Sports betting bankroll management: units, Kelly, variance and records

Bankroll management does not create an edge. If your bets lose money on average, no staking plan turns them into winners; it only changes how fast and how unevenly you lose. What a staking plan does do is control the size of the swings, stop a bad run from wiping you out, and give you records honest enough to tell whether you have an edge at all. This guide covers unit sizing, the Kelly criterion and fractional Kelly with worked numbers, how much variance to expect, and what to record.

What is a sports betting bankroll?

A bankroll is a fixed amount of money set aside for betting and kept separate from everything else: rent, savings, investments. It should be an amount you could lose in full without it changing your life, because that is a realistic outcome for most bettors.

Two rules make the number meaningful:

1. It is closed. You do not top it up after losses. If it runs out, betting stops until you decide, calmly and on a separate day, whether to start again.
2. It is counted in one currency. For crypto bettors this matters more than it sounds; see the crypto section below.

Every other decision, from stake size to when to stop, is expressed as a share of this number.

How big should a betting unit be?

A unit is your standard stake. The common approach is to set it as a small percentage of the bankroll:

For most bettors, 1% to 2% is the sensible range. At 5%, a run of 10 losses (which, as the variance section shows, is not rare) costs about 40% of the bankroll if stakes are recalculated after each bet, and half if they are not.

There are two ways to apply a unit:

• Flat staking: the unit is fixed in money, say $20, and recalculated only at set points, for example the first of each month.
• Percentage staking: every stake is the same percentage of the current bankroll. Stakes shrink after losses and grow after wins, which makes total ruin mathematically impossible but slows recovery from a drawdown.

Flat staking is easier to track and easier to keep disciplined. Percentage staking is safer when the bankroll is small relative to your bet sizes. Either is better than staking by feel.

What is the Kelly criterion?

The Kelly criterion sizes each bet according to your edge. For a bet at decimal odds d with your estimated win probability p:

• b = d - 1 (the net odds)
• q = 1 - p
• Kelly stake as a share of bankroll: f = (b x p - q) / b

A positive result means the bet has positive expected value on your numbers; a negative result means do not bet.

Worked example. You find a price of 1.95 and estimate the true chance at 55%. The book's price implies 1 / 1.95 = 51.28%.

• b = 0.95
• f = (0.95 x 0.55 - 0.45) / 0.95 = 0.0725 / 0.95 = 7.63%

With a $2,000 bankroll, full Kelly says to stake $152.63. Half Kelly is $76.32 and quarter Kelly is $38.16.

Kelly is designed to maximise the long-run growth rate of a bankroll, and in theory no other strategy beats it over a long enough run if your probabilities are right. That condition is the problem.

Why do most bettors use fractional Kelly?

Kelly is only as good as the probability you feed it, and bettors overestimate their edge far more often than they underestimate it. The penalty for overestimating is severe.

Take the same bet. You believe the chance is 55%, but suppose it is really 52%: still a small positive edge at 1.95. The correct Kelly stake would be 1.47%. Here is what happens to a $2,000 bankroll over 1,000 such bets at each staking level, using the typical (median) outcome:

Every bet here has positive expected value, yet full Kelly on an inflated estimate shrinks the typical bankroll by more than 80%. Betting too large a fraction turns a winning edge into a losing outcome because losses compound faster than gains. Half Kelly still slightly shrinks it. Quarter Kelly lands close to the best possible result even though the estimate was wrong.

That is why experienced bettors who use Kelly at all tend to use a quarter to a half of it, and many cap any single bet at 2% to 3% regardless of what the formula says. If you do not have a tested way of estimating probabilities, flat 1% units are the honest choice.

How much variance should you expect?

More than intuition suggests. Assume a bettor with a genuine 2% edge, staking 1 unit on every bet at odds of 1.95. That edge means winning 52.31% of the time instead of the 51.28% needed to break even.

• Each bet has a standard deviation of about 0.97 units.
• Over 500 bets, expected profit is 10 units, with a standard deviation of about 21.8 units.

So a bettor with a real edge has roughly a one in three chance of being down after 500 bets, and nearly a one in five chance after 2,000. Losing streaks follow the same logic: in our simulation of 500 bets at a 52.31% win rate, about half of all runs contained a streak of eight or more straight losses, and about one in seven contained ten or more.

Two conclusions follow. First, a few hundred bets cannot tell you whether you have an edge; results alone are mostly noise at that sample size. Second, stake size must be small enough that a ten-bet losing run is uncomfortable rather than fatal. For more on how outcomes spread out over time, see our glossary entry on volatility.

What records should a sports bettor keep?

Because profit and loss are so noisy, you need records that measure something more stable. A spreadsheet with these columns is enough:

Closing line value (CLV) is the most useful number you can track. If you took 2.10 and the market closed at 1.95, you beat the close by 2.10 / 1.95 - 1 = 7.7%. A stricter version compares your price with the no-margin closing price: if the close was 1.95 on both sides, fair odds were 2.00, and your edge on that measure is 2.10 / 2.00 - 1 = 5.0%. Bettors who consistently beat the closing line tend to show a profit over large samples; those who do not, rarely do. CLV settles faster than profit, so it tells you much sooner whether your process works.

Record every bet, including the impulsive ones. A log that leaves out the bad bets is worse than no log, because it tells you a story that is not true. Our bookmaker margin calculator guide shows how to strip the margin from closing odds.

Bankroll management with crypto

Crypto adds three traps that a fiat bettor does not face.

• Mixed currencies. If your bankroll sits in BTC, a winning month can show as a loss in dollars because the coin fell, or the reverse. Keep the bankroll in a stablecoin, or at least record every bet in one fixed currency.
• Fees. Every deposit and withdrawal costs a network fee, and sometimes a cashier fee on top. Log them as betting costs; over a year they add up.
• Spread across books. Holding balances at several books makes it easy to lose track of the total. Add them up weekly and compare with your log.

Keep coins you are holding as an investment in your own self-custody wallet, never on a sportsbook. Our guide to how crypto sportsbooks work covers balances and fees in more detail.

When should you stop betting?

Set your stop rules before you need them, when you are calm:

• A daily or weekly loss limit, in units, after which you stop for the period.
• A rule that you never increase your unit size to recover losses.
• A monthly review of CLV and results, with a decision on whether to continue.

Use the tools books provide: a deposit limit, a cooling-off period and, if you need it, self-exclusion. If betting is affecting your money, sleep or relationships, our responsible gambling page lists free and confidential support in many countries. No staking method promises a profit, and the safest bankroll is one you can lose without regret.