Augur is an open-source, decentralized prediction market where users can both create markets and vote on other market’s expected outcomes.
Augur is a prediction market platform built on Ethereum. In a prediction market, users can express their opinions on the outcomes of different issues by buying shares in a given outcome. By allowing participants to win, and lose, real money, prediction markets incentivize individuals to express their convictions on a given subject. Augur’s REP token is used to incentivize honest and timely reporting on prediction market outcomes; REP holders who participate as reporters can earn a portion of platform fees. The system operates on the principle that the ‘wisdom of the crowd’ is a superior forecasting tool than individual views. It was one of the first Ethereum-based ICOs, raising $5 million in October 2015, and launched a beta in 2016.
Augur was designed as a decentralized prediction market that no one entity controlled or managed. In a traditional prediction market, such as Hollywood Stock Exchange or Nadex, a company is responsible for opening markets, managing bets and funds, and resolving outcomes. Augur creators wanted to enable this functionality without having market participants trust the solvency and good intentions of a managing entity. Co-founded by Joey Krug and Jack Peterson, Augur has been under development since 2014, with both alpha (2015) and beta (2016) versions of the platform having been released on the Ethereum testnet. Open Zeppelin’s Zeppelin Solutions team completed security audits of Augur Core and the first official trading market is planned for early 2018, with the full platform slotted for staged deployment pending the initial market’s successful release.
Notable advisors for the team include Vitalik Buterin, founder of Ethereum, Elizabeth Stark, founder of Lightning, and Dr. Robin Hanson, an economist at George Mason who has published works on prediction markets.
Unique Technologies: Decentralized Prediction Markets
Prediction markets are held to generate accurate odds, informing users how likely it is that a given market’s outcome will occur. Augur is a platform where users can create and participate in prediction markets on Ethereum. Within the system of smart contracts, users wager on the outcome they believe will take place, winning money when correct and losing money when incorrect. At any given time, an outcome’s share price can be interpreted as an estimate of the probability of that event occuring. Prediction market advocates claim such estimates are accurate, or are superior to other forms of forecasting, such as experts, models, or polling.
At a high level, the life cycle of an Augur prediction market includes the following:
Creation: A market creator spends REP, Augur’s token, to create a market for a question, such as ‘Will the Spain win World Cup?’ The creator defines the possible outcomes that can be wagered on, such as ‘Spain wins, Spain loses, draw, or match cancelled’, and all questions include a default ‘invalid’ outcome. Market creators are incentivized to form markets based on well-defined events with unambiguous outcomes via a ‘validity bond’, a sum set aside by market creators at the market’s creation and forfeited should market reporters return an outcome of ‘invalid’. Additionally, the market creator picks a resolution source, such as ESPN.com, that reporters will treat as providing the truth about the event’s outcome.
The market creator also chooses a designated reporter, who first declares the event’s tentative outcome within three days of the event’s completion by staking REP to one of the market’s outcome. For expediency, a market will be resolved using the tentative outcome unless that outcome is subsequently successfully disputed by reporters–REP holders–during a seven day dispute period. A ‘no-show’ bond, funded by the market creator and forfeited on a tardy designated reporter response, incentives a responsible choice of the designated reporter.
Creators fees compensate market creators for their efforts and investment. Market creators set a fee between 0 and 50% of settled shares––Augur’s settlement fee, paid by market participants closing shares, is a combination of this creator determined ‘creator fee’ and a reporting fee that is paid to reporters and set by an automated Augur market. The creator fee can later be lowered, but cannot be raised; this provides market creators a response to undercutting markets.
Trading: Users buy and trade shares in a market’s outcomes. Markets can support Boolean, scalar and categorical outcomes, each having different associated sets of shares. Augur’s REP token is not used as a prefered currency in such wagering. Augur plans to use Ether at first, then integrate ‘stablecoins’, coins whose value is tied to fiat currency, for ease of use. Prediction markets built on volatile cryptocurrencies could produce skewed data, such as when participants’ decisions are influenced by their speculative beliefs about the future price of their token holdings and not only their local knowledge. Augur’s plans to integrate stablecoins should help mitigate this issue.
Forecasting: Forecast generation occurs simultaneously with trading. An outcome’s share price is interpreted as the odds of that event occuring. For instance, in a simplified win/lose scenario, a 20 cent share price for Spain losing is interpreted as the market’s judgment that a Spain loss has a 20% chance of occuring. Augur claims that the superiority of such estimate is justified in light of current research on prediction markets, citing studies such as ‘The Real Power of Artificial Markets’ (2001) and ‘Predictions Without Markets’ (2010).
Reporting: Part of what makes Augur a decentralized prediction market is how an event’s actual outcome is determined. Rather than have a central entity state whether Spain won or lost, Augur’s protocol relies on distributed reporters to verify what occured. REP holders’ collective participation in the process is designed to ensure the emerging outcome’s accuracy.
