Bankroll Management for Sports Bettors: Kelly Criterion and Flat Staking

A bettor hitting 55% at -110 (1.91 decimal) generates a theoretical edge of roughly 5%. That is a solid, professional-grade number. But run that same bettor through 1,000 bets staking 10% of bankroll per play, and the probability of a 50% drawdown at some point in the sequence exceeds 80%, even though the long-run expected value is strongly positive. The math of ruin does not care that the picks were good. It cares about variance and stake size. Compare that to a bettor with an identical 55% win rate staking 2% per bet. The same simulation drops the probability of a 50% drawdown to under 10%. The edge is unchanged. The outcome distribution is completely different. This is the core lesson bankroll management teaches: two bettors with identical predictive skill can produce wildly different survival outcomes purely from stake sizing. Most bettors spend 90% of their time on pick selection and 10% on stake sizing. The data suggests that ratio should be closer to even. A 6% edge bet staked at 15% of bankroll is a worse decision than a 6% edge bet staked at 2%, even though the pick itself is identical in both cases.

The Kelly Criterion formula is f = (bp - q) / b, where b is the decimal odds minus 1, p is your estimated true win probability, and q is 1 minus p. Take a bet at 2.10 (b = 1.10) where your model gives a true probability of 52% (p = 0.52, q = 0.48). Full Kelly says stake f = (1.10 x 0.52 - 0.48) / 1.10 = 0.0945, or 9.45% of bankroll. That number is mathematically correct and practically dangerous. Full Kelly assumes your probability estimate is exact. In practice, model error of even 3-4 percentage points is common, and full Kelly amplifies that error into stake sizes that produce 30-40% drawdowns with uncomfortable frequency. This is why every serious quant shop runs fractional Kelly, typically 25% to 50% of the full calculation. Applying 25% Kelly to the example above gives a stake of 2.36% of bankroll instead of 9.45%. On a $10,000 bankroll that is $236 instead of $945. The expected growth rate drops slightly compared to full Kelly, but the variance drops by roughly 75%, which is the trade almost every professional bettor takes. Half Kelly (50%) sits in between, at 4.7% stake size, and is the most common setting among bettors who have moved past flat staking but are not yet ready for full Kelly's swings. The critical dependency in all of this is the accuracy of p, your true win probability. If that number is wrong, the entire Kelly output is wrong, which is why edge validation has to happen before stake sizing, not after.

Flat staking assigns a fixed percentage of bankroll to every bet, regardless of perceived edge size. The standard range is 1% to 2% per bet, with 1% common for higher-variance markets like correct score or first goalscorer, and 2% reserved for more stable markets like Asian Handicaps on major leagues. On a $10,000 bankroll, that is a $100 to $200 stake, unchanged whether the bet is a 4% edge or an 8% edge. The appeal of flat staking is that it removes the dependency on precise probability estimation. Kelly punishes bad probability inputs severely, because the stake size scales directly with the perceived edge - overestimate your edge by 5 percentage points and Kelly can nearly double your stake. Flat staking caps that damage at the fixed percentage no matter how wrong the model is. For bettors still calibrating their models, or betting across markets where historical data is thin (lower-league totals, esports handicaps), flat staking is the more defensible choice. The trade-off is efficiency. A flat 2% stake on a 15% edge bet and a flat 2% stake on a 6% edge bet are mathematically identical in size despite very different expected values. Over a large sample, this leaves value on the table compared to a well-calibrated Kelly model. The practical resolution many professional bettors use is a hybrid: flat staking as the default, with a modest multiplier (1.5x to 2x base stake) applied only to bets clearing a defined high-confidence threshold, which keeps the simplicity of flat staking while still respecting edge size at the margins.

