Prediction Market Resolution— A Risk Analysis
Market description | Resolution source
Peer to peer prediction markets are resolved through an oracle that provides data about the final outcome of a certain market. This kind of resolution mechanism is needed for a decentralized prediction market platform to exist and maintain the level of trustless functionality that its open nature requires.
The purpose of this post is to analyze the intrinsic risk framework of the resolution system for Augur prediction markets specifically.
The term human oracles is used to refer to token networks where outcomes are decided by token holders weighing their vote for a certain outcome or data point. They are essentially a mechanism to achieve consensus. This is how Augur’s oracle works (also similar to MakerDAO’s price feed oracles for DAI).
Augur prediction markets are resolved by Augur’s own oracle system. Dispute rounds happen between token holders as they stake REP on the outcome they believe is true for a market. If there’s divided opinions about the outcome of a certain event, as can happen when a prediction market has not been worded precisely or some unexpected event brings uncertainty into the process, the oracle mechanism allows for a fight for consensus to happen. The amount of REP needed to stake grows in consecutive dispute rounds, and the conflict escalates until a decision on one outcome is reached, or a fork of Augur universes occurs. It’s important to note that while Augur’s interface does show a “Resolution source” field, this doesn’t interact with the smart contracts directly (although doing so will lead to disputes, a reporter technically can report in dissonance with the specified source).
To provide various examples of companies building products that currently interact with Augur’s oracle, we’ll look at Guesser and Veil. Both teams create some of the markets listed on their product, and unforeseen issues might arise around their resolution.
What does it mean for the oracle to resolve a market to a wrong outcome? A simple answer could be that it resolves to a lie or to a false statement, rather than to the truth. But because prediction markets are usually worded by people, they can have issues. Humans can be imprecise; prediction markets can be imprecise.
On Veil, the recent market “Will GRIN/USD be listed on CoinMarketCap by March 16, 2019?” mentions “GRIN/USD”, what some understood as not just Grin being listed on CMC, but the coin having a USD pair mentioned in its markets. It seems the team’s intention when creating the market referred to Grin being listed with a USD price for it, like most coins have on CMC, an issue that’s causing debate in the trader community.
This type of misunderstandings can happen, and might cause disputes and oracle delays for as long as humans define and stand as the resolution source for markets. We must then analyze and try to minimize the risk involved in the resolution process, which can lead to consequences such as:
- Long delays for the market to be resolved and the funds unlocked.
- Markets resolving to a wrong outcome.
- Augur’s network forking.
Needless to say, no stake holder in the Augur ecosystem benefits from any of the above happening rather than a market successfully resolving.
Market Description Risk
We can see the risk of resolution disputes in this model lies on the definition of the market’s description. Not only in the title, but also in their explanation of what should happen in the case of unforeseen scenarios (a website that acted as the selected data source being down, for example), or the choice of resolution source to guide reporters.
Some ways to de-risk in this type of network:
- Being highly precise when defining market details.
- Fair creators can own a significant stake in REP, and therefore a significant weight in dispute votes.
- Making the resolution source/reporter more precise (see below) to inherently improve 1.
A popular type of prediction markets are those whose resolution source is directly tied to a data point. Common examples are Augur markets about the price of an asset, which usually resolve to “the last price reported by…”, citing as the source a reputable API or website maintained by a company whose business depends a lot on the reputation around the quality and consistency of that data.
As mentioned above, in Augur this type of resolution source is indicative: the oracle does not directly access the API, it’s really a guidance for the reporter as well as making traders have precise information about what data the market will resolve to. Other oracle projects are working on directly connecting smart contracts to these APIs, so in a near future we could see these oracle smart contracts act as reporters for the Augur network in a combined mechanism.
Whatever the mechanism, directly connecting to data for prediction market resolution has a couple of advantages. The first one is that since the resolution source is precise and clear, like “the USD price of the last trade recorded on Coinbase PRO”, and usually refers to reputable data sources or APIs, the definition of the market is inherently better. As mentioned above, this makes the risk of resolution disputes lower. Relying directly on data can also improve the speed of some parts of the reporting process (this depends on the architecture of each oracle system).
But what if a certain API is hacked? Or the company shuts down and stops providing that API before the market ends? One could argue that to lower that risk, more than one data source should be provided. In that case, a market could resolve to the average value provided by 5 different sources, for example. But if one source is tampered with, the average value would be tampered with as well. To avoid outliers, the resolution source could provide a range of values in which the data will be accepted. So what should that range be?
On Guesser, the market on “What will the Total Value Locked in DeFi be on Monday March 11, 15:00:00 UTC, according to defipulse.com” considered that DeFi Pulse syncs data at 13:00:00 UTC, so we provided two hours in between. We did not notice Daylight saving time started on Sunday March 10, so if we had defined a smaller time gap and the site’s data synced following the old schedule, we could have faced a mismatch.
Resolution Source Risk
The risk in this model is lowered on the market description, as this is inherently improved by providing a data point as resolution. Thus, in this type of markets the risk of outcome disputes lies directly on the specification of the resolution source.
Some ways to de-risk can be:
- Provide various resolution data points, either aggregated or as backup in case of unavailability/tampering of the first choice.
- Use data sources/APIs whose businesses’ reputation depends on quality data.
The data model has its risks as well, they just lie on a different part of the system. In a protocol like Augur’s, where even if the resolution source leads to a certain data point the reporting is done by token holders, this type of market can encounter issues as well.
Also, some markets are based on topics whose outcome is not generally available data (i.e. ”Will Andreas Antonopoulos say the word “technology” in his ETHDenver speech?”), so we shouldn’t apply a one-size-fits-all approach.
For market creators the decision about which of these two types (or others that might work) to choose when creating a prediction market is essentially a risk evaluation. The decision comes down to which model’s risk you consider lower for the type of market in question. Then, when creating the market, you must work on de-risking either the market description or market resolution source as much as possible.
For market participants the analysis is not very different, it’s just seen from a different point of view. Whether a market will resolve without issues or not is important for your funds, so this risk framework is something to keep in mind when placing forecasts.
Finally, market reporters participating in disputes in the oracle network are important stakeholders in this. The token holder community might verbally or through disputes express a preference for one risk framework or another, and market creators and participants should take that into account when evaluating the risk linked to a market’s resolution.