Why Soccer Analytics Works Like Volatility Arbitrage Trading

Soccer analytics and volatility trading share a structural similarity because every aspect of soccer, from on-pitch performance to seasonal outcomes, is distribution-based. Mike Tracy, head of risk at Apex FinTech Solutions and former volatility arbitrage trader at Peak Six, argues that soccer requires making highly levered bets on imperfect information that is not necessarily predictive in the way baseball statistics are, and that promotion and relegation makes club finances highly variant year over year. The 2026 FIFA World Cup is projected to generate more than 90 petabytes of data across 104 matches, a 45-fold increase over 2022, driven largely by skeletal tracking data that adds roughly 27 body coordinate points per player per frame, making it approximately 27 times denser than standard positional tracking data.

Possession percentage and most traditional television overlay statistics are not predictive of match outcomes. Expected goals only registers when a shot occurs and therefore misses possession threat and match momentum. Corners may carry some signal. The field has evolved from on-ball event data through positional tracking at 10 or 25 frames per second to full body pose and skeletal movement data, and AI capability improvements made this level of analysis tractable in a way that was not previously possible.

Red cards are routinely removed from predictive model datasets because data from irregular game states produces highly skewed per-90-minute statistics that create poor inputs for player evaluation. A Chelsea versus Tottenham match illustrated this problem when Tottenham played a high defensive line with nine men and conceded multiple goals, and Nicholas Jackson scoring three goals in that single match represented approximately 20 percent of his total goals for the entire season, making that data misleading for evaluation purposes.

The biggest unsolved problem in soccer analytics is translating model outputs into language coaches can understand and act on. Most clubs use an expert data analyst as an intermediary who converts model outputs into video clips rather than feeding raw data directly to coaches. Model outputs used to find relevant clips include classification of team buildup structure, expected goal probabilities, and expected possession value models, which measure the probability a team will score within the next 30 seconds or the next possession. Most live model-based adjustments happen pre-match, post-match, or at halftime rather than in real time. In-game win probability models begin with a pregame team strength such as an Elo rating and update based on expected goals or momentum data, though five minutes of in-game information does not move the win probability needle much given the weight of prior information.

Analytics can quantify the value of events that do not happen, such as a player closing a passing lane, and measures player value across two dimensions: in possession, tracking progression of the ball into threatening areas, and out of possession, tracking prevention of the opponent from doing the same. Jude Bellingham is identified under this framework as the best current soccer player because he can perform effectively across four or five different positions and adapt to whatever role is needed at each stage of a match.

Tracy argues that MLS analytics adds a third vector beyond the recruitment and first team analysis that dominates European soccer, which he calls portfolio management. The MLS salary cap requires evaluating each player from a relative value perspective based on their cap slot, meaning a player who would be a poor designated player could be a top-percentile talent at a senior minimum salary slot. Lionel Messi may earn over 20 million dollars in salary while his designated player cap charge could be as low as 750,000 dollars. Tracy notes that Austin FC is only nine months into their portfolio management project and has had sporting department turnover, so the thesis has not yet been fully integrated or tested.

Joe Wiesenthal argues that Goodhart's law applies directly to sports analytics because players and coaches begin optimizing for stats rather than winning, and that a single Premier League team may simultaneously be optimizing for profit, avoiding relegation, and wins, which are distinct objectives with different answers. Basketball became dominated by three-point attempts after analytics showed they were underutilized, potentially making the game less entertaining despite being more optimal, and there is already criticism of convergence in football play styles due to over-optimization. The standard criticism of published predictive models also applies: if a model genuinely predicted outcomes better than the market, its owner would use it to beat bookmakers rather than publish it, and sell-side World Cup prediction notes are described as generally poor, frequently predicting favorites like England or the United States incorrectly.