The Lab · The math · updated 2026-07-28

Why baseball favorites only win 54%

The number surprises everyone who sees it. It isn't the market getting beaten by variance — it's the market being right about a sport that is closer to a coin flip than anyone wants to admit.

Across every game our board has graded, the favorite has won 54.7% of the time.

People react to that number two ways. Either the books must be bad at baseball, or variance is so wild that nobody can predict anything. Both are wrong, and the real answer is more useful than either.

The market said 55.9%. Favorites delivered 54.7%.

Start with the part that isn't in dispute. Those same games carried an average de-vigged market probability of 55.9%. The favorites won 54.7%. A gap of 1.2 percentage points across 311 games.

That gap is statistically invisible. It sits about four-tenths of a standard error from zero — the kind of miss a perfectly calibrated model produces constantly by chance. To even measure whether a 1.2-point bias is real, you'd need somewhere north of twenty thousand games. We have three hundred.

So the books didn't miss. They said "these teams win about 56% of the time," and they won about 56% of the time.

The favorite is barely a favorite

Here's the part that actually explains the number, and it's almost embarrassingly simple: most baseball games don't have a real favorite.

Sorted by how confident the market was, our graded games land like this:

Two things jump out. First, every band with a real sample landed inside or within a whisker of its own range — that's what calibration looks like when nobody is cheating. Second, and more to the point: half of all games sit in the 50–55% bucket. Four out of five are under 60%.

That's the whole mystery solved. The average favorite wins 54.7% because the average "favorite" is a team the market gives a coin flip and a nudge. You're not averaging a pile of dominant teams. You're averaging a pile of near-even games with a handful of genuine mismatches sprinkled in.

Why baseball is built this way

Compare typical single-game favorite win rates across sports:

Same sportsbooks. Same sharp bettors. Wildly different rates. Baseball isn't priced worse — it's structurally closer to even, for reasons baked into the sport:

• The best player is capped. An ace throws about six innings, then hands the game to a bullpen much closer to league average. A dominant pitcher simply cannot own a game the way a dominant basketball player owns forty minutes.
• The talent band is narrow. Excellent baseball teams win around 60% of their games; bad ones win around 40%. In basketball that spread runs from 80% down to 20%.
• Scoring is low and lumpy. One bloop single or one error decides a 3–2 game. Fewer scoring events means less room for the better team to prove it.
• Even the at-bat is nearly a coin flip. An elite hitter succeeds around 30% of the time, a poor one around 20%. Ten points of separation, compounding very slowly over nine innings.

The clinching evidence is the schedule itself. Baseball plays 162 games precisely because one game tells you so little. Football settles its standings in 17. If a single baseball game reliably revealed the better team, the season wouldn't need to be ten times longer.

The trap: accurate is not the same as profitable

This is where the 54% number gets genuinely dangerous, because it's easy to read it as "favorites win more than half, so bet favorites."

Favorites cost more. A favorite's price already demands a win rate:

Read the −150 row twice. The favorite needs to win 60% for you to break even. It wins about 55%. That five-point hole is not a bad beat or a cold streak — it is the arithmetic of the bet, every single time you make it.

Which is why a model can be right and still lose money. Being accurate gets you to the market's number. Profit requires being more accurate than the market, by more than the fee it charges to play. Those are enormously different achievements, and almost every betting pitch you'll ever read quietly swaps one for the other.

What this doesn't tell you

Three honest limits.

Our 311 games are a real sample but not a large one — enough to say the market is roughly calibrated, nowhere near enough to detect a one-point bias, and useless for slicing. Two of our confidence buckets hold 4 games and 1 game; they read 25% and 100%, and both numbers are pure noise that we publish rather than hide.

The cross-sport rates above are approximate industry ranges, not measurements from our own board — we grade MLB, so treat those as orientation rather than evidence.

And none of this says favorites are bad bets or underdogs are good ones. It says the price already knows. The only question worth asking about any bet is whether the price is wrong, and the answer is usually no.

Model output is informational and entertainment content, not betting or financial advice. If you bet, bet what you can afford to lose.

← Back to The Lab · How the model works · The record