The Hold-Up Problem in Protocol Economies
Why specific investment under mutable rules creates a commitment problem — and why "you can always fork" is not an answer
There is an old result in industrial organisation that economists have been refining for half a century. When a buyer and a seller make investments that are valuable only inside their relationship, and one party retains discretion to revise terms after the other has committed, the party with discretion can extract rents at the committed party’s expense. This is the hold-up problem. It is the reason firms integrate vertically, write long-term contracts, take ownership stakes, and design elaborate governance structures around what looks, from the outside, like ordinary commerce. The problem is not exotic. It is the central problem of relationship-specific investment under incomplete contracts, and it determines a great deal of how the real economy is organised.
Protocol economies recreate this problem in a new form, and a substantial part of the field has not noticed. The argument of this essay is direct: when participants make protocol-specific investments and a coalition retains effective discretion over base-layer rule revision, the situation is a textbook hold-up problem. Investment is discounted ex ante. The discount obtains whether or not the discretion is ever exercised. Forks do not solve it, because the cost of forking is the asset specificity of the original investment. The widespread response that “exit is always available” misunderstands the structure of the problem.
I will set out the mechanism formally enough to be useful, address the standard objections directly, and end with what the analysis implies for how protocol economies should be evaluated by participants who have not yet committed and by analysts who are trying to explain the patterns of those who have.
1. What asset specificity actually is
The concept of asset specificity is older than its formal name. Williamson (1975, 1985) systematised it as the central object of transaction-cost economics, but the underlying observation goes back to Coase (1937) and earlier institutional economics: not all capital is fungible, and the degree to which capital loses value when redeployed outside its original use is a critical determinant of how the relationship around it must be governed.
Williamson distinguished several forms of specificity. Site specificity arises when assets are physically located to support a particular relationship. Physical asset specificity arises when equipment is engineered for particular counterparties or processes. Human asset specificity arises when workers acquire skills useful only inside a particular relationship. Dedicated asset specificity arises when capacity is built to serve a particular customer at a scale that would not be justified by general demand. Brand specificity arises when reputational capital is tied to particular partners. Temporal specificity arises when timing of delivery is critical and substitution is impossible within the relevant window.
In each case, the economic content is the same. The asset has higher value inside the relationship than outside it, and the difference between those two values — the quasi-rent, in Klein, Crawford and Alchian’s (1978) formulation — is what the counterparty with discretion can attempt to appropriate. Once the investment is sunk, the original cost is no longer relevant to the bargaining; what matters is the asset’s salvage value in alternative uses, which by hypothesis is lower than its in-relationship value. The party with discretion can therefore extract any amount up to the difference without driving the committed party to exit.
The classical example is the supplier who builds a plant adjacent to a single buyer’s facility. Once the plant exists, the buyer can renegotiate the price downward toward the supplier’s marginal cost rather than the originally contracted price. The supplier’s only alternative is to walk away from a plant that is worth substantially less anywhere else. The supplier knows this in advance. The supplier therefore does not build the plant — or builds a smaller one, or demands a contract with strong commitment provisions, or insists on integration into the buyer’s firm — unless the governance terms address the problem.
This is the classical mechanism. The interesting question is whether protocol economies reproduce it, and the answer is straightforward.
2. Protocol-specific investment
Participants in a protocol economy make several categories of investment that are specific to that protocol’s rule set. Consider the actors who operate above the consensus layer:-
Application developers build software whose logic depends on the protocol’s transaction format, state-transition rules, fee structure, settlement guarantees, and timing semantics. The software is not transferable to a protocol with different semantics without substantial rewriting. The development cost is sunk against a specific rule set.
-
Exchanges and custodians integrate wallet infrastructure, key management systems, withdrawal logic, address validation, and compliance reporting against a particular protocol. The integration cost is real. Migration to a different protocol is not a software switch; it is a re-engineering programme with operational risk.
