2.1 The Hold-up Problem and Its Impacts

In order to illustrate the costs of the hold-up problem and, later, its possible theoretical solutions, we use a recent example that should be familiar to readers, at least those familiar with smart phones. The buyer in our example is Apple, an original equipment manufacturer (“OEM”) of the iPhone smartphone, and the seller is Corning, an important component manufacturer that produces “Gorilla Glass.” Our contention is not that Apple and Corning’s contract actually had a hold-up problem but that it might have and that we can therefore use it to illustrate some of the impacts of hold-up and solutions to hold-up. The numbers we employ in this example were chosen not because they reflect reality but because they help demonstrate how hold-up can deter efficient investment.Footnote ⁹

Suppose Apple and Corning want to enter into a relationship wherein Corning provides 1 unit of a good (“Gorilla Glass” or “glass”) at price, $p$ , per unit.Footnote ¹⁰ The value generated by the trade depends on Apple’s valuation for the glass, $v$ , and Corning’s cost of producing the glass, $c$ .

The timing of the relationship is as follows. First, the parties contract. Second, they make their investments noncooperatively and simultaneously. Third, they both learn $v$ and $c$ . Finally, the contract is executed.

At the time of contracting, the precise values of $v$ and $c$ are not known since they depend on investments made by Apple and Corning. We propose some numbers that have no special significance but can help the reader see how the hold-up problem operates. Suppose that $v$ can be either $40 or $32 and correspondingly c can be $16 or $10.Footnote ¹¹ Apple’s investment affects the probability that v is high (or low) and Corning’s investment affects the probability that c is high (or low). These investments are privately costly to Apple and Corning, costing each $5. One can think of Apple’s investment as marketing of the iPhone, thus increasing sales volume, or as Apple making other features of the iPhone more complementary with strong glass – say by having a smaller form-factor, reduced bezel or adding a face recognition system that eliminates the need for a thumbprint ID-enabled home button and allows the screen to take up the entire front of the smartphone. One can think of Corning’s investment as improving the strength of the glass, especially larger pieces of glass, or lowering the cost of production.

The parties do not write fully state-contingent contracts that specify $p$ and the probability of trade $q$ for each combination of $v$ and $c$ and they only invest after they contract because, in reality, there are unanticipated innovations, opportunities or challenges that arise after contracting but before delivery. Real-world examples include Samsung’s surprise introduction of a glass screen that folds around the edges of their Galaxy phone (see, e.g., Samsung Galaxy 6). Corning felt pressure to match that. Another real-world example is Apple’s surprise filing of a facial recognition system that increases the value of a larger glass plate front from Corning (Reference CrookCrook 2017). Apple would not want to announce the patent earlier for fear that competitors or even suppliers might file their related patents earlier.

With the numbers we have chosen in our example, the socially efficient thing to do is for both parties to make the investments, since both have a marginal value above their marginal cost ($8 compared to $5 for Apple’s investment and $6 compared to $5 for Corning’s). If the investments are made, then the total surplus is $40 − $10 − $5 − $5 = $20.

The essence of the hold-up problem, however, is that Apple and Corning will underinvest in the absence of the ability to contract on the investments (or values/costs) if they cannot prevent renegotiation. To see this, consider Corning, and suppose that they make their privately costly investment. Once Apple’s value for the glass and Corning’s cost of producing it is realized, the parties will renegotiate the price, since contracting was incomplete at the start of the relationship. Assuming, for simplicity, that the parties evenly split the incremental surplus generated (as would arise as a result of the Nash bargaining solution), then the price will be adjusted so that Corning only gets half of the $6 that it increased total surplus by, that is $3. Anticipating this at the investment stage, Corning compares the $3 benefit with the $5 cost and will not invest. Analogously, Apple will compare a $8/2 = $4 benefit with a $5 cost and likewise will not invest. This means that, in the presence of hold-up, neither party will invest, Apple’s valuation will be low and Corning’s cost high, and total surplus will be $32 − $16 = $16.

As our numerical example illustrates, hold-up reduces economic surplus. It is thus natural, therefore, that economists have spent considerable effortFootnote ¹² exploring how its impact can be mitigated, in circumstances where it cannot be avoided due to limitations of the contracting environment.

2.2 Using Renegotiation-Design Mechanisms to Address Hold-up

The difficulties with hold-up derive from the subsequent renegotiation of prices. Economists have designed a number of contracting solutions – called “mechanisms” – to tackle the hold-up problem. The logic behind these mechanisms stem from the observation that if this renegotiation could be structured differently, then perhaps the social optimum could be obtained despite hold-up. Reference ChungChung (1991) and Reference AghionAghion, Dewatripont and Rey (1994) are leading examples of this approach. We illustrate how renegotiation design mechanisms work using the mechanism in Reference AghionAghion, Dewatripont and Rey (1994), henceforth “ADR.” In Appendix A, we illustrate another important renegotiation mechanism based on options contracts, attributable to Reference Noldeke and SchmidtNoldeke and Schmidt (1995).

