In one sentence
The central limit theorem says that when you add up lots of independent bets, the total profit follows a bell-shaped normal curve, whatever the odds of the individual bets.
How it works
A single bet has a lumpy outcome: you win a fixed amount or lose your stake. Add up hundreds of those lumpy results, though, and the total settles into a smooth bell curve. The centre of the bell is your expected profit and its width depends on how much each bet swings.
This is very useful. Once you know the bell's centre and width, you can say how likely a losing season is, or what range of results is normal, without simulating anything.
The law of large numbers tells you where results head. The central limit theorem tells you how widely they scatter on the way.
The maths
- S n: total profit after n bets.
- n: number of bets.
- μ: expected profit per bet.
- σ: standard deviation of profit on a single bet.
- z: how many standard deviations break-even sits below the expected total; look it up on the normal curve to get the chance of a loss.
In plain English: total profit is roughly bell-shaped, centred on n times your EV, with a width that grows with the square root of n.
Worked betting example
You back Over 2.5 goals 500 times, £10 a bet at evens (2.00), and each bet truly wins 52% of the time. Ignore commission for now.
- EV per bet = 0.52 × £10 − 0.48 × £10 = £0.40.
- SD per bet = £20 × √(0.52 × 0.48) = £9.99.
- Expected total = 500 × £0.40 = £200.
- SD of total = £9.99 × √500 = £223.43.
- 95% range = £200 ± 1.96 × £223.43, so from about −£238 to +£638.
- Chance of a loss: z = −200 ÷ 223.43 = −0.895, which gives about 18.5%.
The exact binomial answer for finishing strictly down is 17.4%, with a further 2.4% chance of exactly breaking even, so the bell-curve shortcut is close. With a 4% edge and 500 bets, nearly one bettor in five still shows a loss.
Where it's good
- Putting realistic ranges on expected season profit.
- Working out the chance of a losing month or year for a given edge.
- Building confidence intervals for a strategy's ROI.
- Comparing strategies with the same EV but different odds profiles.
- Quick sanity checks before running a full Monte Carlo simulation.
Limitations and pitfalls
- It needs enough bets. For a handful of bets, or longshots at 50.0 and above, the total is still lopsided and the bell curve understates the chance of a long drought.
- Bets must be roughly independent. Correlated positions (several correct scores in one match, many bets on one match) make the real spread wider.
- Staking must be fairly stable. Big changes in stake size or Kelly-style compounding break the simple formula.
- The tails are thinner than reality when edges or conditions change over time.
- It describes the range for a known EV; it does not tell you what your EV is.
- Commission changes both the mean and the width, so include it in the per-bet numbers.
How to build it
- scipy.stats.norm for the curve and scipy.stats.binom to check it against the exact answer.
- Data: your stake sizes, average odds and a realistic estimate of win probability.
- Practical tip: when your odds vary a lot, compute σ from your actual list of bets (each bet's own variance, summed) rather than using an average price.
Related methods
- Law of large numbers: where the bell curve's centre comes from.
- Normal distribution: the bell curve itself.
- P-values and confidence intervals: using this spread to judge skill.
- Expected value: the per-bet mean in the formula.
- Drawdown analysis: the swings you live through along the way.