-
Merchants and payment processors build acceptance infrastructure, train staff, configure point-of-sale systems, and adjust accounting and tax workflows. These investments are specific to the protocol’s transaction model, confirmation semantics, and refund mechanics.
-
Compliance infrastructure includes transaction monitoring, sanctions screening, suspicious activity reporting, and audit systems. Each of these is calibrated to a particular protocol’s data structures and identity model.
-
Liquidity providers and market makers deploy capital into automated systems whose strategies depend on the protocol’s fee structure, ordering semantics, and oracle infrastructure. Their capital is specific in the sense that retiring it from one protocol incurs migration costs that do not appear in standard return calculations.
-
Users and holders accumulate balances, records, and counterparty relationships that have value within the protocol’s accounting and that lose value to the extent they cannot be carried across.
And consider the actors operating at the consensus layer:-
Miners, in proof-of-work systems, deploy specialised hardware whose value collapses outside the algorithm it was built for. ASIC investment is among the most extreme cases of physical asset specificity in the modern economy. A change in the proof-of-work algorithm can render an entire fleet worthless.
-
Validators, in proof-of-stake systems, post bonded stake and operate infrastructure designed for a particular protocol’s slashing rules, key management requirements, and uptime expectations. The bonded stake is, by design, specific to the protocol.
-
Mining pools and staking services build operational infrastructure — payout logic, monitoring, customer interfaces — that is specific to the underlying protocol.
This is, by any reasonable transaction-cost analysis, a population of actors making asset-specific investments. The investments are specific in Williamson’s precise sense: they have higher value inside the relationship with the protocol than outside it. The salvage value is lower than the in-relationship value. There is a quasi-rent. There is, therefore, an opening for opportunism by any party with discretion to revise the terms under which the investment was made.
The terms, in protocol economies, are the protocol rule set. The discretion to revise them is the governance layer. The actors who hold discretion are not always identifiable as a single coalition, but the previous essay in this series argued — and I will not re-argue here — that effective rule-change authority exists in any operational protocol, even where it is informal and distributed across implementation maintainers, sponsors, validators, exchanges, and infrastructure providers. What follows treats that authority as given and asks what its existence implies for the actors whose investment is specific to the rule set the authority can revise.
3. The hold-up mechanism, formally
Let xi denote the protocol-specific investment of participant i. Let Pt denote the protocol rule set at time t. Let rt denote a base-layer rule revision, so that Pt+1 = Pt + rt in the case where the rule set is mutable. Let Bi(Pt, xi) denote the gross benefit to i from operating with investment xi under rules Pt, and ci(xi) the cost of making the investment.
Under a credibly fixed rule set — that is, a regime in which Pt+1 = Pt with probability one — participant i‘s expected payoff is straightforward:
ΠiF = Bi(Pt, xi) − ci(xi)
The participant chooses xi to maximise this expression, subject to whatever resource constraints apply. The investment problem is a standard optimisation, and the comparative statics are familiar.
Under a mutable rule set, the situation changes. The participant must now form an expectation over rt. Let Pr(rt) denote the probability that a rule revision occurs in the relevant period, and let Hi(rt, xi) denote the loss imposed on i by the revision, conditional on its occurrence. The loss Hi is a function of the participant’s investment because the revision damages investment to the extent that the investment was specific to the pre-revision rule set. A rule change that affects the transaction format damages the application developer who built against the old format; it has no effect on the developer who built against the new format.
Let Oi(rt) denote the option value of beneficial rule change to participant i. Some rule changes are improvements: they fix bugs, accommodate new uses, reduce costs. To represent the situation honestly, the option value of beneficial revision must enter the calculation alongside the hold-up cost of opportunistic revision.