The ADR mechanism has two components. The first component is a default trade that can always be requested by one party, even if it is held-up. This default option is structured to give that party (say the seller) a full return to its investment for sure. This needs to be enforceable with a specific performance remedy or a liquidated damages remedy that strongly incentivizes specific performance if the party with the default option requests the default trade. The second component gives all the bargaining power in renegotiation to the other party (say the buyer), by allowing that party to make a take-it-or-leave-it offer. The timing of the renegotiation game is as follows: (1) the parties make investment decisions; (2) the buyer makes a take-it-or-leave-it offer; and (3) the seller can accept the offer (in which case trade takes place on those terms) or can trigger the default trade. This two-stage mechanism achieves the socially optimal level of investment by both parties. To see why, note that the seller, despite having no bargaining power at the take-it-or-leave-it stage, is the residual claimant on their investment due to their access to the default option in the second stage of the game, and thus has appropriate incentives to invest optimally. Inducting back to the take-it-or-leave-it-stage, the buyer, because it has all the bargaining power, has the requisite incentives to invest optimally (Reference AghionAghion, Dewatripont and Rey 1994: 263–266).

To see how this renegotiation game plays out in the Apple–Corning example,Footnote ¹³ there is a single unit of glass to be traded, so it is convenient to think of setting the default probability that a widget will be traded. The first component of the ADR mechanism is the default trade that Corning can trigger. We set the default probability of trade at 5/6 and the default price at $23 1/3. We will not go through a derivation of the numbers we have chosen here,Footnote ¹⁴ but they are constructed to do two things: split the ex ante surplus evenly between Apple and Corning, and ensure that Corning has the appropriate incentives to make the cost-reducing investment. The second component of the ADR mechanism is to give Apple the right to make a take-it-or-leave-it offer.

What is the best offer for Apple to make when they get the opportunity? Since Apple is making a take-it-or-leave-it offer, they have all the bargaining power and will thus want to trade the efficient amount of glass, which is 1 unit. Apple, possessing all the bargaining power, will extract Corning’s entire surplus. This leaves Corning indifferent between the default option and accepting Apple’s offer. Corning will thus anticipate getting the same payoff as under the default option, whatever happens.

Now work backward, and consider Corning’s decision whether to invest at the earlier stage. If Corning invests, they get a payoff equal to $23 1/3 − (5/6) × $10 − $5 = $10 (the price minus probability of trade multiplied by the cost of production, minus the investment cost). If Corning does not invest, they get a payoff equal to $23 1/3 − (5/6) × $16 = $10. So, Corning is willing to invest.Footnote ¹⁵

Now consider Apple, who is the residual claimant on their investment. They obtain $8 if they make the investment, but bear a cost of just $5. So, Apple also makes the efficient investment. Thus, total surplus is $40 − $10 − $5 − $5 = $20. Remarkably, this is the first best – the same as if Apple and Corning could contract on all relevant contingencies.

It is, at first glance, rather surprising that Corning finds it optimal to invest despite having none of the bargaining power in the renegotiation. The key is their ability to reject Apple’s offer and trigger trade under the terms of the default option. The default option becomes more appealing when the glass is low cost, which happens precisely when Corning invests. In other words, the presence of the default option makes Corning’s payoff sensitive to their investment.

Of course, there are also legal doctrines that alter the ex post bargaining power of each party. Notable examples of this are highlighted in Reference GuerrieroGuerriero (2020) and Reference Guerriero and PignataroGuerriero and Pignataro (2021). In our preceding example, this might involve increasing Corning’s bargaining power by strengthening remedies for “unfair contract terms,” “abuse of right” and “compulsory licensing.” And Apple’s bargaining power is affected by “non-disclosure,” “work for hire” and “non-compete” agreements. When the investments by the two parties are sufficiently complementary, this can play an important role.Footnote ¹⁶

2.3 Mechanisms to Address Nonverifiability More Generally

So far, we have examined the problem of hold-up when the parties make relationship-specific investments, like Alaska Packers’ chartering of a ship to take plaintiff fishermen to Alaska, or in our Apple and Corning example. However, there may be important ex post inefficiencies that arise from the inability to write complete, state-contingent contracts even in the absence of relationship-specific investments.

The economics literature has only recently made progress in providing formal models of such ex post inefficiencies, typically through the use of ingredients from social psychology such as “reference points” (see Reference Hart and MooreHart and Moore 2008). We will not delve into those formal models here. However, we do highlight solutions to these ex post inefficiencies that do not restructure renegotiation so much as to get the parties to truthfully report nonverifiable information to the court, making that information verifiable.

Beginning with Reference MaskinMaskin (1977), a large literature has explored how carefully crafted mechanisms that require players to make announcements, and face payoffs dependent on those announcements, may be able to cause information observable to players but not to outsiders to become observable and thus verifiable by outsiders. This is particularly useful in the hold-up setting, because once the parties know that the true state will be revealed, then they can ex ante contract on $v$ or $c$ , even if they cannot contract on relationship-specific investments per se.Footnote ¹⁷

An illustration of Maskin’s theorem in the Apple–Corning setting is as follows. To make Apple’s value $v$ – which both Apple and Corning know once it is realized – verifiable to a court, each party is asked to simultaneously announce 40 or 32. The mechanism specifies that if both parties agree in their announcement, then that is the stipulated value of $v$ . If the parties disagree, then each pays a very large fine to a third party.

It is straightforward to see that both Apple and Corning announcing truthfully is a Nash equilibrium. Suppose $v$ is actually 40. Conditional on Apple announcing “40,” the Corning’s best response is to announce “40” as they avoid the large fine. Unfortunately, both parties not announcing truthfully is also a Nash equilibrium. Suppose again that $v$ is actually 40, conditional on Apple announcing “32,” the best response of Corning is to announce “32” to avoid the fine.

Troubled by this multiplicity of equilibrium, Reference Moore and RepulloMoore and Repullo (1988: 1208, 1212) show that by using a multi-stage mechanism, set forth in Appendix B, it is possible to implement any social objective as the unique (subgame–perfect) equilibrium of the game induced by that mechanism.

