GT99 Variance in Gambling: Why Short Runs Mislead

The coin-flip model

Flip a fair coin ten times. Getting seven heads is not unusual; it happens in about one case in six. Flip it ten thousand times and the share of heads will sit very close to half. The average has not changed, but the size of the sample has, and the relative spread shrinks as the sample grows.

Gambling works the same way. The edge is a long-run property. A session of a hundred bets is the equivalent of ten flips: dominated by noise.

Volatility versus return

These are different dials, and mixing them up is the commonest conceptual error.

A worked comparison

Imagine two games with the same long-run return, illustration only. Game A pays small wins often. Game B pays rarely but in large amounts. Over 100 bets, Game A will usually end near the average, with a gentle drift. Game B will often end well below the average, with an occasional session far above it.

Both have the same expected loss, yet the experience differs: Game B can empty a small budget before its large payouts appear. Hence the practical rule, to match volatility to budget. A small budget fits low-volatility games, because it lasts long enough to approach the average.

Why streaks look meaningful

Randomness produces clusters. In a long sequence of independent results there will be runs of wins and losses, and people notice them. Our minds search for causes, such as luck, a hot machine or a due result. These are explained away in gambling myths, tested against the maths. The streak is the output of variance, not evidence of a pattern.

Sizing stakes sensibly

A rough working method: divide your session budget by the stake to get the number of bets. For high-volatility games you want that number large, often several dozen at the least, so that a typical dry run does not end the session. If the number is small, reduce the stake. This does not change the edge, but it improves your chance of seeing what the game does before the budget is gone.

See reading RTP the right way for where to find the return figure, and keep records using the spending tracker.

What variance cannot do

It cannot rescue a negative expected return. A game with a house edge loses in the long run, and variance only decides how the losses are scattered. A lucky session is real, but it is not evidence that the edge has gone. The only reliable result of long play is the average, and the average favours the house.

Consider a numerical picture, with invented figures. Suppose a bet loses 1 unit with probability 0.52 and wins 1 unit with probability 0.48. The expected result per bet is minus 0.04. After 100 bets the expected total is minus 4, yet the typical swing around it is about 10 units either way, so finishing ahead by a few units is entirely ordinary. After 10,000 bets the expected total is minus 400, while the typical swing is about 100, so the loss now dominates. The edge was constant; only the sample changed.

This is why a player who wins for a week has learned very little, and why a player who loses for a week has not proved anything about the game either. Judge over long samples or not at all.