4,026 games. 3,887 moneyline bets. Every closing line, every final score. The market won — but not by as much as you’d think, and there’s one price band that keeps refusing to lose.
There’s a piece of folk wisdom that circulates every spring: baseball is the underdog sport. Any team can beat any team on any given day. A journeyman starter throws six shutout innings, a bullpen implodes, a 100-win team drops two of three in Pittsburgh. The sport has less predictive separation between good and bad teams than any other major American league, and the moneylines reflect it — a heavy MLB favorite is -200, where an NFL favorite can be -1400.
So the theory goes: back the dogs. You’ll lose more often than you win, but the prices are generous enough to carry you.
I wanted to know if that’s true, so I stopped theorizing and pulled the data. Every regular season game of 2025. Every 2026 game through July 27. Closing moneylines on both sides, final scores, no exceptions and no gaps. Then I ran the dumbest possible strategy through it: $100 flat on the underdog, every single game, no judgment applied.
Here’s what 4,026 baseball games have to say.
The Setup
Before the results, the plumbing — because a backtest is only worth what its data is worth.
Games. 2,430 games in the 2025 regular season (including the two Tokyo Series openers between the Dodgers and Cubs in March). 1,596 games in 2026 through July 27. Total: 4,026 games, 36,025 runs scored, an average of 8.95 runs per game. That number is a useful sanity check — it lines up with league-wide scoring, which tells you the game set isn’t missing chunks.
Odds. Closing moneylines on both sides of every game, pulled game-by-game from ESPN’s public odds feed. For 2025 that’s ESPN BET’s closing number. For 2026, ESPN switched its default provider, so those are DraftKings closes. Every one of the 4,026 games matched a closing price — no interpolation, no substitution from a different book to fill a hole.
Why closing lines matter. The close is the sharpest number a book will ever hang. It has absorbed every injury scratch, every lineup card, every dollar of steam. Beating closing lines is the standard test of whether a betting approach has any edge at all, and it’s a hard test — deliberately harder than betting into openers or midday numbers. Everything below is measured against the toughest available benchmark.
What counts as an underdog. The side with the longer price. Usually that’s a plus-money team, but in 220 games in 2025 and 304 in 2026, both sides were negative — a -105 / -115 market, say. In those, the -105 side is technically the dog, and I treated it as one.
Pick’ems get no bet. 109 games in 2025 and 30 in 2026 closed with identical prices on both sides (-110 / -110, mostly). There is no underdog in those games, so no bet was placed.
Payouts. Standard American odds. A +140 winner returns $140 profit. A -105 winner returns $95.24. Every loser costs the full $100. No pushes exist on an MLB moneyline.
Finding One: The Market Is Very, Very Good at This
2025, betting every underdog: 1,013–1,308. A loss of $2,190.28 on $232,100 wagered. ROI: -0.94%.
Sit with that number for a second, because it’s the most interesting result in this entire piece and it looks like nothing.
You placed 2,321 bets. You put nearly a quarter of a million dollars through the window. You applied zero skill, zero handicapping, zero situational awareness — you just mashed the dog button 2,321 times in a row. And you finished down less than one percent.
The reason that’s remarkable: the vig on a standard -110 / -110 market is about 4.55%. If underdogs had performed exactly as the market priced them, blind dog-betting should have bled roughly the juice. It didn’t. It bled a fifth of the juice.
Here’s the mechanism. Across those 2,321 bets, the average underdog closed at +114.3, which carries an implied win probability of 43.79%. The underdogs actually won 43.64% of the time.
Fifteen hundredths of a percentage point. Over 2,321 games, the aggregate closing line on MLB underdogs was wrong by fifteen hundredths of a percentage point.
That is not a market with an exploitable structural bias. That is a market that has solved the problem. And it’s worth naming clearly, because a lot of betting content is built on the premise that the books are lazy or the public is dumb enough to move lines off true. On the MLB moneyline, at the close, in aggregate — they’re not, and it isn’t.
The Ride Was Still Miserable
A -0.94% ROI sounds almost pleasant. Living it would not have been.
The bankroll bottomed at -$7,847 on May 8. Maximum drawdown from a running high: $7,998. Longest losing streak: 13 straight. The strategy was in profit on exactly one day all season — September 10, at +$212 — before fading back to -$2,190 by the final Sunday.
