Kelly Criterion for Baseball Betting: A UK Bettor's Bankroll Manual

Kelly is not a formula — it is a discipline

The first time I ran a full Kelly stake on an MLB underdog I felt physically ill. The number was 4.7% of my bankroll. The bet was on a +145 home underdog in a Wednesday-night divisional game in May. I had built the model, I had plugged in the inputs, I had double-checked the maths twice. I still hesitated for fifteen minutes before pressing the slip, because 4.7% felt larger than my normal stake by a wide margin and the bet looked, from the outside, like an ordinary value play. I placed it. It lost. The next forty Kelly bets returned me a comfortable profit and I have not looked back since.

That story is the entire lesson of Kelly applied to baseball. The formula gives you the optimal long-run stake size given your edge and the bookmaker’s price. It does not tell you the bet will win. It does not protect you from a 1-7 streak. It does not feel comfortable when the stake size jumps. Kelly is a discipline you adopt because you have made peace with variance, not because you have eliminated it.

This article is the working version of that discipline applied to MLB betting in 2026 from a UK bankroll. It assumes you are betting through UKGC-licensed operators, that your bankroll lives in pounds, and that you face the affordability-check regime that has been standard since 28 February 2025. It explains the formula plainly, walks through two worked examples on the run line and the moneyline, and lays out why most serious bettors I know — including me — run fractional Kelly rather than full. It is not a magic system. Done correctly, it is the seatbelt that keeps your bankroll alive long enough for your real edges to compound.

Why flat stakes fail at MLB scale

Flat staking — the same unit on every bet — is the default for most British punters who have come up through football and horse racing. It is intuitive, it is easy, and on a few thousand bets at modest edge it works adequately. It fails at MLB scale for one specific reason: MLB betting offers wildly different price points across the market on any given night, and a flat stake on a +180 underdog and a -180 favourite represents radically different exposures.

Consider the maths. Only 3 to 5% of sports bettors are profitable long-term, and at standard -110 odds a bettor needs a 52.38% win rate to break even. The reason most of the rest fail is not that they cannot pick winners. It is that they cannot survive the variance even when they are picking winners at a profitable rate. A 53% bettor at -110 has positive expected value, but a 200-bet sample produces a standard deviation of 7 bets either way — meaning a perfectly skilled bettor can land anywhere between 46 wins and 60 wins across that sample purely by chance. Flat staking through that variance, the bettor’s bankroll either compounds modestly or vanishes; the difference depends on the path, not the skill.

The 2024 hold figures tell the same story from the bookmaker side. U.S. sportsbooks retained $13.71 billion from $149.8 billion in handle, a 9.3% hold rate (up from 7.0% in 2019). The trend matters because it tells you bookmakers have increased their structural margin meaningfully over five years, mostly through parlay and same-game-parlay product expansion, and the flat-staking bettor is the customer keeping that margin alive. Stake-sizing matters more in 2026 than it did a decade ago precisely because the margins are wider and the variance bites harder.

The Kelly framework solves the staking problem by tying the bet size to the actual edge on each bet. A 4% edge bet gets a larger fraction of bankroll than a 1.5% edge bet; a -150 favourite at fair price gets nothing at all; a +180 underdog at 41% true probability gets a meaningful stake. Across a season, this produces a fundamentally different bankroll trajectory from flat staking — smaller losses in unprofitable bets, larger profits in the genuinely valuable ones, and a survival rate through cold streaks that flat staking simply cannot match.

The Kelly formula in MLB terms

The standard Kelly formula is f = (bp – q) / b, where f is the fraction of bankroll to stake, b is the decimal odds minus one (the net win on a £1 stake), p is your estimated probability of winning, and q is 1 – p. That looks abstract until you plug in baseball numbers, at which point it becomes the working tool of the rest of this article.

For an MLB bet at decimal 2.40 (+140 in American odds, a typical home underdog price), b = 1.40. If your model gives the underdog a 45% chance of winning, p = 0.45 and q = 0.55. Plug in: f = (1.40 × 0.45 – 0.55) / 1.40 = (0.63 – 0.55) / 1.40 = 0.08 / 1.40 ≈ 0.057. Kelly tells you to bet 5.7% of bankroll on this play. On a £2,000 bankroll that is a £114 stake.

