title: "Prediction-Market Factor Research: Polymarket"
slug: polymarket-factors
summary: Cross-sectional factor research on 4,000 Polymarket markets over 574 days, built around removing the five mechanical artifacts that dominate prediction-market backtests.
research_question: Do behavioral biases leave predictable cross-sectional structure in prediction-market prices once clipping, return-base, skip-day, early-settlement, and stale-quote artifacts are removed?
why_it_matters: Naive prediction-market backtests overstate predictability by an order of magnitude because prices are bounded in [0, 1], shorting YES means buying NO at 1 - p, and early resolution creates survival bias.
research_fields:
  - Asset pricing
  - Prediction markets
  - Market microstructure
project_type: empirical-research
status: public-research-snapshot
authors:
  - Repository owner and maintainer
original_source: PolymarketFactors workspace project
original_sample_period: 2024-09-01 through 2026-03-28 (4,000 markets x 574 days, including resolved markets).
updated_sample_period: Factor signs set on 2024-10 to 2025-06; out-of-sample evaluation across 54 five-day periods from 2025-06-28 through 2026-03-20.
data_sources:
  - Polymarket Users dataset (Akey, Gregoire, Harvie, and Martineau 2026, CC BY 4.0)
  - Polymarket-v1 on-chain OrderFilled and CTF database (Qin and Yang 2026, CC BY 4.0)
  - Polymarket Gamma and CLOB public APIs
data_license: Raw market archives are not redistributed; scripts read the public CC BY 4.0 academic datasets and public APIs. Published summary tables and figures are included under reports/.
code_license: MIT
reproduction_command: make reproduce-sample && make test
expected_runtime: Deterministic verification and unit tests run in seconds; full 4,000-market panel recomputation takes about 5 minutes locally.
outputs:
  - src/polymarket_factors/
  - reports/
  - data/
  - docs/
  - tests/
  - portfolio/
github_url: https://github.com/YangXiaoShawn/open-economic-quant/tree/main/PolymarketFactors
site_url: https://yangxiaoshawn.github.io/projects/polymarket-factors/
dataset_url: https://huggingface.co/datasets/ShawnChamberlain/open-economic-quant-research-data/tree/main/PolymarketFactors
space_url: https://huggingface.co/spaces/ShawnChamberlain/open-economic-quant-research-observatory?project=polymarket-factors
last_updated: 2026-09-30
limitations: Backtest association, not live execution. Capacity is roughly $100-200 per position; daily-close fills omit intraday slippage; the 50/50 combined-strategy weight was chosen in sample and is not validated out of sample.
catalog:
  field: prediction-markets
  accent: amber
  tags:
    - Prediction Markets
    - Factor Investing
    - Backtest Hygiene
  metric: 5 artifacts removed (-0.44 to -0.054 IC)