RTP tells you the game’s average return over a very large body of play. Volatility tells you how widely individual results can spread around that average. One is a long-run mean; the other is a description of dispersion and prize shape.
That distinction explains a common slot contradiction: two games can both advertise 96% RTP, yet one produces frequent small returns while the other produces long losing stretches interrupted by rare large awards. The games have the same expected return per unit wagered in this simplified comparison, but they do not create the same bankroll experience.
The shortest correct comparison
| Measure | Main question answered | What it does not answer |
|---|---|---|
| Theoretical RTP | How much is expected to be returned per unit wagered over the game’s full mathematical cycle or model? | What will happen in one visit |
| House edge | What percentage of total action is retained on average? | How quickly results will swing |
| Volatility | How dispersed are the possible returns? | Whether the average return is favorable |
| Hit frequency | How often does any defined winning result occur? | The size or profitability of those wins |
| Actual RTP | What percentage was returned in a measured sample? | Whether the next spin is more likely to win |
RTP and volatility should be read together, not substituted for each other.
RTP is an expected-value percentage
For a game with possible returns (r_i) and probabilities (p_i), the theoretical return per unit wagered is:
[ \mu=\sum_i p_i r_i ]
The percentage RTP is:
[ RTP=100\mu% ]
Here:
- (p_i) is the probability of outcome (i);
- (r_i) is the total amount returned, including the stake when the paytable defines it that way;
- (\mu) is average return per unit wagered.
If (\mu=0.96), the theoretical RTP is 96%. The corresponding house edge is:
[ House\ Edge=1-\mu=0.04=4% ]
On $10,000 of total action, the mathematical expected loss is $400. That does not mean every player loses exactly $400, nor that the game pays $96 after every $100 inserted. It is an average across a large and properly defined sample.
The amount inserted into a machine is also not always the same as total action. A player may recycle returned credits through many additional spins. Expected loss is based on cumulative wagered amount:
[ Expected\ Loss=Total\ Action\times(1-RTP) ]
At 96% RTP, 500 spins at $2 each create $1,000 of action and a $40 expected loss. A $200 starting balance can support more or less than 500 spins depending on the actual sequence of returns.
Volatility comes from the whole payout distribution
Variance measures how far outcomes spread around the expected return:
[ \sigma^2=\sum_i p_i(r_i-\mu)^2 ]
Standard deviation is the square root of variance:
[ \sigma=\sqrt{\sigma^2} ]
The symbols mean:
- (\mu): expected return per spin;
- (r_i): return for outcome (i);
- (p_i): probability of that outcome;
- (\sigma^2): variance per spin;
- (\sigma): standard deviation per spin.
A larger standard deviation means outcomes are more widely dispersed. It does not automatically mean lower RTP. A high-volatility game can have a higher, equal, or lower RTP than a low-volatility game.
Two toy games with the same 96% RTP
Consider two deliberately simplified one-unit games.
Game A: frequent small return
- 48% chance of receiving 2 units;
- 52% chance of receiving 0.
Expected return:
[ \mu_A=(0.48\times2)+(0.52\times0)=0.96 ]
Variance:
[ \sigma_A^2=0.48(2-0.96)^2+0.52(0-0.96)^2=0.9984 ]
Standard deviation:
[ \sigma_A\approx0.9992 ]
Game B: rare large return
- 1% chance of receiving 96 units;
- 99% chance of receiving 0.
Expected return:
[ \mu_B=(0.01\times96)+(0.99\times0)=0.96 ]
Variance:
[ \sigma_B^2=0.01(96-0.96)^2+0.99(0-0.96)^2=91.2384 ]
Standard deviation:
[ \sigma_B\approx9.5519 ]
Both games return 0.96 units on average. Game B’s per-spin standard deviation is more than nine times larger. A player can therefore encounter dramatically different drawdowns and peaks even though the RTP is identical.
Real slot games have many more outcomes, bonus states, multipliers, jackpots, free-spin sequences, and line or ways combinations. The example is not a model of a commercial game. It isolates the difference between mean return and dispersion.
Why a higher RTP game can still empty a bankroll first
Suppose Game X has 97% RTP and high volatility, while Game Y has 95% RTP and low volatility. Game X has the lower long-run expected cost per dollar wagered. It can still produce a harsher short session because its return is concentrated in rarer awards.
