The Retrodictions & Draft Model Success, Part 2

Part 1 was long and rambly, which I am prone to – especially when lacking sleep. I’m going to be uploading a speadsheet with updated 2015 draft model rankings soon (couple hours) – with some tweaks & a whole bunch of stuff included (projecting the player careers season by season, similarity scores, skillset breakdowns at peak age compared to past draftees, etc.)

Anyway, for now, here’s the updated spreadsheet with the last 17 seasons of draft model retrodictions compared to the actual drafts:

1998-2014NBAdraftRetrodictionResultsVersion2

I will go over (like before) the most important thing I want someone to take from this work – my draft model finds better players on average, and ALSO finds them MUCH later in the draft that NBA teams have done the past 17 seasons. I put the data in the last sheet to make it easier to understand.

The work – I split every past player into the NBA GM group (if the player was drafted higher than my model ranked them), the HoopsNerd group (my model had the player ranked higher than their actual draft position), and the Draws – they were ranked the same by both. I ignore the draws.

491 players in the GM group. 500 in the HoopsNerd group. I created a factor for each player, how much that player’s info was factored in by the ratio of how much better one model had a player ranked than another. Olowokandi was a 25.0 for the gms in 1998, #1 by gms, #25 that season by HoopsNerd (25/1). Brad Miller was a 25.5 for HoopsNerd model – he was #2 by HoopsNerd, #51 (undrafted, but I ranked undrafted for gms by career minutes played – giving him 51 since the last guy drafted was 50), 51/2=25.5.

This methodology should kill my model – since ANY player’s career it ranks extremely high but yet he goes undrafted or late second will count much more toward the model score. Any player drafted extremely high (players teams guarantee money to & are very vested in) that my model doesn’t like gets bigger weight for the gms.

Yet – when it’s all said and done – the players my draft model favored ended up on average with a career WAR of 15.1, while the players that were drafted higher than my model thought they should be ended up with an average career WAR of 13.3.

While that might not seem like much – but overall that’s almost 1000 more Wins Above Replacement for the HoopsNerd side. But, the most important point – the average draft position of the gms was 22.7, while the average draft position for the draft model was 42.4. The HoopsNerd draft model found BETTER players than the actual NBA teams – on average almost 20 draft picks later PER PLAYER.

All this was done without ANY other info outside of projected production based on college performance (adjusted for pace/SoS/team quality, etc.). No scouting rankings (mocks), no Rivals high school rankings, no player height/weight info, no combine results, no pre draft workouts, no interviews – nothing.

OK, I think I hammered that point again – a little more concisely than my last article, so hopefully the message doesn’t get lost.  On to finishing up (tweaking) the 2015 Draft Model spreadsheet.

Dan

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>