2.4 Information Problems That Hamper Mechanisms

Rarely, if ever, are the renegotiation design or the revelation mechanisms described seen in practice (Reference ShavellShavell 2007: 347–348). This begs the question: Why not? One reason may be that these mechanisms seem too complex to implement as they require two or three stages of structured bargaining. Reference Holden and MalaniHolden and Malani (2014: 155) show that these hurdles cannot be too large because, while uncommon, procedures like the ADR mechanism can be found in, for example, variable quantity contracts. Moreover, even the multi-stage procedures in Moore–Repullo mechanisms are less complicated than some arbitration procedures to which contractual parties agree, let alone actual trials and the civil procedure rules that govern contracts in the absence of arbitration provisions.

3.1 Public Solutions

Historically, courts took a dim view of modifications. They applied the preexisting duty rule, which stated that doing what a party had already promised to do is insufficient consideration for a new promise.Footnote ²⁰ This is formally the rule that the Ninth Circuit applied to invalidate the modification in Alaska Packers’ Ass’n v. Domenico (1902: 102), though as we will see, it has nonetheless come to be interpreted as standing for a different rule.Footnote ²¹

The problem with the preexisting duty rule is that it made it difficult to implement contract modifications that were efficient responses to changed circumstances (Reference Graham and PierceGraham and Pierce 1989: 14). It was also easily gamed with small amounts of consideration, which largely preserved the rents earned by exploitative contracting parties (Reference John Edward MurrayMurray 1974: 179; Reference Calamari and PerilloCalamari and Perillo 1983: § 4–9, at 207; Reference Graham and PierceGraham and Pierce 1989: 14). As a result, the rule has largely been abandoned (Reference Graham and PierceGraham and Pierce 1989: 14–15).

In its stead, courts have largely followed three different rules. The first and most widespread in common law is the duress doctrine, under which courts may rule a modification unenforceable if they determine it was secured under duress. Duress is typically defined as induced by an “improper threat,” including a threat of breach (see, e.g., Hartsville Oil Mill v. United States 1926; Sistrom v. Anderson 1942; Steinberg Press, Inc. v. Charles Henry Publications, Inc. 1947; Reference JollsJolls 1997: 207), especially if the threat leaves the victim with “no reasonable alternatives” (Restatement (Second) of Contracts 1979: §§ 175(1), 176(1)(d)). The definition of duress has expanded over time and has been criticized for being ambiguous (Reference Graham and PierceGraham and Pierce 1989: 10).

The second rule is good faith. The comment to Uniform Commercial Code § 2–209, which governs modification to contracts for the sale of goods only, explains that “modifications … must meet the test of good faith.” However, this requirement largely overlaps with the duress test; the comment explains that “the extortion of a ‘modification’ without legitimate commercial reason is ineffective as a violation of the duty of good faith.”

The third rule is the “changed circumstances” rule. Under this rule, courts look for unanticipated changes in commercial conditions that would make the original contract unappealing for the parties (Angel v. Reference John Edward MurrayMurray 1974). If found, the court would enforce a modification. In contrast with the duress doctrine, in which courts look for evidence (duress) that a modification is bad, with the changed circumstances rule, courts look for evidence that a modification is good.

The Restatement (Second) of Contracts effectively endorses all three rules at once. While the background rule is that contracts (or contract modifications) lacking consideration are unenforceable (Restatement (Second) of Contracts 1979: § 73), the Restatement carves an exception: modifications unsupported by consideration are still enforceable if they are “fair and equitable in view of circumstances not anticipated by the parties when the contract was made” (Restatement (Second) of Contracts 1979: § 89). However, a modification is not enforceable if it is induced by an “improper threat,” which includes a threat to breach in violation of the duty of good faith (Restatement (Second) of Contracts 1979: §§ 175(1), 176(1)(d)).

Scholars have proposed alternative rules, although these have not gained traction in case law. For example, Reference ShavellShavell (2007: 326) has advocated a rule that allows modifications, so long as the change in price is reasonable. Reference GillBar Gill and Ben Shahar (2004: 392) have proposed a rule that would allow modifications even if induced by threats so long as those threats are credible. Reference JollsJolls (1997: 204) has gone so far as to suggest that courts allow the parties to contract to disallow any future modifications to a contract.Footnote ²²

The different rules enforced by courts, as well as alternative rules proposed by scholars, all suffer from a basic problem: non-verifiability. For example, the duress doctrine depends on demonstrating improper threat and the lack of alternatives. Likewise, Bar Gill and Ben Shahar’s proposal requires distinguishing between credible and non-credible threats. The changed circumstances rule requires proof of unanticipated alteration in economic conditions surrounding the contract. There may be disagreements about whether there was an improper or credible threat or an increase in costs, as exemplified by Reference ThreedyThreedy’s (2000) exegesis of the facts behind Alaska Packers’ Ass’n v. Domenico (1901).

What makes matters more complicated is that the existence of, say, changed circumstances is not sufficient evidence that a modification should be enforced from an economic perspective. What the court should do, as Shavell explains, is compare the increase in costs due to changed circumstances with the increase in price. If the increase in price is greater than the increase in costs, then there may be hold-up even in the presence of changed circumstances. Yet, these careful economic calculations are not the comparative advantage of courts.Footnote ²³

Even if ex post courts can overcome these disagreements about facts and reach a decision, ex ante there will be uncertainty among the parties about what rule the courts will apply or which facts the court will find. This uncertainty implies that there is a positive probability that remedies will be inadequate.Footnote ²⁴ That in turn creates opportunities for hold-up.

