The power of prediction: an experiment in vote buying


This week, early cryptocurrency adopter, Shayne Coplan, created several Augur-based prediction markets that test the limits of on-chain governance processes, adding extra-protocol incentives in order to alter themechanism designs developed by protocol architects.

While still in their infancy, these market experiments may eventually threaten the value proposition of on-chain voting as efficient coordination infrastructure, transforming governance processes from ostensibly democratic demonstrations into purely plutocratic procedures.  

Stopping Sybil

Designing effective governance mechanisms in permissionless networks is inherently difficult.

In networks where any one person can create multiple identities at zero cost, the democratic protocol of ‘one man one vote’ loses its value. Exercises in gauging community sentiment through polling are ultimately futile as anonymous, adversarial parties can easily participate at scale, thus manipulating the outcome to their advantage.

With ‘one man one vote’ impossible to enforce, what coordination mechanisms are afforded to permissionless blockchains when deciding whether to pursue novel features and upgrades? Consider a scenario like theDAO hack of 2016 – what is the optimal method to gauge community sentiment regarding a recovery of funds?  

Several ‘next generation’ networks have proposed formal on-chain governance as the panacea to this coordination problem. Opting for a ‘one coin one vote’ system, on-chain governance mechanisms claim to serve as a single, transparent platform for decision making, thereby eliminating the inefficiencies and imprecision of governance by ‘rough consensus’.

On the surface, on-chain governance sounds pretty sensible — what better way to gauge sentiment than to literally quantify it through stake? If $10m votes ‘yes’ and $20m votes ‘no’ on Proposal X, it should be resoundingly clear that the ‘no’ outcome is preferable, and the protocol should proceed along those lines.

However, where on-chain governance makes up for coordination, it lacks in ‘fairness’. By conflating sentiment with stake, these networks suffer from a Tyranny of the Wealthy, with outcomes continuously resolving in the direction favoured by their most capitalized stakeholders.   

In many ways this is a feature rather than a bug – proponents of on-chain governance will argue that those with the most ‘skin-in-the-game’ should legitimately have outsized influence over protocol direction: they stand to win and lose the most, and thus should be rationally incentivized to act in the best interests of the network.

Fighting economics is futile

Ultimately, this argument fails to acknowledge the various ways in which majority stake holders can abuse their dominance to profitable ends, even when taking ‘skin-in-the-game’ into account.

Consider, for example, the implementation of on-chain governance in EOS. With a fixed set of 21 block producers, stakeholders can use their coins to vote for their preferred delegates. These delegates then partake in the block production process and, in return, receive block rewards, currently calculated as 1% of yearly inflation. In theory, the voting process is designed as a means to hold block producers accountable: if they fail to perform their duties, stakeholders can vote them out of their jobs.

In practice, these delegates are also large stakeholders, who can collude and thereby guarantee their permanence as block producers. As a result, the ‘checks and balances’ process is rendered redundant: block producers can feasibly act maliciously with no apparent risk of usurpation.

Or consider exchanges, which custody the vast majority coins on behalf of users. An exchange operator can leverage customer deposits to vote on proposals. However, because exchanges directly profit from volatility, they may be incentivized to vote in a way that induces community contention, thus driving customers to buy and sell coins at a rapid pace and, as a result, netting outsized fees from the ensuing chaos. One might presume the market will demand that exchanges don’t vote on behalf of users, but there is nothing legally preventing exchanges from pursuing this strategy, and, even in the face of this behaviour, users will likely continue using exchanges as long as they offer competitive liquidity.

Finally, consider a proposal to allocate some percentage of block rewards to a group of developers, who promise to fulfill some function for the network. Indeed, such proposals are the norm in Decred’sPoliteia system.


Let’s assume that these bounties are worth $1 million. An enterprising group of non-developers should be willing to bribe stakeholders anywhere up to $1 million in order to vote them in as recipients of the grant. Assuming that stake ownership is relatively distributed, this should not be particularly difficult to do: voting processes traditionally see low turnout, with most stakeholders apathetic as to their ability to influence outcomes. Consequently, they should be receptive to a bribe, which allows them to monetize their otherwise insignificant stake at no perceived marginal downside.  