After the designated reporter reports the tentative outcome, there is a seven day dispute period where REP holders individually confirm or deny the report’s veracity by staking REP to a given outcome; by staking REP to the ‘Spain won’ outcome, a REP holder reports that ESPN did in fact state ‘Spain won.’ If sufficient REP is staked to an alternate outcome, ‘Spain lost’, the tentative outcome is disputed and the alternate outcome becomes the new tentative outcome, subject to a new dispute period. The market resolves using that tentative outcome which ultimately emerges from such cycles of proposal and dispute–at least, in normal circumstances.
For performing such verification services, REP holders become eligible for a portion of the platform’s trading fees, collected during 7-day long, consecutively occuring fee windows. At the end of the fee window, reporters who participated, such as by staking during an initial report or disputing a tentative outcome, are compensated from reporting fees pooled in such windows in proportion to amount of REP staked during that period. An automated Augur market determines the reporting fee, increasing or decreasing the default market fee of 1% in relation to the current market capitalization of REP. (The minimum reporting fee is 0.01% and the maximum is 33.3%.)
Forking: In special circumstances, market resolution involves a further process, so-called ‘forking.’ A ‘market fork’ occurs when REP holders disagreement over a market’s outcome reaches a certain threshold; if the size of the filled dispute outcome is greater than 2.5% of all REP, the market ‘forks.’
A market fork is a disruptive process whereby all non-resolved prediction markets are put on hold while REP holders determine the disputed market’s outcome. In the event of a fork, one ‘universe’ is generated for each possible market outcome; such universes are the ‘children’ of the generative ‘parent’ universe. REP holders then have the opportunity to resolve the market by migrating their REP tokens from parent to child universe. Such migration is permanent–tokens moved to a child universe cannot be transferred to any other child universe. Migrating REP to a given universe counts as ‘voting’ on that outcome; leaving one’s REP in the parent universe is to abstain.
The child universe receiving the most REP is the outcome that the market settles on; if a majority of REP migrates to ‘Spain lost’, that outcome is what the market uses to settle bets. Crucially, however, REP migrated to a ‘minority’ universe continues to exist and potentially has value, because each child universes can support its own subsequent prediction markets, which would be verified using that universe’s REP. The purpose of this is to give market creators and users the opportunity to vote with their feet; users could decide that majority outcome was incorrect, so doubt the competence or honesty of reporters who supported that resolution and take their business to a minority universe containing better reporters. Whether a given universe’s REP continues to have value thus depends on whether the universe contains active markets.
Market forks, by design, have significant usability consequences, which Augur hopes will make them function more as ‘nuclear option’ incentivizing honesty than as a general means of resolving market outcomes. Besides disrupting all markets for up to 60 days, the generating universe’s REP token would become locked, spawning at least three new tokens, one for each possible market outcome (minimally outcome A, outcome B, and the invalid outcome.) Exchanges and wallets would need to decide which of these new tokens to support. Depending on whether market creation and trading continue to occur in a universe, a spawned universe’s token could become worthless. Should trading continue in multiple universes, the available work that a token holders has a right to perform would be a fraction of what it previously was.
Settlement: When the market’s outcome is determined, winning shares become redeemable for the total staked currency and losing shares become worthless.
Augur’s token, REP, is used primarily to incentivize honest and timely reporting on prediction markets. It is not a prefered currency used in the actual prediction market, but rather grants holders the right to contribute to the prediction market’s resolution process. By performing this service, REP holders will receive a percentage of a given market’s trading fees, proportional to the quantity of REP tokens staked. Diligence in reporting is incentivized by this setup, as tokens staked to non-consensus outcomes are effectively lost––such REP are redistributed to users staking for the consensus outcome. In the optimal case, the sum value of work, performed exclusively by token holders through reporting, will grow with overall platform growth.
The process by which anonymous REP holders resolve markets is intended to add security to the platform. Rather than having a single entity determine a market’s outcome, Augur envisions thousand of reporters participating and, ideally, independently verifying outcomes, making the platform more resilient to mistakes and manipulation. The thesis is that REP holders who don’t want to lose their tokens should stake them to the outcome that they think other REP holders will also report; all things being equal, this should be the generally agreed upon truth that can resolve the market.
While Augur’s code is open-source, the Augur platform’s development is supported via bounties through The Forecast Foundation, a non-profit, research entity set up by Augur’s founders. The Forecast Foundation ran Augur’s Initial Coin Offering (ICO), and retained 4% (440,000 REP)––Augur’s founders retained 16% (1,760,000 REP). The team currently works fully remotely.
Game Theory Applications
Augur’s project is notable for its application of game theoretic incentives. The project employs various bond structures that work together to economically incentivize honesty in all parties. Some significant incentive mechanisms employed include:
Validity bonds: Validity bond incentivizes market creators to create markets based on events with objective, unambiguous outcomes. A validity bond is paid in ETH and is returned to the market creator if the market resolves to any outcome other than invalid.
