The future of maximal extractable value (MEV)

In the past year, the phenomenon of maximal extractable value (MEV; formerly known as miner extractable value) has captured the public consciousness, in part because of the apparent high level of technical skill required to extract MEV and in part due to the lucrative nature of successful MEV extraction. However, the discussion of MEV is often confused and imprecise, despite the increasingly critical role that MEV plays in blockchain ecosystems. With total extracted MEV plausibly in the billions of dollars (600M in tracked MEV on Ethereum mainnet alone via MEV-Explore), it is unsurprising that the bulk of MEV dialogue focuses on the profits of MEV extractors; however, the scope, evolution, and management of MEV are far-reaching topics with potentially existential consequences for blockchain security in the very long term.

In this article, we will aim to clarify a number of important topics around MEV. We will first present and illustrate a precise definition of MEV. Subsequently, we will discuss how MEV has evolved over the past year and extrapolate forward in order to understand the key questions and concerns posed by the continued growth of both MEV and the broader cryptoeconomy in the coming years. We will pay close attention in particular to the incentive structures (both extant and novel) which motivate different actors in the MEV ecosystem. Finally, we close with a brief survey of directions for future research.

The interlocked roles of block producers, searchers, protocols, and users in the creation and extraction of MEV affect each other in varied and dynamic ways, often rendering discussions of MEV somewhat confusing. Through this article, we will seek to analyze MEV from the perspective of how different systems or proposed solutions benefit and harm the different actors in the system. We will see that this is a clear and effective framework through which we may begin to derive the long-term end states of MEV in different cryptoeconomic systems.

What is MEV?

MEV is precisely defined by as “the maximum value that can be extracted from block production in excess of the standard block reward and gas fees by including, excluding, and changing the order of transactions in a block.” At first glance, this appears fairly distinct from the common conception of MEV, which is used almost interchangeably in the popular lexicon with “running trading bots.” However, if we carefully examine several common examples of MEV, we will readily understand how they relate back to the formal definition.

Recall that MEV was first defined by Daian et al. (2019), Flash Boys 2.0: Frontrunning, Transaction Reordering, and Consensus Instability in Decentralized Exchanges by first looking at “the widespread and rising deployment of arbitrage bots in blockchain systems” before extrapolating to a more general phenomenon of value extraction via priority ordering of transactions within and across blocks. Beyond arbitrage alone, another typical example of MEV with which many users have personal experience is the phenomenon of sandwich attacks.

Many other forms of MEV exist beyond arbitrage and sandwich attacks, particularly so-called “exotic” or “long tail” MEV. For example, the phenomenon of generalized frontrunning was vividly illustrated in a very popular post, Ethereum is a Dark Forest, by Dan Robinson of Paradigm, and more comprehensively studied by Qin et al. (2021) in Quantifying Blockchain Extractable Value: How dark is the forest? In this article, we will not seek to comprehensively catalogue every form of MEV; curious readers are advised to refer to the Flashbots Research Vault.

By analyzing arbitrage and sandwich attacks, we may see that both accrue value from block producers’ ability to arbitrarily reorder transactions, and are hence properly considered MEV:

  • Arbitrage. An arbitrage opportunity is characterized by a sequence of trades such that the trader ends the sequence of trades with a larger quantity of the initial asset. When performed atomically (i.e., the entire sequence of trades is contained within a single transaction, each part of which only executes if the entire arbitrage succeeds), this profit is riskless (net of transaction fees).

    Suppose that prices on two markets (including, e.g., two separate liquidity pools on the same AMM) differ sufficiently for the same asset, such that a profitable arbitrage opportunity exists. This ‘unbalanced’ state must be the end result of users transacting with the relevant markets. Suppose that a user creates an arbitrage opportunity (e.g., with a very large purchase or sale); then a miner may insert their arbitrage transaction immediately after the arbitrage-creating transaction to capture the arbitrage.

    In this situation, it is possible in principle that the arbitrage opportunity could go unclaimed for many blocks. However, the block producer is privileged by being able to “backrun” the arbitrage-creating transaction, which makes the nature of their profits riskless. In contrast, a non-block producer entity trying to capitalize on the arbitrage opportunity must either pay the block producer for the privilege of inserting their arbitrage transaction immediately after the arbitrage-creating transaction or run the risk of transaction failure if the arbitrage opportunity disappears before their transaction. Hence we see that in this common situation, even if an external (non-block producer) user is capturing some of the arbitrage profits, they are fundamentally relying on the ability of the block producer to reorder transactions to ensure riskless profits.
  • Sandwich attacks. A “sandwich attack” is a phenomenon where a user’s trade is “sandwiched” by two transactions. In general, block producers have the ability to monitor pending transactions (which have not yet been ordered and assembled into a validated block, and which reside in a location known as the “mempool,” short for “memory pool”) and, therefore, to insert their own transactions before certain target transactions, a practice known as “frontrunning.” Upon noticing that a user is about to purchase a given asset, the sandwicher (1) inserts a large buy for the same asset immediately before the target transaction (a frontrun) and (2) inserts a sale of all the purchased quantity immediately after the target transaction (a so-called “backrun”). The sale executes at a better price than the buy due to the execution of the target transaction in between, resulting in a profit for the sandwicher.

    We see immediately that the ideal execution of a sandwich attack is reliant on transaction ordering privileges. Without the ability to order transactions as desired, other trades may happen in between the two halves of the sandwich attack, potentially resulting in losses for the sandwicher. Similarly as with arbitrage opportunities, many sandwich attacks are conducted by non-block producing entities; however, these entities are still reliant upon the special privileges of block producers, and compete to capture the value of those privileges through fee payments to block producers.

