Abstract: A prediction market is a useful means of aggregating information about a future event. To function, the market needs a trusted entity who will verify the true outcome in the end. Motivated by the recent introduction of decentralized prediction markets, we introduce a mechanism that allows for the outcome to be determined by the votes of a group of arbiters who may themselves hold stakes in the market. Despite the potential conflict of interest, we derive conditions under which we can incentivize arbiters to vote truthfully by using funds raised from market fees to implement a peer prediction mechanism. Finally, we investigate what parameter values could be used in a real-world implementation of our mechanism.
Introduction
Prediction markets are powerful tools for aggregating information, but they rely on a "ground truth" to pay out winners. In centralized markets, a trusted entity (like a company or a judge) provides this truth. However, in the rising world of decentralized prediction markets (e.g., Augur or Gnosis), there is no central authority to settle bets.
This creates three major hurdles for decentralized systems:
- Outcome Manipulation: If the outcome is decided by a vote, and the voters (arbiters) also hold shares in the market, they have a blatant conflict of interest.
- Ambiguity: Real-world events are often messy. A binary "Yes/No" settlement can be unfair if the event was poorly defined or the outcome was ambiguous.
- Incentive Alignment: There is a need to pay arbiters enough to ensure truthful reporting without requiring an outside subsidy.
Mechanism Design
The proposed solution is a two-stage mechanism that integrates market activity with a robust arbitration process funded by trading fees.
Non-Binary Payoffs
Instead of a winner-take-all $0 or $1 payout, the security pays out the fraction of arbiters who report the event occurred. This "smooths" the incentive structure; a single arbiter can only change the payout by $1/m$ (where $m$ is the number of arbiters), making manipulation significantly less profitable than in a binary "swing vote" scenario.
The 1/prior with Midpoint Mechanism
Standard "peer prediction" rewards agents for agreeing with others based on their private signals. This paper adapts the "1/prior" mechanism by using a "midpoint" between potential posterior beliefs. This ensures that the incentive to report truthfully is symmetric, regardless of whether the arbiter's signal was "Yes" or "No."
Funding via Trading Fees
The mechanism charges a fee on every trade (as shown in Figure 1, step 1b). These fees are pooled and used to pay the arbiters for their work in the second stage. By assuming a maximum budget ($B$) that any single arbiter can spend in the market, the authors calculate the exact truth-telling reward needed to outweigh the potential profit from a $B$-sized lie.
Theoretical Analysis
Several key insights emerge from the mathematical model:
- The "Discontinuity" Problem: By making the payout continuous based on vote share, the authors drastically lower the "bribe" needed to keep arbiters honest compared to binary markets where one vote can flip the entire payout.
- Independence from Arbiter Count: Surprisingly, the required trading fee does not depend on the number of arbiters ($m$). While more arbiters mean each one has less influence, it also means the fee revenue is split more ways; these two effects perfectly cancel out.
- Update Strength ($\delta$): The most critical variable is the clarity of the question. If arbiters are highly likely to see the same signal (high $\delta$), it is much cheaper to incentivize them. Subjective or confusing questions (low $\delta$) require higher fees.
Results and Calibration
While this is a theoretical mechanism design paper, the authors use mathematical proofs and a "Parameter Calibration" section based on the Logarithmic Market Scoring Rule (LMSR) to demonstrate feasibility.
Theorem 1 & 2 derive the exact conditions under which truthful reporting is a best response for arbiters, even in the "worst-case" scenario where an arbiter spends their entire budget to manipulate the market. Calibration shows that for realistic market parameters, a trading fee of roughly 5% is sufficient to fully fund the arbitration process and guarantee a truthful Nash equilibrium. For example, if a single arbiter represents 0.1% of the total market spend, a 4% fee can subsidize truthful reporting for a clear event.
Discussion and Conclusion
While the mechanism provides a strong foundation for decentralized settlement, some challenges remain for future work:
- Uninformative Equilibria: Peer prediction mechanisms can suffer from "collusion" equilibria where everyone agrees to vote the same way regardless of the truth.
- Risk Neutrality: The current model assumes agents are risk-neutral, whereas real-world traders may be risk-averse.
- Reputation Systems: Integrating these one-shot mathematical guarantees with long-term reputation systems (like those used in Augur) could further strengthen the mechanism's resilience.
- Entry Models: The paper notes that if traders can enter the market multiple times to "chip away" at the price, the required fees to maintain incentives are higher than in a single-entry model.
Summary updated