Crash games: what science says about your odds (2026 study)

The UNLV 2026 study, in brief

The study in brief — Title: "Cash or Crash: Return to Player Percentages and Expected Value of Crash Games"
Authors: Robert H. Scott III, Mikhail M. Sher, Jonathan Daigle
Journal: UNLV Gaming Research & Review Journal, February 4, 2026
Read the study: https://oasis.library.unlv.edu/grrj/
Method: history of crash games, theoretical derivation of expected value, analysis of real player data on two popular crash games.⁠^[1]
Key result: the theoretical calculations confirm that very large multipliers are mathematically rare; the study then measures player behaviour and observed RTP.
Limit: data on two games only; it names no "good" game and validates no strategy.
Transparency note: at the time of writing, only the public abstract was accessible — the authors' exact figures could not be verified; the numerical illustrations on this page come from standard crash-game mathematics, not from the study's results.

The question: "do crash games pay well?"

RTP is the theoretical share of stakes returned to players over the very long term (98% = €98 returned on average for €100 staked; €2 = the house edge). But it says nothing about how often you win, what one evening costs, or whether a "strategy" changes anything. That is what a team of American researchers examined.

The mathematics underneath — expected value, house edge, the law of large numbers — is explained with worked examples on our gambling mathematics page.

What the researchers did

In three parts: the history (Bustabit, the first crash game — a crypto casino; then DraftKings' Rocket: the niche that became a format in its own right); the mathematics (deriving the theoretical values: the expected value of a round — negative for the player, like every casino game); the real data (real players' behaviour and observed RTP on two popular crash games — the calculations-versus-facts confrontation that makes the study interesting).⁠^[1]

What they found — in plain language

Very large multipliers exist, but the calculations show they are rare.⁠^[1] Illustration (standard mathematics, NOT the study's results): on a classic crash game with a ~1% house edge — reaching ×2: ~1 round in 2; ×10: ~1 in 10; ×100: ~1 in 100. The crash can also occur at ×1.00 (stake lost outright) — that is the mechanism by which the operator takes its margin. These orders of magnitude vary with each operator's parameters.

What the study does not prove

It does not say which game to choose (two games studied, no generalisation). It validates no strategy (martingale, auto cash-outs, "waiting for the big hit": none makes the expected value positive; they change the distribution, never the sign of the long-term average). It does not predict your rounds (round independence — the ball has no memory).⁠^[1]

What you take away

1. RTP is an average over millions of rounds, not a promise for your evening (variance).
2. No strategy beats the mathematics.
3. The crash can occur at any moment, including immediately; big multipliers are real but rare.
4. Check the parameters of the game you play.

Adults only (18+). Gambling must stay entertainment: never stake a sum you cannot afford to lose.

Our analysis: "one round in 100" does not mean "one ×100 per evening"

What follows is our analysis, and it mixes three things that we distinguish at every step: the findings of Scott, Sher & Daigle's study, credited to it and cited; our calculations on the frequencies published in the table above, flagged as ours, with their margins; and our interpretations, labelled as such.

Our calculation, on this page's frequencies. The study works at the scale of the single round; we convert this page's per-round frequencies to the scale of a session — a conversion neither the study nor the table above makes. Basis: the standard orders of magnitude stated on this page (roughly 1 round in 100 reaches ×100; roughly 1 in 10 reaches ×10) and the independence of rounds. For an event with probability p per round, the probability of seeing it at least once in n rounds is 1 − (1 − p)ⁿ.

The ×100: expected once per hundred rounds, absent from a third of hundreds. Over 100 rounds, the average number of ×100 outcomes is 1 — but the probability of seeing at least one is only 63.4%: 36.6% of 100-round blocks contain none. The median wait for the first ×100 is 69 rounds: half the time, you wait longer than that. Over a short session of 20 rounds, the probability of seeing one falls to 18.2%.

The ×10, at the same scale. With p ≈ 1/10, the probability of at least one ×10 in 10 rounds is 65.1%, and the median wait for the first is 7 rounds. Even at this "common" frequency, about a third of 10-round blocks (34.9%) contain none.

Our reading: "Rare per round" and "absent from most short sessions" are the same number read at two scales. The gap between the two readings is, in our view, where the feeling that a big multiplier is "due" comes from: the per-round frequency is exact, but it promises nothing at the scale where the game is actually lived — the session. This is our reading of these calculations, not a result of the study.

What does the UNLV 2026 study say about crash-game RTP?

That the player starts out losing on average: the expected value of a round is negative, and RTP is read over millions of rounds — it promises nothing for any given evening.⁠^[1]

How often does a ×100 multiplier appear?

As a standard order of magnitude: about one round in 100. And the crash can also occur as early as ×1.00 — these very large multipliers exist, but stay rare.

Can a martingale change the expected value?

No. These approaches redistribute outcomes differently; they do not make the expected value positive, and every round stays independent of the previous ones. The study concludes there is no way to beat these games over the long run: an auto-bet set at any multiplier will not yield a positive expected value.⁠^[1]