In both examples, it is apparent how maximal value extraction is reliant upon the ability to re-order transactions, a privilege given to block producers.

It is occasionally claimed that this definition of MEV is “too broad,” a criticism which is more applicable to the example of arbitrage than to sandwich attacks; however, even in the case of arbitrage, there is clearly nonzero value in the ability to order transactions such that the arbitrage transaction is placed directly after the transaction which creates the arbitrage opportunity. As such, we may perhaps distinguish between two alternative definitions of an “MEV opportunity”:

  1. Strict definition. An MEV opportunity is characterized by the majority or totality of the value being captured via transaction ordering privileges.
  2. Permissive definition. An MEV opportunity is characterized by at least a small fraction of the value being captured by transaction ordering privileges.

In both cases, however, there is at least some value (even if relatively small) being captured via transaction ordering privileges. It is likely that failure to distinguish between MEV, strictly defined MEV opportunities, and permissively defined MEV opportunities is responsible for some of the confusion about what “is” or “is not” MEV.

An even more restrictive criterion for MEV is occasionally postulated, where the value extracted must necessarily be close to riskless, with minimal warehousing of risk before profits are realized. This definition excludes so-called “probabilistic MEV,” where the value of the MEV opportunity is not exactly calculable but is rather sampled randomly from some distribution. Although readers are free to define MEV how they like, we do not believe that excessively strict definitions of MEV are practically useful. Ultimately, considerations about the risks and benefits of MEV apply not only to riskless profits derived from complex transaction reordering schemes where the bulk of the value is inaccessible without reordering privileges, and correspondingly it is the broader, more inclusive definitions of MEV which prove to be of the greatest practical utility. Whenever a user is not wholly insensitive to the exact placement of their transaction within a given block or even a series of contiguous blocks, their willingness to pay to reduce uncertainty about their transaction’s relative position represents the presence of cryptoeconomically significant MEV.

Beneficiaries of MEV extraction

As alluded to above, although MEV is intrinsically tied to block production privileges, and, correspondingly, the ability to arbitrarily reorder transactions, a complex economy has evolved around MEV extraction such that block producers are not the sole beneficiaries of MEV extraction. Due to the complexity of identifying and acting upon MEV extraction opportunities, the vast majority of MEV is currently extracted by external “searchers,” who submit transactions to block producers for inclusion in a future block. In some (perhaps many) cases, block producers may themselves be searchers. When they are not, the searcher will typically pay the block producer to have their transactions placed in the desired position within the block (typically at the top), with payment most commonly determined either through a priority gas auction (PGA) or sealed envelope auction (e.g., through Flashbots).

Beyond the block producer and the searcher, the broader ecosystem may derive general benefits or suffer various costs as a result of MEV extraction. For example, especially prior to the implementation of EIP-1559, PGAs would frequently bid up gas prices on Ethereum mainnet to tremendously high levels, significantly degrading usability of the network for ordinary users due to expensive and unpredictable transaction costs. At the same time, however, efficient arbitrage between AMM pools ensures consistency of asset prices across markets and propagation of price discovery. Furthermore, some protocols rely on arbitrageurs to function ‘correctly,’ such as Balancer, where external arbitrage is the mechanism which rebalances users’ fixed-weight asset portfolios, or Primitive, where external arbitrage evolves users’ option positions toward the correct payoffs. Consequently, the design of an appropriate MEV extraction that incentivizes positive-sum behavior and accrues rewards to the right actors is of far-reaching importance to the long-term health of any given blockchain.

Block producers and searchers

Due to the large number of actors in the MEV ecosystem, it is simplest to begin with an analysis of the benefits which accrue to block producers, as they play essential roles in blockchain functionality. Earlier blockchains such as Bitcoin and Ethereum rely on proof-of-work as a consensus mechanism, in which miners are the block producers. However, as blockchain architecture has evolved, we have seen the development of proof-of-stake blockchains in which validators, incentivized toward good behavior through their staked tokens, serve the role of block producers. (Ethereum mainnet itself is slated to transition to proof-of-stake in 2022, an event known as “The Merge.”) The increasing popularity of proof-of-stake blockchains is what motivated the change from “miner extractable value” to “maximal extractable value;” similarly, considering block producers generally rather than miners alone will broaden the applicability of our analysis. As we will see, examining the perspectives of block producers will also allow us to understand the dynamics that drive searchers, due to the substantial advantages of integrating both roles together.

Block producers benefit in two primary ways. First, block producers may extract MEV themselves by running software to search for extraction opportunities whenever they propose a block. Second, they may sell transaction reordering rights to searchers. In the first case, they capture all the value extracted; notably, in the second case, they currentlycapture an increasing fraction of the value extracted as competing searchers submit higher bids (i.e., competing searchers are willing to accept increasingly lower shares of the extracted value for the right to extract any value at all).

A number of fascinating dynamics are at play here:

  • Increasing network dominance. Separate block producers which implement different MEV searching strategies are strongly incentivized to merge and form increasingly larger entities. Through the combination of their MEV strategies and greater ability to invest into searcher R&D, mergers allow both parties to extract more value than they would otherwise be able to on their own (i.e., MEV extraction is subject to economies of scale). In particular, small block producers which are not well capitalized enough to develop competitive MEV strategies are likely to be acquired by larger blocs of integrated searcher-block producers, threatening the decentralization of the entire blockchain. Although auction mechanisms such as Flashbots auctions mitigate this risk by allowing even small validators to capture a large share of MEV revenue, increasing MEV complexity may lead to the differential performance of integrated searcher-block producers (discussed below), which will in time exacerbate the risk of centralization via block producer mergers and acquisitions.

