Law of large numbers
Average enough independent trials and the average settles on the true mean. The typical distance from the truth shrinks like one over the square root of the number of trials.
01The formula
- the result of trial i, here: 1 if trade i wins and 0 if it loses
- the running average after n trials, here: the win rate so far
- the true average, here: the true chance of a win p
- the standard deviation of one trial, √(p(1 − p)) for a win-or-lose trade
02See it move
03Where it is used
- Market makerRelies on it: each trade's edge is tiny and noisy, but over many thousands of trades the average edge shows up reliably.
- Quant researcherWorks out how many trades a backtest needs before its win rate or average return means anything.
- Retail traderLearns why a handful of trades says almost nothing about whether a strategy works.
04Why traders use it
It is why a small edge repeated many times becomes a business, and why a short track record proves so little. The √n tells you how much data you need: four times the trades for half the error.
05Worked example
Each trade wins with chance 55%. After 500 trades the orange trader's win rate is 54.4%. The 95% band around the truth is ± 4.4% wide at that point, and 90% of the 20 traders are inside it.
06How it is built
- The average has mean , because expectations add.
- For independent trials the variances add, so .
- The standard deviation of the average is therefore , which goes to 0 as grows.
- Chebyshev's inequality then shows the chance of being more than any fixed distance from goes to 0 too.
07The common mistake
The gambler's fallacy: thinking a run of losses makes a win more likely next. The average converges because new trials swamp the old ones, not because the coin corrects itself.
08In Python and LaTeX
import numpy as np
p, n = 0.55, 500
half = 1.96 * np.sqrt(p * (1 - p) / n)
print('95% band: +/-', round(half, 4)) # 0.0436\bar X_n = \frac{1}{n}\sum_{i=1}^{n} X_i \;\longrightarrow\; \mu \quad \text{as } n \to \infty, \qquad \text{typical error} \approx \frac{\sigma}{\sqrt n}09Where 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.