How we predict the FIFA World Cup 2026 with a Dixon-Coles bivariate Poisson model

Short summary

Onside Arena's World Cup 2026 prediction model uses Dixon-Coles bivariate Poisson (correcting independence assumptions in standard Poisson models) with Bayesian-shrunk team ratings from 32K historical matches and confederation-based priors. Live accuracy: 75% on MD1, Brier score 0.179; the model recalibrates after every match and exposes predictions via free JSON API and MCP plugin for Claude/Cursor.

• •Dixon-Coles bivariate Poisson corrects football goal distributions (0-0, 1-1 over-predicted by standard Poisson)
• •Hierarchical Bayesian model trained on 32K+ matches with confederation-cluster priors, live recalibration after each WC game
• •75% accuracy on MD1 (24/32 correct); free API + MCP integration for LLMs

Generated with AI, which can make mistakes.