Batter total bases: both sides lose

Batter total bases is the first market measured where neither side comes back near fair. Overs realized about -6.9% and unders about -3.5%, both outside their noise floor.

Written by Jesse, NegativeEV. Last updated 1 August 2026.

• Sample: 19,320 over props and 18,950 unders, from 1,089 games over 88 slates
• Window: 2026-04-19 to 2026-07-31
• Priced at: best price across books on each side, at the consensus point
• Measured: 2026-08-01

By side	Priced at	Hit	Realized EV
Over	49.4%	46.1%	-6.9% (z = -8.1)
Under	55.4%	53.6%	-3.5% (z = -4.6)

The mean consensus line is 1.08, so the market is effectively the 0.5 and 1.5 rungs. Both rungs run in the same direction on both sides.

By line	Realized EV
0.5 over (n = 8,249)	-6.2%
0.5 under (n = 8,248)	-4.7%
1.5 over (n = 11,024)	-7.2%
1.5 under (n = 10,659)	-2.6%

Most of that cost is the price of the market rather than a wrong side. The average book's two-sided margin on these quotes was 6.99%, and taking the best available number on each side across books it was still 5.10%. The hold is the book's margin; the realized figures are what a graded bet actually returned. They are different numbers, and one side lands worse than the margin alone would explain while the other lands better.

The over is negative in all four months of the window (-3.5, -7.7, -5.4, -8.0) and on both odds snapshots read, so the result is not one hot stretch or an artifact of when prices were captured. An under was archived at the same consensus point on 98% of props, so the under figure is graded on real quotes rather than inferred from the over.

Method

Every archived batter total-bases prop is graded against the batter's real total bases in that game from the mapped play-by-play corpus (1B + 2x2B + 3x3B + 4xHR). For each prop the consensus point is taken, then the best over price and the best under price across books at that point - deliberately the most favorable framing available, so the figures are the floor of the cost rather than the ceiling. Two odds snapshots are read and de-duplicated, any snapshot taken after first pitch is dropped, and doubleheaders are resolved by game id order. Standard errors are reported iid, game-clustered and slate-clustered; all three land at 0.7-0.8 points. Batters with no plate appearance in that game and games missing from the corpus are excluded, never defaulted.

Inside the noise floor

Two figures from this census are deliberately not stated as results. The under's margin over what the hold alone would predict is inside its noise floor, so the publishable statement about the under is its realized EV and nothing more. The 2.5 line graded on about 45 props and is far too small to read in either direction.

Written up for readers: https://negativeev.com/about/player-props All findings: https://negativeev.com/research