    Additionally, block producers which are able to more efficiently extract MEV will, all else held equal, acquire increasingly dominant shares of the network. In a world with no MEV and a fixed set of block producers with constant hashrate or stakes, rewards are approximately proportionally distributed and the relative power of different block producers stays constant over time. Therefore, if some block producers are effectively compensated at a higher rate than others through superior MEV extraction, they will asymptotically dominate the network.

    Rewards from MEV extraction can themselves be used to acquire a larger fraction of the network; furthermore, in proof-of-stake blockchains, they can induce users to delegate tokens to their stake by offering them liquid staking derivatives that capture some fraction of the MEV profits generated on the delegated stake, such as with the recent release of Eden Network’s yyAVAX. Note that MEV extraction itself scales superlinearly with network dominance, with a straightforward linear component from ability to reorder transactions directly scaling with share of hashrate or stake, and an additional component from new MEV opportunities arising from reordering transactions across multiple contiguous blocks. That being said, these winner-take-all dynamics may take some time to play out, as evidenced Ethermine’s maintenance of a high fraction of total Ethereum hashrate (currently 30%) despite their ban on bundles containing DEX transaction frontrunning

    In sum, these constitute the commonly discussed centralization risk of MEV. With increased centralization, blockchains become exposed to adverse actions such as 51% attacks or malicious reorganizations. It is valuable to note, however, that as network dominance of any given block producer increases, they are increasingly incentivized to protect the value of the network as a whole, which perhaps mitigates the risk of truly destructive attacks against the chain.
  • Integration of searchers with block producers. If the block producer is not near-maximally competent at MEV extraction, then they are heavily incentivized to sell the right to reorder transactions to searchers, hence the increasing popularity of MEV-Geth, which supports a system of closed bid auctions for transaction bundles known as Flashbots auctions.

    It is conceivable that a competitive market for MEV extraction will lead to a situation where the greatest profits for the vast majority of block producers will come from selling their reordering rights in this market, rather than extracting MEV themselves. As such, some have hypothesized that Flashbots or Flashbots-like mechanisms will dominate in coming years, due to competition among searchers asymptotically reducing the searcher profit margin to very low levels and correspondingly allowing block producers to capture the vast majority of MEV with near-zero marginal investment.

    However, as noted by Doug Colkitt, this is only applicable when the value of a given transaction reordering is agreed upon by all participants. This is currently the case with the vast majority of MEV opportunities; e.g., the value of an atomic arbitrage is straightforward to calculate. However, as the complexity of blockchain transactions increase, it is natural to expect that searchers will become increasingly differentiated in their ability to value the total extractable MEV in any given set of transactions. In such a situation, it becomes strongly advantageous to be an integrated searcher-validator rather than to be a searcher alone, because if other searchers arrive at a higher valuation of the reordering opportunity than you do, they will bid accordingly and you will be able to extract zero value. In contrast, if you are an integrated searcher-validator (or you establish an exclusive private relationship with a block producer, etc.), you will be able to act on your private information and capture the associated value.

    In essence, the above situation is analogous to the “winner’s curse” in classic auction theory where participants receive private information about the value of the item being bid upon. As discussed above, private information, i.e. diverging valuations of any given transaction reordering opportunity, is likely to come about with the increasing complexity of blockchain transactions, where sophisticated searchers will have substantial advantages over naive searchers. Beyond the complexity of transactions within any given blockchain, however, diverging valuations may also come about as a result of statistical MEV, where searchers probabilistically diversify risk across time and/or space (e.g. executing transactions on multiple blockchains) rather than focusing purely on atomic, guaranteed profit opportunities. Finally, the increasing popularity of low-fee blockchains such as Solana or appchains in the Cosmos ecosystem effectively makes high-complexity, low-profit MEV opportunities increasingly viable, whereas blockchains with high transaction fees such as Ethereum mainnet set a higher floor for the profitability of MEV opportunities and hence a lower ceiling for their complexity (under the reasonable assumption of a complexity-versus-profit tradeoff in the space of MEV opportunities). Empirical evidence for this hypothesis is supplied by the dominance of Jump Capital among Solana validators; they compose approximately 20% of total staked Solana, and it is quite likely that they are leveraging their high levels of human capital to extract the vast majority of available MEV.
  • Generalized security benefits. Although block producers’ ability to capture MEV may result in long-term centralization risk where dominant producers have increasing ability to mount attacks on the network as described above, the existence of MEV incentivizes new network entrants, which directly counteracts the centralization risk from highly competent searcher-validators. Additionally, the accrual of financial value to block producers in and of itself increases the security of the network as a whole against external attackers, who have to deploy larger amounts of capital to dominate a majority of the network’s hashrate or stake.

    As a result, MEV has both positive and negative effects on the security of a given chain. The general accrual of MEV to block producers increases network security, whereas the specific accrual of MEV to particular block producers decreases network security.

Fundamentally, as transaction complexity and searcher ability increase over time, it is highly plausible that differences in block producers’ abilities to extract MEV will exacerbate. Therefore, one can imagine a world where the most competent searchers integrate with block producers and eventually make up a very large proportion of the network’s hashrate or stake, and almost all other searchers (with relatively little private information compared to the top of the pack) capture diminishing amounts of MEV through Flashbots-like auction mechanisms. Although this may be an acceptable tradeoff for increased total network security, a number of further strategies to mitigate centralization risk have been proposed, which we will discuss in the next section.

