). Quite unusually (at least for an economics paper), we begin with a line-by-line analysis of key passages and citations from these classic works. This foray into intellectual history suggests that the informal relationaladaptation theory of the firm developed by Simon and Williamson is distinct from later work by Klein, Crawford, and Alchian (1978), Williamson (1979, 1985), and Grossman and Hart (1986) that emphasizes hold-up motivated by specific investments. To prove that the informal Simon-Williamson theory is distinct from the subsequent specific-investment theories, we develop a formal theory of the firm that has no specific investments.1 Instead, the relational-adaptation theory of the firm analyzes the incentives for adaptation as uncertainty is resolved. We do not argue that great effort or ingenuity was required to develop this formal theory. To the contrary, the theoretical section of this paper defines a simple model and analyzes it using familiar techniques. Indeed, the simplicity of this resuscitation of the relational-adaptation theory of the firm presents something of a puzzle: why was the theory dormant for so long?2 Having revisited and resuscitated the basic relational-adaptation theory of the firm, we then ask how this theory can be extended, applied, and tested. We describe extensions 1 As Whinston (2002) emphasizes, there are two different theories of the firm that involve specific investments: the “transaction cost” theory of Klein, Crawford and Alchian (1978) and Williamson (1979, 1985), and the “property rights” theory of Grossman and Hart (1986), Hart and Moore (1990) and Hart (1995). See Gibbons (2002) for more on how these two theories compare to the “relational adaptation” theory described here and the “incentive system” theory of Holmstrom and Milgrom (1991, 1994), Holmstrom and Tirole (1991), and Holmstrom (1999). 2 Krugman (1995: 27) may provide part of the answer: “Like it or not, the influence of ideas that have not been embalmed in models soon decays.” But Coase (1937) is a counter-example to Krugman’s rule, so we may need more than its informality to explain the long dormancy of the relational-adaptation theory of the firm. SEPTEMBER 20, 2002 S HAPPENS 2 to richer settings than the classic make-or-buy problem we consider here, such as the “hybrid” governance structures we consider in Baker, Gibbons, and Murphy (2002a). As we discuss in that paper, there is already empirical work motivated by an informal relational-adaptation theory of the firm, such as Gulati (1995) and Robinson and Stuart (2002). Thus, while the basic relational-adaptation theory is being formalized only now, it is also being extended and applied at the same time. Finally, discussing these extensions and applications of the relational-adaptation theory of the firm leads naturally to a broader discussion of relational contracts in the theory of the firm, such as in property-rights models and agency models. We explain that, in contrast to these property-rights and agency streams of research, the relational-adaptation approach we develop here has no ex ante actions, so relationships have no role to play in improving ex ante incentives. More broadly, we discuss these complementary propertyrights and agency models, in an effort to organize the burgeoning theoretical and empirical literature on relational contracts in the theory of the firm. In sum, we believe it is possible to separate the core argument in the relationaladaptation theory initiated by Simon (1951) and Williamson (1975) from the core arguments in the specific-investments theories initiated by Klein, Crawford, and Alchian (1978), Williamson (1979, 1985), and Grossman and Hart (1986). We also believe it is useful to separate these arguments. For example, clarifying these arguments may allow us to understand their boundary conditions, to test their competing predictions, and to integrate their elements into a richer theory. But so far the specific-investment arguments have received much more attention. Thus, in this paper, we seek to resuscitate the longdormant relational-adaptation theory of the firm. We present intellectual history in Section 1 and theory in Section 2. We then discuss extensions and evidence in Section 3 and the broader literature on relational contracts and the theory of the firm in Section 4. 1. Intellectual History In this section we undertake a detailed analysis of key passages and citations from Simon (1951) and Williamson (1971, 1975, 1979). The first step in Williamson’s (1975) argument is now familiar: formal contracts (i.e., those that can be enforced in a court) are incomplete. This is not to say that all contracts are useless: some one-shot transactions can be efficiently governed by spot contracts, some long-run transactions in stationary environments can be efficiently SEPTEMBER 20, 2002 S HAPPENS 3 governed by long-term contracts, and other long-run transactions in fluctuating environments can be efficiently governed by sequences of spot contracts. But there remains an important class of transactions that cannot be well governed by sole reliance on any formal contract. Simon’s model of an employment relationship analyzes a labor transaction that cannot be well governed by sole reliance on a formal contract. In Simon’s model, a limited version of adaptive, sequential decision-making can be achieved by giving the boss decision-making authority. More specifically, Simon solved for the subgame-perfect equilibrium of a two-stage game, and showed that the boss’s self-interested state-dependent decision ex post might be superior to a state-independent decision negotiated ex ante. But Simon also informally discussed how a repeated game could improve on this one-shot outcome (p. 302), perhaps achieving first-best adaptive, sequential decision-making. In this sense, Simon’s conception of an employment relationship, including the possibility of a repeated game, was richer than his formal model of an employment relationship. The second step of Williamson’s (1975) argument involves something akin to Simon’s richer conception of an employment relationship. This second step is by analogy: if certain labor transactions should be governed by giving the boss authority, then analogous transactions between firms should be governed by giving one of the firms authority (i.e., by integration). To see this analogy in more detail, consider Chapters 4 and 5 of Markets and Hierarchies. In