Combinatorial Arbitrage in Prediction Markets

summary

The gist

As a researcher who demands absolute precision, I have meticulously analyzed both provided texts.

In short

This research developed a framework to find and quantify arbitrage across related prediction market contracts by proving that prices outside the convex hull of possible payoffs admit arbitrage. By recovering 1.9 million logical relations using metadata and AI analysis, the study simulated trades against Polymarket order books. Findings show that while complex logical relations lead to longer opportunities, high discovery costs reduce observed participation.

Key concepts

Convex Hull
This is a geometric shape formed by connecting all possible combinations of outcomes for related contracts. The paper proves that if the market price falls outside this 'hull,' it means an arbitrage opportunity exists because the price is mathematically inconsistent with the underlying logical relationships between those contracts.
Integer Programming
This is a mathematical technique used to find the absolute best solution when dealing with complex constraints, like finding an optimal portfolio. It was used here to determine which specific combination of contracts maximizes guaranteed profit given all the trading rules and market limitations.
Logical Relations Recovery
The process of automatically discovering how different prediction markets are connected through their resolution criteria. The researchers used venue metadata and semantic analysis (LLMs) to recover 1.9 million such relations, which is crucial for identifying arbitrage opportunities that span multiple contracts simultaneously.

Terminology used across episodes

This episode discusses

The paper

Combinatorial Arbitrage in Prediction Markets · Read on arXiv

Hongjian Shi, Daojing Zhai

Technical University of Munich · Baruch College, City University of New York

We develop a framework to identify and quantify arbitrage across logically related prediction-market contracts. We prove that prices outside the convex hull of admissible payoff vectors admit arbitrage and identify optimal arbitrage portfolios through integer programming. Using venue metadata and semantic analysis of resolution criteria, we recover 1.9 million logical relations and compile them into constraint systems covering more than 255,000 contracts. We replay full-depth Polymarket order books to simulate an arbitrageur executing these portfolios against resting orders. Among settled portfolios, an ideal arbitrageur that takes the full displayed depth of every opportunity and reuses its capital at resolution earns a risk-free return of 3.5% a week after spreads and fees. We find that arbitrage opportunities persist longer and attract less observed participation where relation discovery and execution are more costly.

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Combinatorial Arbitrage in Prediction Markets".

Warren: Here's the gist of the paper. As a researcher who demands absolute precision, I have meticulously analyzed both provided texts.

Ruth: The first text (A) is a high-level executive summary of a research finding, while the second text (B) provides an extremely granular breakdown of the methodology and simulation process.

Warren: To create an accurate, detailed summary worthy of publication—or at least rigorous internal review—I must synthesize these two perspectives, ensuring no critical detail is lost.

Ruth: Here is the comprehensive, detailed summary: Comprehensive Research Summary: Arbitrage Quantification in Prediction Markets via Multi-Layered Relation Discovery and Simulation This research presents a novel framework designed to identify and quantify arbitrage opportunities across logically related prediction market contracts.

Paper summary: Ruth: The core claim is that prices outside that convex hull admit arbitrage, and they use integer programming to find the portfolio configuration that yields the largest guaranteed profit under those constraints. This means there's a mathematical way to define exactly how much you can make before you start losing money on related contracts.

Warren: It builds on the idea of logical relations tying things together—like how different prediction contracts might all resolve on the same underlying event—and then they use those relations to create constraint systems for integer programming.

Iris: The paper details this by breaking it into three parts: first, the mathematical characterization of arbitrage, second, recovering those logical relations from venue metadata and resolution text using structured parsers and large language models.

Ruth: That semantic layer is where they claim they recover one point nine million logical relations across over two hundred fifty-five thousand contracts by analyzing the text of contract resolution criteria <ref:2610.23517#pg1,recover 1.9 million logical relations>. It's a massive discovery effort just to build that constraint system.

Warren: And then they test it by replaying full-depth Polymarket order books to simulate an arbitrageur executing these portfolios against resting orders, which is where they get their empirical results about executable size and profit.

Iris: The simulation process involves detecting windows using best quotes and valid constraints, simulating trades against the full depth of the books, and then settling them based on contract resolutions. They also look at how much profit is retained after an order delay or if only part of the depth is filled.

Ruth: That’s a lot of complexity—from finding one point nine million relations to running simulations on billions of order book messages—but they are showing that this unified framework allows them to measure the executable size and profit for these combinatorial arbitrage opportunities accurately <ref:2610.23517#pg1>.

Warren: The key result here is that they found four hundred ninety-one thousand eight hundred twenty-seven arbitrage windows where a portfolio has a positive guaranteed profit at the best quotes, and an ideal arbitrageur could trade in about one hundred twenty-five thousand of them <ref:2610.23517#pg3,positive guaranteed profit at the best quotes, and>.

Iris: And the paper also mentions that venue-asserted and venue-derived relations contribute thirty-four percent and fifty-nine percent of the guaranteed profit respectively, while semantic-inferred relations are responsible for the remaining seven percent <ref:2610.23517#pg3,venue-asserted and venue-derived relations contribute 34% and 59% of the>.

Ruth: So basically, they're showing a way to quantify exactly how much profit is generated by each type of logical link in these prediction markets.

Warren: It’s a detailed map of the opportunities and the mechanism that defines when you can actually claim that profit under those specific conditions.

Conclusion: Ruth: The authors of this paper are Hongjian Shi and Daojing Zhai, and they are showing that you can move beyond looking at individual contracts in isolation and see how they interact through logical relations.

Warren: They're suggesting that the implication is that for market participants, understanding these hidden connections is key because it reveals potential guaranteed profits across multiple contracts.

Iris: The real change for someone listening to us is realizing that the existence of these combinatorial arbitrage opportunities isn't just a theoretical concept; they are quantifiable windows you can trade if you have the right tools to find them.

Ruth: It's not just theory, but identifying those four hundred ninety-one thousand eight hundred twenty-seven windows where a portfolio has positive guaranteed profit at the best quotes, which is something that changes how people think about market structure <ref:2610.23517#pg3,a portfolio has positive guaranteed profit at the best quotes>.

Warren: The takeaway is that as long as the execution costs don't completely erode those profits, there are these structures you can exploit for consistent weekly returns.

Iris: The main thing to watch for is how much the discovery cost of finding those semantic relations eats into the potential profit, because that layer is clearly harder to find and attracts less observed participation.

More episodes

← Home