Ordinary blockchain users

Beyond block producers and searchers, ordinary users may also derive benefits from efficient MEV extraction (beyond the generalized security benefits previously discussed):

  • Cross-market price stability. Arbitrage between liquidity pools ensures that asset prices do not wildly diverge across different DEXes and blockchains, allowing users to freely transact without needing to check prices on dozens of markets beforehand.
  • Conditional transactions. Some systems rely on the timely execution of certain transactions when specific conditions are met. For example, lending platforms rely on users to liquidate positions which have fallen below their minimum collateralization ratio. Protocols might also conceivably want certain utility functions to be run after fixed intervals of time. In both cases, there is a competition among searchers to bribe the block producer such that the block is restructured in a manner so that the searcher is the first user to capture a reward associated with the desired transaction. Therefore, these opportunities, although dismissed by some as simply being “on-chain botting,” ultimately do qualify as MEV to some extent.

Naturally, of course, users may also suffer negative consequences from the proliferation of MEV:

  • High transaction costs. As previously mentioned, depending on the structure of a given blockchain’s transaction fees, priority gas auctions (essentially, open auctions where searchers repeatedly submit successively higher bids for the same transaction based on their observation of competing bids) can drive up transaction fees for ordinary users. This results in unpredictable spikes in fees for mundane transactions, substantially reducing quality of life and rendering less well capitalized participants unable to send any transactions at all.
  • Network spam. In contrast, if transaction fees are unusually low, scaling perhaps excessively slowly or not at all with complexity of computation, then searchers are incentivized to spam the network with enormous numbers of low-value transactions to capture MEV opportunities as soon as they arise. Even if only a fraction of these opportunities are realized for any given searcher, the underpricing of transactions results in net profitability. We observe this in practice with multiple blockchains such as Polygon and Solana. Similar to priority gas auctions, this also reduces quality of life for the ordinary user, except rather than facing high transaction fees, non-searchers are simply unable to have their transactions confirmed with any reasonable probability, as they are simply crowded out by the vast quantities of network spam.
  • Extractive frontrunning. Finally, some forms of MEV are purely value-extractive from users and provide zero benefit to the broader cryptoeconomy. A simple example is the presence of sandwich attacks, which are pure value transfers from users to parties which capture MEV; it should be patently clear that no entity aside from direct MEV beneficiaries benefit from the existence of sandwich attacks. More generally, almost all forms of frontrunning are purely extractive and reduce quality of life for the end user through increased costs and unpredictability.

Consequently, the net effect of MEV for ordinary users is the sum of a large number of different factors, and whether it is positive or negative in sign is often unclear. As we will soon see, different systems have been proposed which attempt to shift this calculation toward the direction of net benefits for users.

Innovations in MEV management

Over time, blockchain developers have come to appreciate the complexity of MEV and its integral part in modern cryptoeconomic systems. Correspondingly, they have sought to restructure blockchain architectures and incentive systems to mitigate the negative effects of MEV while preserving or magnifying the positive effects. These attempts largely fall into two categories:

  • Distribution of MEV equally across block producers, to avoid centralization risk while reaping the network security benefits of MEV extraction
  • Mitigation of the negative effects of MEV on ordinary users through reduction of purely extractive MEV and/or distribution of MEV profits to the blockchain ecosystem

These strategies have been attempted at both the base architectural layer as well as at the protocol or application layer. To illustrate, the implementation of EIP-1559 is an architectural change which seeks to mitigate the negative effects of priority gas auctions on ordinary users, but does little to change the distribution of MEV profits across block proposers. In contrast, Flashbots-style auctions for transaction ordering privileges allow block producers to take advantage of a competitive searcher market to produce an effective floor on their MEV extraction effectiveness, therefore narrowing the gap between the worst and best block producer in terms of MEV extraction, but do nothing to prevent extractive MEV. In the following, we will touch upon some newer systems or proposed changes as well as their relative strengths and weaknesses.

Fair sequencing

Naively, the simplest way to eliminate transaction frontrunning is to impose a first-in-first-out rule for transaction processing. This is straightforward to implement if a single centralized party possesses the right to order all transactions; in this case, there is a well-defined arrival order, and so long as the centralized party can be trusted, frontrunning is effectively impossible. For example, this is currently the case with Arbitrum One, an optimistic rollup on top of Ethereum mainnet, which has a single permissioned full node run by Offchain Labs with total transaction sequencing privileges. (Note that usage of sequencers is an optional part of the Arbitrum rollup technology which allows for near-instant confirmation of transactions.) However, centralization of transaction ordering to a single sequencer naturally exposes all users to the risk of malicious activity from that sequencer, and it is therefore desirable to eventually shift to a decentralized model.

However, in a decentralized setting where thousands of nodes may receive transactions at different times, implementing a precise notion of a fair arrival order is not trivial. Some progress in this direction was made by Kelkar et al. (2020), Order-Fairness for Byzantine Consensus, which proposes formal definitions of “fair ordering” as well as a family of protocols, termed Aequitas, which provide various guarantees of fair ordering. At a very high level, these protocols seek to ensure that if many nodes receive transaction A before they receive transaction B, then the resulting ordering should place transaction A before transaction B. Arbitrum One plans to eventually implement such a fair ordering protocol with the assistance of a decentralized network of Chainlink oracles.

