How live betting patterns could flag match‑fixing: a study of Italian Serie B
This paper looks at whether unusual patterns in live betting can help spot possible match‑fixing in football. The authors study second‑by‑second betting records from the Italian Serie B across the 2018/19 to 2020/21 seasons. Their idea is simple: build a statistical model of normal betting behaviour and then flag large, unexplained departures from that behaviour as potentially suspicious.
To do this they use a state‑space modelling framework. In plain language, that means they model an underlying level of market activity that evolves over time and is influenced by things like goals, red cards, and changing odds. The dataset they analyse includes the amounts staked over time and the bookmaker’s live odds, provided by a major European bookmaker. The model predicts how much betting volume one would expect, given the match situation and recent events.
After estimating expected volumes, the researchers apply outlier detection methods. These methods look for unusually large positive deviations from the model’s predictions. For example, one concerning pattern would be many bets placed on an otherwise unlikely outcome very late in a match, when odds change and manipulators could cash in. By isolating such unexplained spikes, the method aims to flag matches or periods that deserve further investigation.
This work matters because live betting now makes up roughly half of total sports betting volume, but it has been less studied than pre‑game betting. Live markets react quickly to in‑match events, so they could reveal manipulation that pre‑game data would miss. The authors position their study as a proof‑of‑concept: it shows how advanced statistical tools can contribute to early detection systems for betting irregularities.
Important caveats remain. The paper notes a major challenge: there is very limited verified “ground truth” data on confirmed match‑fixing cases. That means flagged anomalies are signals of unusual market behaviour, not proof of wrongdoing. The modelling also depends on the quality and coverage of the betting data and on how well typical responses to events are captured. The authors control for known drivers of betting activity, but unexplained deviations still require human review and corroborating evidence.