Who wins Most Improved Player in 2027?
My model's pick is Brandon Miller at 16.7%, more than three times anyone else. Charlotte traded LaMelo Ball this summer, and the player who used 31% of the Hornets' plays left a lot of shots behind.
Ppg is 2025-26. Market = betting-market odds on Oct 18, before the season (— means he wasn't among the 26 names listed). Miller and Coby White together give the Hornets a 23% chance of winning the award.
Watch every version get built, the lab and the countdown. Open on YouTube →
Brandon Miller, No. 2 pick in 2023, in his 4th season. He already scored 20.2 a game, and the model still has him ahead by more than three to one.
Seasons since 2008 where the model's favourite won, using only what was known on opening night: Danny Granger, Brandon Ingram, Lauri Markkanen and Tyrese Maxey.
On the 8 seasons with preseason odds (2017-18 on, no odds for 2018-19), the Vegas favourite never won. My model's favourite won 3.
Dylan Harper is +900, but he's a second-year player, and no second-year player has won since 2007. The model doesn't consider him.
Every player's chance
All 180 players the final model considers. Market odds are from Oct 18, 2026 (26 names listed); "Implied" is the chance those odds suggest with the bookmaker's margin taken out. Search a name or tap a column to sort.
Model vs the betting markets
The three shortest prices on Oct 18, and what my model says about each.
Dylan Harper
The market's favourite is in his second season. Voters haven't given this award to a second-year player since Monta Ellis in 2007, so the model leaves him out.
Payton Pritchard
He's 29 and already won Sixth Man, so voters likely see him as a player who's already arrived. My model gives him 0.5%, 48th on the board.
Brandon Miller
The market has him around 8%. My model gives him 16.7%, about double, and makes him the clear favourite.
The field leaves out second-year players, past All-Stars (none has ever won this award), anyone under 15% usage last season and anyone who played less than 45% of last season's games.
The voters changed
There's no definition of "improved" anywhere, so the voters decide what it means. Since 2008 they've rewarded a player who was already pretty good and then became a star, so the model only learns from those 19 seasons.
7 in 22 seasons from 1986 to 2007, the last one Monta Ellis. None in the 19 since.
Almost a third of all the votes before 2008. About one in seven since.
Winners since 2008, in today's scoring. The old-era winners went from 10.6 to 20.2.
Five versions
Each version is backtested: it goes back to each of the 19 seasons from 2008 to 2026, sees only what was known on opening night, and I check how often its favourite actually won. It's like giving the model old exams with the answers covered up.
7 became 4
I picked version five's ideas while looking at all 19 seasons, so I don't fully trust the 7. I ran the whole lab again 19 times, each time hiding the season it was about to predict. Done that way, it picks 4 winners. That's the honest number, and it's the one the report card and the Vegas comparison below use.
Teammates don't split the vote
Two players on one team taking votes from each other matters for Sixth Man. Here it does nothing: 16 of the last 41 winners had a teammate who also got votes. A winning team didn't matter either.
Every winner since 2008
Where the honest version of the model ranked each winner before the season. Green: the favourite won (4). The last two winners broke out on defence and in new roles, which no model built on scoring was going to see coming.
Did the favourite win?
Preseason odds for this award only go back to 2017-18, and 2018-19 has none. On the 8 seasons where both exist, the Vegas favourite won 0 and my model's favourite won 3. Several winners weren't even on the preseason board.