Markets & strategy
Summary + sample · full document is 1,406 words
The strongest public model in each category, plus the metric it claims and where the claim comes from. The point is to set a target: if our Phase 3 baseline can't match these, we don't have an edge yet.
Sample sizes, leagues, and time windows differ across these references — direct head-to-head comparisons are not always clean. Where comparison is muddy, the entry says so.
| Model | Owner | Method | Public access | Claimed metric / validation |
|---|---|---|---|---|
| g+ (Goals Added) | American Soccer Analysis | VAEP-family on event data | Free, MLS + USL + NWSL public ratings | Decomposed into six action types; correlates with team-level goal difference at R² ≈ 0.6+ over season samples. Self-published, MLS-focused. |
| OBV (On-Ball Value) | StatsBomb | VAEP-family on event data with proprietary feature set | Free club-level ratings published; per-player ratings sold | Per-player and per-action; documented in StatsBomb whitepapers. Validated against subjective expert evaluation and sale price. |
| VAEP | KU Leuven (Decroos et al.) | P(score) − P(concede) over next k actions | Free socceraction library | KDD 2019 paper; correlates with manager and scout assessments better than expected goals alone. |
| xT (Expected Threat) | Karun Singh | Possession-grid Markov | Free, public methodology | Simpler than VAEP/OBV; baseline in many academic studies. |
…
Pro
Incumbent Baselines runs 1,406 words. Pro members get every research note in full, plus the arbitrage feed, model outputs, and weekly updates.