Paper: arXiv 2610.06412
Authors: Philipp D. Dubach
Abstract
Trade-sign errors can change measured trading costs even when classification accuracy is high. We validate the side of Polymarket’s public trade prints against the taker leg of each print’s on-chain settlement. On twelve selected days between April and August 2026, spanning both exchange generations, 92.1% to 100.0% of prints match a settled taker leg, with exact side and token agreement on all 24.5 million matched pairs. Mint-and-merge settlement makes that taker leg essential: pooling maker and taker legs changes the measured buy share. Signing the full cached tape before settlement selection yields 16.6 million prints signed by every rule; equal-day balanced accuracy is 0.942 for Lee-Ready, 0.768 for the tick test and 0.670 for retrospective bulk volume classification. On 16.4 million identical eligible fills, Lee-Ready raises effective spread by 0.429 cents per share under equal-fill weights; its realised-spread difference is -0.209 cents. Under share-volume weights, taker five-minute midpoint impact is 0.689 cents, while tick and bulk classifications give -0.374 and -0.119 cents. That aggregate sign reversal disappears when tied print rows are excluded. Distortion depends on error-weighted signed outcomes, sample selection and weighting. Receipt-time ordering and unobserved future-quote age limit these delivered-quote accounting quantities; they do not identify causal impact or private information. Supporting venue and collector analyses provide descriptive diagnostics.
Complexity vs Empirical Score
- Math Complexity: 3.0/10
- Empirical Rigor: 8.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: This paper demonstrates strong empirical rigor through extensive data validation and statistical analysis of trade direction. While the mathematical complexity is moderate, the novel application of blockchain ground truth to a critical microstructure problem is highly valuable.
Research Flowchart
flowchart TD
A["Research Goal: Measure Trade Direction in a Prediction Market"] --> B{"Key Methodology Steps"};
B --> C["Validate Polymarket Trade Prints vs. On-Chain Settlement"];
C --> D["Apply Trade Classification Algorithms (Lee-Ready, Tick Test, Bulk Volume)"];
D --> E["Data/Inputs: Polymarket Public Trade Prints, On-Chain Settlement Data, Cached Tapes"];
E --> F{"Computational Processes: Matching, Classification, Spread & Impact Calculation, Error Analysis"};
F --> G["Key Findings: High Print-Settlement Match Rate, Trade-Sign Errors Impact Trading Costs, Algorithm Performance Differs"];
G --> H["Outcomes: Quantified Impact of Sign Errors on Effective Spread and Midpoint Impact, Importance of Settlement Ground Truth"];
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 measures trade sign accuracy and its impact on trading‑cost metrics in Polymarket.
Source sentence
This paper measures the trade sign and makes no forecast with it.
Assumes 3
- Mint‑and‑merge settlement makes the taker leg essential for sign measurement. market
Source sentence
Mint-and-merge settlement makes that taker leg essential: pooling maker and taker legs changes the measured buy share.
- Receipt‑time ordering does not identify the execution‑time venue book. execution
Source sentence
Receipt-time ordering does not identify the execution-time venue book, and a five-minute signed midpoint change is not an identified causal price impact or measure of private information.
- The twelve selected days are not a random sample but were chosen to span venue changes. statistical
Source sentence
The twelve days were selected to span the sample and venue changes rather than sampled at random.
Data
- feed archive order-book public
Source sentence
The order-book data are a tick-level archive of Polymarket’s public WebSocket market channel, captured by a third party, PMXT [PMXT, 2026], in two formats: the provider’s first format from 21 February to 15 April 2026 and its second format from 15 April to 10 August, 171 UTC days in all with the seam day present in bot
Costs
Unclear - sources disagree
Source sentence (the authors' words)
On identical eligible fills, Lee–Ready raises effective spread by 0.429 cents per share under equal-fill weights; its realised-spread difference is -0.209 cents.
Finds 5
- Between 92.1% and 100.0% of prints match a settled taker leg, with side and token agreement on all 24.5 million matched pairs. statistical
Source sentence
On twelve selected days between April and August, including two V1‑exchange days, 92.1 % to 100.0 % of prints match a settled taker leg under their transaction hash. Side and token agree on every one of the 24.5 million matched pairs.
- Lee‑Ready achieves a balanced accuracy of 0.942 on the common scored sample. statistical
Source sentence
Lee–Ready on the quote strictly before the print recovers it with balanced accuracy of 0.942 on the common scored sample.
- Lee‑Ready adds 0.429 cents per share to effective spread under equal‑fill weights. performance
Source sentence
On identical eligible fills, Lee–Ready raises effective spread by 0.429 cents per share under equal-fill weights
- Realised‑spread effect of Lee‑Ready differs in sign between equal‑fill and share‑volume weighting schemes. performance
Source sentence
Its realised‑spread effect differs in sign between equal‑fill and share‑volume weights.