Every staking formula on this page is downstream of one number: your estimated true probability versus the market-implied probability. A market price of 1.91 implies a 52.4% break-even probability once you strip out the bookmaker margin. If your model says 55%, your edge is 2.6 percentage points - not enough to clear our internal threshold. If your model says 59%, your edge is 6.6 percentage points, which does clear it. Our staking recommendations across VoxSports only apply above a 6% edge threshold, detailed fully on /methodology/. Below that line, the bet is treated as a coin flip with fees attached, regardless of how favorable the story around the matchup sounds. This threshold exists because backtesting shows that edges under 6% are frequently artifacts of model noise rather than genuine mispricing, and staking real money on noise is how disciplined-looking bettors still lose money over a season. The practical workflow is sequential: first calculate implied probability from the market price, then compare it to your model probability, then check whether the gap clears 6%, and only then move to Kelly or flat staking to size the position. Skipping the first three steps and going straight to "how much should I bet" is the single most common bankroll management error we see, more damaging than picking the wrong staking formula entirely.

A bettor with a genuine 6% edge at average odds of 1.95, staking 2% flat, should still expect to see a losing stretch of 8-10 bets in any 100-bet sample purely from variance. This is not a sign the model is broken. Binomial variance at a 51% true win rate produces losing streaks of that length with a probability exceeding 60% across a season of betting volume. The risk of ruin formula, R = ((1-e)/(1+e))^(B/s), where e is edge, B is bankroll units, and s is stake size, quantifies this precisely. At a 5% edge, staking 5% of bankroll per bet produces a risk of ruin near 30% before the edge has time to compound. Drop the stake to 1.5% and risk of ruin falls below 2%. This single variable, stake size relative to edge, explains more bankroll failures than bad picks ever do. Drawdown tolerance should be set before the betting session starts, not during it. A bettor who defines in advance that a 20% bankroll drawdown triggers a stake reduction (say, cutting from 2% to 1% flat) removes emotional decision-making from the moment it matters most, which is precisely when tilt-driven stake increases historically do the most damage.

Stake sizing assumes the price you are betting is close to fair. A traditional fixed-odds book baking a 6-8% margin into a market erodes your edge before your staking formula even applies. If your model shows a 6% edge against a fair market but the book's margin already ate 4 of those points, your real edge is closer to 2%, well under any threshold worth staking on. This is where exchange betting changes the calculation. The Betfair Exchange (https://www.betfair.com/exchange/plus/?PLACEHOLDER) typically operates on margins closer to 2-5%, and lets you both back and lay a position, which matters for two reasons. First, tighter margins mean your calculated edge is closer to your real edge, so Kelly and flat staking outputs are more reliable. Second, lay betting lets you hedge a position mid-event, effectively adjusting your stake in real time rather than locking it in at the moment of the original bet. For bettors running fractional Kelly specifically, the margin difference matters more than it first appears, because Kelly's stake output is directly proportional to the edge input. A market with a 7% margin versus one with a 3% margin can shift a calculated Kelly stake by 30-40% relative terms on the same underlying pick. Lower margin does not create edge out of nothing, but it stops the market from quietly deducting edge you have already found.

Start with bankroll segmentation. A $10,000 bankroll dedicated to betting should be treated as fully separate from personal savings or checking funds, sized at an amount whose complete loss would not affect financial obligations. This sounds obvious and is the step most frequently skipped. Next, choose a base staking method. Bettors with under 200 settled bets of model-tracked history should default to flat staking at 1% to 1.5%, because the sample size is too small to trust probability estimates enough for Kelly's sensitivity. Bettors with 500+ tracked bets and a documented calibration (predicted probabilities matching observed outcomes within 2-3 percentage points across buckets) can move to fractional Kelly at 25-50%. Third, set a hard edge floor at 6% before any bet enters the staking calculation, consistent with the threshold detailed on /methodology/. Fourth, define drawdown triggers in writing: a 20% bankroll decline cuts stake size in half until 10 consecutive settled bets confirm the model is still tracking. Fifth, route action through lower-margin venues like the Betfair Exchange where liquidity allows, since margin directly inflates or deflates the effective edge every formula in this guide depends on. Running this as a five-step sequence, rather than jumping straight to "how much do I bet on this one game," is what separates a staking plan from a staking habit. The math in Kelly and flat staking is only as good as the discipline applied around it.