So the honest summary of 2025 isn’t “blind dog-betting is nearly break-even.” It’s “blind dog-betting is nearly break-even if you can survive a five-figure hole for five months on a strategy with no edge.” Almost nobody can. The variance eats the bankroll long before the expectancy has a chance to be nearly-fine.
Finding Two: The Obvious Fix Will Destroy You
Every bettor’s instinct, staring at that result, is the same one I had: the small dogs are just coin flips with juice. Bet the real dogs. That’s where the value is.
So I re-ran it with a single filter: only bet underdogs priced +150 or longer.
Result: 181–383. A loss of $5,895 on $56,400 wagered. ROI: -10.45%.
Cutting to the long dogs turned a $2,190 loss into a $5,895 loss on a quarter as many bets. The ROI got eleven times worse. The win rate collapsed from 43.6% to 32.1%. Longest losing streak stretched to 17.
And it got monotonically worse the longer the price:
| 2025 underdog price | Bets | Record | P&L | ROI |
|---|---|---|---|---|
| +150 to +199 | 382 | 140–242 | -$1,080 | -2.83% |
| +200 to +299 | 177 | 41–136 | -$4,315 | -24.38% |
| +300 or longer | 5 | 0–5 | -$500 | -100.00% |
Underdogs of +200 or longer went 41–141 and lost $4,815. That’s roughly a quarter of every dollar wagered, gone.
This is longshot bias — the oldest documented inefficiency in gambling markets, first measured in horse racing in the 1940s and confirmed in basically every betting market anyone has looked at since. Bettors systematically overpay for long prices because a big payout is more fun to imagine than a small one. Books know this and shade the long end accordingly. The bias survives because the appetite for lottery tickets is structural, not informational — no amount of public data makes people stop wanting the 12-to-1 shot.
The practical lesson: on MLB moneylines, “value” and “long price” are close to opposites. The longer the dog, the worse the deal, and the relationship is close to linear.
Finding Three: There Is a Band That Keeps Working
If the long dogs are poison and the near-coin-flips are grinding out the vig, what’s left?
The 2025 buckets, all of them:
| 2025 underdog price | Bets | Record | Win % | P&L | ROI |
|---|---|---|---|---|---|
| Both sides negative | 220 | 122–98 | 55.45% | +$1,814.72 | +8.25% |
| +100 to +119 | 750 | 351–399 | 46.80% | -$2,350.00 | -3.13% |
| +120 to +149 | 787 | 359–428 | 45.62% | +$4,240.00 | +5.39% |
| +150 to +199 | 382 | 140–242 | 36.65% | -$1,080.00 | -2.83% |
| +200 to +299 | 177 | 41–136 | 23.16% | -$4,315.00 | -24.38% |
| +300 or longer | 5 | 0–5 | 0.00% | -$500.00 | -100.00% |
A clean +5.39% ROI over 787 bets, sitting in the middle of the price distribution, flanked by losers on both sides.
Now — this is exactly the point in a betting article where you should get suspicious, and I want to be explicit about why. I did not go looking for this band. I found it by slicing 2025 six ways and noticing which slice was green. That is textbook data mining. If you cut a random walk into six pieces, one of them is going to look like a strategy. Finding a profitable bucket after the fact proves almost nothing.
The only thing that would mean anything is testing it on data I hadn’t looked at yet.
The Out-of-Sample Test: 2026
So I pulled the 2026 season — 1,596 games through July 27 — and applied the +120 to +149 rule with no adjustments.
Result: 175–224. A profit of $816 on $39,900 wagered. ROI: +2.05%.
It held. Barely, but it held. Here’s the full two-season comparison:
| Underdog price band | 2025 P&L | 2025 ROI | 2026 P&L | 2026 ROI |
|---|---|---|---|---|
| Both sides negative | +$1,814.72 | +8.25% | -$4,002.90 | -13.17% |
| +100 to +119 | -$2,350.00 | -3.13% | -$1,389.00 | -2.11% |
| +120 to +149 | +$4,240.00 | +5.39% | +$816.00 | +2.05% |
| +150 or longer | -$5,895.00 | -10.45% | -$589.00 | -2.86% |
| All underdogs | -$2,190.28 | -0.94% | -$5,164.90 | -3.30% |
Two things jump out.
