WNBA Betting Model: Possession Monte Carlo, Four Factors

Olympus Bets models WNBA betting with a purpose-built possession Monte Carlo engine (v0.5), rebuilt for WNBA structure rather than adapted from our NBA engine. It uses opponent-adjusted iterative SRS team ratings, Bayesian shrinkage toward the league mean for small early-season samples, Four Factors-based possession scoring, and antithetic variates for variance reduction. WNBA projections are currently free and informational while the tracked record accrues — we publish before we charge. Free daily WNBA projections are on the site and via the MCP server.

Engine Overview

It would have been faster to take our NBA possession engine and relabel it for the WNBA. We didn't do that. The NBA engine carries machinery that doesn't transfer: 30-team altitude tables, 13-man rotation modeling, and NBA-specific play-type tables. The WNBA is a 12-team league with a 40-game season and its own scoring and pace profile, so the WNBA Possession Monte Carlo engine (v0.5) was built from the ground up around that structure.

The result is a leaner engine than our NBA model, deliberately so — it's sized to the amount of real signal a 40-game season actually provides, rather than carrying complexity the data can't support.

How the Simulation Works

1. Opponent-Adjusted Team Ratings

Team ratings are computed with an iterative strength-of-schedule adjustment (SRS-style), so a team's rating reflects who they played, not just raw scoring margin. Ratings update as the season progresses and more games are observed.

2. Bayesian Shrinkage for Small Samples

A 40-game season means early-season ratings are built on very few observations, and unshrunk ratings from small samples overfit to noise. Each team's rating is shrunk toward the league mean, with the shrinkage weight tied to effective games played — teams with more games keep more of their raw rating, while teams with only a game or two played are pulled harder toward league average.

3. Possession-Based Scoring with Correlated Variance

The engine generates scores using a shared game-environment multiplier (which induces realistic home/away score correlation — some nights both teams shoot well or poorly together) plus independent team multipliers (which drive margin variance separately from total-score variance). These are calibrated so the simulated output's statistical moments — total standard deviation and margin standard deviation — match the values fitted from real 2025 WNBA game data.

4. Four Factors Foundation

Effective field goal percentage, turnover rate, offensive rebound rate, and free throw rate are the core inputs that drive possession-level scoring in the engine, rather than a single blended efficiency number standing in for offensive and defensive identity.

5. Variance Reduction and Win-Probability Shrinkage

The engine uses antithetic variates — a technique also used in our golf simulation engine — to reduce simulation noise for a given number of iterations. Win probabilities also carry a conservative shrinkage factor (0.15), the same guard used in our NBA engine, to avoid overstating confidence from a model still building its track record on a shorter season.

6. Lineup-Aware Modeling

The engine is lineup-aware, incorporating which players are actually available for a given game rather than treating a team's roster as static across the season.

Data Sources

• ESPN box scores — game results and box score data used to build opponent-adjusted team ratings
• WNBA team data pipeline — iterative SRS rating construction from observed results
• Injury/availability data — feeds lineup-aware adjustments

What the Model Outputs

For each WNBA game, the simulation produces:

• Win probability — the percentage of simulations each team wins, with conservative shrinkage applied
• Projected total — the simulated combined score with its standard deviation
• Projected margin — the simulated margin (home minus away) with its standard deviation

These are published free and informational. WNBA has not been folded into the premium tier — the plan is to let the tracked, resolved-game record speak for itself first.

WNBA Engine Facts

Why We're Publishing WNBA Free Right Now

Most prediction products ask you to trust a track record before you can see it. We're doing the opposite for WNBA: the possession engine runs daily, the projections are public, and they stay free and informational until the resolved-game record demonstrates the model holds up. That record is the same kind of auditable ledger we maintain for every other league.

This isn't a marketing gimmick — it's a direct consequence of the engine being new and purpose-built rather than a mature, multi-season model. We'd rather let the data build the case for WNBA than charge for it before that case exists.

Explore Other Models