It is plausible that we will see further advances in development of fair ordering algorithms in the coming years, and that practical usage of these consensus protocols by blockchains such Arbitrum One will significantly diminish the severity of extractive transaction frontrunning in certain venues. However, it is valuable to note that reliance upon a first-in-first-out paradigm is not without drawbacks:

  • Latency advantages. Actors with the lowest latency to nodes will be able to extract more MEV than those without. This favors highly capitalized entities, which are able to invest resources in co-locating near nodes and building out fast network connections. In general, differences in network latency will systematically disadvantage regions of the world with lower connectivity, which are also precisely those areas most likely to be disproportionately lacking in economic resources.
  • Network spam. To increase the probability that their transactions are broadcasted through the network as soon as possible, users and particularly MEV searchers are heavily incentivized to spam the network with identical transactions, repeatedly sending them to many different endpoints and dramatically increasing the probability that a normal user’s single transaction will be dropped or delayed.
  • Intermediaries between users and sequencers. Depending on how users typically send transactions to sequencers (or decentralized networks of oracles for fair sequencing, etc.), the intermediary may itself be a source of risk. For example, if users send transactions to an Arbitrum-based blockchain through RPCs, those RPCs could in principle reorder transactions and extract MEV before passing them along to sequencers.

At the end of the day, “fair sequencing” is only fair relative to a particular set of priorities; ultimately, current proposed implementations can simply be thought of as an alternative set of tradeoffs relative to other MEV economies.

Auctions for N-block sequencing rights

Instead of reliance upon a predetermined set of entities to sequence transactions (e.g., a decentralized Chainlink network implementing fair sequencing), it has been alternatively proposed that the ability to arbitrarily reorder transactions within a contiguous window of N blocks can be auctioned off by block producers. This mechanism creates a competitive market for MEV extraction rights while ensuring that users have a guarantee that their transaction will only be delayed by ~N blocks at most. The most well-known implementation of this strategy is by Optimism (an optimistic rollup on top of Ethereum mainnet), which calls these auctions “MEVA” (MEV Auctions) and aims to use MEVA revenues to fund the development of public goods.

It is informative to analyze the impact of MEVA from the standpoint of individual beneficiaries:

  • Block producers should in principle be able to capture the majority of the value through the competitive bidding process for transaction sequencing rights. However, their short-term profits may be reduced to the extent that the blockchain requires a fraction of the auction proceeds to be diverted to public goods funding. This reduction is potentially mitigated, in part or in full, by the ability for searchers to capture higher amounts of total MEV.
  • Searchers will be dramatically affected by the introduction of MEVA; total searcher profits may increase due to increased ease of extracting multi-block MEV, but the distribution of those revenues will likely become enormously unbalanced, with the vast majority accruing to the most skilled searchers.

    Suppose, for example, that a searcher wins an auction to be the sequencer for a chunk of N blocks. The searcher will extract what MEV they can with their expertise; however, it is plausible that there is MEV remaining in the block that they have not extracted. They can therefore either sell the rights to extract the remaining MEV to other searchers or expand their capabilities to more fully extract every single type of MEV. However, it is extremely logistically complex to auction off the rights to extract MEV when the searcher is themselves not aware of what MEV remains in the block (because if they were, they would extract it themselves). Therefore, the introduction of MEVA will accelerate the formation of a small number of monolithic MEV groups skilled at extracting every single form of MEV which consistently win the N-block auctions.

    This may be contrasted with the Flashbots sealed envelope auctions, which allow searchers to only bid for the rights to reorder selective bundles of transactions. Although it is possible in principle for bundles to contain unextracted MEV, the relatively targeted nature of bundle submissions means there is relatively less incentive for searchers of different types of MEV to combine into monolithic entities compared to a setting with multi-block MEVA.
  • Ordinary users experience slight long-term gains from the funding of public goods that benefit all blockchain users. However, the explicit introduction of a competitive market to extract multi-block MEV and the overall centralization of MEV extraction may result in a higher level of short-term losses.

Interestingly, due to the “winner’s curse” of auctions where participants have different private signals as described in a prior section, if all transaction ordering privileges must be obtained through MEVA, the degree of complex MEV extraction may be capped at a relatively low level.

As with fair sequencing, it appears that MEVA is yet another alternative set of tradeoffs. Blockchain users benefit from the diversion of some MEV revenues to public goods funding; however, that is at the cost of MEV extraction centralization, leading to higher levels of overall MEV extraction. Additionally, there is potentially a small tradeoff in network security commensurate with the degree of revenues extracted for public goods funding, although may be counterbalanced by higher overall MEV revenue. Whether the MEVA model of MEV management proves to be more appealing than others remains to be seen in practice.

Proposer/block-builder separation

One natural extension of the optional usage of auctions to democratize MEV capture across block producers is to enforce a necessary separation between the role of the block proposer, which is the entity that assembles a complete block, and the block builder, which is the entity that attests to the validity of the assembled block. Currently, in the majority of blockchains, block proposers are also block builders, which fundamentally gives them the ability to extract MEV from the blocks, even if many may voluntarily opt to instead derive MEV revenue through auction-based mechanisms such as MEV-Geth. In this proposed scheme, known as proposer/block-builder separation (PBS), the block producer (alternatively, the block builder or attester) must accept the highest bid from block builders. Builders may attempt to extract MEV themselves; alternatively, they may accept smaller transaction bundles from searchers, which they assemble into a full block.

