Model Accuracy — MLB

How well do ranking models predict regular-season game outcomes?

Fraction of games where the predicted favourite won

higher is better

Mean squared error of probability predictions (lower is better)

lower is better

Cross-entropy loss of probability predictions (lower is better)

lower is better

Results by Season

Season	Games	Pick Accuracy	Brier Score	Log Loss
2026	2439	55.1%	0.2460	0.6851
2025	2430	56.6%	0.2433	0.6794
2024	2429	56.2%	0.2439	0.6808
2023	2430	56.4%	0.2442	0.6815
2022	2430	58.5%	0.2403	0.6736
2021	2429	58.2%	0.2407	0.6742
2020	898	56.6%	0.2447	0.6826
2019	2429	59.3%	0.2377	0.6681
2018	2431	56.8%	0.2413	0.6755
2017	2430	55.8%	0.2453	0.6837
2016	2427	55.7%	0.2459	0.6848
2015	2429	54.5%	0.2462	0.6854
2014	2430	55.1%	0.2473	0.6877
2013	2431	55.9%	0.2439	0.6808
2012	2430	56.8%	0.2440	0.6809
2011	2429	54.8%	0.2456	0.6844
2010	2430	56.4%	0.2431	0.6792
2009	2430	56.9%	0.2437	0.6804
2008	2428	55.4%	0.2449	0.6829
2007	2431	55.4%	0.2467	0.6866
2006	2429	54.5%	0.2459	0.6848
2005	2430	56.4%	0.2448	0.6826
2004	2428	57.0%	0.2431	0.6792
2003	2429	58.7%	0.2403	0.6733
2002	2425	58.2%	0.2390	0.6707
2001	2428	56.1%	0.2440	0.6809
2000	2428	55.1%	0.2467	0.6865

Each point is a 5% probability bucket. On the diagonal = perfectly calibrated. Above = model underestimates; below = overestimates.