RTP vs House Edge vs Volatility: What Casino Percentages Mean

RTP describes expected prizes relative to stakes. House edge describes the corresponding expected loss. Volatility describes how widely results can vary. None of these numbers promises what one person will receive. This reference separates those concepts, shows the arithmetic and explains which claims the numbers cannot support.

The examples below are our own simplified mathematical models, not measurements of real casinos. Regulatory references are from Britain’s Gambling Commission; they are identified as British guidance rather than universal legal requirements.

RTP, house edge and volatility compared

The Commission’s technical glossary distinguishes designed RTP from observed RTP and commonly represents volatility using standard deviation. A marketing label such as “high” does not, by itself, specify the numerical spread or complete payout distribution.

The formulas, with the denominator made explicit

Actual RTP (%) = total prizes ÷ total stakes × 100. The Commission’s calculation guidance uses this win-to-turnover relationship. “Prizes” here means returned amounts, not net profit. A payout of 0.50 units on a one-unit stake is still a loss of 0.50 units.

House edge (%) = 100% − theoretical RTP (%), provided both measures use the same rules, expected payouts and stake denominator. In a model with 96% RTP, the corresponding edge is 4%. On 250 units of total stakes, the expected prizes are 240 units and expected loss is 10 units. Those are expectations, not a prediction that an observed account will contain those amounts.

This complement can be misleading when figures use different assumptions. A percentage based on initial stakes is not automatically comparable with one based on all stakes, and results under one set of rules should not be attributed to another without evidence.

Why deposits and turnover are different

The Commission’s definition of turnover includes stakes made using previously returned winnings. Consider this invented accounting record; it is an illustration, not a suggested pattern of play:

1. An account starts with 10 units. A 10-unit stake returns 6 units.
2. A subsequent 6-unit stake returns 3 units. The account then contains 3 units.

Total stakes are 16, total prizes are 9, and actual RTP for these two transactions is 9 ÷ 16 = 56.25%. The balance fell by 7, or 70% of the original 10 units. There is no contradiction: the percentages have different denominators. The accounting identity is starting balance + prizes − stakes = ending balance: 10 + 9 − 16 = 3.

Same 96% RTP, very different outcomes

Here are two invented one-unit-stake models. Every listed payout includes any returned stake; an unlisted outcome pays zero. These distributions exist only to illustrate the distinction between a mean and a spread.

The expected payout is the sum of each payout multiplied by its probability. Variance is the sum of squared payouts multiplied by their probabilities, minus the squared mean. Thus A has variance 0.80 × 1.20² − 0.96² = 0.2304. B has variance 0.08 × 12² − 0.96² = 10.5984. Taking square roots gives the standard deviations shown above. Subtracting the constant one-unit stake changes the mean but does not change the variance.

Both models have identical average returns, yet their possible outcomes are distributed very differently. This comparison is a demonstration of probability, not a ranking or a recommendation. Neither model provides a way to avoid the expected loss.

Why a short record cannot certify the advertised RTP

An observed return is a statistic about a particular sample. Its distance from a theoretical value cannot be interpreted from the percentage alone. The Commission’s monitoring guidance explains that an assessment needs both volume of play and volatility-dependent tolerances. There is no single number of plays that establishes every game’s performance.

For example, a reported 110% actual RTP tells us that prizes exceeded stakes in the measured data. Without the game model, period, dataset size and relevant uncertainty, that figure cannot establish that the designed RTP is 110% or that future customers will receive the same return. A disappointing individual result likewise does not, on its own, establish a software defect.

Does a losing streak mean a win is due?

For random machines described in the Commission’s consumer guidance, previous wins and losses do not change the current game’s win probabilities. The same guidance distinguishes compensated machines, which can operate differently. The product and rules matter; independence should not be asserted about every gambling product.

In a model with independent rounds and unchanged probabilities, past losses create no mathematical debt that the next round must repay. A long-run expectation is not a schedule, a promise of recovery or a reason to continue gambling. No amount of additional play guarantees an individual return.

A verification checklist for readers and writers

• Identify the exact product: record the game, version and applicable rules rather than copying a percentage attached only to a title.
• Label the measure: distinguish theoretical RTP from an observed return.
• State the basis: identify what counts as a stake and payout, and whether special features or jackpot components are included.
• Describe the dataset: for actual RTP, give the period, total stakes, prizes and number of plays where available. Say when these details are unavailable.
• Keep uncertainty visible: do not infer fairness or malfunction from an unexplained percentage difference.
• Separate other claims: RTP does not measure withdrawal reliability, licence status, bonus restrictions or whether gambling is affordable for an individual.

Short answers to common questions

Is 96% RTP a 96% chance of winning? No. In our examples, both models have 96% RTP but positive-payout probabilities of 80% and 8%. An average amount and the probability of an event are different quantities.

Does 96% RTP mean losing exactly 4% of a deposit? No. The model’s edge applies to the specified stake basis. Deposits and accumulated stakes are different quantities, and actual results vary.

Does lower volatility remove the house edge? No. Our two models have different standard deviations and the same 4% edge. Dispersion and expected loss describe separate properties.

Method and scope

Definitions were checked against the official sources below on 31 October 2026. All worked examples were calculated for this article, with complete probabilities stated so the results can be reproduced. No operator results, player transactions or proprietary test data were analysed. The examples omit fees, bonuses and taxes. This is statistical education, not a certification of any game or advice on how to win.

• Gambling Commission: definitions of theoretical RTP, actual RTP, volatility, turnover and win
• Gambling Commission: consumer explanation of machine payouts, randomness and house edge
• Gambling Commission: sample volume and volatility-dependent monitoring tolerances
• Gambling Commission: how actual RTP is calculated

Official sources support the definitions and scoped regulatory explanations. Numerical models, accounting examples and the verification checklist are original explanatory work by Affiliates Directory; they are not findings attributed to the regulator.