RTP and Volatility: What the Distribution Actually Looks Like

Return to player is quoted everywhere and understood almost nowhere. Volatility is understood even less, largely because it is described in words — "high", "medium", "low" — rather than in numbers. Both are properties of a single object: the probability distribution of what a game pays back per unit staked. This guide builds that distribution explicitly, derives the return-to-player figure and the standard deviation from it, and then answers the question the two numbers exist to answer: how long does it take for the house edge to outrun the noise?

RTP is a property of the paytable, not of your session

A game's return to player is the mean of its payout distribution. Take every possible outcome of one spin, multiply each payout by its probability, add them up, and that sum is the RTP. Nothing about it refers to time, sessions, or luck. It is an arithmetic property of the paytable and the reel weightings, fixed when the game was designed.

The house edge is its complement. A 96% game has a 4% edge, which means that across turnover T the expected cost is 0.04 × T. Ten thousand euro of turnover has an expected cost of 400 euro whether it happens over an evening or a decade.

What RTP emphatically does not describe is what happens to you. That is governed by the second moment of the same distribution, and the gap between the two is the entire subject of this page.

Building a paytable you can actually check

Published volatility ratings are qualitative, so rather than assume a figure, here is a fully specified distribution. Stake is one unit; the table gives the amount returned and its probability. This is a simplified model rather than any real title, but it is a legitimate distribution and every number below follows from it.

• Returns 0 — probability 0.6879
• Returns 1 — probability 0.18
• Returns 2 — probability 0.09
• Returns 5 — probability 0.02
• Returns 10 — probability 0.02
• Returns 100 — probability 0.002
• Returns 1,000 — probability 0.0001

The probabilities sum to 1.0000. The mean return is:

(1 × 0.18) + (2 × 0.09) + (5 × 0.02) + (10 × 0.02) + (100 × 0.002) + (1,000 × 0.0001)
= 0.18 + 0.18 + 0.10 + 0.20 + 0.20 + 0.10 = 0.96

So this game runs at exactly 96% return to player and a 4% house edge. Note where the return comes from: the two rarest outcomes, at probabilities of 0.2% and 0.01%, supply 0.30 of the 0.96 — just under a third of the entire return sits in events you will see roughly once in five hundred and once in ten thousand spins. That single observation explains most of what players find confusing about slots.

Where the standard deviation comes from

Variance is the mean of the squares minus the square of the mean. Squaring the payouts:

E[X²] = (1 × 0.18) + (4 × 0.09) + (25 × 0.02) + (100 × 0.02) + (10,000 × 0.002) + (1,000,000 × 0.0001)
= 0.18 + 0.36 + 0.50 + 2.00 + 20.00 + 100.00 = 123.04

Then:

Var = 123.04 − 0.96² = 123.04 − 0.9216 = 122.1184
σ = √122.1184 = 11.05 per unit staked

Look at the composition of that 123.04. The single 1,000-payout outcome contributes 100 of it — 81% of the variance comes from an event with a probability of one in ten thousand. The top prize is not a bonus feature sitting on top of the game; mathematically it is the volatility.

Over n spins, the mean scales with n and the standard deviation with √n. For one-euro spins on this game:

• 100 spins: expected net −4 euro, standard deviation 110.51 euro
• 1,000 spins: expected net −40 euro, standard deviation 349.45 euro
• 10,000 spins: expected net −400 euro, standard deviation 1,105.07 euro
• 100,000 spins: expected net −4,000 euro, standard deviation 3,494.54 euro

At a thousand spins the expected loss is 40 euro and the typical deviation from it is nearly 350 euro — the noise is roughly nine times the signal. Any conclusion drawn from a session of that length about whether a game is "paying" is a conclusion about the noise.

The same RTP at one eighth the volatility

Now a second distribution, deliberately built to the identical 96% return with far smaller prizes:

• Returns 0 — probability 0.454
• Returns 1 — probability 0.30
• Returns 2 — probability 0.20
• Returns 5 — probability 0.04
• Returns 10 — probability 0.006

Mean: 0.30 + 0.40 + 0.20 + 0.06 = 0.96. Identical. And the second moment:

E[X²] = 0.30 + 0.80 + 1.00 + 0.60 = 2.70
Var = 2.70 − 0.9216 = 1.7784, σ = 1.33

Same return to player, same 4% edge, same expected cost per unit of turnover — and a standard deviation of 1.33 against 11.05, a factor of 8.3. Over a thousand one-euro spins:

• High-volatility game: expected net −40 euro, standard deviation 349 euro
• Low-volatility game: expected net −40 euro, standard deviation 42 euro

Two games that cost exactly the same to play, producing experiences that have nothing in common. This is why "which has the better RTP" is only half a question.