The band is the only slice green in both seasons. Everything else lost money in at least one year.
The “both sides negative” bucket is a perfect illustration of why one season proves nothing. It was the best bucket in 2025 at +8.25%. In 2026 it’s the worst at -13.17%. Same rule, same market, opposite sign, 524 bets between them. If I’d published this article after the 2025 season alone and told you to hammer the -105 side of pick’em-adjacent games, you’d have lost thirteen cents on the dollar this year. That bucket was noise wearing a costume, and I would have had no way to know.
But the 2026 Trend Is Not Encouraging
The band’s 2026 profit is front-loaded, and that matters:
| Month (2026) | Bets | Record | P&L |
|---|---|---|---|
| March (25th–31st) | 25 | 10–15 | -$171 |
| April | 110 | 55–55 | +$1,865 |
| May | 100 | 46–54 | +$695 |
| June | 92 | 35–57 | -$1,040 |
| July (through 27th) | 72 | 29–43 | -$533 |
April was the entire season. A 50% win rate on dogs averaging +132.8 is a hell of a month, and it accounts for more than double the full-season profit. Since May 27 — when the bankroll peaked at +$3,017 — the strategy has given back $2,289. June and July are a combined 64–100.
You can read that two ways. Either April was the noise and the band’s true edge is the modest, grinding thing the full-season number suggests. Or the last two months are the signal, the market has adjusted, and April was a gift that isn’t coming back. The data cannot distinguish between those two stories, and anyone who tells you otherwise is selling something.
The Part Where I Do the Statistics Honestly
Combined across both seasons, the +120 to +149 band is:
1,186 bets
534–652 (45.03%)
Break-even win rate at these prices: 43.26%
Profit: +$5,056.00
ROI: +4.26%
Expectancy: +$4.26 per $100 bet
A 4.26% edge over the closing line would be a genuinely excellent result. Professional sports bettors operate on edges in that neighborhood. So: is it real?
Run the test. The standard deviation of a single $100 bet in this band is about $115.30 — that’s the inherent volatility of a bet that either wins ~$131 or loses $100. Over 1,186 bets, the standard error on the mean is $3.35.
t = 4.26 / 3.35 = 1.27.
That’s a p-value of roughly 0.20 two-sided, or 0.10 one-sided. In plain English: if this strategy had exactly zero edge, you’d see a two-season swing at least this large about one time in five.
That is not significant. Not close to it.
Here’s the number that really frames the problem. To distinguish a 4.26% edge from zero at conventional significance (t = 2), given this level of per-bet variance, you would need roughly 2,900 bets in the band. At ~590 qualifying games per full season, that’s five seasons of data. You’d finish the test in 2030.
This is the fundamental, unfixable problem with betting backtests, and it’s worth stating plainly because so much betting content pretends it away: the signal-to-noise ratio of an individual moneyline bet is about 0.037. The edge you’re hunting is four dollars. The noise on each attempt is a hundred and fifteen. You need enormous samples to see anything through that, and by the time you’ve collected them, the market has usually moved.
For contrast — all 3,887 underdog bets across both seasons produced a -1.89% ROI, t = -1.04. Also not significant. The correct read on the whole exercise is: nothing here is statistically distinguishable from a market that is priced correctly and taking its cut.
Six Reasons This Would Be Worse in Practice
Even granting the band a real edge, the gap between a spreadsheet and a betting account is wide:
- You cannot bet the close. Everything above assumes you got the closing number on all 1,186 games. You can’t. The close exists for a moment, at a price the book is willing to hang because it’s the sharpest number they have. Betting earlier means worse prices on average.
- The book changed between seasons. 2025 is ESPN BET, 2026 is DraftKings. A cent or two of difference in where each book hangs its number is enough to move a 4% edge around meaningfully. This is not a clean apples-to-apples two-season sample and I won’t pretend it is.
- No line shopping. One book, one price, every game. Real bettors shop, which would improve these numbers — but it also means the “edge” I measured isn’t the edge you’d get; it’s a single book’s error, if it’s an error at all.
- Limits and account life. Anyone flat-betting the same identifiable band 590 times a season gets noticed. Retail books limit profitable accounts fast. A strategy that requires 2,900 bets to prove itself will get you restricted around bet 200.