Chapter 4, Williamson gives a detailed discussion of how real employment relationships, such as those in internal labor markets, may deliver “consummate” adaptive, sequential decision-making. Chapter 5 then focuses on intermediate products rather than labor transactions, and contains probably the most important paragraph in the book: “The argument here really parallels that of Chapter 4 in most essential respects. What one wants to devise is a contractual relation that promises fair (competitive) returns [and] promotes adaptive efficiency... Inside contracting is deficient ... Shifting inside contractors from a quasiautonomous bargaining status to an employment relation has advantages...” (1975:99). In short, the second step of Williamson’s (1975) argument asserts that what works for people will work for divisions: to achieve adaptive, sequential decision-making without a formal contract, aggregate units of the firm should be governed by employment SEPTEMBER 20, 2002 S HAPPENS 4 relationships (i.e., their authority should be taken away and given to a boss; they should be run as divisions rather than independent contractors). In partial contrast to Markets and Hierarchies, Williamson (1979, 1985) emphasized an equally important but ultimately distinct argument. This second argument again begins by observing that formal contracts are incomplete, but its second step is new: specific investments generate appropriable quasi-rents, which inspire socially counterproductive rent-seeking; integration is the efficient governance structure if it can eliminate such rent-seeking. As is well known, Klein, Crawford, and Alchian (1978) also forcefully articulated this “hold-up” argument, and it has been further developed by Klein (1991, 2000). It may be less well known that the seeds of this argument can also be found in Williamson (1971). Thus, we are not suggesting that in 1975 Williamson knew nothing of specific investments and hold-up. Instead, we are asserting that these issues were not crucial to the main argument in Markets and Hierarchies: Chapter 4 (and hence Chapter 5) of that book gives a prominent role to Simon’s conception of an employment relationship; in contrast, Simon’s paper is not even cited in Williamson (1971, 1979) and receives only scant mention in Williamson (1985). 2. The Relational-Adaptation Theory of the Firm In this section we develop first the staticand then the relational-adaptation theory of the firm. Our agenda owes much to Simon: he informally provided the repeated-game conception of an employment relationship, and he also posed the make-or-buy problem for employment transactions by asking when decision rights should be allocated to the worker instead of to the boss (p. 304). But Simon’s formal model implicitly assumes that not only decision rights but also decisions themselves are contractible. As we discuss below, one cannot construct a relational-adaptation theory of the firm from this assumption, so we motivate and impose an alternative assumption. We intend our model to be canonical – a base case that can be extended and tailored to various applications, rather than a second-generation model that explains stylized facts or delivers testable predictions for a specific environment. We therefore keep the model simple but abstract: there are only two parties and one asset, but the states, decisions, and payoffs are quite general. Hart and Holmstrom (2002) and Aghion, Dewatripont, and Rey (2002) present richer static adaptation models; we emphasize relational adaptation. SEPTEMBER 20, 2002 S HAPPENS 5 2.A The One-Shot Game In a model of adaptation, uncertainty must be resolved and then a decision must be taken. Let s denote the state, drawn from the finite set S according to the probability density f(s), and let D denote the set of feasible decisions, with generic element d. Let πA(d, s) denote the payoff to the owner of the asset, A, if decisions d are taken in state s; similarly, let πi(d, s) denote the private benefit to party i (= 1,2) if decisions d are taken in state s. The state, the decision, and the payoffs are observable but not verifiable.3 In the tradition of Grossman-Hart-Moore (GHM), we assume that the owner of the asset controls the decision (say, over the way the asset will be used), and we assume that asset ownership is contractible. Maskin and Tirole (1999) propose a reinterpretation of this GHM framework: there is no asset, but the decision right can be allocated to one party or the other via contract. As we describe in Section 3, we envision applications of the relational-adaptation theory under each interpretation – asset ownership and contractible decision rights. In both cases, we intend our model to capture something like Simon’s notion of a boss’s authority: control over a specified set of feasible decisions, D. While our focus on asset ownership is in the GHM tradition, we depart from their framework by imposing the following assumption: decision rights (i.e., who makes the decision) are contractible, but decisions (i.e., what decision is made) are not, even ex post. (In the asset-ownership interpretation, our assumption can be restated as: asset ownership is contractible, but asset utilization is not, even ex post.) We discuss this departure from the GHM framework below, but to clarify this discussion, we turn first to the timing and analysis of our one-shot game. The timing of the one-shot game is as follows: (a) asset ownership is determined; (b) the state is publicly revealed; (c) the asset owner chooses a decision; and (d) payoffs are received. The parties are risk-neutral. In this environment, the first-best asset-utilization decision d(s) solves 3 The assumption that the state, the decision, and the payoffs are not verifiable allows us to omit court-enforceable contracts from our analysis. The usual argument given for this omission is that the assetownership analysis pertains to the inevitable gaps in the court-enforceable contracts. But this argument ignores the possibility that the effects of asset ownership may interact with the terms of court-enforceable contracts, as in our (1994) paper. Thus, a superior approach would include the court-enforceable contracts in the analysis, along with asset ownership and relational contracts, as we began to do in a simple setting in our (2001) paper. We hope to pursue this approach in future work. SEPTEMBER 20, 2002 S HAPPENS 6 (1) max d ∈D 1 (d,s) + 2(d,s) + A (d,s),