Paper: arXiv 2508.03474
Authors: Oriol Saguillo, Vahid Ghafouri, Lucianna Kiffer, Guillermo Suarez-Tangil
Abstract
Polymarket is a prediction market platform where users can speculate on future events by trading shares tied to specific outcomes, known as conditions. Each market is associated with a set of one or more such conditions. To ensure proper market resolution, the condition set must be exhaustive – collectively accounting for all possible outcomes – and mutually exclusive – only one condition may resolve as true. Thus, the collective prices of all related outcomes should be $1, representing a combined probability of 1 of any outcome. Despite this design, Polymarket exhibits cases where dependent assets are mispriced, allowing for purchasing (or selling) a certain outcome for less than (or more than) $1, guaranteeing profit. This phenomenon, known as arbitrage, could enable sophisticated participants to exploit such inconsistencies. In this paper, we conduct an empirical arbitrage analysis on Polymarket data to answer three key questions: (Q1) What conditions give rise to arbitrage (Q2) Does arbitrage actually occur on Polymarket and (Q3) Has anyone exploited these opportunities. A major challenge in analyzing arbitrage between related markets lies in the scalability of comparisons across a large number of markets and conditions, with a naive analysis requiring \(O(2^{“n+m”})\) comparisons. To overcome this, we employ a heuristic-driven reduction strategy based on timeliness, topical similarity, and combinatorial relationships, further validated by expert input. Our study reveals two distinct forms of arbitrage on Polymarket: Market Rebalancing Arbitrage, which occurs within a single market or condition, and Combinatorial Arbitrage, which spans across multiple markets. We use on-chain historical order book data to analyze when these types of arbitrage opportunities have existed, and when they have been executed by users. We find a realized estimate of 40 million USD of profit extracted.
Complexity vs Empirical Score
- Math Complexity: 3.5/10
- Empirical Rigor: 6.0/10 (capped from 8.5: no out-of-sample test; see penalty flags)
- Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.5)
Why this score: The paper relies heavily on empirical data from on-chain order books and quantitative profit estimates (40M USD), but the mathematics is primarily combinatorial heuristics and probability concepts rather than advanced derivations, resulting in high empirical rigor with moderate math complexity.
Research Flowchart
flowchart TD
A["Research Goal<br>Identify & Analyze Arbitrage<br>in Polymarket"] --> B["Data & Preprocessing<br>On-Chain Order Book Data"]
B --> C["Methodology: Heuristic Reduction<br>Timeliness, Similarity, Combinatorics"]
C --> D["Computational Analysis<br>Scalable Arbitrage Detection"]
D --> E{"Arbitrage Types Found?"}
E --> F["Market Rebalancing Arbitrage"]
E --> G["Combinatorial Arbitrage"]
F --> H["Outcome: $40M+ Profit Extracted"]
G --> H
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper conducts an empirical arbitrage analysis on Polymarket data, identifying two distinct forms of arbitrage and estimating extracted profit.
Source sentence
In this paper, we conduct an empirical arbitrage analysis on Polymarket data to answer three key questions: (Q1) What conditions give rise to arbitrage? (Q2) Does arbitrage actually occur on Polymarket?, and (Q3) Has anyone exploited these opportunities?
Assumes 5
- Each market’s condition set is exhaustive and mutually exclusive, so the sum of YES token prices should equal 1. market
Source sentence
Each market is associated with a set of one or more such conditions. To ensure proper market resolution, the condition set must be exhaustive -- collectively accounting for all possible outcomes -- and mutually exclusive -- only one condition may resolve as true.
- Arbitrage profit is calculated without accounting for any trading fees. execution
Source sentence
It is important to clarify that we do not take into account the fees because Polymarket currently does not charge per trade executed.
- The LLM can correctly infer logical dependencies between market conditions. model
Source sentence
together with market description data to capture the state space of possible market resolutions to then infer semantic dependency between pairs of
- Reducing markets to at most four high‑volume conditions plus a catch‑all preserves the essential logical dependencies. data
Source sentence
We thus pre-process markets with more than 4 conditions and reduce them to 4 or fewer conditions with the most total traded volume (in YES, and NO tokens), and a fifth condition to catch all other outcomes.
Show 1 more
- A time window of 950 blocks (≈1 hour) captures the relevant bid matching delay for arbitrage execution. execution
Source sentence
We set T to 950 blocks (approx. 1 hour) to capture some delay in bids being matched (75% of bids fall within this window – see App. I.1 for the distribution).
Data
Polymarket prediction markets · crypto
- Polymarket historical bid data order-book public
Source sentence
Our data collection efforts span markets that were resolved over the period of one year, from April 1, 2024, to April 1, 2025.
- On‑chain historical order book data order-book public
Source sentence
We use on-chain historical order book data to analyze when these types of arbitrage opportunities have existed, and when they have been executed by users.
Twitter / X
Costs
Not modeled
Source sentence
It is important to clarify that we do not take into account the fees because Polymarket currently does not charge per trade executed.
