Do MLB Favorites Win Enough to Profit?
9,144 games · 2023 to 2026Baseball has the flattest moneylines in major American sport. The best team in the league is often barely a 180 favorite against the worst, where a football mismatch would be laying double digits. That makes it a clean place to ask a simple question: if you had backed every favorite, or every underdog, would you have made money? We took every MLB game from 2023 through 2026 that carried a closing moneyline, 9,144 in all, and measured it. The answer is no in both directions, and the way it fails is the part worth understanding.
Read the third number first. Backing both sides of the same game cannot win. You are buying the same event twice and paying the house margin on each ticket, so it has to come back at roughly minus the vig. It does, at -3.46%. That is not a finding about baseball. It is the proof that the measurement itself is sound. If it had landed near zero, or above it, the study would be broken and nothing else on this page would be worth reading.
Every bar below is a price range. The filled bar is how often favorites in that range actually won. The outlined bar is the win rate the price demanded just to break even. Favorites win most of their games, as you would expect. They simply do not win often enough to cover what you pay for them.
The same view from the other side. Underdogs lose most of their games and get paid more when they win. Across the whole sample the two effects very nearly cancel, which is what an efficient market looks like.
| Closing price | Games | Won | Needed to break even | Return per dollar | 95 percent range | Verdict |
|---|---|---|---|---|---|---|
| -100 to -119 | 1,718 | 50.81% | 53.13% | -4.33% | [-8.94%, +0.05%] | Not distinguishable from chance |
| -120 to -149 | 3,628 | 54.00% | 56.98% | -5.17% | [-8.07%, -2.31%] | Loses |
| -150 to -199 | 2,581 | 58.27% | 62.73% | -7.11% | [-10.15%, -4.09%] | Loses |
| -200 and heavier | 1,217 | 68.86% | 70.52% | -2.52% | [-6.32%, +1.13%] | Not distinguishable from chance |
| Every favorite | 9,144 | 56.58% | 59.68% | -5.21% |
Every one of the 9,144 games has a favorite, so this table covers the full sample with nothing left over. The heaviest range combines prices from 200 through 299 with everything past 300. On its own the past 300 group held only 126 games, too few to publish as a rate.
| Closing price | Games | Won | Needed to break even | Return per dollar | 95 percent range | Verdict |
|---|---|---|---|---|---|---|
| +100 to +119 | 3,017 | 46.57% | 47.99% | -3.01% | [-6.56%, +0.72%] | Not distinguishable from chance |
| +120 to +149 | 2,802 | 44.33% | 43.19% | +2.63% | [-1.51%, +6.90%] | Not distinguishable from chance |
| +150 to +199 | 1,555 | 38.14% | 37.47% | +1.82% | [-4.59%, +8.20%] | Not distinguishable from chance |
| +200 and longer | 543 | 24.49% | 30.62% | -19.85% | [-31.80%, -7.75%] | Loses |
| Near pick em, dog at a minus price | 1,227 | 48.66% | 51.17% | -4.91% | [-10.34%, +0.48%] | Not distinguishable from chance |
| Every underdog | 9,144 | 43.42% | 44.13% | -1.71% |
The final price row is the one most studies quietly drop. In 1,227 games the market was so close to even that the book priced both sides at a minus number, so the underdog does not sit at a plus price at all and cannot fall into any plus range. Rather than discard those games we give them their own row. The longest price range combines 200 through 299 with everything past 300, because the past 300 group held only 16 games, which is far too few to report as a rate.
Nine price ranges are published above. Only three of them can be told apart from chance at all, and every one of those three loses money.
The three reliable results are all losses. Favorites from 120 through 149 return -5.17%, favorites from 150 through 199 return -7.11%, and underdogs priced 200 and longer return -19.85%. In each case the 95 percent range sits entirely below zero, so the loss is real rather than noise. The longest shots being the worst priced is a well known market pattern called the favorite longshot bias, seen across many sports, and this data is a clean example of it.
Two ranges look like value. We are not calling them edges. Underdogs from 120 through 149 returned +2.63% and underdogs from 150 through 199 returned +1.82%. Both look tempting and both have a 95 percent range that crosses zero, at [-1.51%, +6.90%] and [-4.59%, +8.20%]. A range that crosses zero means the data cannot separate a small edge from no edge from a small loss. Publishing either as a system would be presenting noise as signal, which is the most common way betting content misleads people. We are not going to do it.
Put together, this is a market doing its job. The price already contains what the public knows about who is better, and the margin sits on top. Backing a whole category of team, whichever category you pick, mostly buys you the margin. Beating baseball is not a matter of picking a side and repeating it. It takes an actual edge on a specific game, and a price that has not caught up yet.
Three plain language guides that explain what the numbers above are doing.
- Sample. 9,144 completed MLB games from 2023 through 2026 that carried a closing moneyline. Regular season and postseason are both included. Seasons 2021 and 2022 are absent because no market data exists for them in our source.
- Closing price. The last price captured before first pitch, taken as the median across roughly 11 sportsbooks. The median is computed in decimal odds because American odds jump discontinuously either side of even money, which would distort a straight average of the raw numbers.
- Prices carry the margin. The return figures use real prices with the house margin still in them. A no vig price is fair by construction and would force every return toward zero, which would make the whole exercise circular.
- Live prices excluded. Every snapshot taken after first pitch is dropped. Those are in play prices, and they were the entire source of implausible numbers in the raw feed. Applying that one rule left every remaining price inside a normal range with no further filtering needed.
- Cross check. Removing the margin from our prices reproduced an independently maintained consensus probability to within 0.001 on average across 9,143 shared games.
- Ranges. Every figure marked as a 95 percent range comes from 5,000 bootstrap resamples of the games in that group.
- Home field. The 53.0% figure uses a wider sample of all 14,224 completed games from 2021 through 2026, since it needs no market data.
- Data through 17 August 2026.