The expected payoff under mutable rules is then:
ΠiM = Bi(Pt, xi) − ci(xi) − Pr(rt) · 𝔼[Hi(rt, xi)] + 𝔼[Oi(rt)]
Comparing the two regimes:
ΠiF > ΠiM if and only if Pr(rt) · 𝔼[Hi(rt, xi)] > 𝔼[Oi(rt)]
This is the central inequality. Fixedness dominates when the expected hold-up cost of discretionary revision exceeds the expected option value of adaptive revision. Mutability dominates when the inequality reverses. Neither is universally superior. Whether one regime or the other is preferable depends on parameters — and on the credibility of the commitment to whichever regime has been chosen.
Several features of the inequality deserve attention. First, the hold-up term is multiplicative: it is the product of probability and expected loss. A small probability of a large loss is functionally equivalent to a moderate probability of a moderate loss, holding the option value fixed. Participants who treat low-probability rule revisions as economically negligible are making an error of valuation, particularly when the loss conditional on revision is large relative to the participant’s total investment in the system.
Second, the hold-up term scales with xi. The more the participant invests, the more the participant has at risk from rule revision. This is the precise sense in which mutable governance creates an investment-discouraging feedback: the larger the investment, the larger the exposure, and therefore the lower the marginal benefit of additional investment compared with what it would be under a fixed rule set. The result is underinvestment, not in absolute terms, but relative to the level that would obtain if commitment were credible.
Third, the option value term is also a function of the participant’s position. A rule change that benefits one class of participants may impose hold-up losses on another. The application developer who built against an old transaction format experiences a rule change as a loss, while the developer who built against the new format experiences it as a gain. This asymmetry is a standard feature of distributive rule changes and is part of what makes governance over rule revision politically contested. The aggregate option value across all participants may be positive even when individual hold-up costs are large, but aggregation does not compensate the participant who bears the loss.
Fourth, and most importantly, the hold-up term does not require that revision occur. The expected loss is a forward-looking expectation, formed before any actual rule change has happened. The participant who anticipates a non-zero probability of revision will discount investment regardless of whether the revision ever occurs. The damage to the equilibrium is done by the possibility, not by the realisation. This is exactly the structure of the time-inconsistency problem identified by Kydland and Prescott (1977): rational expectations of future discretionary action change present behaviour, even when the future action does not in fact materialise.
4. Why “you can always fork” is not an answer
The most common response to the hold-up framing, when it is raised, is that protocol economies have a feature that classical industrial organisation does not: participants can fork. If the rule-changing coalition imposes terms a participant rejects, the participant can run the old client, maintain the old chain, transact under the old rules. There is, the argument goes, no real lock-in because exit is always technically available.
This argument treats the problem as a coordination problem and misses its economic structure. To see why, consider what forking actually costs.
When a rule change is adopted by the rule-changing coalition and a participant chooses not to follow it, the participant is not preserved in the previous equilibrium. The participant is moved to a new and smaller equilibrium consisting of whoever else has chosen not to follow. The two networks now have separate hash power or stake, separate liquidity, separate user bases, separate exchange listings, separate application ecosystems, separate brand recognition, separate developer attention, and separate operational infrastructure. The participant who has remained on the old chain has not, in any economic sense, retained the original investment. The participant has accepted a smaller, less liquid, less secure, less integrated version of it.
The cost of this acceptance is precisely the asset specificity that made the original investment vulnerable to hold-up in the first place. The application that depended on the old rule set still functions, but it functions for a smaller user population, with less market depth, less price discovery, less network effect. The merchant who continues to accept transactions under the old rules accepts them from fewer counterparties, against thinner markets, at higher operational cost. The miner who continues to operate the old algorithm operates against lower fees and lower coin value, with no straightforward path to recovering the original investment.
This is what Hirschman (1970) was getting at when he distinguished exit from voice, and it is what subsequent work on lock-in and switching costs has documented across many domains. Exit is a constraint on opportunism only to the extent that exit is costless or near-costless. Where exit imposes a substantial cost on the exiting party — the cost of stranded asset specificity, of lost network effects, of operational re-engineering — the threat of exit does not effectively discipline the party with discretion. The discretionary party can revise terms within a range bounded above by the participant’s exit cost, and the participant rationally accepts the revision rather than incurring the higher cost of leaving.
