Cash-Out Button Tapped 2.1x More When the Odds Freeze Mid-Play
Operators running in-play markets have known for years that bettors behave differently once a line stops moving. What they have not had until recently is a clean number for how differently. According to internal data shared by three U.S.-licensed sportsbooks and one European exchange, the cash-out button is tapped roughly 2.1 times more often during the window in which an in-play price is frozen — suspended for a review, held during a VAR check, or locked while a settlement engine recalculates after a scoring play — than during comparable live windows when the price is actively updating. The figure holds across soccer, basketball, and tennis, and it survives when you control for the size of the position and the time remaining.
The 2.1x number is not a marketing statistic. It comes from session logs, and it describes a behavior that operators have a commercial interest in either encouraging or suppressing, depending on which side of the book they sit. That ambiguity is why the finding has circulated quietly among trading desks rather than appearing in a press release.
What "frozen odds" actually means in the data
The term is loose in common usage, so it is worth pinning down before the number means anything. In the datasets behind the 2.1x figure, a "frozen" state is any interval of 4 seconds or longer in which the displayed price for a market does not change and the market remains open to cash-out but closed to new stakes. That definition excludes the ordinary jitter of a liquid market, where prices tick every few hundred milliseconds, and it excludes full suspensions where cash-out is disabled entirely.
The distinction matters because the two events look identical to a casual bettor — the price stops moving — but they produce very different behavior. When cash-out is disabled, bettors mostly wait. When cash-out stays live but the price is frozen, they act. The 2.1x multiplier applies only to the second case.
Three categories dominate the frozen windows in the logs:
- Integrity reviews. A suspicious pattern in the order book triggers a manual or automated hold. These are typically short, 4 to 12 seconds, and are the most common trigger in the sample.
- Event-driven holds. A goal, a red card, a challenge flag, a review. The market freezes while the trading model re-prices. These run longer, often 15 to 45 seconds in soccer and 20 to 60 seconds during a multi-flag NFL sequence.
- Settlement recalculation. Less a freeze than a lag, where the cash-out quote is stale because the underlying model has moved but the display has not refreshed. These are the shortest, usually under 6 seconds, and the hardest to isolate cleanly.
The 2.1x figure is an average across all three. Broken out, the effect is strongest in the second category — event-driven holds — where cash-out taps run closer to 3.4x the baseline. Integrity reviews show a smaller but still measurable lift, around 1.6x. Settlement lags are noisy and the sample is thin, so the honest answer there is that the effect is real but the point estimate is not stable.
Why the baseline is the hard part
Every operator who has looked at this has hit the same methodological wall: what do you compare against? The obvious control is "the same market 30 seconds earlier," but that window often contains the event that caused the freeze, so you are comparing a calm period to a chaotic one and calling the difference a freeze effect.
The cleaner approach, used in the dataset behind the 2.1x number, is a matched-window design. For each frozen interval, the analysts pulled a comparable interval from a different match in the same league on the same day, matched on score differential, time remaining, and position size, where the price was moving normally. That controls for the fact that freezes cluster around high-variance moments. It does not fully solve the problem — freezes are not randomly assigned, and the events that cause them are exactly the events that make people want to cash out — but it gets the estimate into a range where the 2.1x figure is defensible rather than an artifact.
The behavioral mechanics behind the 2.1x
There are at least three plausible explanations, and they are not mutually exclusive. The data cannot cleanly separate them, which is part of why the finding is interesting rather than decisive.
Frozen prices read as information
A moving price tells a bettor that the market is doing its job. A frozen price tells them something is wrong, and "something is wrong" is a signal to reduce exposure. This is a well-documented pattern outside gambling: when a quote goes stale in equities or FX, order flow shifts toward exits. The bettor is not necessarily reasoning about the underlying event. They are reasoning about the market's inability to price it, and treating that inability as a warning.
That interpretation predicts the effect should be larger for bettors with more experience, since they are more likely to notice a stale quote. The logs show the opposite, weakly. Casual accounts — those with fewer than 40 lifetime bets — show a slightly higher lift than the heaviest users. The heavy users appear to have learned that a 6-second freeze is usually noise and often resolves in their favor.
The freeze creates a decision window
A second explanation is mechanical. During a moving market, cashing out requires catching a price before it changes. During a frozen market, the price is stable, so the decision is easier to execute. Some of the 2.1x is not increased desire to exit — it is increased ability to act on a desire that was already there.
This is testable, and the data offers partial support. When the freeze is long enough that the bettor has time to see it and respond — roughly 8 seconds or more — the lift is larger than for short freezes. That is consistent with a decision-window effect, though it is also consistent with longer freezes being caused by more dramatic events.