The “edge” you can read off this calculation is the bookmaker’s implied probability versus yours. Decimal 2.40 implies 41.7% probability; you have estimated 45%. The gap — 3.3 percentage points — is your edge, and Kelly converts that gap into the optimal stake. As the gap narrows, the stake shrinks. At a 1% edge, Kelly tells you to bet roughly 1.7% of bankroll. At a 5% edge, the stake jumps toward 9% — uncomfortably large for most British bettors used to 1-2% units.

The variant that matters for run-line betting is the same formula applied to shorter prices. For a -1.5 favourite at decimal 1.95, b = 0.95. If your model gives the favourite a 53% chance of covering the spread, plug in: f = (0.95 × 0.53 – 0.47) / 0.95 = (0.5035 – 0.47) / 0.95 = 0.0335 / 0.95 ≈ 0.035. Kelly says 3.5% of bankroll. The shorter the price, the lower the optimal stake for the same edge — counterintuitive at first, mathematically inevitable.

The key input — and the place where every Kelly application lives or dies — is the probability estimate p. Get the probability wrong and the stake size is wrong; if you systematically over-estimate your win rate by even 2 percentage points, full Kelly grinds your bankroll into the floor over a few hundred bets. This is not a hypothetical risk; it is the dominant failure mode of Kelly bettors. Probability calibration matters more than any other input, and this is the single biggest reason most professionals run fractional Kelly rather than full, which I get to in section six.

One UK-specific note. Kelly assumes decimal odds, which is the British default and matches the formula directly. American odds need to be converted first: a +140 line becomes decimal 2.40 by adding 1 + (140/100); a -150 line becomes 1.667 by computing 1 + (100/150). If your model speaks American and you bet British, build the conversion into your spreadsheet rather than running it in your head at the slip. The cumulative damage from arithmetic errors at 11pm on a Friday night is not trivial.

Worked example: Kelly stake on a plus 130 underdog

Take a midweek matchup. Tigers at Twins, Tigers as the road underdog at +130 on the moneyline. Decimal odds 2.30. Your model gives the Tigers a 47% chance of winning the game outright; the implied probability of +130 is 43.5%. Edge is 3.5 percentage points.

Kelly maths. b = 1.30. p = 0.47. q = 0.53. f = (1.30 × 0.47 – 0.53) / 1.30 = (0.611 – 0.53) / 1.30 = 0.081 / 1.30 ≈ 0.062. Full Kelly says 6.2% of bankroll. On a £2,500 bankroll that is £155.

That is uncomfortably large for most British bettors and it should be. Full Kelly is calibrated for an exactly-correct probability estimate, infinite bankroll, infinite time, and zero correlation between bets. Real bettors have approximate probability estimates, finite bankrolls, finite seasons, and some correlation between bets. For all those reasons, applying full Kelly with a real-world probability estimate is functionally equivalent to leveraging up your variance.

The realistic application drops to half Kelly: 3.1% of bankroll, £77 on the £2,500 example. That is still meaningful — three or four times the size of a typical 1% flat stake — and it captures the differentiation Kelly is designed to provide. If your edge had been smaller, the stake would have shrunk; if larger, it would have grown.

Walk through the variance. A 47% bettor at +130 has positive expected value (0.47 × 1.30 – 0.53 = +0.081 per unit), but the standard deviation on a single bet is roughly 1.07 units. Across 50 such bets the bettor expects +4.05 units of profit but with a standard deviation of about 7.6 units — meaning the realistic outcome range is roughly -3.5 to +11.5 units across that sample. Half Kelly compresses the dollar variance proportionally; full Kelly expands it. A bettor running full Kelly on a string of 47% bets at +130 is signing up for swings of ±15% of bankroll in any given month, which most people psychologically cannot handle.

The bet itself is the simple part. The hard part is having calibrated the 47% estimate honestly. If you are systematically overconfident — if your “47%” plays actually win 44% of the time — full Kelly is mathematically negative-EV and will erode your bankroll. The probability calibration is the entire game; the formula is just the lever that translates calibration into bet size.

