Suspicious betting: how to analyse an odds move before drawing conclusions

A home win shortens from 2.10 to 1.80. For a bettor, two questions matter immediately: what information changed, and does the available price still offer value? Calling the move “sharp money” or “a fix” answers neither.

The news context is substantial. Sportradar reported 1,116 suspicious matches among more than a million events monitored in 2025, including 618 football matches. These are monitoring findings, not 1,116 proven fixes. The report provides industry context; it cannot classify the particular odds move on your screen. Sportradar, February 2026.

For an analyst, the useful task is to reconstruct the market’s response to information. The framework below is our proposed approach to that task, rather than a description of Sportradar’s proprietary detection rules.

1. Measure the repricing, not just the shortening

A single selection’s odds do not show whether the bookmaker changed its assessment, its margin, or both. Capture the complete market at comparable timestamps.

Consider this hypothetical pre-match 1X2 market, with identical settlement rules in both snapshots:

Here the simple proportional adjustment is:

Normalized probability = (1 / selection odds) / sum(1 / odds for every outcome).

The home side’s normalized probability increased by 8.66 percentage points. This is a material redistribution within the market, alongside a reduction in overround. It is more informative than saying the odds “fell by 0.30.” Proportional normalization is an approximation: it assumes the margin is distributed proportionally and does not establish the true probability of a home win. Our margin calculator applies this calculation.

For Asian handicaps and totals, record the line as well as the price. Over 2.5 at 1.80 and over 2.75 at 1.95 are different contracts. A price-only comparison loses the change in settlement and risk.

2. Reconstruct who moved first and what was actually available

“Ten bookmakers shortened” is not equivalent to ten independent signals. If their changes follow the same price source, the observations may be related. Without evidence about their pricing relationships, treat independence as unknown.

Build a timeline with at least these fields:

A displayed maximum stake is not betting turnover. Likewise, cumulative exchange volume does not tell you how much traded during a particular move. If those data are missing, record them as missing rather than inferring “large money” from the length of a red arrow.

As a working hypothesis, a change that persists after new quotes appear and is supported by executable prices elsewhere deserves more attention than an isolated quote that immediately reverses. That makes it a candidate for investigation, not proof of informed betting. Compare it with the normal behaviour of the same competition, market and time before kick-off; a universal “10% drop” threshold ignores that context.

3. Test a football explanation against related markets

A sporting explanation should imply more than “the favourite got shorter.” Write down what you would expect to change, then check whether the available markets support it.

Suppose the away goalkeeper is ruled out. A plausible hypothesis is an increase in expected home goals. Home team totals and the handicap can help assess whether the repricing is consistent with that hypothesis. A broad move towards a lower-scoring game would require a different or additional explanation. Related markets are correlated, so agreement is a consistency check, not independent proof.

The next step is to quantify the news. “Three starters missing” is not a model input until you consider their roles, expected minutes, replacements and tactical effect. Season-average xG can be an inadequate baseline when the projected lineup changes substantially. Separate information already known when the opening price was posted from information that arrived later.

If the price moved before your news feed updated, you have established an information gap in your observation. You have not established that the information was illicit. Team announcements, local reporting and your data provider may publish at different times.

4. Separate the quality of the signal from the quality of the bet

Even a correctly interpreted move can leave a bad entry price.

Assume, purely for illustration, that your updated model gives the home side a 52% chance. For a win-or-lose bet, expected net return per unit staked is probability × decimal odds − 1:

• At 2.10, the model-implied expected return is +9.2%.
• At 1.80, it is −6.4%.

The same team and the same forecast produce opposite decisions at different prices. The earlier 2.10 is irrelevant if you could not obtain it after receiving the information. A backtest that selects a move using later observations but books the pre-move price contains look-ahead bias.

The 52% estimate also needs scrutiny. Did the news change your forecast, or did you adjust it merely because the market moved? Using the move both to generate the probability and to “confirm” it is circular. If a plausible probability range straddles the break-even point of 1 / available odds, a positive point estimate alone is weak support for an entry. The EV calculator makes the arithmetic explicit; it cannot validate the probability you enter.

5. What an integrity investigation adds

Public price analysis is useful, but it lacks much of the evidence needed to investigate manipulation. Sportradar describes combining odds monitoring with account-level staking data from participating operators, operator reports and intelligence, followed by expert assessment. Those additional inputs can connect market activity to accounts and circumstances that a public chart does not reveal. Sportradar’s detection methodology.

Its 2025 release also reports a 56% increase in suspicious matches flagged through AI analysis, while total suspicious matches fell 1%. The first figure describes a detection channel, not a 56% increase in fixing or a published accuracy rate. Without labelled outcomes and false-positive data, it cannot tell a bettor how reliable a particular alert is. Sportradar’s 2025 findings.

The Jontay Porter case shows the difference in evidence. In April 2024, the NBA said operators and a monitoring organization had alerted it to suspicious bets on his performance. The league’s investigation reported disclosure of confidential health information and deliberate restriction of participation for betting purposes. It also said an $80,000 player-prop parlay linked to the March 20 game was frozen and unpaid. NBA’s original announcement.

The case involved specific bets, information and conduct. It did not rest on the proposition that a large odds move must be a fix.

6. Keep an analytical record you can evaluate later

For each flagged move, save the information available at the time: your explanation, model probability, executable price and decision. After the event, add the closing market and result in separate fields. This keeps hindsight out of the original assessment.

Evaluate the process across a sample that includes rejected signals and losing bets. Check forecast calibration, realized returns and entry prices relative to a consistently defined close. Closing-line value is a useful price benchmark, not proof that an individual bet was profitable or a match was clean. A changed line must be accounted for before comparing entry and closing odds.

A useful analyst’s note ends with an actionable finding: a data error, a repricing consistent with identified news, an unexplained move requiring more information, or a price that no longer fits the model. “Someone knows something” is not a finding you can test.