What SoccerSim Is
SoccerSim is an applied decision laboratory that explores how risk and capital can be allocated responsibly in high-friction, high-uncertainty environments.
Rather than focusing on prediction accuracy alone, the project studies decision quality under constraints — including limited budgets, noisy signals, execution frictions, and correlated outcomes.
Football serves as a transparent and observable environment in which these dynamics can be made explicit, tested, and explained.
Automation, Judgment, and Governance
SoccerSim combines automated inference with explicit human judgment.
Machine-learning models and simulations are used to surface scenarios, risks, and trade-offs at scale.
Critical decisions — such as risk appetite, capital allocation, and system overrides — remain deliberate, reviewable, and accountable.
A core focus of the lab is understanding why results occur.
Performance is examined through systematic attribution that separates signal from noise, edge from variance, and decision quality from execution effects.
Why a Sports Market?
Sports markets provide a rare combination of real uncertainty, observable outcomes, and binding constraints.
This makes them a useful laboratory for demonstrating how actuarial risk management, portfolio thinking, and governance principles translate beyond traditional insurance contexts — supporting the broader objective of making actuarial judgment visible outside the profession.
Current Capability (v1)
The current implementation models matches and full league seasons using probabilistic factor models and Monte Carlo simulation.
At a high level, the system supports:
- Calibration of team strength under uncertainty
- Matchday and season-level scenario simulation
- Portfolio-style aggregation of decisions under budget and variance constraints
- Reproducible, configuration-driven workflows
Technical details are intentionally kept out of the project overview.
Research Direction
Ongoing development focuses on extending SoccerSim as a governed research platform, with emphasis on decision quality rather than model complexity.
Current research priorities include:
-
Risk-adjusted attribution and portfolio diagnostics
Establishing robust separation of signal and noise, edge and variance, and decision quality versus execution effects.
This completes the system by enabling governance, validation, and disciplined learning over time. -
Integration as a service within agentic systems
Exploring how automated inference can be embedded into human-centered decision workflows, including review, override, and accountability mechanisms. -
Hierarchical and dynamic representations of latent team and player effects
Refining signal quality through multi-level and time-varying structures, once attribution and governance foundations are in place.
These directions prioritize system discipline and explainability over incremental predictive gains.
Repository
https://github.com/bvonboyen/Soccer-predictions
(Code, experiments, and technical documentation available on request.)
The goal is not to predict the unpredictable — but to make disciplined decisions when uncertainty dominates.