Pythagorean Win Expectancy: Spotting MLB Teams Due to Regress
In late June 2019, I watched a 47-29 team blow a 5-run lead on a Tuesday night, lose 8-7, and continue to look like a juggernaut in the standings while their underlying numbers told a different story. Run differential said they were closer to 45-31. Strength-of-schedule-adjusted run differential said 43-33. The standings said elite. The pythagorean wins mlb model said good-not-elite, with a meaningful gap between record and skill. They were 78-84 over the rest of the season. I’d bet against them at +120 in a divisional matchup the following week and started winning regularly betting unders on their games for the next two months. The regression came. It always does, eventually, and the framework that anticipates it is one of the oldest and most useful tools in baseball analytics.
This piece walks through Bill James’s original Pythagorean win expectancy formula, the modern exponent adjustments, how to identify teams whose actual record meaningfully diverges from their expected record, and how to translate that information into betting decisions for both individual games and futures markets. The framework is mathematically simple, structurally powerful, and astonishingly underused by casual UK bettors who are still picking games based on recent record rather than underlying performance.
Available in USThe Formula and Its History
Bill James, working in the 1980s before sabermetrics had a name, noticed that team winning percentage correlated more tightly with the ratio of runs scored to runs allowed than with any individual team metric. He proposed a formula: expected winning percentage equals (runs scored)² divided by ((runs scored)² + (runs allowed)²). The squared exponent gave the relationship a non-linear shape that matched empirical data better than a simple linear ratio.
Modern analytics has refined the exponent. The originally-proposed value of 2.0 worked but slightly overestimated the win probability for extreme run differentials. Subsequent analysis, particularly work by Clay Davenport and others through Baseball Prospectus, identified that an exponent closer to 1.83 fits modern MLB data better. Some refinements adjust the exponent dynamically based on the team’s run environment – higher-scoring eras need slightly lower exponents – but the 1.83 default works for nearly all practical purposes in current MLB.
Applied: a team that has scored 450 runs and allowed 380 runs has an expected winning percentage of (450^1.83) / (450^1.83 + 380^1.83), which calculates to roughly .582. Over 162 games, that’s an expected 94-68 record. If the team’s actual record is 88-74, they’re underperforming their run differential by about 6 wins. If their actual record is 100-62, they’re overperforming by about 6 wins. Either gap signals likely regression toward the expected line.
The framework’s elegance is that it works without complex stats. You need two numbers – runs scored and runs allowed – both publicly available for every team and updated daily. That’s it. The interpretive complexity comes in deciding when the gap is meaningful and when it’s noise, but the core calculation is one line of arithmetic.
Identifying Lucky and Unlucky Teams
Once you have expected versus actual win totals for every team, the diagnostic question is which teams are diverging meaningfully and why. A team three or fewer games above or below their Pythagorean expectation is essentially aligned – the gap is within random noise for a baseball sample. A team five-plus games on either side is showing meaningful divergence that’s worth understanding.
Teams overperforming their Pythagorean expectation typically share one or more characteristics. They’ve been clutch in one-run games (winning a higher percentage than the league baseline of around 50%). They’ve had a strong bullpen converting save chances at above-average rates. They’ve benefited from sequence-of-events luck – bunching their hits in scoring situations rather than spreading them across games. None of these characteristics are fully sticky. Clutch performance in one-run games is the noisiest year-to-year metric in MLB; teams that overperform one season typically regress to league baseline the next.
Teams underperforming their Pythagorean expectation show the inverse profile. They’ve lost one-run games at above-baseline rates. They’ve had bullpens that converted save chances poorly even when the underlying ERA was solid. They’ve had unfortunate sequencing – losing 8-0 and winning 2-1, rather than the more efficient 3-2 and 5-4 split. These teams have been unlucky, and unlucky is typically temporary.
The 2025 MLB season produced 71.4 million in attendance – the third consecutive year of growth and the first such streak since 2005-2007 – and that engagement level reflects the kind of long-season variance that creates Pythagorean divergence in the first place. The 162-game schedule produces enough samples for the underlying skill to express itself, but enough single-game variance that record can diverge from skill across a half-season or even a full season.
The practical signal: a team five-plus wins above their Pythagorean expectation is a structural fade. A team five-plus wins below their expectation is a buy candidate. The market typically catches up to the regression over the next 30-60 games, but the catch-up isn’t instant, and the bettors who position before the market adjusts collect on the regression as it unfolds.
How to Bet the Regression
The application to game-level betting is direct. A team overperforming their Pythagorean expectation by 6 wins is, on any given individual game, priced slightly higher than they should be. The moneyline reflects their actual record more than their underlying skill, which means the favoured price overstates the real win probability. Backing the opponent at the inflated price is a small but persistent edge.
The opposite is true for underperforming teams. A team 6 wins below their expectation is priced slightly lower than they should be. The moneyline reflects their actual record, which understates their underlying skill. Backing them at the depressed price is the same edge in reverse.
Doug Upstone, a betting analyst, framed the systematic mindset cleanly: «I have long been a proponent of betting systems in all sports. The beauty of systems is not how a particular team is trending, rather, a system is a specific set of parameters and a team either fits it or it doesn’t in a positive or negative way. I tend to look for systems that win 75 percent or more of the time over at least a five-year period.» The Pythagorean regression bet doesn’t win 75% of the time – it wins maybe 52-54% on individual games, which over a betting career is the kind of small-edge systematic angle that compounds meaningfully.
