Trading on a Rigged Game: Outcome Manipulation in Prediction Markets

Mithun Chakraborty, Sanmay DasIJCAI 2016

Abstract

This paper investigates the integrity of prediction markets when participants possess the power to influence the event's outcome. Using a game-theoretic model, we analyze how the probability of participation among "outcome-deciders" dictates market behavior. We find a critical "crossover probability" that separates two regimes: a High Participation Phase (HPP) where traders collude to manipulate results for profit, and a Low Participation Phase (LPP) where the risk of non-collusion forces traders to reveal private information, maintaining market informativeness.

Introduction

Prediction markets are designed to aggregate private information into a single price that reflects the likelihood of a future event. Standard economic models usually assume the outcome is "exogenous"—meaning the traders cannot change what actually happens. In the real world, this assumption often fails. A politician might trade on a policy outcome they have the power to vote on, or a corporate insider might bet on a project delay they can influence. This creates a "rigged game" scenario. This paper addresses whether we can still trust market prices to provide useful information if the participants are manipulating the outcome.

Methodology

The authors develop a two-stage game-theoretic model involving two agents, Alice and Bob.

  • Stage 1 (The Market): Alice and Bob receive private signals and trade sequentially in a market governed by a Market Scoring Rule (MSR). Crucially, there is a probability ($\pi$) that Bob might not participate in the market.
  • Stage 2 (The Voting): Both agents take an action (a "vote") that determines the final outcome. If Bob didn't trade in Stage 1, he is assumed to be "honest." If he did trade, he acts strategically to maximize his market payout.

The researchers use the Perfect Bayesian Equilibrium (PBE) solution concept to analyze three common scoring rules: the Logarithmic (LMSR), Quadratic (QMSR), and Spherical (SMSR) Market Scoring Rules.

The Participation Dichotomy

The market's behavior hinges on a "crossover probability" ($\pi_c$). This threshold separates the market into two distinct regimes based on the likelihood of Bob's participation.

Comparison of crossover probabilities across LMSR, QMSR, and SMSR scoring rules.
  • High Participation Phase (HPP): If Bob is very likely to trade, the market becomes a tool for collusion. Alice and Bob coordinate to move the price to an extreme and then vote together to ensure that outcome occurs, effectively extracting profit from the market maker.
  • Low Participation Phase (LPP): If there is a high enough chance Bob won't trade (and thus will vote honestly), Alice cannot risk colluding. To protect her own payout, her market trade ends up revealing her private information.

Key Insights and Results

The study reveals that the threat of non-participation forces strategic traders toward honesty. In the LPP regime, Alice’s private signal can actually be recovered by observing her price report, even though she is attempting to be strategic.

Alice's best response and expected vote, illustrating the transition from HPP to LPP.

The paper also quantifies how different MSRs distribute "manipulation profit." In a collusive HPP equilibrium under LMSR, Alice (the first mover) captures approximately 67.81% of the profit, while Bob receives 32.19%. Under QMSR, Alice's share increases to 75%.

Conclusion and Future Work

While the model is currently stylized, it provides a foundational understanding of how outcome manipulation affects prediction markets. It proves that market prices are only "correct" in the HPP regime because the outcome is rigged. In the LPP regime, prices reflect a "Bayesian estimate"—a weighted average of the traders' knowledge. Future research could extend this to larger player counts, non-binary outcomes, and scenarios where agents strategically choose when to trade.

Summary updated

BibTeX
@inproceedings{chakraborty2016trading,
  author = {Mithun Chakraborty and Sanmay Das},
  title = {{Trading on a Rigged Game: Outcome Manipulation in Prediction Markets}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2016},
}
Figure