{ "schemaVersion": "vultax-public-research-data-v1", "articleUrl": "https://vultax.com/research/prediction-market-fair-value-calibration", "datasets": [ { "title": "A fresh check: 9,132 markets read a day before they closed", "columns": [ { "name": "Price a day before close", "unit": null }, { "name": "Markets", "unit": null }, { "name": "Events", "unit": null }, { "name": "Average price", "unit": null }, { "name": "Came true", "unit": null }, { "name": "95% interval (by event)", "unit": null } ], "rows": [ [ "under 2¢", "2,240", "760", "0.4¢", "0.6%", "0.3–1.0%" ], [ "2–10¢", "1,025", "622", "5.3¢", "6.9%", "5.2–8.6%" ], [ "10–30¢", "1,793", "1,320", "20.6¢", "28.1%", "26.1–30.1%" ], [ "30–50¢", "1,627", "1,269", "40.2¢", "42.7%", "40.7–44.9%" ], [ "50–70¢", "1,187", "1,165", "58.4¢", "58.1%", "55.3–60.9%" ], [ "70–90¢", "469", "437", "78.8¢", "79.3%", "75.7–82.9%" ], [ "90–98¢", "251", "227", "94.8¢", "98.8%", "97.3–100.0%" ], [ "98¢ and over", "540", "276", "99.5¢", "99.8%", "99.4–100.0%" ] ], "measurementContext": "Intervals resample whole events (2,000 draws), so the markets of one event move together. 102 markets had no price within two hours of the reading and are left out.", "sources": [ { "label": "Brier score (background)", "url": "https://en.wikipedia.org/wiki/Brier_score", "note": "The proper scoring rule used for validation." }, { "label": "Favorite-longshot bias (background)", "url": "https://en.wikipedia.org/wiki/Favourite-longshot_bias", "note": "The market-structure effect the calibration captures." }, { "label": "Kalshi Research — Prediction market calibration: how accurate is Kalshi? (20 Aug 2026)", "url": "https://news.kalshi.com/p/prediction-market-calibration-how-accurate-is-kalshi", "note": "The venue's own study over more than 2.2 million data points: forecasts improve with time, participation and volume; markets with about $10,000 or twenty traders were already informative; the comparison offered is superforecaster Brier scores of 0.10–0.15." }, { "label": "Krause — Calibration and Forecast Accuracy of Macroeconomic Prediction Markets: Evidence from Kalshi, 2025–2026 (SSRN)", "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7117919", "note": "594 Kalshi macro markets, 1,860 observations, nine series: overall Brier 0.0987; federal-funds-rate markets near-perfectly calibrated (0.0001); unemployment-rate markets the weakest (0.1302); one-hour horizon 0.0812 against 0.1587 at thirty days, as summarised in the abstract." }, { "label": "Gizmodo — Study finds around a quarter of Polymarket trades are fake (Columbia, Nov 2025)", "url": "https://gizmodo.com/study-finds-around-a-quarter-of-polymarket-trades-are-fake-2000683231", "note": "About 25% of Polymarket volume over three years flagged as wash trading by counterparty structure, peaking near 60% in December 2024; roughly one wallet in seven flagged; many flagged wallets made no profit." }, { "label": "CoinDesk — Polymarket's trading volume may be 25% fake, Columbia study finds", "url": "https://www.coindesk.com/markets/2025/11/07/polymarket-s-trading-volume-may-be-25-fake-columbia-study-finds", "note": "Sports markets carried the highest flagged share (about 45% of historical volume); election markets about 17%." }, { "label": "Polymarket — accuracy methodology", "url": "https://polymarket.com/accuracy", "note": "Read 6 October 2026. Used for forecast-horizon context, not as a directly comparable score or a new Vultax measurement." }, { "label": "Polymarket Gamma API: markets and events", "url": "https://docs.polymarket.com/developers/gamma-markets-api/overview", "note": "Resolved Yes/No markets scheduled to end 1 July to 30 September 2026, and every market of each football match; read 19 October 2026" }, { "label": "Polymarket CLOB: price history", "url": "https://docs.polymarket.com/developers/CLOB/timeseries", "note": "Yes price 24 hours before each market closed, hourly points within two hours of the reading" } ], "measurement": { "kind": null, "temporalCoverage": null, "population": null, "methodology": null }, "id": "fresh-check-2026-table", "kind": "published-table", "description": "Original Vultax research: in an experimental test, a calibrated fair-value model beat raw Polymarket prices by ~9.6% on held-out Brier score. Where the favorite-longshot effect lives, what the model can honestly flag, and why it still is not a trading edge.", "articleUrl": "https://vultax.com/research/prediction-market-fair-value-calibration", "sectionUrl": "https://vultax.com/research/prediction-market-fair-value-calibration#fresh-check-2026", "publishedAt": "2026-09-08", "modifiedAt": "2026-10-20", "version": "df97dfcda69a1a039008d01f5c29f323749f20db3aea6490641a4a94f1bbd964", "isLive": false, "limitations": "Published research snapshots and calculations; these exports do not contain current quotes or live account data. Observation windows and populations are stated in the source captions and methodology. Missing metadata is unknown, not zero. Vultax articles are market intelligence and education, not investment advice. This is a description of an experimental, backtested calibration result with explicit limitations; it is not a claim of live tradeable edge, and no market flagged by the model is a recommendation.", "downloads": { "json": "https://vultax.com/research/prediction-market-fair-value-calibration/data.json?dataset=fresh-check-2026-table", "csv": "https://vultax.com/research/prediction-market-fair-value-calibration/data.csv?dataset=fresh-check-2026-table" } } ] }