How the model works

Kubota projects hockey players by finding the skaters in history they most resemble, then ageing them the way those players actually aged.

The approach comes from PECOTA, the baseball system Nate Silver built in the early 2000s. Kubota uses k-nearest-neighbour matching to find comparables for almost every rostered and NHL-affiliated skater going into the season.

Age curves are not universal, which is the whole reason to do it this way. Sidney Crosby at 37 has not declined the way an average 37-year-old declines. Run one league-wide curve over everybody and you will be most wrong about the players you spend the most time thinking about. So Kubota matches first and ages second.

The 2023–24 release was the base model. The underlying model has been reworked since, and an NHL-equivalency model added on top. That last piece is what puts players from different leagues on the same scale.

The three models

Different data availability, different coverage

Kubota NHL

The core model

The big one. Finds comparables back to 1963 and projects any NHL player with games in each of the three prior seasons.

Projects

  • Games played, goals, assists, points
  • Shots on goal, penalty minutes, plus/minus
  • Power-play goals and assists
  • Short-handed goals and assists

Kubota NHL Peripherals

Post-2007 data only

Hits and blocks were not tracked before 2007, so these two run through the same process against a much shorter history.

Projects

  • Blocked shots
  • Hits

Kubota NHL Equivalencies

The most ambitious

Estimates what a player with fewer than three NHL seasons would do in the NHL, working from his production in lower leagues. Fewer categories here, because there is less to go on.

Projects

  • Games played, goals, assists, points
  • Penalty minutes

Reading a projection

NHLe
NHL-equivalent. Production translated onto the NHL scale so you can compare across leagues.
Similarity score
How closely a player from history matches. Higher is closer. The spread across the top ten tells you more than the single best match does.
VORP
Value over replacement. Fantasy points above the worst starter at the same position in your league. This is the number that should drive your draft and your trades.
Games played
The current-season pages let you pick how games played is set: a flat 84 (with known injuries already reduced by their projected missed time), each player's best season over the last three years, or a blend of the two. That keeps the comparison about production rather than who stays healthy, on whichever basis you prefer.
Value index
Cap hit per projected point, set against the league median. Over 100 beats the median, under 100 trails it.

What's coming

Available now

  • Skater projections for the current season
  • NHLe projections for NHL-affiliated skaters
  • Seven-year projections and organisational outlooks
  • Contract value and a fantasy trade analyzer

In progress

  • Goalie projections, current season and seven-year
  • Roster and line builder
  • SHL, Liiga, DEL, NL and KHL projections