The fork option is not absent from the analysis. It is part of the analysis. The relevant question is not whether forking is available — it usually is — but how much it costs the participant to use, relative to the cost of accepting the rule change. If the cost of accepting is lower than the cost of forking, the participant accepts, and the discretionary party has effectively appropriated value up to that limit. The participant’s “choice” to remain on the new chain is a choice in name only. It is the choice the participant makes because the alternative is more expensive.
This dynamic is reinforced by the fact that fork costs are not symmetric across participants. A small holder with little integration can fork at low cost. A large exchange with deep operational integration, regulatory commitments, and counterparty relationships cannot. The asymmetry means that rule changes effectively extract more value from participants with deeper specific investments, which is exactly the participants whose investment was hardest to attract in the first place. The result is a perverse selection effect: protocol economies under mutable governance discount investment most heavily from the participants whose investment matters most.
None of this is unique to protocol economies. The same dynamic operates wherever exit is costly and discretion is concentrated. What is distinctive about protocol economies is the rhetorical insistence that exit is costless, despite the economic structure that makes it costly. The insistence does not change the structure. It only obscures it.
5. The ex ante discount
The most consequential implication of the hold-up framing is that the equilibrium effect of mutable governance is felt before any rule revision has occurred. Participants who anticipate non-zero Pr(rt) reduce their xi. The reduction is a continuous function of the expected hold-up cost, not a step function triggered by an actual revision event.
This has several consequences that are observable in the patterns of investment across protocol economies, and that are unexplained by analyses that treat consensus security as the only relevant variable.
Migration toward credibly fixed base layers. When participants must choose between settlement layers with comparable consensus security but differing degrees of base-layer commitment, they will tend, all else equal, to specialise their high-specificity investments toward the layer with the more credible commitment. This is not because they have computed the option value of mutability and found it negative. It is because the hold-up exposure of high-specificity investment is large, and a credible commitment reduces that exposure to a level the participant can bear. Less specific investments — short-term capital, transient applications, experimental deployments — can tolerate higher mutability because their exit costs are lower.
Differential pricing of equivalent technical services. Two protocols that offer technically similar functionality but differ in governance credibility will not price equally. The protocol with the more credible commitment can attract investment at lower compensation for hold-up risk. The protocol with weaker commitment must compensate participants for accepting greater rule-change exposure. This compensation appears as higher fees demanded by service providers, higher discounts on integration partnerships, higher risk premia on capital deployments, or simply lower transaction volumes at any given fee level.
Concentration of long-duration commitments. Investments with longer expected duration are more exposed to Pr(rt) integrated over time. Participants making long-duration commitments — large infrastructure builds, regulated applications, multi-year integration programmes — therefore concentrate disproportionately on protocols where Pr(rt) is credibly low. Short-duration commitments distribute more freely, because the integral over a short horizon is small even when the per-period probability is non-trivial.
Patterns of regulatory engagement. Regulators evaluating protocol economies are themselves participants whose engagement involves specific investments — analytical capacity, reporting infrastructure, supervisory frameworks. The regulator’s tolerance for engagement with a protocol depends partly on whether the regulatory framework will remain coherent across rule changes. A protocol whose base layer is mutable presents the regulator with a moving target. The regulator’s response is to engage more cautiously, to apply more conservative classifications, or to refuse engagement until the rule set stabilises. None of these responses are arbitrary. They are predictable consequences of asset specificity in the regulator’s own investment in supervision.
Each of these patterns is consistent with the hold-up framework and difficult to explain from a consensus-only framework. They are not predictions in the sense of novel forecasts; they are interpretations of patterns that the field already observes but does not always connect to their economic source. The hold-up framework provides the connection.