No-Show Bonds: No-show bonds incentivizes market creators to choose a reliable designated reporter, improving markets resolution speed. A no-show bond is returned to the market creator if the designated reporter reports within the allotted 3-day window following the market event’s conclusion. Should the designated reporter fail to report within that window, the no-show bond
Dispute bonds: Dispute bonds incentivize reporters to dispute incorrectly reported tentative market outcomes. Participants noting that a tentative outcome is incorrect can stake REP to an alternative outcome, which disputes the tentative outcome if enough REP is staked. Dispute bond sizes are proportional to the total amount staked, determined in such a way as to ensure a 50% rate of investment for reporters who successfully dispute false outcomes
Forking: Market forks are at the heart of Augur’s incentive system. Augur maintains that, since its market forking process can be trusted to determine the truth, the threat of a fork incentives participants to behave honestly. The outcomes that wins in a fork should correspond to ‘objective reality’ when attackers are economically rational and the market cap of REP is large enough—Augur’s whitepaper offers detailed formal arguments for this conclusion. Forking itself has significant negative effects on the platform, such as a 60 day suspension of market activity and a splintering of the REP token, so stakeholders have strong reasons to resolve market outcomes without escalating to a fork.
Prevailing Concerns with Augur and Prediction Markets:
Decentralized Verification and Token Distribution
Augur, like other token economies, relies on sufficient token distribution to ensure the benefits of a decentralized platform. In Augur’s case, token holders are entitled to perform work that adds value to Augur’s decentralized platform, value lacked by centralized prediction markets’ verification processes––Augur’s reporters’ collective participation ensures the process’ integrity and the outcome’ accuracy.
Whether reporters’ work will add value in the ways Augur intends remains to be seen, and crucially depends on how many end up performing the labor entitled by REP holding and how such reporters actually go about resolving market outcomes, factors that are out of Augur’s control to an extent.
At first glance, the question of how Augur will ensure that REP holding remains sufficiently diversified looks pressing. Presuming that a REP token holder will perform the reporting service, the outcome resolution processes of a centralized prediction market and Augur become more homogeneous as more REP tokens are concentrated in the hands of a few individuals or organizations. Should a few entities end up holding most REP tokens, this would undermine Augur’s claim to a distributed resolution process.
Understood this way, the challenge for Augur is ensuring that REP tokens are widely distributed. Yet the presumption that the REP holder will perform the reporting service may be inaccurate. While the Augur white paper states that “Anyone who owns REP may participate in the reporting and disputing of outcomes”, it is unclear what would prevent REP owners from delegating that work to other individuals. The issue arising from delegation is that, if a thousand REP holders delegate the reporting task to the same entity, then the resolution process is not clearly disanalogous to that of a centralized prediction market So, if delegation is possible, then Augur’s challenge is less a matter controlling token distribution and more a matter of controlling labour distribution.
The Augur platform claims to be more secure because the process is decentralized. REP holders delegating work can undermine how decentralized the process is. Yet the process still could arguably be considered secure for other reasons: REP owners are incentivized to delegate an honest central entity because their staked tokens are at risk of being burned. In this case, Augur is secure because of the game theoretic incentives the platform builds into the reporting system and not because of the process’ decentralization.
Justification of Market Accuracy
A prediction market, like a poll or a survey, is a tool for gathering information. While theorists are generally optimistic about prediction market’s performance as compared to polls and other forms of forecasting, there are competing explanations of their apparent success and ongoing disputes on the extent of the differences. User looking to make good use of the tool that is a prediction markets would do well to research the precise sense in which market’s are accurate, the conditions under which they generate reliable forecasts, and the ways data might be manipulated.
Augur claims that an outcome’s share price can be interpreted as an estimate of the probability of an event occuring––a 20 cents share price tells users a Spain loss has a 20% chance of occuring. Augur appeals to the so-called ‘Wisdom of the Crowds’ principle, which holds that the average prediction made by a group is superior to that made by any of the group’s constituents. While economists have appealed to such principles to explain prediction market’s success, there are some reasons to doubt that the principle supports such interpretation of share price.
For instance, one might question whether share price really reflects an aggregated judgment attributable to ‘the crowd’, since the aggregation method doesn’t appear to guarantee proportional representation. As such, share price can reflect more the judgment of a few participants with deep pockets and less the judgment of the ‘outvoted’ majority.
The argument also depends on interpreting share price as an aggregate of individual predictions, yet the aggregation method doesn’t guarantee this either. Users can mistakenly buy shares, can stake for outcomes they can make or prevent from happening, and can stake to influence other users, all of which affects share price but none of which constitutes predicting. Gnosis, a rival project, illustrates this in one of its use cases. Gnosis suggests that market creators needing software testing can stake for ‘no exploits will be found’ in a market asking ‘will software exploits be found’, thus incentivizing debugging.
Finally, even assuming that a share’s price represents the group’s prediction, there’s a question of how to interpret this collective judgment’s content. A share price, 20 cents, could be interpreted as a judgment that ‘Spain has a 20% chance of losing’, but it can also be interpreted as ‘ESPN has a 20% chance of reporting a Spain loss’, among other things. The question then is how to support a given interpretation of what the crowd’s wisdom actually holds.