Worked example: Kelly stake on a minus 1.5 run line

Take a different matchup. Yankees at home, heavy moneyline favourite at -180, with the -1.5 run line priced at +110 (decimal 2.10). Your model gives the Yankees a 67% chance of winning outright but only a 49% chance of covering the -1.5 spread. Implied probability of +110 is 47.6%. Edge is 1.4 percentage points.

Kelly maths. b = 1.10. p = 0.49. q = 0.51. f = (1.10 × 0.49 – 0.51) / 1.10 = (0.539 – 0.51) / 1.10 = 0.029 / 1.10 ≈ 0.026. Full Kelly says 2.6% of bankroll. On the same £2,500 bankroll that is £65.

Notice how the run-line stake is much smaller than the moneyline stake despite the favourite being a strong team. The reason is the edge: 1.4 percentage points on the run line versus 3.5 points on the underdog moneyline. Kelly reads the edge, not the team. A “small edge on a star team” gets a smaller stake than a “bigger edge on a dog.” The formula is indifferent to narrative; it cares only about the gap between your probability and the bookmaker’s.

Half Kelly on this run-line bet drops the stake to £32. That is closer to a typical British flat-staking unit, which gives you a sense of why so many UK bettors instinctively bet small amounts on run-line favourites — it turns out their gut is roughly aligned with half Kelly on a low-edge bet.

One subtlety on run-line maths. The decimal odds 2.10 may seem similar to 2.30 from the previous example, but the implied probabilities differ enough to matter. Run-line markets cluster in the 1.85 to 2.15 range, which means Kelly stakes on run lines tend to fall between 1.5% and 4% of bankroll on half Kelly. Moneyline underdog stakes can range from 2% to 7% on half Kelly when prices stretch above +150. The variability of stake size is wider on the moneyline than on the run line, which is structurally why moneyline betting carries higher variance even at the same edge.

The other lesson is the asymmetry of edge measurement. A 1.4-point edge on a 50/50 market is different from the same edge on a 30/70 market in variance terms. Kelly handles this through the formula naturally — the b term shrinks for shorter odds and grows for longer ones — but the bettor still feels the variance differently. Long-odds bets at small stakes feel “flatter” in P&L terms; short-odds bets at larger stakes feel sharper. Both are correct stakes by the formula, both produce identical long-run growth at calibrated probability estimates. The feel is psychological; the maths is not.

Half Kelly, quarter Kelly, and why most bettors should choose them

I have not run full Kelly on a single bet in five years. Neither has any serious bettor I know. The reason is that full Kelly assumes inputs are exactly correct, and they never are. Even a small systematic error in probability estimation — over-estimating wins by 2 percentage points across all bets — turns full Kelly from optimal-growth into bankroll-destruction over a long enough sample.

Half Kelly is the practical compromise. It produces about three-quarters of the long-run growth rate of full Kelly with about one-quarter of the variance. The math is straightforward: variance scales with the square of stake fraction, while expected growth scales linearly. Cutting stakes in half cuts variance by 75%; expected growth drops by only 25%. That is a trade most bettors should take without hesitation, because the dominant survival risk in betting is variance-induced ruin, not slow growth.

Quarter Kelly takes the trade further. It produces about half the long-run growth rate of full Kelly with about 6% of the variance. For a bettor with a small bankroll, an unproven model, or a low risk tolerance, quarter Kelly is the realistic starting point. You give up half your theoretical edge to keep the bankroll alive through the inevitable cold patches.

The choice between half and quarter Kelly comes down to three honest questions. How calibrated is your probability model? If you can demonstrate that your estimated 53% bets win 53% of the time across hundreds of historical bets, half Kelly is appropriate. If you have not validated calibration to that standard, quarter Kelly is correct. How psychologically tolerant are you of drawdown? If a 20% bankroll dip would cause you to abandon the system, halve again. How urgent is the bankroll growth? Most British bettors are betting at recreational scale where compounding does not need to be aggressive; quarter Kelly is fine and probably better than fine.