At a very high level, one may think of PBS as roughly analogous to requiring that all block producers run (some version of) Flashbots auctions where they are bound to accept the highest bid and where a bundle comprises an entire block’s worth of transactions. In essence, it is a strengthened version of Flashbots auctions. In that sense, PBS is likely to further democratize MEV extraction, allowing small validators to remain somewhat competitive. However, in the presence of substantial economies of scale and the proliferation of complex, probabilistic MEV, the dynamics that tend toward centralization of block producers are only dampened rather than eliminated.

Within the purvey of PBS, several different implementations have been proposed, as described in a recent Flashbots post, Why Building the Most Profitable Block is Important. Broadly speaking, these implementations take different approaches to the problem of block builder privacy, which is a key barrier in successful implementation of PBS. In essence, if the block producer selected for a given block is able to observe the contents of block builders’ submissions and submit their own blocks on the basis of that information, they can simply copy the contents of the highest bidder’s block but bid an arbitrarily high fee, capturing all MEV in the process and ultimately disincentivizing the construction of profitable blocks. Solutions largely fall into three categories:

  • Transaction obfuscation. Cryptographic techniques can be applied to obfuscate the contents of proposed blocks from block producers. For example, transactions and bundles could be compiled by block builders within a secure enclave such as Intel SGX. In theory, because usage of the secure enclave can also be cryptographically verified, this would prevent block producers from observing transactions. (However, Intel SGX specifically is known to be susceptible to multiple attacks.)

    Alternatively, more direct encryption schemes can be used to protect the privacy of users’ transactions, such as timelock encryption (where decryption requires the passage of time) or threshold encryption (where decryption requires the private keys of some threshold proportion of block producers). Regrettably, the former results in poor composability and user experience, whereas the latter is vulnerable to the collusion of multiple block producers.
  • Precommitments to proposed blocks. Instead of cryptographic barriers, block producers can be made to precommit to a certain set of block headers (each corresponding to a proposal from a block builder) before block builders are willing to publish full block contents. Block producers are then subject to a slashing rule if they attest to a block with a header not among those to which they previously committed. As such, block producers cannot observe built blocks and then resubmit a bid using that information. This proposal is described by Vitalik in further detail in Proposer/block builder separation-friendly fee market designs.

    Although elegant in its permissionless nature, this solution requires careful consideration of its design properties in order to effectively guard against attack vectors. For example, malicious block builders could submit bundles with high fees for block producers which they refuse to publish after commitment, potentially crowding out legitimate block proposals if the block producer has any cap on the number of blocks to which they will commit. The slashing mechanism also requires a careful calculation of potential failure modes; if improperly designed, it is conceivable that attack vectors could remain open to block producers, either individually or in collusion.
  • Permissioned relays. If one is willing to accept the introduction of trusted parties into the system — which could be an intermediate step in the transition toward fully decentralized PBS — then the implementation becomes much simpler. Just as how Flashbots auctions currently require the submission of bundles to trusted relays, which are assumed to not steal users’ bundles, introduction of trusted relays between block builders and block producers ensures that block builders’ proposals are not leaked to block producers. One specific implementation of PBS in this vein is MEV-Boost from the Flashbots research group.

Beyond the technical details of particular PBS systems, one unexpected benefit deserves special mention. Enforcement of PBS at the base layer means that block producers may be able to credibly claim total neutrality with respect to inclusion of users’ transactions, especially if they do not also participate in the open market for MEV extraction. It is possible that some forms of MEV may be classified as illegal by regulatory agencies, much in the same way that brokerages frontrunning their customers’ orders in traditional finance is considered a potential breach of fiduciary duty. Although this concern remains largely theoretical, it is notable that Ethermine, one of the largest Ethereum mining pools, stopped accepting DEX frontrunning bundles half a year ago for “compliance.” If such concerns continue to hold ground, PBS may allow centralized exchanges to continue offering staking services at competitive rates, as they would still derive revenue from all forms of MEV without being as exposed to potential enforcement action.

Protocol-level reduction of MEV opportunities

Certain MEV opportunities may be understood as flaws either in user behavior or in protocol design which allow for purely extractive MEV as a result of normal user-protocol interaction. These MEV opportunities are likely to disappear over time as new protocols emerge which prevent their generation.

For example, imbalances in liquidity pool are typically caused by users performing large atomic swaps within single pools. In principle, the user can spread out their transaction across multiple DEXes to reduce overall price impact and execute their trade at lower cost; however, doing so manually is slow and cumbersome. Consequently, DEX aggregators such as 1inch, ParaSwap, and Rango, which identify optimal paths for trade routing to deliver users superior trade execution across many different DEXes (and, in the case of Rango, across multiple chains as well), have become increasingly popular. Eventually, as more trade volume moves to these aggregators, the space of available arbitrage opportunities will commensurately diminish. (That being said, it is valuable to note that the individual routed trades of a larger order through an aggregator can still be sandwiched, and that arbitrage between aggregated and non-aggregated DEXes remains possible.)

Similarly, the introduction of concentrated liquidity on Uniswap V3 has led to the phenomenon of “just-in-time (JIT) liquidity,” where searchers insert very narrow ranges of deep liquidity immediately prior to a user’s trade and withdraw the liquidity immediately after, therefore capturing a large part of the associated trading fees. Although this leads to lower price slippage for executed orders, it strongly disincentivizes the provision of private liquidity and in the most pathological case may settle into an equilibrium where all liquidity is JIT and where the absence of any passive liquidity forces traders to request quotes from JIT liquidity providers. This may be prevented by the introduction of protocol-level mechanisms that make JIT liquidity essentially impossible, such as a “time-to-live” requirement proposed by CrocSwap which enforces a lower bound on how fast users can mint and subsequently redeem liquidity positions.

