Forensic Sports Analytics : Detecting and Predicting Match-Fixing in Tennis

Despite some degree of lingering friction between sports leagues and wagering, the eradication of corruption in sports is one issue in which the interests of sports governing bodies, reputable gambling businesses, entities affiliated with the sports industry, and law enforcement are allied. This paper aims to: (i) detect match-fixing corruption in tennis and (ii) predict such match-fixing before it occurs. We first compare proxy measures of effort in fair matches versus possibly unfair matches – matches played in the first round, where wagering-induced manipulation is more likely. The results show that players exert less effort (tanking) in the first round, even when adjusting for the slightly greater difference in skill level in such matches. We next determine whether the gambling markets were able to identify fixing, tanking, or other types of manipulation before the match was played. Using two predictive tennis models (ELO, Common-Opponent) to determine the fair trading price of a tennis match, we find prima facie evidence of the betting markets being affected, with an average of 23 matches per year likely being manipulated or outright fixed each year. Finally, we determine whether fixed matches can be identified before they are played using predictive modeling and by observing real-time market price changes. We conclude that when the betting price has a large irrational move away from the fair model-predicted price, the move is indicative of a fixed match before it is played