The one variant I avoid is “dynamic” Kelly — adjusting the fraction up after wins and down after losses. The idea is intuitive but it produces strictly worse long-run results than fixed fractional Kelly. Mathematically, the optimal stake at every bet is set by the edge on that bet, not by the bankroll trajectory leading up to it. Adjusting in response to recent variance is gambler’s-fallacy territory in disguise.

The other thing to keep firmly in mind is the variance you cannot escape. Even a fully calibrated quarter-Kelly bettor with a 4% edge runs through downswings of 15-20% of bankroll across a typical season. The variance is structural; the discipline is in not abandoning the system mid-cold-streak. The deeper version of that argument lives in the MLB betting downswings framework, which I treat as the necessary companion to Kelly. Kelly tells you what to bet; downswing-management tells you how to keep betting it.

How big should a UK MLB bankroll be?

The honest answer is “whatever you can lose without it changing your life.” That is unhelpful as a guide to operations, so let me sharpen it. The minimum functional bankroll for serious MLB betting using fractional Kelly is around £2,000 to £2,500. Below that, Kelly stakes on individual bets fall below £20, which puts you below most operators’ minimum bet thresholds and constrains the bets you can actually place. Above £10,000, Kelly stakes start to bump against per-bet limits at smaller UK operators, and you may need to spread action across two or three books to get the full theoretical stake placed.

The middle range — £2,500 to £8,000 — is where Kelly works most cleanly for a UK bettor. Stake sizes are large enough to matter but small enough to fit comfortably under operator limits. Affordability checks become a real consideration in this band but not a crippling one if your betting pattern is consistent.

The other bankroll question is the one nobody asks publicly: how much of your overall savings should sit in a betting bankroll. The answer that has held up for me is “no more than you can comfortably write off in full.” A bankroll is not an investment; it is a working float for an entertainment activity that occasionally produces returns. Treating the bankroll as savings that need to be preserved is exactly the wrong frame, because the bankroll inevitably suffers drawdowns and “preservation” instincts cause bettors to abandon Kelly stakes at the worst possible moments.

The 22.5 million UK adults who gamble regularly span every imaginable bankroll size. Andrew Rhodes, the Gambling Commission’s Chief Executive, framed the participation context bluntly: “Some 22 and a half million consumers gamble on a regular basis in this country. It is a significant economic and social activity that people take part in and continues to be a mass participation exercise but one we all know brings its challenges.” That framing matters because it puts your individual Kelly bankroll in a regulatory and social context. You are one of 22.5 million; the operators you bet with are governed by rules built to protect that whole population, and your bankroll size sits inside those rules.

One last operational note. Keep your bankroll in a separate account from your day-to-day finances. Not because anything will go wrong if you do not, but because the mental discipline of “bankroll” depends on the visible separation. Topping up from a current account in the heat of a losing weekend is the single most common path to losses that exceed what was intended.

Kelly and UKGC affordability checks

The financial vulnerability threshold in UK gambling fell to £150 net loss per 30-day rolling period from 28 February 2025, requiring operators to act on light-touch checks. That figure is the operational ceiling that interacts with Kelly maths in a way most bettors miss.

A bettor running quarter Kelly on a £3,000 bankroll typically stakes £15 to £75 per bet. Across a 30-day period with normal MLB volume — say 30 bets — that bettor’s variance can easily produce a £150 net loss in a cold patch even if the underlying edge is real. The £150 trigger is not “you have lost £150 in a single session”; it is “your net result across 30 days is -£150 or worse.” The Kelly bettor in this position can be hitting a perfectly normal statistical downswing and still trip the operator’s documentation request.

What the operator typically asks for, when the trigger fires, is documentary evidence of source of funds — a payslip, a bank statement, occasionally a written confirmation of how the bettor’s gambling fits within their disposable income. The check is light-touch by design; for most bettors it is one round of paperwork and back to normal action. For some it triggers a deposit limit or a betting cool-off. The operator’s discretion within UKGC rules is wide.

The operational implication for Kelly bettors is twofold. First, run quarter Kelly rather than half if you are bumping against the £150/30-day net-loss ceiling regularly; the smaller stakes shrink the variance and reduce the trigger frequency. Second, be ready with documentation in advance. A Kelly bettor with a £4,000 bankroll and quarter Kelly stakes is plausibly within disposable-income norms for any UK adult earning above the median; demonstrating that on paper takes ten minutes if you have prepared, and the rest of an evening if you have not.