The power of prediction

Although proponents of on-chain governance might acknowledge the theoretical strengths of these attack vectors, they may be reluctant to wholeheartedly accept their existence without witnessing them in practice.

Enter Coplan, who is attempting to prove that extra-protocol incentives can impact the behaviour of on-chain governance participants through the creation of the following Augur markets:   

‘Will ARK delegate “arkade_delegate” be one of the top 51 ranked ARK delegates for at least 3 consecutive days before the end of February?’

Will a Decred Politeia proposal enter the “Active Voting” stage before February 2nd, 2019?

For those less familiar with the mechanics of prediction markets, the value of an outcome share in Augur represents the market’s estimated probability of that outcome and the price of shares will fluctuate as new information becomes available.

Let’s take the ARK market as an example.

The last ‘Yes’ share was purchased for 0.01 ETH, indicating the purchaser’s belief that there is currently just a 1% chance that arkade_delegate ranks in the top 51 ARK delegates by the end of February. In selling that ‘Yes’ share for 0.01 ETH, the market maker is taking on a ‘No’ share for themselves, which is worth 0.99 ETH, indicating a 99% chance that arkade_delegate will not rank in the top 51 delegates. If the market resolves in favour of ‘yes’, owners of ‘yes’ shares buying at 0.01 ETH stand to gain 0.99 ETH, or 99x their initial investment.

Coplan “deliberately created markets where those capable of influencing the outcome, and thus collecting the bounty, have low opportunity and switching cost.” As a result, it’s possible that a bounty of relatively insignificant value “could prompt activity of exponentially larger sums of value.”

Let’s return to the ARK market.

Like EOS, ARK employs a Delegated Proof of Stake consensus algorithm. In order to participate in the consensus process, or, in ARK terminology, ‘forge blocks’, ARK delegates must be in the top 51 of all delegates by vote. For every block ‘forged’, a delegate receives 2 ARK, and a block is ‘forged’ every 8 seconds.


Arkade_delegate (A_D) is an ARK delegate vying for a position in the coveted 51 list. With just 143,227.72 ARK staked in their favour, A_D is currently ranked 58th. The 51st-ranked delegate currently has 1,434,852.53 total votes in their favour, meaning 1,291,624.82 ARK need to be delegated to A_D in order for them to join the ranks of the top 51. With ARK coins currently trading at $0.41, this interval amounts to roughly $530,000.

However, as Coplan’s market description makes clear, there is little opportunity cost, aside from possible commission differences and transaction fee costs, associated with switching votes to a different ARK delegate. As a result, the prediction market acts as a bounty for a large ARK holder to switch their votes to A_D. If this large ARK holder were to buy up shares in the ‘yes’ outcome or, alternatively, sell shares in the ‘no’ outcome, they stand to see sizable returns from their action.  

The question that arises from the emergence of these prediction markets is which agents are most likely to actually participate? Who stands to gain from participation, and what are the risks involved?

If, as a prediction market participant, you cannot manipulate delegate voting, then you do not have any asymmetric edge. You cannot add stake to the 51st ranked delegate in order to ensure their position is not usurped, and you cannot add stake to A_D in order to facilitate their rise to the 51st rank. Buying either ‘yes’ or ‘no’ shares is essentially gambling.  

However, there are some caveats at play:

First, ARK can be bought on various exchanges. Thus, the bounty may be sufficiently high to warrant the purchase of ARK, which can then be used to influence outcomes. This caveat has caveats of its own: if the market is aware of the prediction market, it should react by pricing up ARK, knowing that there will be a flood of extraneous demand from willing manipulators.

Second, because active delegates receive block rewards for their participation, A_D should be willing to sell shares in the ‘yes’ outcome at a favourable price, thus incentivizing ARK holders to buy and stake coins in their favour. However, the 51st ranked delegate, who has their position threatened by A_D, is similarly incentivized to ensure that they retain their spot. As such, they will likely start selling shares in the ‘no’ outcome. The winner will be the party that is willing to spend more capital on setting bounties in their favour.