The size of the edge depends on the size of the gap. A team three wins above expectation is a marginal fade – the gap is barely outside the noise band. A team eight wins above expectation is a stronger fade. A team ten wins above expectation is a strong fade that approaches «always bet against them» territory in lower-leverage spots.
The pairing with underdog analysis is where the framework gets most powerful. Long-run data shows MLB underdogs win approximately 44% of games – roughly 4 of every 9 contests. An underdog who’s also underperforming their Pythagorean expectation is doubly priced for value: the moneyline is shaded toward the favourite, and the team’s actual record understates their skill. Backing the underdog systematically in this profile produces win rates close to break-even by raw count and positive ROI by price.
Limits and Traps of Pythagorean Analysis
The framework isn’t magic. Several specific traps catch new users.
First, sample size matters. Pythagorean expectation needs roughly 40-50 games to stabilise. Early-season divergences – a team 6 wins above expectation through 30 games – are largely noise. The 30-game sample isn’t large enough for the expected record to converge on the true skill, and the actual record over 30 games is also noisy. Wait until late May or early June to start trusting the divergences.
Second, mid-season trade activity disrupts the framework. A team that adds a star pitcher at the deadline changes their underlying skill, and the pre-deadline Pythagorean expectation no longer applies. Recalculate expectations after major roster moves rather than carrying the pre-deadline projection forward. Similarly, a team that loses a key player to injury has a different post-injury skill profile than their pre-injury Pythagorean number suggests.
Third, persistent bullpen quality can sustain Pythagorean overperformance longer than the framework predicts. A team with an elite closer and elite high-leverage relief corps can win one-run games at above-baseline rates not because of luck but because of structural advantage in the leverage moments. The regression in those cases is slower and smaller. The same is true in reverse for teams with structural bullpen weaknesses.
Fourth, late-season Pythagorean signals get muddier as teams adjust their priorities. Teams in playoff position may rest starters and call up minor-league rotation depth, which depresses both runs scored and runs allowed in ways that don’t reflect playoff skill. The Pythagorean expectation is most informative for projecting the regular-season finish and less informative for projecting playoff performance.
UK Application Beyond Individual Games
The futures market is where Pythagorean analysis pays best for UK bettors. By midsummer, the regular-season win-total markets and division-winner markets have been priced for months, with bookmakers adjusting based on actual record and recent form. Pythagorean-driven gaps between record and expected skill let you identify futures positions that the market hasn’t fully reflected.
A team in second place with an expected record that beats the first-place team’s expected record is a div-winner futures bet worth opening. A team in first place with an expected record that trails the second-place team is a fade on their pennant or division futures. The futures market moves more slowly than individual game lines, which means the Pythagorean signal has more time to express itself before the price corrects.
For UK bettors specifically: the Pythagorean approach pairs naturally with the futures-market analysis I cover in my piece on how MLB futures betting works for UK punters. The two frameworks complement each other because Pythagorean identifies which teams are mispriced relative to their skill, and the futures market is where that mispricing produces the largest expected-value positions for patient bettors.
The UK MLB underdog signal is similar. Long-run data showing that underdogs win approximately 44% of MLB games (about 4 in 9 contests) sets the structural baseline for value betting. Pythagorean analysis identifies which underdogs are most likely to deliver on that baseline because their underlying skill exceeds their reflected price.
The Patient Bettor’s Tool
Pythagorean expectancy is not a system that wins individual games at high rates. It’s a framework that produces small structural edges across many bets, where the edge compounds across a season. The bettor who uses Pythagorean signals as one input among several – alongside pitcher matchups, weather, lineup, and schedule analysis – outperforms the bettor who treats actual record as the primary indicator of team quality. The framework rewards patience because the regression unfolds across 30-60 games, not overnight. The traders who fade the hot team in late May and bet steadily against them through July collect on a signal that the casual market keeps ignoring. The maths is old, the application is straightforward, and the edge persists because most bettors prefer the narrative of recent results to the discipline of underlying numbers.
When in the season does Pythagorean expectation stabilise?
The framework needs roughly 40-50 games to produce meaningful divergence signals. Early-season Pythagorean numbers are noisy because both the actual record and the expected record are calculated from small samples. By late May or early June, the gap between actual and expected wins becomes informative. By mid-summer and into the second half, the signal is at its most reliable.
Should I trust 1.83 or 2.0 as the exponent?
1.83 is the modern empirical default and produces more accurate predictions for recent MLB run environments. The original 2.0 exponent proposed by Bill James works adequately but slightly overestimates win probability for teams with extreme run differentials. For betting purposes, use 1.83 unless you’re working in an unusual run-environment season (extreme deadball or extreme offence inflation) where dynamic exponent adjustment might be worth the additional complexity.
Is the regression edge gone by August?
Not entirely, but it compresses. The mid-season market gradually catches up to Pythagorean divergences as the season progresses, which means the largest edges typically appear in late May through late July. By August, much of the gap has been priced in. The remaining edge in August and September lives in nuanced second-order signals – teams whose recent Pythagorean trajectory is shifting, or teams whose underlying skill has changed due to trades or injuries.
Preparado por la redacción de «mlb Betting Systems».