The bigger Kelly-versus-affordability tension is in the larger end of the bankroll spectrum. A bettor at £15,000 running half Kelly on a +180 underdog at a 6% edge stakes £450 on a single bet. That stake size, several times in a month, will draw operator attention regardless of whether the bets win or lose. UK books are increasingly cautious about high-stake action on individual customers, and the Kelly bettor with significant edge often finds the marginal next bet is the one the bookmaker declines to take. Your edge is rationed by the operator’s appetite. Plan for that constraint; do not assume your model’s stake sizes will always get filled.

Why Kelly breaks on multi-bets and parlays

Kelly is built for single, independent bets. Parlays — multi-bets in British parlance — violate the independence assumption in a way that breaks the formula’s optimality. Each leg’s outcome is independent of the others, but the parlay’s payout depends on all of them landing together, which compounds variance multiplicatively while compounding edge only marginally.

The maths. A two-team parlay at standard prices combines a bookmaker margin of roughly 4-5% per leg into a combined margin of roughly 8-10%. Kelly applied to a parlay must absorb that compounded margin; in nearly every realistic case, the optimal Kelly stake on a parlay is lower than the sum of optimal Kelly stakes on the legs separately. You should almost always bet the legs straight rather than parlay them, because the bookmaker’s margin growth outpaces the variance benefit.

Same-game parlays — the British equivalent of bet builders — are worse. The legs within a same-game parlay are correlated by design (a high-strikeout pitcher and a low total are not independent events), and the bookmaker’s pricing engine charges a meaningful premium for that correlation. The hold rate on parlay-style products has driven much of the 9.3% U.S. industry-wide retention rate, which sits well above the 7.0% rate of 2019. The same dynamic applies to UK bet builders, with similar margin compression.

Kelly’s failure on multi-bets is structural, not a calibration issue. Even with perfectly calibrated probabilities, the optimal stake on a parlay tends to zero faster than the optimal stake on the equivalent straight bets. The only narrow exception is when a bettor identifies meaningful negative correlation between legs that the bookmaker has ignored — the bookmaker priced as if independent but the legs actually anti-correlate. Those spots are vanishingly rare in MLB and the bookmaker, when they appear, usually catches them and adjusts.

The practical rule I run for myself: no Kelly maths on multi-bets, full stop. If I want to play a multi-bet for entertainment, I do so at flat-stake levels treated as outside the bankroll — recreational money rather than working capital. Mixing the two corrupts the discipline.

Kelly is the seatbelt, not the engine

The error most Kelly converts make is treating the formula as the source of edge. It is not. Kelly is the stake-sizing layer that sits on top of an actual edge. Without a real probability advantage over the bookmaker, Kelly accelerates losses rather than gains; with a real advantage, Kelly captures it efficiently across a season.

Build the edge first. That means a probability model you have validated against historical data, a process for generating estimates on each MLB matchup, a tracking system that lets you measure calibration over time, and the honesty to abandon parts of the model that are not working. Once that is in place, Kelly is the simple final step: convert each bet’s edge into a stake fraction, multiply by bankroll, place the bet, log the result.

The bettors who succeed with Kelly over multiple seasons are the ones who treat it as a rule rather than a recommendation. Stake sizes get adjusted only when the bankroll has moved meaningfully — typically at 10% increments, not after every win or loss. Probability estimates get recalibrated quarterly, not after a bad weekend. Fractional choice (half, quarter) gets reviewed annually based on realised variance, not on emotional response to a recent run.

I have run quarter Kelly on MLB through seven seasons. The bankroll has grown each year, with drawdowns of 12-22% in the worst patches and a typical annual ROI in the 4-7% range across all bets. Those are not flashy numbers. They are the numbers a real edge produces when staked correctly. The bettors who post 30% annual ROI on social media are either lying, running unsustainable Kelly stakes, or working with edges so wide they will not survive market adjustment. Sustainable, repeatable, fractional-Kelly betting on MLB is a quiet, slow process. The reward is that it actually works.