NBA Prop Bankroll Management for UK Bettors
Available in US
The Stake Size That Survives Bad Months
Two seasons ago I had a stretch where I lost 14 of 19 bets. The bets were not bad — my closing-line value was positive across the run, and the model that selected them later finished the season profitable. The streak was just variance doing variance things. I survived it because my unit size was small enough that the run did not threaten my ability to keep betting. Three months earlier, with a larger unit, the same streak would have wiped out two-thirds of my bankroll.
That experience is the most useful teacher in this whole exercise. Prop bettors who do not respect variance get eliminated by it. The OddsIndex strategy team recommends staking 0.5-1% of bankroll per prop bet, and I think that range is roughly right for most retail bettors with modest edges. A 100-bet sample of 2% true edge bets, staked at 1% of bankroll, has a comfortable expected return — but the path through that sample can include 20-bet losing streaks that swallow 20% of bankroll if your sizing is wrong.
Defining Your Bankroll Honestly
The first arithmetic that matters has nothing to do with prop pricing. It is the question of what your bankroll actually is. The answer is not the balance in your operator account. It is the amount you have set aside specifically for betting, separated from money required for rent, bills, or any obligation that produces real-life consequence if it does not arrive.
A bankroll of £500 in a separate account, funded once at the start of the season and not topped up from current income, is a real bankroll. A bankroll of £500 that gets refilled from each month’s wages is a substitute for income, not a bankroll, and the maths of unit sizing breaks down because the supply is functionally unlimited. The discipline of treating bets against a finite resource is what makes the sizing logic work.
Set the bankroll once. Top it up only on a planned schedule — quarterly, semi-annually, never reactively after a losing streak. The reactive top-up is the most common pathway from controlled bettor to problem bettor, because it transforms losses from learning data into account-balance noise.
Unit Size as a Function of Edge
Unit size should scale with confidence in the edge, not with confidence in the outcome. A 1% bankroll unit on a 2% edge bet is appropriately conservative. The same unit on a 5% edge bet leaves expected value on the table. The bigger the underlying edge, the bigger the unit can be without compromising the long-run survival profile.
The Kelly criterion is the formal tool here. Kelly recommends a stake equal to the edge divided by the odds, expressed in decimal terms. A 3% edge at decimal odds of 2.00 implies a Kelly stake of 6% of bankroll. That sizing is mathematically optimal for long-run bankroll growth, but the variance it produces is brutal. Full Kelly draws down 50% of bankroll roughly once per decade for a bettor running 3% edge.
Most professional bettors use fractional Kelly — half, quarter, or even tenth of the Kelly recommendation. Quarter Kelly on a 3% edge at decimal 2.00 gives a 1.5% stake. The trade-off is slower bankroll growth in exchange for much lower drawdown risk. Quarter Kelly is where I have settled, and the variance is still substantial.
Tracking Closing-Line Value Instead of Win Rate
Win rate is the wrong primary metric. A 55% win rate at -110 is hugely profitable; a 55% win rate at -150 is breakeven. The price you got at the moment of betting is what matters, and the cleanest way to track that is closing-line value (CLV).
CLV measures the difference between your bet price and the price the same line closed at. Beating the closing line consistently is the signature of a winning bettor. A bettor running positive CLV will be profitable over the long run with high confidence; a bettor running negative CLV may have a hot stretch but will revert.
One published 2025-26 NBA prop-tracking model logs closing-line win rates of 55.7% on points, 63.2% on threes, 69.9% on blocks, 61.9% on steals and 54.7% on PRA. Those numbers are not raw win rates — they are CLV outcomes, the percentage of bets that beat the closing line. A model that produces those CLV numbers will translate into long-run profit even when the raw win rate on any given month is lower.
The Ledger Format I Actually Use
The bet log is the single most important piece of infrastructure in this whole exercise, and most bettors keep one for two weeks before abandoning it. The format that has worked for me is minimal: date, sport, market, line, stake, price, projected probability, closing price, settlement, profit/loss. Nine columns. A row per bet.
The projected-probability column is the discipline-builder. Filling it in forces me to articulate the edge before I bet, in numerical terms. A bet I cannot project a probability for is a bet I should not place. The closing-price column is the post-game check on whether the line moved in my direction or against it. CLV is the difference, computed at month-end across all bets.
Bets without ledger entries do not exist for my season-end review. I treat the unlogged bet as if it never happened, which keeps the temptation to log only winners under control. Most months, my logged bets are profitable on CLV by 1-2 percentage points, and the raw win rate fluctuates between 49% and 58% depending on variance. The CLV stays steady; the win rate breathes.
Bet Sequencing and Same-Game Exposure
Bankroll discipline extends beyond per-bet sizing into the question of how many bets are open simultaneously. Three independent bets each at 1% of bankroll have aggregate exposure of 3%, which is fine. Three correlated bets each at 1% of bankroll have aggregate exposure closer to the worst-case single-bet outcome, which can be much higher than 3%.
Same-game correlation is the most common trap. Betting a player’s over on points, his teammate’s over on assists, and the team total over on points is three bets, but it is essentially one bet on the team scoring above expectation. If the team underperforms, all three lose. The effective stake is much closer to 3% than 1%.
I cap same-game exposure at 2% of bankroll across all related bets. Above that the variance compounds in ways that look ordinary on the surface but break in a losing run. Variance is patient. Cumulative same-game exposure is what makes 14-of-19 losing streaks possible. For the deeper frame on how to size individual prop bets relative to their fair price, see my walkthrough of prop implied probability and vig.
Drawdown Tolerance and the Stop-Loss Discipline
Every bettor has a drawdown number that breaks them. For some it is 20% of starting bankroll; for others, 50%. The number is psychological more than mathematical, and it is worth knowing in advance because variance will test it.
My rule is to reduce unit size by half when bankroll has dropped 25% from peak. The half-stake regime continues until bankroll recovers to 90% of peak, at which point full sizing resumes. The rule looks conservative, but it preserves the ability to keep betting through a deep drawdown without going broke. A bettor who tilts up his unit size during a losing run amplifies variance in the wrong direction; the half-stake rule does the opposite.
The stop-loss is the absolute floor. If bankroll drops to 50% of starting balance, I stop betting entirely until I have reviewed every losing bet in the period and identified what went wrong. Sometimes it is the model. Sometimes it is the bet selection process. Sometimes it is just variance. In any case, the pause is mandatory. Continuing to bet through a stop-loss trigger is how recreational bettors become problem bettors, and the regulatory environment in the UK — including the protections built into licensed-operator deposit limits — exists precisely to interrupt that pathway.
Articles
Created by the "HoopMargin" editorial team.