This is slightly problematic as neither A_D nor the 51st ranked delegate has information on the extent of their foes’ resources and therefore cannot accurately price the risk they are taking and determine whether the reward is sufficient. In theory both A_D and the 51st ranked delegate should only be willing to spend slightly less than they expect to receive from future income, but this is difficult to model as you have to take into account both the future price of ARK and the probability of losing out on the 51st spot over time.   

However, there may be third parties outside the delegate pool willing to participate. In the context of a liquid borrowing market for ARK coins, one could feasibly borrow a large volume of ARK, using some percentage for the governance process and the remaining percentage to open a short position. This third party could then sell shares in the ‘yes’ outcome assuming that the market will determine this transparent vote buying process to negate the value of the ostensibly meritocratic and democratic on-chain governance process, and, resultantly, profit when ARK coins are priced below their pre-short value. Similarly, a competing protocol could allocate some percentage of their native block reward to set prediction market bounties, hoping to influence their competitor’s governance process in a way that destroys trust and, consequently, elevates their own project.

Of course, the existence of a liquid lending market opens up a set of distinct variables in itself, in that the fluctuation of borrow rates will affect the expected value of bets. One might presume that if rates are too high, a short strategy becomes less attractive, and if rates are low, shorting becomes more attractive. However, in the scenario where rates are high, honest agents that would normally participate in the governance process may instead choose to lend their coins, thus providing more ammunition for adversaries to manipulate the outcome in their favour. If the expected value from buying shares in the ‘yes’ outcome outweighs the lending rate, then borrowing at high interest might still be the net profitable strategy.

Implications for broader active participation

Coplan’s third market, ‘Will Livepeer’s (LPT) active participation rate reach 30% or above at any time before February 15th?’, extends the implications of derivatives beyond voting to broader active network participation.

The outcome is ultimately irrelevant – Coplan is ambivalent as to whether active participation breaches the 30% level. Rather, Coplan simply aims to initiate a discussion regarding the way in which prediction markets can be used to change the behaviour of protocol-specific agents: “current real world governance is immune to iterative experimentation, especially from outsiders: the opposite is true for any on-chain mechanisms, so we may as well experiment in creative ways.”  


Unlike the Ark proposal, where, excluding commission fees, switching costs for voters are non-existent, the Livepeer market must duel with multiple variables.

Livepeer’s inflation schedule is inextricably tied to the active participation rate. When 50% of LPT supply has been bonded, the per round inflation rate, with one round lasting roughly one day, will settle at an equilibrium of the then current per round inflation rate. Until then, the inflation rate increases by 0.0003% each round. Due to this additional inflation variable, ceteris paribus, LPT holders are more incentivized to actively participate than not actively participate. However, if the bounty for non-participation is large enough to mitigate the inflation rate, then LPT holders will suddenly be more economically incentivized to un-bond their tokens.

The Livepeer participation market is really just the tip of the iceberg. Near-identical markets could be created for Ethereum staking rates, with adversaries laying attractive odds to incentivize against Ether holders contributing to the network’s security. Although Ethereum’s Proof of Stake model, Casper, contains both a dynamic withdrawal delay based on real-time unbonding demand and a dynamic validator reward based on participation rate, it seems plausible that a motivated adversary could leverage a permanent or reoccurring prediction market to disincentivize staking from the outset. This same security-attack, which would instead target a low hash rate, should presumably be easier to launch against a Proof of Work system like Bitcoin as miners can turn off their hardware at will. However, as with the Ark markets, any adversaries will have to contend with opposition forces, who can similarly leverage prediction markets to provide additional incentives in favour of active participation.       

A worthy experiment

For now, Coplan will be seeding these markets himself, bootstrapping liquidity in order to boost participation in his experiment. While it is unlikely these initial iterations have any material impact, they do brilliantly demonstrate how prediction markets can be leveraged to influence outcomes and that outcomes can be determined by the most well-capitalized agents, forcing protocol designers implement disproportionately high switching costs at the expense of user experience when instituting formal on-chain governance and participation processes.

The post The power of prediction: an experiment in vote buying appeared first on The Block.

The power of prediction: an experiment in vote buying written by Matteo Leibowitz @ January 31, 2019 Matteo Leibowitz

Comments are closed.