Some rules, such as the preexisting duty rule and Joll’s rule to enforce no-modifications clauses, are easy for courts to enforce and entail little uncertainty. Yet, these are also extreme solutions. Each rules out any efficient modifications in response to changed circumstances.

3.2 Private Solutions

Given the inherently limited ability of courts to tackle the problem of hold-up, at least without going overboard and stopping even efficient modifications, we turn to private, contractual solutions that parties may pursue – with less reliance on nuanced judgment by courts – to address the risk or consequences of hold-up.

Ideally, the parties would write a contract that precisely specified the investments each party would make in each state of the world and at the price and quantity that they would trade given the result of those investments. Economists call this a complete state-contingent contract that specifies the parties’ payoffs in all states of the world. In this context, courts would have no reason to allow renegotiation – correctly or mistakenly – because the original contract would account for all possible changed circumstances. That, in turn, would deter hold-up.

Yet, such contracts are nearly impossible to write. Parties may not know all possible states or it may not be cost effective to specify them all ex ante. Moreover, courts often cannot verify what state the parties are in and thus enforce the appropriate contractual provision. Indeed, this is the reason that the starting assumption of the economic literature on the hold-up problem is that parties cannot and do not write, or courts cannot enforce, complete contracts (Reference Hart and MooreHart and Moore 1990: 1126; Reference Aghion and HoldenAghion and Holden 2011: 182).

One alternative solution is for the two contracting parties to merge and conduct the transaction internally within a single firm. This option is why Coase and later Grossman and Hart (and Hart and Moore) suggested that hold-up could provide incentives for the parties to move their transaction from the market, mediated by contract between firms, to organizing the transaction within a single, integrated firm.

While Apple and Corning integrating would address the glass plate supply problem, both parties also deal with other suppliers and purchasers. Corning cannot both integrate with Apple and with, say, Motorola, another smartphone OEM. As a result, integrating with Apple would mean having a less efficient relationship with Motorola, due to hold-up. Nor would Apple want to own every single one of its thousands of suppliers, even though their deals may also entail hold-up risks. Managing all of those relationships internally within a firm simply replaces hold-up problems with internal agency problems.

A second alternative for the parties is to rely on repeat play and reputation. If Apple and Corning transact each year for each new iteration of Apple’s smartphone, they each have a strong incentive not to hold the other up, lest they jeopardize future deals between them (Reference Klein, Crawford and AlchianKlein et al. 1978: 302, explaining the value of long-term contracts). That said, if the power to hold-up is asymmetric, it would affect the aggregate division of gains from trade across all the parties’ contracts in favor of the party with more hold-up power. This in turn could reduce the incentive of the weaker party to participate even in repeat play transactions. More importantly, many parties do not transact repeatedly – or expect that their repeated transactions will one day end.Footnote ²⁵ In that context, parties must rely on market reputation. The challenge is that, if courts cannot verify hold-up, other companies might not be able to either.

A third alternative is for the parties to employ one of the renegotiation design or revelation mechanisms discussed in Section 2. For instance, they could specify default trades that favor one party and give the other party the right to make a take-it-or-leave-it offer, as the ADR mechanisms suggest (Reference AghionAghion, Dewatripont and Rey 1994: 258), or they could specify alternating price announcements and challenges alongside payments to third parties, as the Moore–Repullo mechanism proposes (Reference Moore and RepulloMoore and Repullo 1988: 1196).

The weakness of these mechanisms is that they each require a strong form of commitment, which is a significant barrier to their usefulness. If the weaker party can refuse the take-it-or-leave-it offer and the stronger party would still trade (which it is mutually rational to do), then the ultimatum is not credible and it will not give the stronger party adequate return on its investment. Likewise, if the parties agree to split the penalty payment to the third party rather than hand it over to that party (again which is mutually rational to do), then the parties do not have an incentive to truthfully announce their valuations prior to the penalty round. Yet, it is unlikely that the parties can credibly commit to actions required for renegotiation design or revelation mechanisms if they cannot make such commitments to the original contract to deter hold-up in the first place.

One way to obtain this commitment, ostensibly without smart contracts and blockchain, is to include penalty clauses – liquidated damages larger than economic damages – if the parties do not comply with the mechanisms (Reference Gerrit, Wuyts and Gerritde Geest and Wuyts 2000). For example, the stronger party may have to pay a penalty if it makes a second offer after its take-it-or-leave-it offer is rejected by the weaker party but before the weaker party requests the default trade in the ADR mechanism.Footnote ²⁶ Indeed, penalty clauses could go further and possibly disincentivize hold-up in the first place by penalizing deviations from the original contract terms, if the parties are comfortable with foregoing any flexibility to depart from the original contract.

The problem with this approach is that courts frown upon penalty clauses, although they may be more permissive in cases where both parties are sophisticated businesses rather than cases involving small businesses or individual consumers (Reference FarnsworthFarnsworth 2003: 845). If courts won’t enforce the penalty clause, they will not incentivize parties to commit to their roles in the mechanisms or to their original contract terms.

Even if courts would enforce the penalty clause, it can skew the incentives of the two parties. Such clauses act like reliance damages, which are known to risk overinvestment by the protected parties (Reference RogersonRogerson 1984: 41). The solution to overinvestment is to pay the penalties to a third party rather than the party making an investment. Yet, both contractual parties have a mutual incentive to negotiate around that third-party payment, as they did in the revelation mechanism (Reference HolmstromHolmstrom 1982: 327, showing that relaxing the budget breaker yields an efficient Nash equilibrium of the moral hazard in a team game).