6. The credibility problem
If hold-up risk discourages investment, the question of how to make commitment credible becomes the central institutional question. The economics literature on commitment is extensive, and the basic taxonomy is well established.
The first family of commitment mechanisms operates through cost imposition: making rule revision sufficiently expensive that the rule-changing coalition will not undertake it lightly. Coordination costs across many independent operators, supermajority adoption thresholds, public processes that surface dissent, time delays between proposal and activation — all of these increase the cost of revision and therefore reduce Pr(rt) in equilibrium. The mechanism is not magic; it is the standard insight that an action that is expensive to take is taken less often.
The second family operates through accountability: imposing personal or institutional consequences on those who exercise rule-change authority opportunistically. Where the parties exercising discretion are identifiable, they may be subject to legal duties, professional sanctions, fiduciary obligations, reputational costs, or contractual commitments that punish opportunistic exercise of authority. The accountability mechanism reduces opportunism not by making revision expensive in itself but by making the consequences of opportunistic revision personally costly to the actors who would benefit. The legal-economic literature has explored this at length. The essential point is that accountability requires identification, and identification is sometimes precisely what protocol economies have rhetorically resisted.
The third family operates through structural separation: dividing rule-change authority across multiple institutional centres so that no single coalition can effect revision unilaterally. Bicameralism, separation of powers, federated standards bodies, and multi-client implementation ecosystems all instantiate this principle. The mechanism reduces Pr(rt) by raising the coordination cost of forming a coalition large enough to act, particularly where the centres of authority have divergent interests.
The fourth family operates through fixedness: simply removing the ability to revise certain core elements of the rule set, either through technical commitments that are computationally infeasible to reverse, or through institutional commitments that the relevant actors have publicly bound themselves to maintain. This is the most aggressive solution, and it carries the largest cost in the form of foregone option value. It is appropriate where the option value of revision is small relative to the hold-up cost, and where the credibility of weaker mechanisms is in doubt.
None of these is a panacea. Each has costs. Cost imposition reduces Pr(rt) but also reduces the rate of beneficial adaptation. Accountability requires identification, which conflicts with privacy preferences and pseudonymity norms. Structural separation requires multiple competent institutional centres, which not all protocols have. Fixedness sacrifices flexibility entirely. The right combination depends on the asset specificity of the participants, the option value of revision in the relevant environment, and the costs of operating each mechanism.
What is clear is that the absence of any commitment mechanism — the assertion that “the protocol is permissionless and therefore decentralised, and there is nothing further to discuss” — is not a credible institutional position. It is a refusal to address the problem, which leaves the problem unaddressed. The hold-up exposure remains, the ex ante discount on investment continues, and the system underperforms relative to a comparable system that has done the institutional work.
7. Standard objections
“The probability of rule revision is small, so the hold-up term is small.” The probability of revision is an empirical question, not a theoretical assumption. The history of every operational protocol of significant duration includes contested rule changes, emergency interventions, controversial soft forks, and disputed activations. The relevant Pr(rt) is not zero. Whether it is small enough that the hold-up term is negligible relative to the option value depends on the specifics, but the assumption that it is negligible is not warranted by either theory or evidence. Furthermore, the equilibrium effect on investment depends on participant beliefs, not on the realised probability. If participants believe Pr(rt) is non-trivial, the discount applies even if the actual probability is lower.
“Rule changes are usually improvements, so the option value dominates.” The historical record on this is mixed, and the framing assumes the conclusion. The relevant question is not whether rule changes are usually improvements on average, but whether they are improvements for the specific participant making the investment. A rule change that is welfare-improving in aggregate can still be a hold-up event for participants whose specific investment is rendered obsolete. Aggregate welfare and individual hold-up exposure are different quantities, and the former does not compensate the latter except through redistribution mechanisms that protocol economies typically lack.