Loss aversion meets uncertainty
The third explanation is the least flattering to the product. Cash-out is, structurally, a way for a bettor to pay a fee to convert an uncertain position into a certain one. That fee is highest when the book's model is most uncertain about the outcome — which is exactly when the market freezes. So the 2.1x may partly reflect operators pricing cash-out more aggressively during freezes, and bettors accepting worse terms because the alternative is to sit in an unhedged position with no visible price.
If that is what is happening, the 2.1x is not a story about bettor psychology. It is a story about spread capture during moments of maximum customer uncertainty, and it raises a question operators have mostly avoided: is a cash-out quote offered during a frozen market a fair price, or is it a price set with the knowledge that the customer cannot comparison-shop?
What the number is worth to operators
Three commercial uses show up in the logs and in conversations with trading staff.
Risk management. A freeze is a moment when the book's exposure is least predictable and the model is least reliable. If 2.1x more bettors are trying to exit, the book can use that flow to flatten positions it would otherwise carry through the event. Several of the operators in the sample explicitly widen cash-out margins during freezes — not to exploit bettors, they argue, but because the book's own uncertainty is higher. The two explanations are hard to distinguish from the outside.
Product design. If the freeze is what drives the tap, then reducing freeze frequency should reduce cash-out volume, and increasing it should raise it. No operator in the sample was willing to say on the record whether they had tested this. One trading manager, speaking on background, said the company had "absolutely looked at" whether to hold markets open longer during reviews and had decided against it for integrity reasons. That answer is plausible and also convenient.
Regulatory posture. In the U.S., cash-out is regulated market by market, and the rules on what a book may offer during a suspension vary. Some state regulators treat a frozen market as effectively suspended and expect cash-out to be disabled; others allow it as long as the displayed price is honored. The 2.1x figure is relevant to that debate because it quantifies how much money moves through a channel that some regulators believe should be closed. If a freeze is a period of elevated customer vulnerability, the volume matters.
The margin question
The most uncomfortable number in the dataset is not the 2.1x. It is the average cash-out margin during frozen windows, which ran 1.8 percentage points wider than during normal live play in the same markets. That is a real cost to the bettor, and it is not obviously justified by anything except the book's own uncertainty — which, in a well-run operation, should be priced into the model rather than passed to the customer at the moment they are least able to evaluate it.
Operators counter that cash-out is a convenience product, that bettors are free not to use it, and that the alternative — no cash-out during freezes — would leave customers stuck in positions they want to exit. That is a fair point. It is also the same argument every convenience product makes.
The number is probably not 2.1x
Two caveats should travel with this figure wherever it is cited.
First, the sample is not representative. The four operators who shared data skew toward soccer and basketball, toward markets with deep liquidity, and toward customers who use cash-out regularly. A book whose customers rarely cash out would show a smaller lift, because there is less baseline activity to multiply. The 2.1x is best read as an estimate for the segment of the market where cash-out is already a habit.
Second, the effect is almost certainly not stable over time. Cash-out usage has grown every year since U.S. sports betting went state-by-state, and bettor familiarity with freeze events has grown with it. The casual-user lift that shows up in current data may shrink as users learn that most freezes resolve in under 10 seconds and are not, in fact, signals. Or it may grow, if operators continue to widen margins during freezes and users start treating the freeze itself as a reason to leave.
There is also a selection problem that no amount of matching fully solves. Freezes are caused by events. Events change the value of positions. Bettors who want to cash out during a freeze are disproportionately bettors whose positions just got worse. Some of the 2.1x is the freeze effect. Some of it is the goal, the flag, the red card — and those would drive cash-out activity even if the price kept moving. The matched-window design shrinks this problem. It does not eliminate it.
What to watch
The cleanest test of the freeze effect would be an experiment no operator has run publicly: hold a market's price frozen during a period when nothing is happening, and see whether cash-out taps rise. If they do, the effect is about the freeze itself. If they do not, the 2.1x is mostly about the events that cause freezes, and the button-tapping is just bettors reacting to bad news with a slightly better execution window.
Until someone runs that test, the 2.1x figure is best treated as a description of a correlation that operators understand well enough to price around. The more useful question is not whether the number is exactly 2.1. It is whether a product that pays out most heavily when the customer can least tell what a fair price looks like should be offered at all during the moments the market has stopped working. Regulators in a handful of states have started asking that question. The data suggests they are asking it about a channel that moves more money than most of them assumed.