MLB moneylines, graded against real winners

Graded against the real winner at the best price on each side, 1,111 MLB games returned about -1.8% per bet, which on a sample that size is not far enough from zero to call a measured cost. Favorites came back -0.8% and underdogs -3.1%, and neither of those cuts separates from zero either.

Written by Jesse, NegativeEV. Last updated 2 August 2026.

• Sample: 1,111 MLB games, 2,222 side bets
• Window: 2026-04-19 to 2026-07-31
• Priced at: best price across books on each side
• Measured: 2026-08-02

How much did MLB moneylines return?

About -1.8% per bet across both sides, graded against the real winner rather than estimated from simulations.

Realized EV per bet	Figure
Every moneyline, both sides	-1.8% (± 2.1, z = -0.9)
Favorites only (n = 1,210)	-0.8% (± 2.5, z = -0.3)
Underdogs only (n = 1,012)	-3.1% (± 3.6, z = -0.9)
Mean per-book two-sided hold	3.9%

Why is the hold 3.9% when the realized figure is about -2%?

Because taking the best price on each side removes much of the margin, and because the two numbers answer different questions in the first place. About 4% is the book's two-sided margin in the quotes; taking the best price on each side strips a good part of that out, and what is left over these 1,111 games is around -2%. Calling the -2% "the vig" would conflate the two.

What did favorites and underdogs return?

Favorites -0.8% per bet and underdogs -3.1%, with neither cut far enough from zero on this sample to be read as a cost. The favorite-versus-underdog direction matches every larger census run since: the same longshot bias shows on ATP match odds across 35,346 matches and on the MLB prop board, where the longest prices are the worst priced.

Method

Every MLB moneyline held in the archived odds is graded against the real winner from the mapped play-by-play corpus, betting both sides at the best price across books, and reported as realized win rate and expected value per bet overall, favorites-only, underdogs-only and by price bucket. Games missing from the corpus are excluded rather than defaulted.

Inside the noise floor

None of the three cuts clears its own noise floor. On 1,111 games the overall figure is -1.8% ± 2.1 (z = -0.9), favorites are -0.8% ± 2.5 (z = -0.3) and underdogs are -3.1% ± 3.6 (z = -0.9), so every interval covers zero. The supported statement is that MLB moneylines at the best price are close enough to fair that a thousand games cannot separate them from fair, and that the direction of the favorite/underdog gap matches the larger censuses - not the decimals. The large-sample prop and ATP censuses are the sharper measurements.

What does this page not establish?

Whether any one moneyline on the board is priced well. These are averages over one fixed window, and a market that grades close to fair still holds prices that are not; nothing here says which. The decimals are estimates rather than measured costs, since every interval on this page covers zero, so what the census carries is a direction and a sample size. The window also starts on 19 April, where the archived odds start, so the opening weeks of the season sit outside it.

Written up for readers: https://negativeev.com/about/favorites-underdogs Background: picking a winner at a price All findings: https://negativeev.com/research