Sports Betting Bankroll Management: Units, Kelly, Ruin

Sports betting bankroll management is the set of rules that decide how much of a dedicated pot you risk on each bet, so a normal losing run cannot empty it. It does not create an edge. It decides whether a real edge, or no edge at all, is still there after a bad fortnight. The usual working range is 1–2% of the current bankroll per bet, or a fraction of Kelly with a hard cap. Pinnacle's own ruin analysis (re-read 6 November 2026) puts the same ceiling in different words: to keep ruin well below 1%, stakes should not sit much beyond 1% of the starting bankroll, and smaller still at longer odds.

This page is the sizing. What one unit is sits in what is a unit in sports betting. The formula that sizes by edge is the Kelly criterion. The rules you write down first are betting discipline.

Definition. A bankroll is money set aside only for betting, separate from rent, bills and savings. Bankroll management is how you size each stake as a share of that pot, how you react when it falls, and when you stop. Risk of ruin is the chance the pot hits zero (or a floor you set) before any edge has time to show.

What counts as a bankroll?

Only money you can lose without changing how you live. If a losing week would force a deposit from next month's rent, it is not a bankroll; it is a reason not to bet. The amount can be small. Depth in units matters more than the euro figure: 100 units at €5 is a deeper roll than 20 units at €50.

Keep it in one number you can see. Sportsbook balances, exchange wallets and a cash float all count if they are for betting. Money you will withdraw next week does not.

How much of the bankroll should you bet?

A fixed share of the current bankroll, not a fixed euro amount. On €1,000 at 1%, the first stake is €10. After a loss to €990 it is €9.90; after a win to €1,010 it is €10.10. The adjustment is the point: stakes shrink into a losing run and grow after a winning one.

ValueBetFactory's bankroll guide (published June 2025, last updated 11 August 2026, read 6 November 2026) uses the same 1–2% conservative band and quarter Kelly for most +EV bettors. That is a competitor page, not a proof. The numbers below are ours.

A worked Pinnacle price

A real move from Pinnacle Terminal, 28 October 2026, about 09:37 Madrid (CET): ITF Women Poitiers, Leonard v Hesse, games handicap Leonard −3.5. Pinnacle fell from 1.826 to 1.543 and raised its maximum market limit as it fell; the terminal labelled it Limit-backed. Fair price (Power method): 1.631.

Suppose a slower book still offered 1.75 on that line (illustrative, not a recorded quote).

1. True probability: 1 ÷ 1.631 = 61.3%.
2. EV: 1.75 ÷ 1.631 − 1 = +7.3%.
3. Full Kelly: 7.3% ÷ (1.75 − 1) = 9.7% of the bankroll.
4. ¼ Kelly: 2.4%. On €1,000 that is about €24 — 2.4 units if one unit is 1%.

Nearly 10% on one tennis handicap is the warning, not the target. Full Kelly gets large when the edge looks large. The Kelly calculator does steps 2–4; the devig calculator does the fair price.

Drawdowns from the same edge

We simulated 20,000 runs of 1,000 bets, each with a true +3% edge at 2.00 (seed fixed, 31 October 2026; first published in the Kelly article):

Same edge. The only change is the stake. Full Kelly's median hole is 60% of the peak. A quarter of that stake cuts the 50%+ holes from 75% of runs to 0.1%. Uhrín, Šourek, Hubáček and Železný (Czech Technical University, 2021, arXiv:2107.08827, read 31 October 2026) saw the same pattern on 16,000 NBA games at Pinnacle's close and 32,000 football matches at Pinnacle's opener: full Kelly ruined 100% of 1,000 simulated runs in both sports (bankroll below 0.01% of start); fractional Kelly, tuned on training data, had 0% ruin.

With no edge, the same 1,000 bets and full Kelly sizing on an imagined +3% left 68% of runs below the start, against 54% at ¼ Kelly. Smaller stakes do not create an edge. They limit the damage when you are wrong about having one. Run your own numbers in the variance simulator.

Risk of ruin: original table

Gambler's-ruin approximation for even-money bets, p = 0.515 (the +3% at 2.00 used above), q = 0.485, ruin = (q/p)^N where N is bankroll ÷ stake. Computed 6 November 2026:

Same 51.5% shot. At 1% you are about 100 units deep and ruin is a quarter of a percent. At 5% you are 20 units deep and ruin is about three in ten. The formula assumes even money, independent bets, and a floor at zero. Real books stop you earlier; real odds are not all 2.00. Treat the table as the direction of the effect, not a personal forecast.