The only way to obtain commitment without tilting investment incentives then is to have an automatic payment to a third party that cannot be undone by the contractual parties. For example, the parties could set up something akin to a poison pill, where the contract or mechanism creates an IOU from the penalized party to a third party (Reference Dawson, Pence and StoneDawson et al. 1987). Of course, if there is a gap between the timing of the investment and the payment for that investment, the holding-up party could still petition a court to enjoin the poison pill–type penalty before it is triggered by failure to make payment for the investment. The held-up party is unlikely to invest much to stop the holding-up party because it does not benefit from the payment to a third party.

To avoid both circumvention of the penalty by the parties (directly) and by courts (after being petitioned by the parties) is to have the penalty be something that even the court cannot enjoin. An outlandish but effective solution is to create a machine, have the party that needs to show commitment place a large amount of cash in the machine, then have the machine rigged to burn the cash if that party fails to meet its commitment. The machine must have a dead-hand switch so that it is triggered if anyone – even the court via injunction – tries to shut it down.

This machine must be all-knowing about the parties’ accounts, otherwise the parties could get around even this machine. Specifically, they could renegotiate but transact in two parts: one that appears to the machine to conform to the original contract, and a second that undoes the original contract and consummates the renegotiated contract. If the machine only observed the first transaction, it would release the cash it was holding hostage even though the second transaction consummated the renegotiated contract. Again, investment incentives would be undermined.

But this machine we have described is hard to build and actually looks very much like a smart contract on blockchain. Moreover, the latter are simple and cheaper to implement. The smart contract is just a computer script, not an awkward physical machine; it can work with digital money, which is easier to obtain than actual cash; and can be made all knowing with application program interfaces (APIs) to all of the parties’ accounts (see Section 5). Moreover, blockchain allows irreversible transfers to anonymous accounts.

4.1 The Value of Blockchain

Blockchain, a computer science innovation introduced by Satoshi Nakamoto in 2008, is described as a distributed or decentralized, open and secure ledger, meaning that it verifies transactions are intended, feasible and executed through a decentralized system rather than through a central authority (like a government or bank, which might be costly or untrustworthy), it records transactions in a public way (so as to build reputations and counterparty trust) and ensures transactions are not reversible (to protect parties from certain types of theft).Footnote ²⁷ Although Nakamoto initially intended the term transaction to mean a monetary transfer, a transaction recorded in blockchain can be any set of promises – that is, a contract, or indeed any statement (Reference ButerinButerin 2014: 14).

4.2 The Value of Smart Contracts

So-called smart contracts, as first conceived by Reference SzaboNick Szabo (2021), are a quite general concept: a smart contract is a simple series of actions written in computer script.Footnote ³² When the actions constitute fulfillment of mutual, conditional promises, they are a contract in the traditional sense.Footnote ³³ Smart contracts can fully or partially specify a contract, meaning that they can include all promises made pursuant to a contract or they can contain part of the promises, in which case the code plus a paper agreement constitute the whole contract (Reference StarkStark 2016).

Smart contracts can exist independent of a blockchain network, but they can also be announced, witnessed and automatically executed on such a network. Indeed, the reason Vitalik Reference ButerinButerin (2014) created the Ethereum network was that the Bitcoin network did not sufficiently support smart contracts. Ethereum is a blockchain network but with a scripting language that allows smart contracts. Individuals write their smart contracts in that language and the Ethereum network validates and executes it.

5.1 An Example with the ADR Renegotiation Design Mechanism

Suppose that Apple and Corning wish to write a contract that includes an ADR mechanism to structure renegotiation after a hold-up so that the hold-up does not deter efficient relationship-specific investment. Recall that the ADR mechanism requires one party be given a default option and the other the credibility to make a take-it-or-leave-it offer. Each requires commitment– that is, the latter party must have strong incentives not to a second offer and the former a strong incentive not to renegotiate the default offer. How could the parties implement this with these new technologies?

Learning from Section 2.5.1, we would first ask how much information the parties have. If it is enough to devise a renegotiation design or revelation mechanism, then the parties would want to construct a penalty provision that deters deviation from the mechanism and that cannot be undone by courts or the parties through renegotiation. If the amount of information is not adequate to devise such mechanisms, the only thing that the parties can do is construct a penalty provision that discourages any deviation from the original contract. Of course, such a penalty provision would also bar renegotiation due honestly to changed circumstances and that does not affect ex ante incentives to invest. So the parties should not construct a penalty that locks in the original contract unless the expected cost of hold-up is greater than the expected cost of inflexibility.

To illustrate how a penalty clause could be constructed, suppose the parties have enough information to write an ADR renegotiation design mechanism. Let’s see how the penalty provision would work. First, we examine the default option. In our example, it is Corning that is to be given the option to trade a unit with probability 5/6 at price 23 1/3. We would script a clause in the smart contract code that says Corning can ask for the default trade and if that trade is not consummated, then Apple pays a penalty. Likewise, the smart contract code would make Apple pay a penalty if it communicated a second offer to Corning after Corning refused its first ostensibly take-it-or-leave-it offer.

In order not to skew the incentives of the parties when playing the ADR game, the penalty must flow to a third party. For instance, if the smart contract does not observe the default trade by a certain time or if it observes a second offer from Apple, then the code would irreversibly transfer money from Apple to anonymous third parties.