Other protocols have successfully reduced users’ susceptibility to MEV through the general notion of taking a previously ‘open’ process and increasing the level of ‘privacy’ such that external actors can no longer interfere to extract MEV. For example, CowSwap performs periodic batches of off-chain limit order matching between users’ submitted orders. Because orders are directly matched, it is impossible for these trades to be sandwich attacked, as the execution price has no interaction with external factors such as liquidity pool balances. By restricting the scope of the swap interaction to a direct interaction between buyer and seller, the transaction is protected from typical forms of frontrunning.

Another interesting application of scope restriction, as it were, is borne out by the KeeperDAO system, which seeks to establish permissioned channels between specific searchers, termed Keepers, and platforms where MEV opportunities are generated, such as DEXes where users’ unbalanced swaps create arbitrage opportunities. For example, the Keepers’ addresses may be whitelisted to permit them to swap with lower fees; one can in turn analogous systems for other types of protocols. The Keepers will then be able to profitably capitalize on MEV opportunities before non-Keeper searchers, and because they are not participating in an auction with the much larger class of non-Keeper searchers, they will also be able to capture a larger amount of the MEV rather than necessarily being driven down to extremely slim margins. In return for entry into this ‘walled garden’ of MEV extraction, Keepers then give up some part of their profits to be shared with KeeperDAO and the MEV-generating protocol.

In addition to KeeperDAO, other protocols have also proposed similar MEV-sharing schemes, such as bloXroute’s BackRunMe, which protects users from frontrunning while giving specific searchers earlier backrunning opportunities. In general, these arrangements bear some resemblance to the practice of payment for order flow (PFOF) in traditional finance, with privileged searchers benefiting from protection from the “toxic flow” of the broader searcher-block producer ecosystem driving their profits down to near zero much like how market makers avoid the toxic flow of highly informed HFT trades, and with users in both cases experiencing lower costs of trade execution. The MEV ecosystem created by these protocols shifts profits from block producers (who would otherwise be able to asymptotically capture >99% of the value of these opportunities) to the rest of the cryptoeconomic ecosystem. Reducing non-privileged searchers’ and block producers’ MEV revenue in this way may mitigate MEV-based centralization while modestly decreasing the overall degree of network security.

As we can see, there is substantial interest from both users and developers in building out MEV-resistant blockchain ecosystems at the protocol level. In general, to the extent that MEV opportunities are created from transactions “leaving money on the table,” as it were, we should expect users to heavily favor protocols which deliver greater value by allowing them to easily claw back at least some fraction of those inefficiencies (which would otherwise be captured as MEV). Commensurate with the maturation of cryptoeconomic systems, MEV searchers and block producers should expect “easy revenues” arising from rectifiable inefficiencies to diminish over time. This may drive searchers, who have significant investments into domain-specific expertise and hardware, toward increasingly complex forms of MEV.

Probabilistic MEV extraction

At present, the majority of MEV is captured in a very “low-risk” manner. For example, atomic arbitrage transactions or sandwich bundles submitted through Flashbots auctions are completely riskless; either they are accepted, in which case they are profitable, or they are not accepted, in which case the submitter is no worse off than before. However, as competition for MEV becomes increasingly fierce, both from a continual influx of more searchers crowding a fixed space of MEV opportunities and from persistent attempts from users and protocols to eliminate simple MEV opportunities and capture the value themselves, searchers will likely turn to increasingly complex MEV strategies.

In analogy to quantitative strategies in modern traditional finance, MEV searchers will be able to access a broader range of MEV opportunities if they increase their willingness to warehouse and manage risk. That is to say, searchers will profit from being able to extract value through transaction reorderings which are not necessarily profitable but are profitable in expectation if the associated risks are managed and diversified appropriately.

Although seemingly abstract, one simple form of risk warehousing is liquidity sniping, where searchers compete to purchase an asset immediately after the creation of a liquidity pool. Typically, tokens purchased by a liquidity sniper are not immediately offloaded in the same block, but instead sold over the course of minutes to hours. We make the following observations:

  • Despite superficial resemblance to “simple botting,” we believe liquidity sniping still falls under the umbrella of MEV. The simple evidence is that searchers are generally willing to pay the block producer more to include their buy order as soon as possible after the addition of liquidity. The expression of preferences for intra-block tranaction ordering clearly indicates the presence of MEV.
  • The profits of a liquidity sniper are not guaranteed. Depending on the preexisting token allocation, the price may well decrease under their entry price. However, in favorable market conditions, it is likely that the majority of new projects see dramatic price increases after the addition of liquidity. Therefore, the searcher is taking on inventory risk, even if their transactions have positive expected value.

Recall that the role of the market maker is characterized by the acceptance of inventory risk in return for profits from the bid-ask spread. In low-latency, high-TPS blockchains that support traditional market making strategies on central limit order books, we will likely see the role of the market maker and the block producer merge, as transaction ordering privileges will allow them to apply sophisticated management strategies to the inventory risk of their market making strategies. (This is likely one of multiple motivations for Jump Capital’s heavy investment in the Solana ecosystem, where they supply approximately 20% of all staked SOL.)

In a similar way, we may expect MEV searchers to also begin to diversify risks over the time dimension, in much the same way that modern HFT firms execute hundreds of thousands of trades every day. These trades are not all profitable, but because so many of them are made, the law of large numbers ensures that they are consistently profitable on the timescale of hours or days. There is no particular reason why transaction reordering privileges should only result in MEV extraction opportunities which are profitable on the timescale of several transactions or within a single block and therefore effectively riskless for the searcher; therefore, it will be unsurprising to see searcher strategies incorporate the sophisticated quantitative techniques of modern finance which allow the extraction of lower-certainty forms of MEV, especially on newer blockchains with low transaction fees and fast confirmation times.