Pinnacle, "Understanding and managing your risks of ruin" (re-read 6 November 2026): for skilled bettors, holding ruin significantly below 1% means stakes should not sit much beyond 1% of an initial bankroll, and smaller still at longer odds. Their Monte Carlo is a 1-in-100 skill model, not our +3% even-money case. The conclusion on stake size matches the table.

Ten losses in a row, two sizes

With a 51.5% chance per bet, 10-loss runs showed up in 17.2% of 20,000 simulated sequences of 500 bets (computed 31 October 2026). They are normal.

Start at €1,000. Ten losing even-money bets:

The picks were identical. The 10% sizer is finished; the 1% sizer still has a bankroll. That is the whole argument for small units. How you feel after the tenth loss is tilt; the table is what the stake rule was for.

What to write down

Five numbers, before the next bet:

1. Starting bankroll. One figure, money you can lose.
2. Unit size. Usually 1% of current bankroll. Recalculate on a schedule (weekly, or every 20 bets), not after every result. See units.
3. Stake rule. Flat 1–2%, or ¼ Kelly with a cap of 2–3%. The Kelly calculator shows the uncapped number so you can cap it.
4. Stop. A daily or weekly drop (for example 5% in a day, 10% in a week) that pauses betting. When it hits, you stop for the period you set.
5. Review. CLV over a few hundred bets, not last night's P&L. Sample size is in how many bets to know if you beat the close; ROI versus CLV is a separate page.

Log every bet before kick-off. Columns: bet tracking spreadsheet. The product tracker does CLV against Pinnacle's fair close: CLV tracker.

How Pinnacle Terminal handles this

Pinnacle Terminal does not size stakes. It supplies the two prices a stake rule needs from the sharp side: Pinnacle's live price and the fair price (Power devig by default), and it grades every pre-match move 0–10 (Big move, Sharp drop of about 20% or more from the recent high, Limit-backed, Downtrend, Rebound, Standard). Live in-play football drops are shown and filtered, without a grade. Take the fair price to the Kelly calculator or a 1% unit. The terminal does not accept bets and is not affiliated with Pinnacle. Product: Pinnacle dropping odds. Plans: pricing.

FAQ

What is bankroll management in sports betting?

Rules for a dedicated pot: how much of it you stake, how you resize after wins and losses, and when you stop. It does not find bets. It keeps a small edge, or no edge, from turning into an empty account through stake size.

What percentage of my bankroll should I bet?

Most disciplined value bettors use 1–2% of the current bankroll, or a quarter of Kelly with a 2–3% cap. Pinnacle's ruin piece (re-read 6 November 2026) treats much beyond 1% of starting bankroll as too large if you want ruin well below 1%.

Can bankroll management make a losing method win?

No. If every bet is −EV, smaller stakes only slow the decline. Sizing is not an edge. Edge is a price above a fair price; see what is a value bet.

What is risk of ruin?

The chance the bankroll hits zero, or a floor you set, before any edge shows. It is driven mainly by how many units deep the pot runs, not by how sure you feel about the next pick.

Should I use the Kelly criterion?

As a ceiling, yes, once you have a fair price you trust. As the actual stake, use a fraction: our 20,000-run table puts full Kelly's median drawdown at 60% on a true +3% even-money edge. Quarter Kelly is 18%.

How do I handle a long losing streak?

Keep the unit. Do not raise it to "get even". If a stop you wrote in advance is hit, stop. Review CLV, not the last ten results. Ten losses in a row at 51.5% showed up in 17.2% of our 500-bet runs.

Sources

• Drawdown table and 10-loss frequency: our simulation, 20,000 runs × 1,000 bets (drawdowns) and 20,000 × 500 bets (streaks), +3% / 2.00, computed 31 October 2026.
• Ruin table: gambler's-ruin (q/p)^N at p = 0.515, computed 6 November 2026.
• Leonard −3.5 prices: Pinnacle Terminal capture, 28 October 2026. Book price 1.75 is illustrative.
• M. Uhrín, G. Šourek, O. Hubáček and F. Železný, "Optimal sports betting strategies in practice: an experimental review", arXiv:2107.08827, 2021 (read 31 October 2026).
• Pinnacle, "Understanding and managing your risks of ruin", re-read 6 November 2026 (stakes not much beyond 1% of initial bankroll to hold ruin well below 1%).
• ValueBetFactory, "Bankroll Management for Sports Betting", published June 2025, updated 11 August 2026, read 6 November 2026 (1–2% conservative band; quarter Kelly).

Try it

Size each stake against Pinnacle's live fair price, with every pre-match move graded 0–10: start the Pinnacle Terminal 7-day free trial. Nothing is charged before day 8. Plans: pricing.

18+ · Bet responsibly. Only bet money you can afford to lose.