That does not make RTP meaningless. Over the same total action, the difference between 97% and 95% is two expected-loss percentage points. On $5,000 of action:
- 97% RTP implies $150 expected loss;
- 95% RTP implies $250 expected loss.
The $100 expected-cost difference is real. Volatility explains why actual session results may overwhelm that difference for a long time.
The practical comparison therefore has two stages:
- use RTP to compare average cost per unit wagered;
- use volatility and bet size to judge whether the bankroll can tolerate the distribution of results.
The RTP comparison tool addresses the first question. The variance simulator illustrates the second, provided its assumptions match the game being studied.
Hit frequency is not a substitute for either measure
Hit frequency usually means the percentage of spins producing a result classified as a win. The definition can be misleading because a “win” may return less than the total amount wagered.
On a 20-line game, a $2 spin might display a 50-cent line win. The machine can celebrate the event, but the bankroll fell by $1.50. That result may count toward a manufacturer’s hit-frequency statistic even though it was a net loss for the spin.
A high hit frequency can coexist with:
- low RTP if the hits are too small;
- high volatility if rare jackpots dominate the return;
- frequent loss-disguised-as-win events;
- fast bankroll decline when the total bet is large.
Read the slot paytable for the actual return attached to each symbol combination. Do not infer value from animation frequency.
Session results and actual RTP
Actual RTP for a measured sample is:
[ Actual\ RTP=\frac{Total\ Amount\ Returned}{Total\ Amount\ Wagered}\times100% ]
If a machine records $1,200,000 in action and returns $1,085,000, actual RTP is:
[ \frac{1{,}085{,}000}{1{,}200{,}000}\times100%=90.42% ]
That observed percentage is below a hypothetical 91.68% target, but the raw difference cannot be judged without the game’s volatility and sample size. The UK Gambling Commission’s RTP monitoring guidance explicitly notes that volatility determines the acceptable tolerance around theoretical RTP and that the tolerance narrows as play volume grows.
For a player, a few hundred spins are an extremely small performance sample. A session RTP of 40%, 120%, or 300% does not change the certified theoretical model and does not make the next spin “due.” Actual RTP is descriptive of the observed sample, not predictive of the next random outcome.
Standard deviation over many spins
Under a simplified model of independent, identically distributed spins at a constant wager, expected total return after (n) spins is:
[ E(S_n)=n\mu ]
Total standard deviation scales approximately as:
[ SD(S_n)=\sigma\sqrt{n} ]
Relative uncertainty shrinks because expected turnover grows with (n), while standard deviation grows with (\sqrt{n}). This is one reason actual RTP tends to stabilize with very large play volume.
The shortcut has limitations. It can become unreliable when:
- wager size changes;
- bonus states alter the distribution;
- a progressive meter changes value;
- the game has persistent states;
- different titles or configurations are mixed;
- the manufacturer’s published volatility measure uses another scale;
- spins are not independent under the assumed model.
The volatility index page explains numerical measurement. The broader casino volatility definition focuses on practical interpretation.
Volatility labels are not universal units
“Low,” “medium,” and “high” are useful descriptions only within a known classification system. One studio’s medium-volatility title may be another studio’s high-volatility title. A lightning-bolt icon, five-bar scale, or marketing phrase does not establish a cross-manufacturer standard.
Better evidence includes:
- a documented variance or standard-deviation measure;
- prize-frequency and paytable data;
- the share of RTP assigned to rare jackpots or features;
- manufacturer documentation defining the scale;
- a sufficiently large observed sample from the exact configuration.
Even then, theoretical volatility describes the game, not a guarantee about a particular session.
A practical selection method
When comparing two slots, check in this order:
- Total bet per spin. Denomination alone can hide line, ways, or multiplier exposure.
- Exact RTP configuration. The same title may exist in more than one approved return setting.
- Volatility evidence. Prefer a defined measure over a marketing label.
- Prize concentration. Determine whether a large share of return sits in rare features or jackpots.
- Eligibility. Maximum-bet, side-meter, or feature rules can change what the displayed top prize means.
- Session limit. Choose a cash and time limit that remains acceptable even if no major feature appears.
A higher RTP is mathematically preferable when other conditions are equal. Other conditions are rarely equal. Bet size, speed, volatility, feature eligibility, and stopping behavior determine how the percentage reaches the bankroll in practice.
RTP answers, “What is the average price of the game?” Volatility answers, “How unevenly can that price arrive?” A serious comparison needs both.