March 5, 2011
Benchmarcs
Determining the Value of Unrealized Draft Picks

by Marc Foster

Every season as the trade deadline approaches, teams on the playoff bubble are faced with the same question: blow up the roster by trading their best players for draft picks, or soldier on and hope they make the playoffs and the short term financial gain of two or more home playoff dates is worth the opportunity lost in the long term.

But what are draft picks worth? A quick glance at the standings says that Boston's 2011 first round selection from Toronto as part of the Phil Kessel trade is worth more than the first rounder they just sent back to Toronto in the Tomas Kaberle deal. But say the season is over and the slots are locked in—what are the picks worth relative to each other? How do you determine the relative values of unrealized selections for teams looking to move up or down the draft board?

A few years ago, I asked myself how I'd come up with this. Since draft histories are easy to acquire, the next part was to determine on what basis to evaluate players. For the initial pass, I decided to go with games played.

For each draft year, I started with the skater and goalie that has played the most games. They each get a GP Value score of 1.000. Everyone else gets a GP Value score based on the ratio of their GP to the highest value. I split goalies out and scored them against themselves because they don't play a full 82 games. When I first developed this methodology the highest ratio of a goalie GP against the highest skater GP was Patrick Roy, and he was only 0.719, but I suspect Martin Brodeur's ratio vs. skaters might be higher.

By using GP, I'm allowing market forces to drive this analysis. It's market forces telling us if someone is good enough to be on an NHL roster and play the game at a given point in their career. Scoring production isn't fair to defensemen or grinders who otherwise have a valuable role on a roster. This summer, I plan on switching to GVT and make the analysis completely position neutral, but GVT wasn't at my disposal when this was first conceived.

Once I had the values determined for all seasons, I simply summed them up for each selection number and graphed it (Raw Points Chart). The Draft Relative Value (DRV) is the Y axis, and the draft number is the X axis. I set the labels to correspond with draft rounds.

As expected, the slope is generally negatively and somewhat inverse logarithmic. The other thing I did was tease Excel into giving me a best-fit smooth trend line (see the blue line). I haven't done a polynomial regression since I was an undergrad, and back then I had to do it BY HAND. I did this one to the sixth order, which is as far as Excel will go (not to mention much easier to do in Excel than by hand). Since Excel yields the equation, I can reverse-engineer it to give me the DRV of the smoothed trend line at any specific point (draft selection number). Curiously, the line generated using the equation does not yield an exact match, but I'm guessing that at the sixth order polynomial, going out "only" 13 decimal places gives a slightly different return as x increases.

I decided to smooth out the tail starting at the third round. I substituted the actual value found at the end with the minima which occurs at selection 130, and starting with the 61st pick assumed the slope to be linear (see the red line). I don't see the manual gnashing of the later rounds to be of much consequence, as the major purpose of the exercise is to determine the slope and DRVs for the first two rounds and the relative value of the later rounds. As an alternative, I can leave the slight wave found throughout the 3rd and 4th round, then at the minima (selection 130) assume slope=0 for the rest of the draft (orange line).

Back on the raw points chart, you'll notice something of a shelf that runs from around the 17th to 27th pick. I think this is a reflection of stronger organizations, selecting later (only 21 picks in first round until SJ in 1991) due to their better records, but still able to scout/select effectively. Before I'm convinced, however, I need to examine the first rounds for trades and see if a significant number of trades were made that might have impacted the order.

While many draft years the draft went past 210 selections, since that's the length of the draft now that's the end tail of the charts.

I also ran a comparison after I pulled out the Euros (defined as those drafted from Euro teams), but as you can see, the difference does not appear to be significant. The Euros line (blue) does appear above the Euro-free line (red) in the third and fourth rounds, then again in the middle of the sixth round through the end of the draft. This is likely a function of the lower risk, higher reward drafting of teams like Detroit that enjoyed significant production from their Euro drafting strategy back in the heyday of the Euro-drafting era.

Due to the practical applications of DRV, I can't imagine someone in the NHL hasn't done something like this already, and if they have, I'd love to see their methodology. Again, I plan on rebuilding this data over the summer based on career GVT data, though I don't expect any significant changes in the shape of the smoothed line. Since this GP-based analysis was just an attempt to establish a proof-of-concept, I'm not including a detailed table of DRVs for each selection. The other "next step" is to develop comparisons of GVT per draft pick to predicted future GVTs for established players. This way, trade analysis can be expanded from draft pick swaps to draft swaps for existing players with a predicted remaining value to their careers.Marc Foster is a frequent contributor to Hockey Prospectus.