- Multiple comparisons. I tested six buckets. At a 20% false-positive rate per bucket, finding one that looks good across two seasons is roughly what you’d expect from pure chance. The out-of-sample confirmation helps, but it doesn’t erase the selection.
- Market drift. MLB betting markets are more efficient every year. Even if a +120 to +149 mispricing was real in 2025, the 2026 second-half decline is consistent with it being arbitraged away.
What I’d Actually Take From This
The MLB closing moneyline is close to perfectly efficient in aggregate. 2,321 blind underdog bets missed fair value by 0.15 percentage points. That’s the headline, and it should recalibrate how much you expect to find by staring at prices.
Longshot bias is alive and it is expensive. +200-or-longer dogs went 41–141 in 2025 for a -24% ROI. Whatever you do, the long end of the board is not where value lives. This is the single most actionable thing in the data, and it’s actionable in the negative — it tells you where not to bet.
The +120 to +149 band is the most interesting thing here and it still isn’t proof. +4.26% over 1,186 bets, green in both seasons, sitting exactly where you’d theoretically expect an edge to hide — meaningful enough to keep tracking, nowhere near enough to bet the mortgage on. If you want to follow it, follow it forward, on paper, and see whether the second half of 2026 keeps eroding it.
Variance dominates everything at this sample size. The 2025 strategy lost less than 1% and still spent five months eight thousand dollars underwater. Any real-money version of any of this needs a bankroll and a temperament that most people describing these strategies do not have.
And the meta-lesson, which is the one that generalizes past baseball: the first cut of the data gave me a beautiful, clean, profitable rule. The second season cut its edge in half. The significance test said the whole thing is inside the range of chance. If I’d stopped at step one — as most published backtests do — I’d have written a much more confident article, and it would have been wrong.
Appendix: Full 2025 Results by Price Band
| Price band | Bets | Wins | Losses | Win % | Wagered | P&L | ROI |
|---|---|---|---|---|---|---|---|
| Both sides negative | 220 | 122 | 98 | 55.45% | $22,000 | +$1,814.72 | +8.25% |
| +100 to +119 | 750 | 351 | 399 | 46.80% | $75,000 | -$2,350.00 | -3.13% |
| +120 to +149 | 787 | 359 | 428 | 45.62% | $78,700 | +$4,240.00 | +5.39% |
| +150 to +199 | 382 | 140 | 242 | 36.65% | $38,200 | -$1,080.00 | -2.83% |
| +200 to +299 | 177 | 41 | 136 | 23.16% | $17,700 | -$4,315.00 | -24.38% |
| +300 or longer | 5 | 0 | 5 | 0.00% | $500 | -$500.00 | -100.00% |
| Total | 2,321 | 1,013 | 1,308 | 43.64% | $232,100 | -$2,190.28 | -0.94% |
2026 (Through July 27) by Price Band
| Price band | Bets | Wins | Losses | Win % | Wagered | P&L | ROI |
|---|---|---|---|---|---|---|---|
| Both sides negative | 304 | 135 | 169 | 44.41% | $30,400 | -$4,002.90 | -13.17% |
| +100 to +119 | 657 | 306 | 351 | 46.58% | $65,700 | -$1,389.00 | -2.11% |
| +120 to +149 | 399 | 175 | 224 | 43.86% | $39,900 | +$816.00 | +2.05% |
| +150 or longer | 206 | 71 | 135 | 34.47% | $20,600 | -$589.00 | -2.86% |
| Total | 1,566 | 687 | 879 | 43.87% | $156,600 | -$5,164.90 | -3.30% |
Method Notes
Game results and closing moneylines were pulled game-by-game from ESPN’s public scoreboard and odds endpoints, then cross-checked against independent aggregates: 21,614 runs across 2,430 games in 2025 (8.90/game, home 10,910 vs away 10,704) and 14,411 runs across 1,596 games in 2026 (9.03/game, home 7,224 vs away 7,187). 2026 timestamps were converted from UTC to US Eastern so night games carry their correct calendar date. Every P&L figure was computed twice — once in the extraction pass, once independently in the workbook — and reconciled to the cent.
This is a historical analysis of publicly available data, not betting advice, and past results don’t predict future ones. If betting has stopped being entertainment for you, the National Problem Gambling Helpline is at 1-800-522-4700.