Price Evolution in a Continuous Double Auction Prediction Market With a Scoring-Rule Based Market Maker

Mithun Chakraborty, Sanmay Das, Justin PeabodyAAAI 2015 · DOI

Abstract

This paper explores the intersection of automated market making and traditional exchange mechanisms in the context of prediction markets. By introducing a new way to integrate the popular Logarithmic Market Scoring Rule (LMSR) into a Continuous Double Auction (CDA), we provide insights into how market makers influence price discovery, liquidity, and trader welfare. We propose an "Integrated" (INT) market maker that participates in a limit order book as a regular trader. Simulations show that while this model tightens spreads and increases trader surplus, it surprisingly slows the speed of price convergence to the true value.

Introduction

Most prediction markets rely on automated market makers (AMMs) to provide liquidity, particularly in thin markets. The Logarithmic Market Scoring Rule (LMSR) is the standard for these AMMs, designed for "dealer" markets. However, modern exchanges typically use a Continuous Double Auction (CDA) with a limit order book. Bridging these two mechanisms has historically required "Parallel" (PAR) models where the market maker has special system-level privileges. This study introduces the Integrated (INT) market maker, which operates transparently within the standard CDA framework.

The Integrated Market Maker Mechanism

The INT agent calculates its own bid and ask prices based on the current state of the limit order book and places its own limit orders. Unlike the PAR model, which requires peeking at incoming orders, the INT model is transparent and easy to implement. As the book state changes through trades or order cancellations, the INT agent dynamically updates its positions.

Interaction cases between the market maker and the limit order book.

The mechanism handles various market states, ensuring that the market maker's liquidity is always reflected in the order book, thereby improving market depth and transparency for all participants.

Experimental Methodology

To evaluate the INT model, we simulated four market microstructures:

  1. pureCDA: A standard auction without a market maker.
  2. PAR (Parallel): Hanson’s model where the MM and order book coexist with special rules.
  3. INT (Integrated): The proposed model where the MM acts as a standard trader.
  4. pureLMSR: A pure dealer market with no order book.

The simulations utilized Zero-Intelligence (ZI) traders—agents that place random orders within a range dictated by a noisy private signal of the "true" value. This setup isolates the effects of the market mechanism itself. We ran 1,000 simulations per scenario, varying the precision of trader beliefs ($N_{trials} = 20, 40, 100$) over 500 trading episodes.

Results and Discussion

Liquidity and Spreads

A primary goal of introducing a market maker is to improve liquidity. The INT model effectively reduced the bid-ask spread compared to a pureCDA. Interestingly, the PAR model exhibited significantly higher spreads. This is because the PAR model does not place orders on the book until a trade occurs, making its liquidity "invisible" and resulting in a wider effective spread for incoming traders.

Comparison of bid-ask spreads across different market mechanisms and information levels.

Price Discovery and Efficiency

A key finding is the trade-off between liquidity and informational efficiency. While one might expect a market maker to accelerate price discovery, the results show that adding an LMSR-based market maker to a CDA actually slows down convergence. The market maker acts as a buffer, absorbing trades that would otherwise move the price more aggressively toward the consensus value. Consequently, pureLMSR and pureCDA converged faster than the hybrid INT and PAR models.

Trader Welfare

Both market-making models significantly increased "Trader Surplus"—the difference between a trader's valuation and the execution price. This confirms that the market maker effectively subsidizes the trading population. Furthermore, the INT model generally incurred lower losses for the market operator than the PAR model, suggesting it is a more sustainable choice for real-world implementation.

Conclusion

The Integrated market maker offers a transparent and effective way to provide liquidity in CDA-based prediction markets. While it improves trader welfare and reduces spreads, market designers must be mindful of the impact on price discovery speed. Future work should explore these dynamics with more sophisticated agent models that can react strategically to the market maker's presence.

Summary updated

BibTeX
@inproceedings{chakraborty2015price,
  author = {Mithun Chakraborty and Sanmay Das and Justin Peabody},
  title = {{Price Evolution in a Continuous Double Auction Prediction Market With a Scoring-Rule Based Market Maker}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2015},
  doi = {10.1609/aaai.v29i1.9313},
}

Loading related content...

Figure