Volatility and Variance: Why the Same RTP Can Feel Different

Volatility describes how uneven a sequence of gambling results can be. RTP describes a long-run average return. Those are different dimensions: two games can share the same theoretical RTP while producing very different patterns of wins, losses and balance changes.

This distinction matters because people experience sequences, not long-run averages. A smooth-looking sequence and a sharply fluctuating sequence can be built around the same expected value, yet feel nothing alike during a short session.

Volatility in plain English

In gambling product descriptions, volatility usually refers to the pattern and spread of outcomes:

• lower volatility generally means smaller prizes occurring more often
• higher volatility generally means larger prizes occurring less often

The UK Gambling Commission uses this distinction in its live RTP monitoring guidance. It also notes that volatility is commonly expressed using standard deviation. A description such as “low”, “medium” or “high” is therefore a broad category unless the provider also publishes the underlying calculation and assumptions.

Volatility does not state whether a particular session will win or lose. It indicates how widely results may move around the average and how those returns are distributed among outcomes.

Mathematical variance and industry volatility

Variance has a specific statistical meaning. Start with the expected value, measure how far each possible result lies from it, square those distances, and weight them by their probabilities. The result is the variance. Standard deviation is the square root of variance and expresses spread in the original unit of measurement.

In formula form, for possible results x with expected value μ:

Variance = Σ probability(x) × (x - μ)²

Industry use of “volatility” is less precise. It may refer to variance, standard deviation, prize frequency, prize size, or an internal category based on several features. That does not make the label useless, but it means a high-volatility badge is not a universal numerical standard across providers.

Expected value and dispersion answer different questions

Expected value asks: what is the average net result per wager in the model?

Variance asks: how far can individual results spread around that average?

Consider two simplified £1 games. Each has an expected return of £0.90, or a net expected value of -£0.10 per play.

Game A

• 90% chance of receiving £1
• 10% chance of receiving £0

Game B

• 10% chance of receiving £9
• 90% chance of receiving £0

Both return an average of £0.90 per £1 in the mathematical model. Their experience is very different. Game A usually returns the stake and occasionally loses it. Game B usually returns nothing and occasionally produces a much larger payment. Game B has the wider dispersion of outcomes.

This example is illustrative rather than a description of a real product. It shows why RTP alone cannot communicate how results are distributed.

Why games with similar RTP can feel different

Theoretical RTP compresses the entire prize model into one average percentage. It does not reveal:

• how often any prize occurs
• whether a prize is smaller or larger than the stake
• how much of the theoretical return sits in rare outcomes
• the likely size of balance swings over a limited number of plays
• the longest possible or probable run without a particular prize

As a result, similar RTP figures can coexist with very different hit patterns. A game returning many amounts close to the stake may preserve a balance for longer in some sequences. A game concentrating more return in rare outcomes may show repeated losses interrupted by occasional sharp increases. Neither pattern changes the long-run average by itself.

Our guide to RTP explains why that average is not a personal return promise. The house-edge guide covers the corresponding expected-loss percentage.

Sample size and convergence

Theoretical RTP is established from a game's probability model. Actual RTP is calculated from observed prizes divided by observed turnover. The two need not match over a short sample.

The Gambling Commission's guidance explains that a volatile game has a wider expected range of actual RTP over a limited number of plays. As the quantity of play increases, the acceptable statistical range becomes narrower and actual RTP is expected to move closer to theoretical RTP. This is a statement about aggregate statistical behaviour, not a guarantee for a particular player or a fixed number of bets.

Three qualifications matter here. A larger sample reduces the influence of individual unusual results on the average, but it does not eliminate randomness. Convergence does not require every short block of play to resemble the theoretical percentage, nor can a player infer that the next outcome must compensate for an earlier run.

A useful way to see the first point is to compare one rare prize with the sample around it. In 20 plays, one large result can dominate the observed return. In two million plays, the same single result has far less influence. The larger sample is more stable because each individual outcome carries less weight.

Streaks and clustering can occur naturally

Independent random events do not have to alternate neatly. Several losses can occur in a row, as can several wins. Random sequences often contain clusters that look meaningful even when each event is generated under unchanged probabilities.

Suppose a simplified event has a 50% probability on every independent trial. The probability of three specified losses in a row is:

0.5 × 0.5 × 0.5 = 0.125, or 12.5%.

That sequence is less common than one loss, but it is not evidence that the process changed. Nor does completing the three-loss run alter the probability of the fourth trial: it remains 50% under the stated independent model.

Real games may have more complicated outcome structures, but the same principle applies where outcomes are independent. The Gambling Commission states that for random machines, the chance of winning on the next game remains constant and is not affected by previous wins or losses.

Where a game uses independent random outcomes, the RNG does not change the next result to compensate for an earlier streak. Our RNG guide explains the technical and regulatory position.

The gambler's fallacy

The gambler's fallacy is the mistaken belief that an outcome becomes more likely because the opposite outcome has occurred repeatedly. After a run of losses, this can appear as “a win is due”. After a run of wins, it can appear as an expectation that a loss must immediately restore balance.

Peer-reviewed research describes this belief in the context of independent events with fixed probabilities. The error is not noticing that long-run proportions can stabilise without the next event correcting the past. A future sequence can bring the overall average closer to expectation while every new trial still begins with the same probabilities.

This is also why a visible streak is not, by itself, a sound basis for increasing a stake. It describes what already happened, not a change to the underlying odds.

What volatility can mean for a session

Higher volatility can produce wider balance swings and longer runs without the outcomes that carry much of the return. Lower volatility can produce smaller, more frequent returns, but still has losing outcomes and a negative expected value when a house edge applies.

For a fixed starting balance, stake size affects how many losses can be absorbed. Larger stakes expose more money to each outcome and can shorten a session after an adverse run. Smaller stakes may allow more plays, but more plays also create more total wagering and therefore more exposure to the game's expected loss. Neither changes a negative expected value into a positive one.

“Bankroll” in this mathematical context simply means the amount allocated to a sequence of wagers. It should not be read as an investment concept or as a method for making gambling profitable. Volatility can help describe the range of possible experiences, but it cannot make the next outcome predictable.

A practical comparison checklist

When reading a volatility claim, ask:

• Is the label supported by a number or only a category?
• Is the comparison between games from the same provider and scale?
• Does the published RTP assume a particular configuration?
• Are rare prizes responsible for a large part of the theoretical return?
• Is the claim about a mathematical model or a short sample of observed results?

You can use the Outcome Volatility Visualiser to explore how simulated sequences can spread around an expected value. It is an educational illustration, not a predictor of real results.

The central point is simple: RTP locates the long-run average, while variance describes dispersion around it. Short sessions can land almost anywhere within the range permitted by the game, and a streak does not make its opposite due.