“Decentralised governance distributes rule-change authority widely enough that opportunism is impossible.” This claim is empirically testable and, in most cases, has not been tested. The previous essay argued that effective rule-change authority in operational protocols is typically more concentrated than the rhetoric suggests, with implementation maintainers, sponsors, and adoption-critical infrastructure providers exercising disproportionate influence. The claim that authority is sufficiently distributed to prevent opportunism is not self-evidently true and cannot be substituted for the analysis it would replace.
“Protocol participants are sophisticated and accept these risks.” Sophistication does not eliminate the equilibrium effect. The discount on investment is exactly the rational response of sophisticated participants to perceived hold-up risk. The point of identifying the discount is not to claim that participants are misled, but to explain why investment is lower than it would be under credible commitment, and to identify the institutional mechanisms that could reduce the discount and increase the investment. The argument is about the equilibrium, not about participant naivety.
“The hold-up framework is borrowed from a different context and does not apply here.” The framework is general. It applies wherever specific investment meets discretionary authority. The original applications were vertical relationships between firms, but the same logic has been extended to long-term contracts, employment relationships, public utilities, federal-state relations, international trade, and constitutional design. The application to protocol economies is not a forced analogy; it is a straightforward instance of the general structure.
“Investors who fear hold-up should simply not invest.” They do not. They invest less, or invest in different protocols, or invest with less specificity, or invest with shorter duration. The aggregate effect is the equilibrium discount described in section 5. The argument is not that no investment occurs under mutable governance — manifestly investment does occur — but that the level and composition of investment is shaped by the hold-up exposure in ways that are observable and consistent with the framework.
8. The empirical signature
If the hold-up framework is correct, certain patterns should be observable in the cross-section and time-series of protocol economies. The framework does not predict every detail of every system, but it generates falsifiable expectations.
The cross-section should show that participants with higher asset specificity concentrate disproportionately in protocols with more credible commitment to base-layer fixedness, after controlling for technical and economic characteristics. This is not a tautology, because asset specificity is observable independently of governance choice — it can be measured by the cost of redeployment, the duration of commitment, the depth of integration, the legal and regulatory exposure, and the substitutability of alternatives.
The time-series should show that contested rule-change events are followed by reduced specific investment in the affected protocol, with the magnitude of the reduction increasing in the perceived likelihood of further such events. The reduction should be largest among participants with the highest specificity and longest expected duration of commitment. It should be smaller, and may even reverse, among participants whose specificity is low or whose investment was structured to benefit from the rule change.
The pricing structure should show that protocols with weaker commitment compensate participants through higher fees, larger token allocations, more favourable integration terms, or greater regulatory accommodation. The compensation is a direct measure of the hold-up risk premium that the protocol must offer to attract investment that would otherwise migrate to a more credibly fixed alternative.
The composition of investment should show that long-duration, regulated, and integration-heavy investments accumulate disproportionately in credibly fixed environments, while short-duration, experimental, and lightly-integrated investments distribute more freely. This selection effect should be observable in the application mix, the institutional participant mix, and the regulatory engagement profile of different protocols.
None of these predictions is novel as an empirical observation. Each is consistent with patterns that practitioners and analysts already note. The contribution of the framework is to connect the patterns to a coherent economic mechanism, and to make explicit the equilibrium logic that generates them. Once the mechanism is identified, the patterns are no longer surprising. They are the predictable consequences of specific investment under mutable rules.
9. What the analysis does not say
The argument so far has been that mutable governance under specific investment generates hold-up exposure, and that hold-up exposure discounts investment ex ante. It is important to be clear about what this argument does not claim.
It does not claim that fixedness is always preferable. The trade-off between hold-up cost and option value is a real trade-off, and the optimal regime depends on parameters. Where option value is high — for example, in a young protocol where bugs and design errors are likely to be discovered and corrected — moderate mutability may dominate. Where asset specificity is low — for example, in experimental applications with short expected lifespan — hold-up exposure is small and mutability is cheap. The argument identifies a cost of mutability that has been under-represented in the literature, not a universal verdict against it.
