Who wins
the 2027 MVP?
A preseason model trained on every MVP vote since 1985. It projects each star's season, scores him the way voters actually vote, then plays the season out 10,000 times.
Tap a name to dig in. Assumes every candidate reaches the 65-game minimum. Injuries are the biggest risk to any pick.
Watch the model get built, version by version, including the ideas that failed.
Look up any player
Every projected starter in the league is in the model. See his odds, why the model likes him (or doesn't) and how each factor moved him.
One factor at a time
The favourite changed hands almost every time a new piece of voter behaviour went in. Tap through the steps.
Would it have worked?
Each season predicted using only what was known that October. No peeking.
What's inside
Project the season
Last three seasons weighted 3-2-1 with an age curve: impact, points, rebounds + assists, games.
Score it like a voter
Voters reward all-around stars, the best player on a top seed and win shares. They also remember last year.
Voter fatigue
Nobody wins three in a row anymore, and nobody 31+ has won since Nash. Both are in the model.
Last year's Finals
Stars from last season's Finals teams get a ×1.3 bump. It was the one late idea that held up in the backtest.
Simulate 10,000 seasons
Team records and stat lines vary in every run, and the model counts how often each player wins.
Ideas that failed
- Extra credit for the scoring champ, triple-doubles, advanced-stat leaders or the best record
- Punishing injury history, boosting young "breakout" candidates
- Vegas win totals, Vegas MVP odds, one giant model that learns everything: all worse
- Other voter fatigue: back-to-back, chasing a 4th, two of the last three
- How well a player played last postseason (scoring, plus-minus): sharper odds, fewer winners
- Blocks, steals and DPOY votes (defensive win shares did help)
The model's blind spot: leaps. Rose in 2011, Curry in 2015, Giannis in 2019 and Jokić in 2021 all came out of nowhere.