Blockchain makes it possible for penalties to flow to third parties. All the smart contract has to do is specify that some large amount would be transferred from Apple’s account (its public address or key) on the network to a randomly generated list of public addresses or keys and that the smart contract would announce to the blockchain network the private keys associated with those public keys. The latter step would allow any node on the network to access and transfer to their account the money at those public addresses. It would be as if the code announced that there was a pile of cash on the corner of Broadway and Fifth Avenue in Manhattan: people would rush to take the money (Reference AntonopolousAntonopolous 2017: 17, explaining how transferring money to a public key for which the private key is publicly known would lead to loss of funds). The entities that collect the money distributed by the smart contract would be able to remain anonymous – as blockchain allows anonymous transactions (Reference AruAru 2017). If that entity were located in a country that did not have a KYC requirement, they could also convert that money into the government currency of their choice. There are plenty of countries happy to do that (US Department of State 2012; PWC 2013; IRS 2021).

The penalized party, and perhaps even the nonpenalized party, would have an incentive to go to court to enjoin the penalty. With the default option, Apple would have an incentive to not be penalized for defying Corning’s request. With Apple’s take-it-or-leave-it offer, Apple has an ex post incentive to avoid the penalty for making a second offer and Corning has an incentive to receive a second Apple offer, meaning both would want to petition a court.

The future commitment feature of smart contracts and the irreversibility of blockchain transactions can render court injunctions ineffective. Once the parties sign the agreement, future exchanges are already booked and cannot be undone – by either the parties or a court – without having either a majority of the computing power or the tokens on the blockchain (see Section 4.1.3). A court does not have that. Thus, a court can no more require a transaction on blockchain be reversed than it can require that the stock price of a company that committed fraud increase to compensate shareholders.

Court-ordered damages would not help undermine the smart contract commitment. For one thing, punishing Corning because Apple has to pay a penalty does not change Apple’s incentives unless Corning has to pay Apple to compensate for the penalty Apple must pay. But if Corning has to pay Apple, then the penalty script in the smart contract can make the penalty contingent on subsequent damages ordered by a court. For example, if the optimal penalty on Apple for, say, making more than one offer is 100 and the damages awarded by the court for making Apple pay a penalty is D, the script could say Apple must pay 100 + D.

Apple may be tempted to hold Corning up but avoid penalties by renegotiating in two steps. First, they would comply with the ADR mechanism in the original contract. Second, they would simultaneously write a separate contract with Corning that functionally puts Apple in the same position as if it had successfully held up Corning in the original contracts. This second contract would say that Corning will sell to Apple one unit of Gorilla glass at a price that is equal to the price the parties would negotiate for Gorilla glass in the abstract (say $p_2$ ) minus an amount equal to how much a successful hold-up in the original contract would benefit Apple. This benefit is equal to the difference between the price that the original contract specified ( $p_1$ ) and the price Apple could extract if it was able successfully to hold Corning up ( $p_H<p_1$ ). In short, the second contract price would be $p_2-\left(p_1-p_H\right)$ .

Blockchain can be used to prevent renegotiating through subsequent contracts as well. The smart contract could crudely specify that, if there were a second exchange between Apple and Corning, the penalty would be triggered. Of course, that approach would have collateral damage: the parties may reasonably want to trade a second time and this penalty trigger would prevent that. To address that, the parties could agree that the penalty would only be triggered if the second contract had a lower price than the first contract, although this would create problems if Corning’s costs fell over time. Alternatively, they could even configure the smart contract to both structure renegotiation on the first trade and to provide a framework for subsequent trades.Footnote ³⁸ That framework includes a revelation mechanism for subsequent trades that would allow the smart contract to compare the price and cost of the subsequent trades to ensure that Corning was fully reimbursed for its investment in prior to the first trade.

A natural question is how the smart contract code would know if Corning or Apple violated the default option or final offer or negotiated a subsequent contract that de facto renegotiated their original contract. The smart contract can certainly monitor the blockchain – the ledger of transactions – on the network, in which case it would know directly if another trade occurs. It is here that the open feature of blockchain is critical. It could also monitor communications or accounts not on the blockchain network by using APIs that gave it access to the parties’ messages or to other accounts the parties may have.Footnote ³⁹ When the parties sign the contract, they want the penalties so that they would have an incentive to allow the smart contract access to all their messages and accounts.Footnote ⁴⁰

Of course, the fact that the parties want to share information does not mean it is easy to do so. It may be difficult to give access to all communications between parties, for example, in person conversations between employees of the two parties.Footnote ⁴¹ It may also be that the parties have accounts at some banks that do not provide APIs that can give the smart contract access and the parties cannot change those banks’ policies. We hope this is a limitation that recedes in time as more accounts become interoperable in the sense of providing API access. We also think it may not be a large limitation as parties certainly have an incentive to keep most of their money in accounts with banks that offer APIs, as those APIs allow the parties themselves to monitor their own accounts more easily.

5.2 Generalizing the Example

Having shown examples of how ADR provisions might be implemented, we can infer how the provisions required for a revelation mechanism or the original contract as a whole can be implemented via smart contract. The smart contract can employ penalties to get the parties to comply with specific steps of the revelation mechanism. To complete each mechanism, the smart contracts must require that parties take actions in a certain sequence, for example, the take-it-or-leave-it offer comes before the default option can be triggered, otherwise the party that was not supposed to move faces penalties. Since the renegotiation design mechanism is the whole contract, the parties that want to employ that mechanism need only specify the steps that that mechanism requires. Parties that wish to employ the revelation mechanism can only finalize their smart contract by adding to their revelation mechanism a schedule of contingent trades that are each triggered by a different combination of valuations $v$ and costs $c$ revealed through the revelation mechanism.