As a matter of fact, one can already imagine how existing MEV extraction might be extended to a probabilistic setting. Currently, a sandwich attack both frontruns and backruns its target transaction; by excluding any other intervening transactions in between the two halves of the sandwich, the sandwicher minimizes risk (of, e.g., the price declining before they can sell off inventory). However, this requires the precise placement of two separate transactions, each with an associated swap fee (0.3% in the case of Uniswap). Recall that in a liquidity pool with two paired assets, A and B, trades are ‘symmetric,’ in that buying A is roughly equivalent to selling B, and vice versa. Consider the following sequence of events with two separate sandwichable target transactions:

  • Sandwicher purchases A
  • Target transaction #1 purchases A
  • Sandwicher swaps A for B
  • Target transaction #2 purchases B
  • Sandwicher sells B

In the above example, the sandwicher only pays a swap fee three times, but if the sandwicher were to sandwich both target transactions separately, they would have to pay a swap fee four times. Because the swap fee is 0.3% of the entire trade size, being able to take on some inventory risk between target transactions #1 and #2 can result in meaningfully higher profits (trading additional variance in their returns distribution for higher expected value). However, the sandwicher must take care to appropriately manage their inventory risk; for example, if target transaction #1 was actually an arbitrage which took advantage of a below-market price for A, then it is less likely that the sandwicher would be able to profitably exit on a sandwich in the opposite direction (i.e., the sandwich trades themselves may carry information about future price movement). Depending on their particular setup, a probabilistic MEV searcher might also impose bounds on position sizes across all open trades to guard against excessive exposure to any single asset’s idiosyncratic risks.

One final form of probabilistic MEV arises in the consideration of MEV across different domains, as discussed in Obadia et al. (2021), Unity is Strength: A Formalization of Cross-Domain Maximal Extractable Value. For searchers which are not themselves block producers across all of the relevant domains, extraction of cross-domain MEV (e.g., arbitrage across two different blockchains) will necessarily involve some degree of uncertainty about the relative order or confirmation status of their transactions. For instance, one could imagine a purchase on one blockchain confirming while an offsetting sale on another blockchain failing to process, leaving the searcher holding inventory, potentially at a loss. Nevertheless, those searchers which are able to competently manage these risks will be able to take good advantage of MEV extraction in an increasingly cross-chain world. (It is valuable to note, however, that the profitability of cross-chain MEV may be a significant driver of overall cryptoeconomic centralization, as the agglomeration of multiple blockchains’ or bridges’ validators, nodes, and miners under the umbrella of a single searcher will allow for supremely efficient, low-risk extraction of both inter-chain and intra-chain MEV.)

A word of caution to the would-be probabilistic MEV searcher: because financial markets are antagonistic, executing on probabilistic MEV may expose the searcher to a very large potential attack surface, depending on the difficulty of deriving the underlying strategy from on-chain data. One can imagine being able to “bait” an algorithm into certain unfavorable trades, much like how “poisonous” liquidity pools are often deployed in the hopes that an overly naive liquidity sniper will purchase their (unsellable) tokens.

The tremendous complexity of MEV is evident from the discussion above, which out of necessity only cursorily touches upon the main points of the situation. However, there is a great deal of space for more detailed studies of MEV, such as:


  • More comprehensive, cross-chain, quantitative analyses of MEV extraction
  • Theoretical and empirical studies of probabilistic MEV and its similarities or differences compared to HFT in traditional finances
  • Application of more sophisticated auction mechanisms to capture and distribute MEV to different ecosystem actors

Future work in these directions is eagerly anticipated.

I will close this article with a brief personal digression of my own views on MEV. Although this is perhaps verging on being meaninglessly abstract or speculative, I have come to believe that the ‘fight’ over MEV — over its extraction, beneficiaries, and mitigation — is a perfect, microcosmic illustration of how cryptoeconomic networks are intrinsically shaped by competitive forces that unceasingly drive the evolution of superior technology. Consider, for example, how the sophistication of transaction running has directly motivated the development of different blockchain architectures and protocols which seek to capture more value for users by internalizing what would otherwise be frontrunners’ profits. In the same way, the general adversarial nature of cryptoeconomic systems, where their permissionless and open nature allows any competent operator to extract value from flaws, forces these systems to prioritize security and robustness from the very beginning.

This is a highly desirable quality for infrastructure which may one day underlie the growth of a new financial system. Compare, for instance, the bulky and unwieldy technology of traditional banks, characterized by broken websites, outdated SMS authentication practices, susceptibility to social engineering, and in general innumerable attack vectors rife for exploitation which are shoddily patched up as an afterthought after sufficiently large amounts of money have been lost. This is the end result of a system which is intrinsically ill-adapted to the brutally adversarial nature of a globalized world. Although it may take time for cryptoeconomic systems to mature, they will be far more durable in part because the only survivors are those who will have successfully adapted to adversarial environments from the first blocks of their genesis.

For these reasons, and more, I find MEV irresistibly fascinating. It is a beautiful game of technical sophistication and intellect, and as actors redouble their efforts time and time again, the entire cryptoeconomic ecosystem ultimately becomes stronger for it. To be able to watch the unfolding of this drama over the coming years is a great privilege.

July 30th, 2022 | Posted in Crypto

Leave a Reply