It does not claim that all rule changes are opportunistic. Many rule changes are welfare-improving in aggregate, and even those that impose hold-up costs on some participants may be defensible on broader grounds. The point is not to delegitimise rule change, but to recognise that the capacity for rule change has costs as well as benefits, and that those costs are concentrated on participants whose investments are most specific.
It does not claim that protocol economies are uniquely vulnerable to hold-up. The same dynamics operate in vertical supplier relationships, long-term contracts, employment, public utilities, and international agreements. The protocol case is interesting because the rhetoric of the field has often denied that the dynamic applies, while the structural conditions for it are clearly present. The contribution is to bring standard transaction-cost economics to bear on a domain that has resisted it.
It does not claim that the institutional fix is obvious. Each commitment mechanism has its own costs, and the right combination is a design problem, not a derivation. What the argument does claim is that the design problem cannot be evaded by denying the underlying mechanism, and that protocol economies that have not done the institutional work are operating with an implicit hold-up exposure that participants are pricing into their decisions even when they do not articulate it in those terms.
10. Closing
The hold-up problem is not new, and it is not exotic. It is the central problem of relationship-specific investment under incomplete contracts, and it has been studied for half a century. The contribution of this essay is to argue that protocol economies are not exempt from it, and that the field’s standard responses — that exit is always available, that decentralisation prevents opportunism, that rule changes are usually improvements — are inadequate as analyses of the equilibrium they produce.
When participants make protocol-specific investments, and a coalition retains effective discretion over base-layer rule revision, the participants face the same structural problem that a supplier faces who has built a plant adjacent to a single buyer. The rule-changing coalition is in a position to revise terms after the participants have committed. The participants anticipate this and discount their investment accordingly. The discount obtains regardless of whether the coalition ever exercises its discretion opportunistically. The damage to the equilibrium is done by the capacity, not by the act.
The fork option is part of the analysis, not a solution to it. Forking is costly precisely because the original investment was specific, and the cost of forking is the asset specificity that made the investment vulnerable to hold-up in the first place. The participant’s “choice” between accepting a rule change and forking is bounded by the relative costs of each. Discretionary parties can revise terms within those bounds and the participant rationally accepts.
The institutional fix is not a single mechanism. It is a combination of cost imposition, accountability, structural separation, and selective fixedness, calibrated to the asset specificity of the participants and the option value of revision in the relevant environment. The combination is a design problem that the field has only partly addressed, and that it cannot evade by asserting that the problem does not exist.
The basic claim of this essay can be stated as a single inequality. Participant investment is discounted by the expected hold-up cost net of the expected option value: Pr(rt) · 𝔼[H(rt, x)] versus 𝔼[O(rt)]. When the first term exceeds the second, mutable governance is welfare-reducing relative to fixedness. Whether one or the other is preferable in any given case is an empirical and design question, not an ideological one. But the question must be addressed, and addressing it requires acknowledging that the hold-up problem applies. The field’s tendency to deny that it applies, or to wave at exit as if it solved the problem, has produced an equilibrium in which investment is lower than it could be, security is weaker than it could be, and credibility is more contested than it needs to be. The economic analysis is available. It has been available for decades. The field’s job is to use it.
References mentioned in passing: R. H. Coase, “The Nature of the Firm,” Economica (1937); A. O. Hirschman, Exit, Voice, and Loyalty (1970); B. Klein, R. G. Crawford, and A. A. Alchian, “Vertical Integration, Appropriable Rents, and the Competitive Contracting Process,” Journal of Law and Economics (1978); F. E. Kydland and E. C. Prescott, “Rules Rather than Discretion: The Inconsistency of Optimal Plans,” Journal of Political Economy (1977); O. E. Williamson, Markets and Hierarchies (1975); O. E. Williamson, The Economic Institutions of Capitalism (1985).