Why Soccer Analytics Works Like Volatility Arbitrage Trading

What It Covers

Soccer analytics consultant Joris Bekkers and volatility arbitrage trader-turned-risk executive Mike Tracy explain how tracking data, neural networks, and expected possession value models are transforming football recruitment, in-game strategy, and player evaluation — drawing direct parallels to derivatives trading frameworks and portfolio management theory.

Key Questions Answered

• •Data Censoring for Irregular Game States: When analyzing player performance, remove data from matches with extreme game states — such as a team reduced to nine men — to avoid skewing seasonal metrics. A single anomalous match, like Nicholas Jackson's three-goal game against nine-man Tottenham, can represent roughly 20% of a player's entire season goal output.
• •Expected Possession Value Models: Move beyond expected goals (xG), which only registers value at the moment of a shot, by using expected possession value models that calculate the probability of scoring within the next 30 seconds of any possession. These models identify momentum spikes and can be used to surface specific video clips for coach review.
• •MLS Portfolio Management Framework: MLS salary cap rules create a three-vector analytics challenge — recruitment, first-team analysis, and portfolio management. Each roster slot carries a designated cap charge rather than actual salary, so player value must be assessed relative to slot cost, similar to allocating capital across asset classes with different risk-return profiles.
• •Tracking Data Structure for Neural Networks: Positional tracking data captures x-y coordinates for all 22 players at 10–25 frames per second, while skeletal body pose data adds 27 coordinate points per player — roughly 27 times denser. This data enables neural networks to quantify off-ball movements, defensive lane closures, and positional value that on-ball event data entirely misses.
• •Analyst-as-Translator Layer: The primary bottleneck in applied soccer analytics is not model quality but knowledge translation. Effective implementation requires a dedicated data analyst who converts model outputs into curated video clips for coaches, since raw model outputs are not directly actionable by coaching staff and halftime or quarter-break windows are the realistic adjustment points.