Flagship research · MLB analytics · 2026
Window-Adjusted WAR (wWAR)
A team-specific framework for prospective MLB player value under competitive windows. Standard WAR estimates player production; wWAR estimates how that projected production translates into championship value for a particular franchise.
Research question: How much should the same projected WAR be worth to different MLB franchises when championship windows, long-run franchise strength, and postseason leverage differ?
Destination-specific results under a fixed player projection
The final case study holds Shohei Ohtani's 2026–2030 projection constant at 19.81 WAR and varies only the destination environment.
Ohtani wWAR by destination
Frozen 50,000-universe production model. The vertical marker denotes 19.81 projected raw WAR.
Model architecture
The framework separates player performance, team context, league simulation, and downstream transaction economics.
Player WAR forecast
Project expected MLB WAR one through five seasons ahead using temporally validated player models.
Team-strength forecast
Blend current competitive state with persistent franchise structure using learned horizon-specific decay.
Joint MLB simulation
Simulate correlated five-year regular-season and postseason outcomes under the 12-team playoff format.
Paired counterfactual
Compare baseline and player-added universes using common random numbers to isolate the player's marginal effect.
wWAR normalization
Convert marginal championships through a fixed, player-independent league reference; compute pWAR separately.
Transaction economics
Apply salary, CBT effects, contract control, roster displacement, and acquisition cost downstream from wWAR.
Out-of-sample and simulation checks
Validation is temporal where possible and compares the framework with simpler baselines before applying it to the final case study.
| Test | Result | Interpretation |
|---|---|---|
| Player forecast vs. WAR persistence | MAE improvement at H1–H5: 0.204, 0.272, 0.326, 0.361, 0.384 WAR | Forecast layer improves on persistence at every horizon. |
| Hybrid team strength, H1 | 9.33-win MAE vs. 9.52 CurrentState, 10.15 Structural | Current-state information improves the near-term forecast. |
| H1 playoff probability | Brier 0.2065; 13.4% skill vs. naive; AUC 0.725 | Probability estimates improve on the equal-probability benchmark. |
| Fixed WS reference stability | Forward mean 10.7% above 2021–2025 historical mean | The normalization reference remains reasonably stable. |
| Independent NYY–LAD replication | +1.86 wWAR; 95% interval +0.67 to +3.04 | The 100,000-universe replication preserved the ordering. |
Production estimates
Values below are taken from the frozen 50,000-universe production run reported in the paper.
| Team | Raw WAR | wWAR | 95% MC CI | Multiplier | Δ expected titles | Δ P(≥1 WS) | pWAR |
|---|
Full methodology and results
The 22-page paper documents the hypotheses, data design, player forecasting, team-strength architecture, postseason simulation, fixed-reference normalization, validation, Ohtani case study, transaction economics, reproducibility, applications, and limitations.
The framework is prospective and model-based. Contract, trade, and prospect costs are not treated as baseball performance; they enter as separate downstream decision variables.