Volatility, in depth

Two games can share the same RTP and feel nothing alike. Volatility is the width of the ride, not the height of the return — and the single-number “volatility index” that vendors publish hides most of what that width actually contains.

Two slots can advertise the identical return-to-player and deliver completely different experiences: one drips small wins steadily, the other pays almost nothing for a long stretch and then, rarely, a great deal at once. Both hand back the same fraction of stakes over the long run. What separates them is volatility — the spread of outcomes around that average. This chapter is about that spread: what it is, how the industry tries to measure it, and why the single number it publishes conceals more than it reveals.

What volatility means for RTP

RTP fixes the centre of the outcome distribution — the mean return per unit staked. Volatility fixes its width. Formally it is the standard deviation of a spin’s return, denominated in bet units: a low-volatility game has a small standard deviation (returns cluster near the mean), a high-volatility game a large one (returns scatter widely, with most spins below the mean and a few far above it). Changing volatility slides no part of the mean — the return figure is untouched — it only stretches or compresses the distribution around it. That is the whole idea, and it is why “high volatility” must never be read as “pays more.” It pays the same, through a different-shaped ride.

This is the honest core of the chapter, so it is worth stating plainly: variance is not opportunity. A wider distribution puts more probability far above the mean, but it does so by putting more probability below it too, and the expected value stays exactly where the RTP left it — below breakeven at every volatility. The gap between RTP and your actual returns is precisely this spread; volatility is its name.

How it’s measured

The practitioner tool is the standard deviation of per-spin returns, sometimes repackaged as a “volatility index” or a star rating. A studio can compute the standard deviation directly from a game’s own pay table and symbol weights (the PAR sheet), because those fix the exact probability of every outcome; academic analyses of real machines have done the same from reverse-engineered PAR data. The catch is that there is no single, standardised, cross-vendor formula for the published index: the underlying quantity — the standard deviation — is well defined, but the scale it is squeezed onto, the confidence level or session length it assumes, and where the rating boundaries fall are all conventions that differ from one studio or reviewer to the next. Treat the standard deviation as solid and the branded index number as an indicative summary, not a measured universal constant.

The explorable below makes the effect concrete. Fix an RTP, then change the volatility class and the session length, and watch what happens to the distribution of a session’s average return. By the Central Limit Theorem the realized average over N spins is approximately Normal, centred on the RTP with standard deviation σ/√N — so more spins narrow the ride, and higher volatility widens it, while the centre never budges.

Two things stand out as you move the controls. First, at the same RTP a jackpot-heavy game’s 95% band is enormously wider than a low-volatility game’s — same edge, wildly different session outcomes. Second, lengthening the session shrinks every band and pulls the probability of ending ahead down toward zero, because the average return converges on the RTP, which sits below breakeven. Neither the standard deviation nor the calculator ever moves the mean above 100%.

Why the published indices hide as much as they show

Here is the payoff, and the reason to distrust a lone volatility number. A single index collapses an entire distribution to one figure, and at least three important things fall through the gaps.

Heavy tails. The Normal model in the explorable is a teaching device: it shows the shape of the volatility effect, but real slot payouts are heavy-tailed — the return is concentrated in rare, very large hits — so the true distribution has far more weight in its extreme tail than any bell curve. A standard-deviation-based index registers that a game is “spread out” but not how the spread is shaped, and it is the shape of the tail that a player actually feels. Hit frequency. Two games can share a volatility rating while one pays something on a third of spins and the other on a twentieth; the rating alone cannot separate them. The maximum-win cap. A capped top prize truncates the very tail that volatility is supposed to describe, so a headline number computed with or without the cap can differ materially — yet the published figure rarely says which. Collapse all of that into one star rating and you have discarded most of what makes two same-RTP, same-rating games feel different.

What you can and can’t infer

A volatility rating is genuinely useful for one thing: setting rough expectations about the texture of play. A high rating fairly warns that most sessions will run below the average with occasional large swings; a low rating that outcomes will cluster more tightly. Used that way, alongside the variance chapter, it is a legitimate planning aid — bankroll and session length feel very different at σ = 2 than at σ = 30.

What it cannot do is anything about a specific spin. Because spins are independent, a rating carries no information about timing, streaks, or a win being “due” or a game being “hot” — those ideas are the gambler’s fallacy, not properties of the distribution. It cannot be compared reliably across vendors, because the scales differ. And it never changes the expected value: no volatility setting turns a below-breakeven game into a favourable one. If you want to see how large a data set it would take to pin a return figure down at a given volatility in the first place, the verification chapter turns that same σ into a spin count — and the numbers are sobering.

Common questions

Sources (3)

• NIST/SEMATECH e-Handbook of Statistical Methods — Normal Distribution (mean μ, standard deviation σ)itl.nist.gov
• NIST/SEMATECH e-Handbook of Statistical Methods — Cumulative Distribution Function of the Standard Normal Distribution (Φ)itl.nist.gov
• Harrigan, K. A. & Dixon, M. (2009). PAR Sheets, probabilities, and slot machine play — Journal of Gambling Issues, 23doi.org

Education, not advice. This chapter explains how return-to-player is defined, computed and checked so you can read the number honestly. It is not a system, and nothing here treats gambling as a way to make money — over enough play the mathematics favours the house. 18+.