Study 01 · R18 in the register · retired
The sportsbook close overprices the over on player props.
Restated 2026-09-24 after the settlement fix: was −1.40pp on 43,209.
Abstract
We asked: Does the sportsbook close overprice the over on player props, by more than it costs to bet the under? Across 44,198 settled NFL player props, over side, regular season (weeks 1-18), carrying a benchmark close, 2023–2025, the close priced the over at 0.4880 and it cleared 0.4637 of the time. At a game-block 95% interval the gap above is distinguishable from zero: the over is overpriced. By market, all 5 intervals sit below zero; by season, 2 of 3 intervals sit below zero, 1 includes zero.
The hypothesis, as the register states it
R18 · entered 2026-09-24 · script research/calibration.py
Does the sportsbook close overprice the over on player props, by more than it costs to bet the under?
Measure realized minus priced over rate; de-vigged DraftKings / FanDuel / BetMGM close; settled regular-season player props 2023-2025, over side; game-block bootstrap
Entered in the register 2026-09-24, after the earliest figure this study records (2026-09-10): the question was written down once a result had been seen, so this is not a pre-registration. No earlier registration and no hash are on file.
The register’s reading of it, verbatim · verdict retired
The close priced the over at 0.4880 and it cleared 0.4637 of the time; half the archive's median 1.0675 overround is 3.38c a side, which is what a sportsbook under costs. Restated 2026-09-24 from the 2026-09-10 figure (-1.40pp, n 43,209): the 2026-09-17 settlement fix now settles a player who played with no stat row at 0. Postseason games are excluded because their props are not yet settled correctly.
Method
The population is settled NFL player props, over side, regular season (weeks 1-18), carrying a benchmark close, seasons 2023, 2024, 2025. The price is the median de-vigged close across draftkings / fanduel / betmgm. The measure is realized minus priced on the over side alone: over and under are exact complements, so an error measured across both sides hides an asymmetry inside its own complement. Intervals are a 2,000-draw bootstrap resampling whole games (seed 29), so outcomes in the same game move together.
Result
It sits below the identity line in 11 of 12 buckets, and 7 are distinguishable from it at a game-block 95% interval (0.10–0.15, 0.30–0.35, 0.35–0.40, 0.40–0.45, 0.45–0.50, 0.50–0.55, 0.55–0.60), all below it. 2 bins under 30 outcomes are not counted.
Scores on the same outcomes, over side: expected calibration error 0.0244, Brier 0.2449.
By market
How this is computed
Realized minus priced, over side, by market, with a game-block 95% interval.
By season
How this is computed
By season, on the same books and the same settlement source.
By price
How this is computed
Realized minus priced, over side, by the close’s own over price in bins 0.05 wide. A bin under 30 outcomes is listed, not drawn: one outcome would set the scale.
Limits
- The price studied is the median de-vigged close across draftkings / fanduel / betmgm. No earlier price is in the file, so how the price moved on the way to the close is not measured here.
- 2,941 postseason props are excluded; the register gives the reason above. Seasons 2023, 2024, 2025 only.
- The file carries 5 markets (receiving yards, receptions, tackles assists, rush attempts, sacks); nothing here speaks for a market outside them, or for an exchange.
- The file publishes no interval for a difference between two markets, two seasons or two bins, so none is compared with another.
It is a finding about the market, not a strategy.
Cite as
Calibrated Sports (2026). Market calibration, R18: “Does the sportsbook close overprice the over on player props, by more than it costs to bet the under?”. calibratedsports.com/studies/market-calibration. Entered in the register 2026-09-24; figures computed 2026-09-26 by research/calibration.py --publish.