Central limit theorem
Add up many small independent random pieces and the total follows a bell curve, whatever the shape of the pieces. Its centre is n times the mean and its width grows like √n.
01The formula
- the sum of n independent random steps, here: how many times a ball bounces right
- the mean and standard deviation of one step
- the standard normal distribution, the bell curve
- converges in distribution: its histogram looks more and more like the bell curve
02See it move
03Where it is used
- Risk managerTreats a portfolio's daily P&L, the sum of many positions' moves, as roughly normal to estimate the chance of a large loss.
- Quant researcherBuilds confidence intervals and t-tests on average returns, which work because the average of many returns is close to normal.
- Options deskModels a log price over a month as normal because it is the sum of many small daily log returns.
04Why traders use it
It is why the normal distribution shows up everywhere in finance, and why statistics on averages work even when single returns are not normal. Knowing its limits matters just as much: fat tails and dependence slow it down.
05Worked example
A ball falls through 11 rows of pegs and bounces right with chance 50% at each one. Its bin has mean 5.50 and spread 1.658. The chance it ends in bin 3 or lower is exactly 11.33%; the bell curve gives 11.39%.
06How it is built
- Each bounce is a step that is 1 (right) with chance and 0 (left) otherwise, so and .
- The bin is the sum , with mean and variance , because means and independent variances add.
- Standardise: has mean 0 and variance 1 for every .
- The theorem says its distribution approaches , so , with correcting for whole-number bins.
07The common mistake
Using it for tail events. The theorem is good near the middle, but far out in the tails, where crashes live, real sums of returns are often much fatter than the normal curve says, especially when the pieces are dependent.
08In Python and LaTeX
import numpy as np
from scipy.stats import binom, norm
n, p, k = 11, 0.5, 3
mu, s = n * p, np.sqrt(n * p * (1 - p))
print('exact', round(binom.cdf(k, n, p), 4), 'normal', round(norm.cdf((k + 0.5 - mu) / s), 4)) # 0.1133 0.1139\frac{S_n - n\mu}{\sigma\sqrt{n}} \;\xrightarrow{d}\; N(0, 1), \qquad S_n = X_1 + \dots + X_n09Where it fits
Rather not code the maths yourself? Describe your trading idea and let AI build the bot.
Build a bot with AgenKitThe worked example and the model run the same code, so every number matches. Models simplify markets on purpose. This is education, not financial advice.