For those parties that do not have the relevant information to devise renegotiation design or revelation mechanisms, blockchain still has value. Such parties have to make a choice: write a contract that never allows renegotiation, or always allows renegotiation, whether due to hold-up or to changed circumstances. If they think the hold-up problem is bigger than the problem of inflexibility, they can bar renegotiation under any circumstance by imposing penalties for such renegotiation. The steps were outlined earlier.

5.3 Robustness of Smart Contracts on the Blockchain

Smart contracts on the blockchain have at least three limitations that we have not yet discussed. One is that government may simply ban blockchain or smart contracts that impose penalties.Footnote ⁴² This seems like an extreme course of action and, while we can only speculate, this does not seem particularly likely. Blockchain has a great deal of value outside of the contractual commitment setting. Entities that obtain that value would likely lobby or litigate against a broad ban of blockchain (Reference MatonisMatonis 2013). A narrower ban on smart contracts with penalties is perhaps more plausible. It would be akin to a ban on penalty clauses (Reference FarnsworthFarnsworth 2003: 845). The main problem is that the ban is difficult to enforce. Both parties to a contract with a penalty want that penalty ex ante, so have little incentive to report that they have written a smart contract with a penalty. And ex post a court would face difficulties in rescuing them from that penalty if the penalty is written correctly or if the parties remain anonymous – as discussed earlier.

One possibility is that KYC rules – versions of which already exist – could be used to pierce the anonymity of the contracting parties. Then the efficacy of the smart contract would hinge on whether courts are willing to enforce contracts that fully ensure one or both of the contracting parties against decisions by the court. Courts may well blanch at such provisions and refuse to enforce them. This would limit the applicability and efficacy of the smart contracts that we have outlined.

Absent this, only a whistleblower, a third party, or a contractual party who was coerced into agreeing to a smart contract penalty clause has an incentive to get courts involved. But legislators would have to authorize bounties for whistleblowers, another type of third-party standing, or criminal penalties. Those may be political hard sells because no outsiders are hurt by the commitment described in this element. The biggest risk comes from coerced parties. Indeed, these are the individuals who were able to undermine penalty clauses in non-smart contracts. Perhaps the difficulty of enforcing a ban on all smart contracts with penalty causes will give courts an opportunity to limit their ban on such agreements to those where one of the parties is unsophisticated and may have been coerced into signing it.

A second limitation concern with smart contracts on blockchain is that, like computer scripts generally, they are sensitive to honest mistakes. For example, if Apple accidentally hits send on its email that contains a draft – but not the final draft – of its take-it-or-leave-it offer, the smart contract will not allow it to recall the message, otherwise Apple could circumvent the smart contract by calling its first offer a message it wants to recall. To address the costs of such mistakes, game theorists sometimes look for equilibria of games that are robust to “trembling hands”; correlatively, mechanism designers might look for game rules that yield trembling hands perfect or robust equilibria (Reference SeltenSelten 1975: 38). We do not know if the mechanisms we have discussed in this Element are robust to errors, although empirical work by Reference AghionAghion et al. (2017: 232) gives us some hope that smaller mistakes might yield smaller deviations from the first best incentives to invest.

It is also possible that even sophisticated parties could simply make a mistake in crafting the smart contract and then find it impossible or extremely costly to reverse. If there is a Pareto improvement to be made from correcting the smart contract, then both parties would have an incentive to renegotiate the contract and reveal themselves to the court. This presents important challenges for how courts should respond to such circumstances. On the one hand, it would be unusual for a court to stand in the way of an agreement that makes both parties better off. On the other hand, refusing to do so might deter the use of smart contracts in general. Indeed, if the court wishes to deter such contracts, it may want to develop doctrine that does just that by being unwilling to correct mistakes at the request of the parties to the smart contract.

Another way to frame this drawback with smart contracts is that they introduce a new form of transaction cost – namely the cost of blockchain adaptation. This suggests a clear trade-off between the benefits of commitment and the costs of being unable to adapt (easily) to changing circumstances. One would naturally expect that in environments where adapting to changing circumstances is very important that smart contracts on a blockchain would be less desirable, and used less extensively.

A third limitation of the smart contract penalties is that they only work if the parties have sufficient assets to pay the penalties required by the smart contract. One way to ensure that smart contract penalties can be implemented and are effective is for the parties to place enough assets in their accounts on the blockchain network to cover penalties until both parties satisfactorily perform on the contract. Note that these assets are available to be employed or spent until penalties are required if the parties take out a loan collateralized by the assets in their account on the blockchain. Another way to ensure penalties can be covered is to not require the parties to put up a bond on the blockchain but to allow the smart contract to create debt on behalf of the penalized party to a lender, with the proceeds from the lender being used to pay third parties pursuant to the penalty. The lender would be willing to do so long as the party on the hook for the penalty had enough assets, on the blockchain or otherwise. Of course, if either party did not have enough assets to cover optimal penalties, optimal penalties would not be available. In that case, penalties would have to be lowered, and hold-up risks would rise commensurately.

A final potential limitation of smart contracts concerns implementation. As we mentioned earlier in this Element, the automated execution facilitated by smart contracts takes as a crucial input information that can be verified digitally. As Reference Bakos and HalaburdaBakos and Halaburda (2019) point out, this information will often come from the so-called Internet of Things. They put it this way: “IoT sensors expand the state space over which the contract can be specified by creating finer partitions of the verifiable states of nature. This typically leads to more efficient trades, but it is still not fully efficient.” Indeed, the proximity of what we call the smart-contract second best to the first-best outcome will be a function of the information required by the first-best contract and the efficacy of the information-providing technology. Indeed, this technological view of the second-best is a nascent and interesting area for future work.