How long until the edge outruns the noise

Here is the calculation almost nobody does, and it is the one that puts the other numbers in proportion. The expected loss grows linearly with n; the standard deviation grows with √n. So there is a spin count at which the edge finally equals one standard deviation. Setting h·n = σ·√n and solving:

n = (σ / h)²

For the high-volatility game: (11.05 / 0.04)² = 276.27² = 76,324 spins. For the edge to reach two standard deviations, four times that: 305,296 spins.

For the low-volatility game: (1.33 / 0.04)² = 33.34² = 1,112 spins, and 4,446 for two standard deviations.

Sit with the first figure. On the high-volatility game, you must play seventy-six thousand spins before the house edge is merely as large as one standard deviation of the outcome — that is, before losing becomes the clearly likelier description of where you are than not. At a leisurely two seconds a spin that is over forty hours of continuous play, and it is the point at which the edge is only just visible, not the point at which it has taken over.

The ratio between the two games is instructive too: 76,324 / 1,112 = 68.7, which is 8.3 squared. Since n scales with σ², a game that is eight times more volatile takes sixty-nine times as long for its edge to become apparent.

Two conclusions follow, pointing in opposite directions, and both are true. Short-run results tell you essentially nothing about a high-volatility game; and the edge is nonetheless real, unavoidable, and collecting the entire time.

What this means for bankroll and session length

The practical translation is that a bankroll has to be sized against the standard deviation, not against the expected loss. The expected loss on a thousand one-euro spins is 40 euro; a bankroll of 40 euro survives that session almost never.

Three numbers, decided before you start rather than during:

1. The amount you are content to lose in full. A ceiling, not a target.
2. Stake per spin. On a high-volatility game, a stake of roughly 0.5% of the session budget gives about two hundred spins of runway once the expected loss and a one-sigma downswing are both accounted for. On a low-volatility game the same budget stretches many times further.
3. Session length in spins, not minutes. Spins are the unit everything above is denominated in, and they are countable.

The same variance arithmetic governs bonus clearing, because a wagering requirement is simply a mandated quantity of turnover — the cost side of which is worked through in our guide to wagering requirement mathematics. And if a bankroll is held in a currency you do not spend, the exchange rate adds a second source of variance on top of the game's own, which is covered in multi-currency bankroll management.

Reading a game's disclosed figures

Two practical points close the loop between the arithmetic and what you can actually look up.

First, many suppliers ship a single title in more than one return-to-player configuration, and the operator chooses which to run. The authoritative figure for the version you are playing is the one in that game's own information or paytable screen, not a figure quoted in a review, including anything on our casino comparison. Over 10,000 euro of turnover the difference between 96% and 94% is 400 euro against 600 euro — a 200 euro gap for opening one screen.

Second, volatility is usually published only as a word, if at all. The proxy you can see is the paytable: the size of the top prize relative to the stake tells you where the variance is, because as shown above the largest payout typically dominates it. A game whose top prize is a thousand times the stake is a high-variance game regardless of what the marketing calls it. Jackpot titles are the extreme case of exactly this structure, which our jackpot comparison lists separately for that reason.

Everything on this page starts from a negative expectation and stays there. A higher return to player slows the rate of decline; it does not change its direction. Set deposit and loss limits before a session rather than after a bad one, and use the tools on our responsible gambling page if that stops being easy to do.

FAQ

Does 96% RTP mean I get 96 back for every 100 staked?

Only as the mean of the payout distribution over its full cycle. Over a thousand spins on the high-volatility game modelled above, the expected loss is 40 units while the standard deviation is 349 — so the average is a number almost no individual session resembles.

Do two games with the same RTP cost the same to play?

Yes, in expectation, per unit of turnover. The two distributions above both cost 4% of turnover. What differs is the spread of outcomes around that cost, which was 349 against 42 over the same thousand spins.

Why does the top prize matter so much to volatility?

Because variance weights payouts by their square. In the high-volatility model, the one-in-ten-thousand outcome contributes 100 of the 123.04 second moment — about 81% of the total variance from a single line of the paytable.

How many spins before the house edge is obvious?

For the high-volatility model, (σ/h)² = 76,324 spins for the edge to reach a single standard deviation, and over 305,000 for two. For the low-volatility model with the same RTP, 1,112 and 4,446. Short sessions are uninformative by construction, not by bad luck.