Finds 5
- The authors estimate that about $40 million of profit was extracted from arbitrage on Polymarket. performance
Source sentence
We find a realized estimate of 40 million USD of profit extracted.
- Total profit summed across all detected arbitrage strategies amounts to $39,587,585.02 (assuming $1 profit per trade). performance
Source sentence
The combined amount extracted from all strategies presented totals $39,587,585.02, assuming an ε = $1 profit per trade.
- Out of 1,578 NegRisk markets, 662 had at least one arbitrage opportunity, averaging about 100 opportunities per market. statistical
Source sentence
In general, we see many more opportunities per market, averaging around 100 per market.
- Only about 1 % of estimated U.S. election arbitrage opportunities were actually exploited by users. performance
Source sentence
About 1% of our estimated U.S. election opportunities were exploited by users.
- The top arbitrageur generated $2,009,631.76 in profit. performance
Source sentence
The user with the highest profits generated a total of $2,009,631.76 through these strategies.
Checked by others
Qualified by
- Arbitrage Analysis in Polymarket NBA Markets
Compare the findings
This paper: The authors estimate that about $40 million of profit was extracted from arbitrage on Polymarket.
That paper: Capped profit from single‑market arbitrage totals $210.19 (median yield 11.0% per episode) while theoretical uncapped profit would be $4,418.44.
6 papers in the archive : ForesightFlow: An Information Leakage Score Framework for … ; Arbitrage Analysis in Polymarket NBA Markets ; SoK: Market Microstructure for Decentralized Prediction …
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (49)
- DeepSeek‑R1‑Distill‑Qwen‑32B (2025)
- Tweet on unusual trading activity (2024)
- Tweet on market dynamics and liquidity (2024)
- Linq‑Embed‑Mistral
- MTEB Leaderboard – Embedding Benchmark
- Journalists Wake Up to the Power of Prediction Markets (2022)
- Presidential Election Winner– 2024 Market (2024)
- Flash boys 2.0: Frontrunning in decentralized exchanges, miner extractable value, and consensus instability (2020)
and 41 more
- Non-atomic arbitrage in decentralized finance — → 2401.01622
- Quantifying blockchain extractable value: How dark is the forest? (2022)
- Prediction markets (2004)
- Polymarket: A Decentralized Information Markets Platform (2025)
- UMA DVM Oracle Whitepaper (2025)
- Attention is all you need (2017)
- Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning (2025) — link
- The Policy Analysis Market (and FutureMAP) Archive (2003) — link
- Accessing Polymarket Data in Python (2024) — link
- py-clob-client: Python client for Polymarket CLOB (2024)
- Linq-Embed-Mistral (2024)
- AI in the Gray: Exploring Moderation Policies in Dialogic Large Language Models vs. Human Answers in Controversial Topics (2023) — link
- APT-Pipe: A Prompt-Tuning Tool for Social Data Annotation using ChatGPT (2024) — link
- Cross-Partisan Interactions on Twitter (2025) — link
- Principles of forecasting: a handbook for researchers and practitioners (2001)
- Expert political judgment: How good is it? How can we know?-New edition (2017)
- How are markets created? (2025) — link
- CLOB Introduction (2025)
- A flash (bot) in the pan: measuring maximal extractable value in private pools (2022)
- Rolling in the shadows: Analyzing the extraction of mev across layer-2 rollups (2024)
- No Fish Is Too Big for Flash Boys! Frontrunning on DAG-based Blockchains (2024)
- Disputed Polymarket resolution on Ukraine rare earth deal (2024) — link
- FIDE World Blitz Chess Champion 2024 (2024) — link
- UMA Voter Analysis in Polymarket Disputes (2024) — link
- Public Node RPC Endpoint for Polygon PoS — link
- SoK: Attacks on DAOs (2024) — link
- Selling and Redeeming Shares (2022) — link
- SoK: MEV Countermeasures (2024)
- Ethereum’s Proposer-Builder Separation: Promises and Realities (2023)
- Automated market making and arbitrage profits in the presence of fees — → 2305.14604
- Pandora’s Box: Cross-Chain Arbitrages in the Realm of Blockchain Interoperability (2025) — link
- Who wins ethereum block building auctions and why? (2024) — link
- Are Longer Prompts Always Better? Prompt Selection in Large Language Models for Recommendation Systems (2024) — link
- LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression (2024)
- LegalBERT: A Unified and Comprehensive Pre-trained Language Model for Legal English (2023) — link
- Harnessing NLP for Indigenous Language Education: Fine-Tuning Large Language Models for Sentence Transformation (2025) — link
- Chain-of-thought prompting elicits reasoning in large language models (2022)
- Self-Consistency Improves Chain of Thought Reasoning in Language Models (2023) — link
- Tree of thoughts: deliberate problem solving with large language models (2023)
- Matic Network: Scalable and Instant Blockchain Transactions (2019) — link
- Conditional Tokens Documentation (2020) — link
2 of 49 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.