6.1 Implications for Size of Firms

One implication of our claim is that blockchain and smart contracts have to the potential to increase the gains from, and thus amount of, relationship-specific investment and trade and thus total output that we observe in the economy (Reference WilliamsonWilliamson 1979). The degree of benefit is proportional to the extent to which hold-up is a drag on investment and trade. In addition, we expect that blockchain should reduce the size of firms. As blockchain increases the returns to market-mediated transactions, firms will move more transactions outside of the firm to the market. In this way, it functions much like prior technological innovations that increased counterparty trust and accountability (Reference CoaseCoase 1937; for an example of how technology can have this effect, see Reference Baker and HubbardBaker and Hubbard 2004). Blockchain may also do this by reducing the role of trust intermediaries, such as banks, in the economy, although that is not a feature we emphasize in this element (Reference AntonopolousAntonopolous 2017: 4).

6.2 Speculation about Effects on Contracting Generally

While our focus has been how smart contracts on the blockchain can tackle the problem of hold-up in contracts, there may be other benefits of these technologies to contract law generally. One that we think is particularly important is that they may reduce the need for centralized entities, especially courts, to resolve contractual disputes. We think this is possible through at least two channels. First, as we explained in Section 5, smart contracts can reduce the need for court interpretation of contracts. A smart contract script, like other computer code, does what it does. The language compiler is the interpreter and it does not permit variation in interpretation. Parties can tackle bugs in the code via simulation, but they get what the code does. In some sense, the compiler is the ultimate textualist interpreter, with little regard for absurdities. Parties do not have to use it, but they have the option. When they do, the role of courts will decline to a greater or lesser extent.

Second, blockchain networks may use their validation mechanism or their consensus rules to adjudicate smart contract disputes. If a smart contract crashes because it is poorly written or if a party feels the counterparty somehow did not honor a smart contract,Footnote ⁴³ the network can appoint a third-party node on the network not just to witness the contract but to adjudicate the dispute. The idea would be that, just as the Bitcoin network replaced banks as the intermediary for payments, it can replace courts as arbiters of smart contract disputes.

Of course, the difficulty is that the work required to adjudicate smart contract disputes may not be as algorithmic as mediating transactions. The latter require only that the third party verify the sender of money has enough money in its account (which is evident from the blockchain ledger up until the date of payment) and that the transfer to the receiver’s account is recorded in the updated blockchain ledger. Resolving contract disputes requires, for example, determining what the parties would have scripted if they had considered the state in which the smart contract crashed. This requires fact finding beyond the information already available on the blockchain.

It is not clear that other nodes, or even a consensus among nodes, is a good way to resolve the problems. The cost of fact finding of the sort required for adjudication is greater than the cost of fact finding required to verify a payment. The third-party node has to be compensated for that higher cost, just as it is compensated for verifying transactions. Moreover, the other nodes may not be well qualified to adjudicate disputes. The idea of using blockchain is not entirely ludicrous, however. Civil trials, including contract trials, in existing courts also impose great costs on parties. Moreover, they may be tried by a jury, which also may not be qualified.

Even if a blockchain-mediated peer-to-peer network does not displace courts, it may be able to complement courts in two ways. First, it may help courts better identify majoritarian default rules. The more that smart contracts are written on the blockchain, the easier it would be for courts to determine true majoritarian default rules. This is currently difficult as selection into litigation determines what judges see (Reference Baker and MalaniBaker and Malani 2014: 1577) and general fact finding is limited or biased by the adversarial process (Reference ChristopherTarver Robertson 2010: 177). As a result, there is reason to suspect that courts may not be filling in gaps in contracts with the right default rule, but because contracts on the blockchain may be public, the court can directly query what most parties want in specific situations.Footnote ⁴⁴

Second, just as parties now can set the jurisdiction that will govern their contract disputes, smart contracts may even allow parties to specify which set of (scripted) default rules should govern the gaps in their contracts. This process would be similar to the way in which people putting up intellectual property (IP) such as photos or images onto the web might simply appeal to the Creative Commons license to govern the usage of their IP. This would lower the demand for courts, but in case disputes that go to trial, it would help courts decide cases.

6.3 The Importance of Blockchain and Smart Contracts

With many new technologies that come along, there is a wave of articles that claim the technology fundamentally changes law. They risk creating what Judge Easterbrook called the law of the horse (Reference EasterbrookEasterbrook 1996). We do not yet think that blockchain and smart contracts will change contract law. Rather, we think it may have a measurable impact on the sorts and amounts of contracts that can be written. To justify our beliefs, we need only to point to the substantial amount of investment going into blockchain. A large fraction of the largest banks, IT companies, and consulting firms are investing in blockchain.Footnote ⁴⁵ The total number and value of transactions on the Bitcoin and Ethereum networks, while still far short of even Paypal, are growing rapidly (Bitinfocharts.com 2012a, 2012b; Altcointoday.com 2017). Finally, the total amount of money raised through initial coin offerings exceeded the total amount of venture funding in the last quarter (Reference KharpalKharpal 2017). While we do not think that there is enough data to conclude that blockchain will change the world, let alone law, we do think that there is enough activity in the technology to begin considering the legal implications of this technology, including, as we do in this article, how blockchain and smart contracts affect contracting.