- Tick and bulk‑volume classifications can reverse the sign of volume‑weighted five‑minute midpoint impact, but this reversal disappears when tied scored rows are excluded. other
Source sentence
Tick and BVC can reverse the sign of volume‑weighted five‑minute midpoint impact relative to the taker, although this aggregate reversal disappears when tied scored rows are excluded.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 32 related papers). Coverage is partial, so this is neither replication nor refutation.
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 (45)
- The Anatomy of a Decentralized Prediction Market: Microstructure Evidence from the Polymarket Order Book — → 2604.24366
- OpenMarket: A Synchronized Polymarket–Binance Dataset for High-Frequency Prediction-Market Research (2026) — link
- Settlement Manipulation in Prediction Markets (2026) — link
- Volatility in Prediction Markets: A Structural Approach (2026) — link
- SoK: Market Microstructure for Decentralized Prediction Markets ( DePMs ) — → 2510.15612
- The Anatomy of a Blockchain Prediction Market: Polymarket in the 2024 U.S. Presidential Election (2026) — link
- Crypto Wash Trading (2023) — link
- Inferring Trade Direction from Intraday Data (1991) — link
and 37 more
- Prediction Markets (2004) — link
- Interpreting the Predictions of Prediction Markets (2006) — link
- Do Prediction Markets Produce Well-Calibrated Probability Forecasts? (2013) — link
- The Price Impact of Order Book Events (2014) — link
- Queue Imbalance as a One-Tick-Ahead Price Predictor in a Limit Order Book (2016) — link
- The Accuracy of Trade Classification Rules: Evidence from Nasdaq (2000) — link
- On the Occurrence and Consequences of Inaccurate Trade Classification (2000) — link
- Short Sales, Long Sales, and the Lee–Ready Trade Classification Algorithm Revisited (2012) — link
- Discerning Information from Trade Data (2016) — link
- Minimum Price Variations, Discrete Bid–Ask Spreads, and Quotation Sizes (1994) — link
- Eighths, Sixteenths, and Market Depth: Changes in Tick Size and Liquidity Provision on the NYSE (2000) — link
- Trade Execution Costs and Market Quality after Decimalization (2003) — link
- Difference-in-Differences with Multiple Time Periods (2021) — link
- Estimating Dynamic Treatment Effects in Event Studies with Heterogeneous Treatment Effects (2021) — link
- Difference-in-Differences with Variation in Treatment Timing (2021) — link
- Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects (2020) — link
- The Advantages of the Matthews Correlation Coefficient ( MCC ) over F1 Score and Accuracy in Binary Classification Evaluation (2020) — link
- The Ghosts of Polymarket : When Off-Chain Matches Meet On-Chain Reverts (2026) — link
- Polymarket -v1 Database (2026) — link
- Network-Based Detection of Wash Trading (2025) — link
- Price Discovery and Trading in Modern Prediction Markets (2026) — link
- Prediction Market Accuracy: Crowd Wisdom or Informed Minority? (2026) — link
- Executable Arbitrage and Market Efficiency in Prediction Markets (2026) — link
- Inferring Trade Directions in Fast Markets (2022) — link
- Evaluating Trade Classification Algorithms: Bulk Volume Classification versus the Tick Rule and the Lee – Ready Algorithm (2015) — link
- The Accuracy of the Tick Rule in the Bitcoin Market (2021) — link
- U.S. Tick Size Pilot (2019) — link
- Tick Size Wars: The Market Quality Effects of Pricing Grid Competition (2023) — link
- Automation, Speed, and Stock Market Quality: The NYSE ’s Hybrid (2011) — link
- Latency, Liquidity and Price Discovery (2012) — link
- Revisiting Event-Study Designs: Robust and Efficient Estimation (2024) — link
- How Much Should We Trust Staggered Difference-in-Differences Estimates? (2022) — link
- Market channel: \texttt tick_size_change message (2025)
- Polymarket Orderbook Archive (v2): data overview (2026)
- Commission Delegated Regulation (EU) 2017/588 of 14 July 2016 supplementing Directive 2014/65/EU of the European Parliament and of the Council with regard to regulatory technical standards on the tick size regime for shares, depositary receipts and exchange-traded funds (2017)
- Continuous Auctions and Insider Trading (1985) — link
- Replication package: Measuring Trade Direction in a Prediction Market (Settlement Ground Truth and Trading-Cost Measurement on Polymarket) (2026) — link
2 of 45 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2601 linked papers.