match outcome modelling

Estimating win probabilities in sports analytics is inherently challenging due to high observational noise and strong multicollinearity among predictors, leading to biased, high-variance binary win/loss models with severely miscalibrated confidence intervals. Method: We construct a stochastic-walk-based football simulation environment with known ground-truth win probabilities and conduct Monte Carlo experiments using multiple machine learning regressors (e.g., logistic regression, gradient boosting) to quantify estimator performance under realistic data dependencies. Contribution/Results: We provide the first empirical quantification showing that observation-dependent structures substantially degrade estimator bias, variance, and nominal coverage. Crucially, effective sample size decays markedly, necessitating substantial widening of conventional confidence intervals to achieve target coverage. This phenomenon is generalizable across clustered sports data, offering both theoretical grounding and empirical benchmarks for characterizing the fundamental uncertainty in win-probability modeling.

This study addresses the limitations of existing match classification methods, which fail to accurately capture variations in team incentives across different match outcomes and rely on overly rigid definitions of “meaningless” matches. The authors propose a novel classification framework based on changes in qualification probabilities: by integrating tournament rules with Monte Carlo simulations and probabilistic modeling, they enumerate all possible match outcomes and quantify how each of the three results (win, draw, loss) affects each team’s advancement prospects. This approach enables a fine-grained categorization of matches into six distinct types, offering the first systematic characterization of incentive intensity. Applied to UEFA’s new 2024/25 competition format, the analysis reveals that while the incomplete round-robin structure reduces the number of mutually meaningless matches, it substantially increases the frequency of matches requiring aggressive play and elevates the risk of potential collusion.

This study addresses the challenges posed by the dynamic nature of team strength, high outcome randomness, and the critical role of draws in football matches—factors that traditional rating systems struggle to model effectively within football-specific contexts. To overcome these limitations, this work proposes an adaptive rating framework built upon Glicko-2, incorporating adjustments for goal margin, home-field advantage, structural shocks, and dominance weighting. An ordered logit model is integrated to explicitly capture the probabilities of win, draw, and loss outcomes. By combining dynamic Bayesian updating with Monte Carlo simulation, the proposed method not only enhances predictive accuracy for match results but also enables high-fidelity full-season simulations, significantly improving both contextual adaptability and the representation of uncertainty inherent in football competitions.

Existing rating systems treat draws as equivalent to half-wins and half-losses, ignoring the empirically observed nonlinear increase in draw probability with player strength—particularly pronounced in strategic games like chess—leading to biased strength estimation. This paper proposes the first Bayesian dynamic rating framework that explicitly embeds a strength-dependent draw mechanism, abandoning the conventional linear simplification. Our method enables efficient online inference via closed-form posterior updates and a single-step Newton–Raphson approximation. Experiments on large-scale correspondence chess data from the International Correspondence Chess Federation demonstrate significantly improved model fit, more stable strength tracking over time, and superior long-term predictive accuracy. The core contribution is the first formal modeling of the draw-generation process as a nonlinear function of player strength, seamlessly integrated into a principled Bayesian dynamic rating paradigm.

Traditional baseball win-probability models neglect the dynamic pitcher-batter interactions and strategic rationality inherent in in-game decision-making. Method: We propose four progressively sophisticated hierarchical Bayesian matchup models that jointly incorporate hierarchical prior modeling, subgame-perfect Nash equilibrium, baserunner advancement probabilities driven by player steal tendencies, and Monte Carlo simulation. The models integrate pitcher-batter attributes, platoon effects, and recent performance to enable real-time, strategy-aware adaptation. Contribution/Results: In simulated 2024 MLB postseason scenarios, the optimal model yields an average of 1.0 (±0.1) additional wins per season relative to baseline. Moreover, its win-probability forecasts align closely with market odds (Spearman’s ρ = 0.92), demonstrating dual efficacy in both strategic decision support and probabilistic win modeling. This work is the first to embed game-theoretic equilibrium concepts within a hierarchical Bayesian baseball framework while explicitly modeling dynamic baserunning behavior.

This study addresses the accurate conversion of betting odds into outcome probabilities for sports forecasting and market efficiency analysis. It proposes two approaches: first, an odds-only expected profit-consistent (OO-EPC) method that requires no historical data and is grounded in the assumption that bookmakers set odds such that their expected profit is equal across all possible outcomes; second, a frequency-learning generalized linear model (FL-GLM) that incorporates historical data and corrects the favorite–longshot bias by estimating only a single calibration parameter. OO-EPC introduces a novel perspective by modeling odds conversion through the lens of bookmaker profit confidence, while FL-GLM enhances interpretability and simplifies existing frameworks. Evaluated on a dataset of 90,014 football matches, OO-EPC outperforms existing odds-only methods, and FL-GLM consistently surpasses traditional multinomial and logistic regression models across multiple bookmakers, demonstrating successful application in six basketball prediction contests.

研究构建了一个基于逻辑回归的赛前胜率预测模型，用于预测《英雄联盟》职业比赛结果，通过结合动态和静态团队实力等因素，提高了预测准确性。

This study addresses the challenge that real-time football match prediction models often fall short of the accuracy offered by betting exchange odds. To bridge this gap, the authors propose a novel approach based on the Weibull accelerated failure time (AFT) model. The method jointly calibrates team strength parameters to Betfair’s 1X2 and over/under markets via least squares and, for the first time within an AFT framework, incorporates expected goals from shots as a time-varying covariate, thereby integrating pre-match market information with in-game dynamics. Evaluated on 140 Premier League matches, the model achieves a win/loss prediction accuracy of 70.2%—nearly matching Betfair’s 70.6%—and yields a 4.5% return on investment (Sharpe ratio: 5.94) across 17,458 simulated bets, demonstrating the effectiveness and novelty of the proposed market-calibration mechanism and time-varying feature integration.

This work addresses the instability or failure of maximum likelihood estimation in the Bradley–Terry model when the comparison graph is disconnected or nearly separable. To mitigate this issue, the authors propose two interpretable data augmentation–based regularization strategies: introducing pseudo-matches between all pairs of players and incorporating virtual players with fixed strengths. The former yields shrinkage estimates of ability parameters, while the latter simultaneously resolves the inherent non-identifiability due to location invariance in a natural manner. Experiments on 2025 Major League Baseball season data demonstrate that, with appropriate tuning, these regularized approaches accurately replicate the effects of ridge regression while preserving intuitive semantic interpretations grounded in the augmented data.

本文解决了大规模相关竞赛中的逆问题，通过使用包括因子、块和层次协方差结构在内的方法，提高了计算效率和准确性。