RTP vs house edge: what the percentages measure

RTP is an average return, not your chance of winning

Return to player, or RTP, describes a theoretical average under specified game rules and assumptions. A 96% RTP is not a 96% chance of winning, a promise to return 96% of your starting balance, or a forecast for one session. First identify the money counted as stakes and the money counted as payouts.

The amounts returned include any stake that comes back; they are not just profit. House edge describes an expected retention on a stated basis. The familiar calculation, 100% minus RTP, works only when the two measures use matching amounts and assumptions. It does not turn an explicit fee, a bookmaker’s overround or a past result into the same measurement.

The UK Gambling Commission’s consumer guide explains RTP as an average over many plays, rather than a return achieved each time. Its guidance concerns particular gambling mechanisms in Great Britain. The arithmetic below is fictional education, not a recommendation to choose a game or keep playing.

Name what is returned before using a percentage

Suppose one attempt costs a stake of 1 unit. Use “gross payout” for everything returned after that attempt, including any returned stake. Net profit or loss is that payout minus the stake, before any separate costs.

“Win” can mean receiving a prize, receiving more than the stake, or finishing a session ahead. Those are different events. A percentage needs a named numerator and denominator before it can answer any of them.

When does house edge equal 1 − RTP?

House edge equals 1 − RTP only when both use the same complete payout, stake base, rules and assumptions. Here, take a fixed positive stake, include every possible outcome and count gross payouts on a stake-inclusive basis. Any strategy assumptions must also match. These are conditions of the calculation, not facts about every product carrying an RTP label.

With a fixed stake of 1 unit, an expected gross return of 0.96 units leaves an expected net result of −0.04 units. On that same basis, RTP is 96% and expected house retention is 4%. The 0.04 is an average loss per staked unit in the model, not a maximum session loss or an extra fee to subtract again.

This retention is not the operator’s accounting profit after its operating costs. Nor can the identity be transferred unchanged to a payout figure that excludes returned stakes, an incomplete outcome list or a different strategy. The Commission’s RTS 3 information requirements distinguish rules, payouts and likelihood information, including strategy assumptions where relevant. These requirements do not establish worldwide legal permission to use a product.

The same 96% RTP can hide different outcomes

Consider two invented distributions. Each attempt has a stake of 1 unit, exactly one listed outcome occurs, and the probabilities in each model sum to 100%. There are no fees, taxes, bonuses, refunds or other payouts. The figures are stipulated, not measured from a real game.

For A, the weighted gross payout is 0 × 0.40 + 1 × 0.40 + 2.8 × 0.20 = 0.96 units. For B, it is 0 × 0.90 + 9.6 × 0.10 = 0.96 units. Both therefore have the same 96% RTP on the specified 1-unit stake.

In A, the 1-unit payout counts as receiving something, but it only restores the stake. That is why the 60% payout frequency differs from the 20% chance of profit. B has rarer, larger payouts. The same mean does not describe that difference in spread or identify a preferable game. Neither model promises a particular result after any fixed number of attempts. For the general weighted-mean calculation, see expected value and probability.

Re-staked prizes change the denominator

The Commission’s monitoring definitions count all stakes in turnover, including winnings staked again. Turnover therefore differs from an initial deposit. Its term “win” means prizes awarded, rather than the player’s profit.

Here is one possible path through model A. Start with 1 unit, make no further deposits or withdrawals, and use no credit or separate fees. Each next stake is paid from the balance left by the preceding row.

Total stakes are 3 units; total gross payouts are 3.8 units. The observed payout-to-stake ratio is 3.8 ÷ 3 = 126⅔%. The closing balance is 1.8 units, or 0.8 above the starting balance. Dividing prizes by the initial 1-unit deposit would instead give 380%; that is not this observed RTP. These ratios describe different amounts, even though they come from the same ledger.

An observed return is a different number

The ledger’s 126⅔% is a description of that small, selected path. It neither changes the fictional model’s theoretical 96% nor estimates a reliable advantage. A path with losses could give a very different observed percentage.

The Commission’s monitoring guidance compares actual payouts and turnover with designed RTP, taking volatility and the amount of play into account. A short observation alone cannot establish that a game is defective. This does not tell a player when a payout is due or justify continuing until results resemble an average.

Do not assume every mechanism has independent, unchanged probabilities. The consumer guide distinguishes random and compensated machines. A history of losses is not a general signal for a forthcoming win; the relevant design and rules still matter.

Add the stakes, not the percentages

Different stake sizes create another denominator problem. In this separate invented two-row record, the payouts are observations, not a probability model or evidence about a real game.

The total ratio is 1 ÷ 10 = 10%. Averaging 100% and 0% equally gives 50%, because it weights the two rows equally instead of weighting the money staked. For a payout-to-total-stakes measure, add the matching payouts and stakes first. Neither two-row calculation supplies a theoretical RTP.

Keep extra costs and other margins separate

Suppose a separate cost of 0.02 units applies outside the 1-unit stake and is not already reflected in the model’s payouts. The expected net cash result is 0.96 − 1 − 0.02 = −0.06 units. Count that cost once. The model’s base RTP remains 96%; 0.96 ÷ 1.02 = 16/17 compares expected gross payout with a different denominator, total cash outlay. It is not the advertised RTP or the original house-edge measure.

A bookmaker’s overround comes from quoted prices across outcomes. Realized hold concerns what was actually retained over a stated period. An explicit fee is a charge on its stated base. None becomes a synonym for theoretical house edge merely because it is written as a percentage. See the odds and overround guide for that price-book calculation, rather than applying an RTP formula to it.

Common questions

Does 96% RTP mean a 96% chance of winning?

No. Receiving any prize and finishing with a profit are different events. At stake 1, a return of 1 restores the stake and leaves net profit at 0. Models A and B both average 0.96 units back per unit staked, but their chances of positive net profit are 20% and 10%, respectively.

Will I receive 96% of my deposit back?

No. The displayed theoretical percentage does not forecast your session or closing balance. Deposits, total stakes, gross payouts and final withdrawable funds are different quantities. Restaking can make total stakes exceed the original deposit.

Can recent payouts show when to play?

No timing rule is established here. A recorded payout ratio describes a chosen past period, and the selection, amount of play and rules affect its interpretation. It is not a live signal, a promise of reversal or a way to recover losses.

Does understanding these percentages establish that a product is suitable?

No. A mathematical explanation does not establish legal permission, age or account eligibility, product availability or affordability. You can lose the entire amount staked and any separate costs. Choosing not to play is a valid outcome of reading this guide.

Sources and scope

The UK Gambling Commission sources explain RTP, turnover, monitoring and specified British technical requirements. Penn State supports the expected-value definition used in the arithmetic. All distributions, balances, costs and stake records here are invented examples, not data from a game, a return forecast or evidence of product quality. Source access dates are not publication dates; jurisdiction and extraction limits are recorded with each source.

1. UK Gambling Commission consumer RTP guide
2. UK Gambling Commission RTS3 rules and likelihood
3. UK Gambling Commission actual RTP calculation
4. UK Gambling Commission monitoring key terms
5. Penn State STAT414 Lesson08 Mathematical Expectation