A few days ago, our old pal Joffrey Lupul made some comments about Corsi on Twitter:
contracts aren't awarded by this CORSI i am hearing all about. They are awarded for an equal value of skill and depth (at a certain position— Joffrey Lupul (@JLupul) July 7, 2013
If you bring certain attributes and you play to win. I'll take you on my team 7 nights a week. Lets not look at this like Moneyball.— Joffrey Lupul (@JLupul) July 7, 2013
This touched off the typical firestorm; it gave people who don't put much stock in advanced stats the opportunity to say "I told you so" and the statguys an opportunity to lecture on how important the stats are. Lather, rinse, repeat.
But here's the thing: Lupul is right, and that's the whole darn point.
Moneyball wasn't a book about advanced stats. Michael Lewis is a financial writer who particularly likes looking at market inefficiencies and how they get exploited. He wasn't writing a treatise on the importance of on base percentage; he was describing how a team that caught on to the importance of on base percentage before other teams was able to use that to their advantage.
The lesson isn't "stats can help you win"; it's "knowing more than the other guy can help you win".
And Lupul is exactly right. Contracts aren't awarded by Corsi. That's what makes it a market inefficiency, something that can be exploited.
That point was emphasized recently when Patrice Bergeron, one of the best two-way guys and drivers of possession in the league, signed a deal that pays just (just!) $52M over eight years. That's a lot of money, but it's nothing compared to what we'd expect a comparably skilled shooter to get -- Ilya Kovalchuk's deal paid $79M over the first eight years and $100M overall, despite being signed at a time when the cap was significantly lower.
But individual examples are no way to make a point. Let's dig into this in more detail and see if there's a general trend in how much teams pay for possession numbers. To do that, I compiled a list of 261 contracts signed by forwards that met three criteria:
- The contract started in 2010 or later (which ensures we can look back at least three years for stats like Corsi and on-ice shooting percentage)
- The contract covered only years in which the player was eligible for unrestricted free agency (so we can compare contracts without adjusting for RFA status)
- The player had played at least 1000 minutes of 5-on-5 ice time over the three years before the contract started (so the stats are reasonably meaningful)
Then I compiled each players' 5-on-5 stats for the three-year period before the contract started. I didn't look up the date that each contract was signed -- if a player signed an extension in February, his stats from February to April will still be treated as having come before the contract for the purposes of this study. That isn't exactly right, of course, but it shouldn't have a significant impact on our analysis.
So now we can look at the correlations between cap hit and various measures of performance -- at how strongly past performance in a given category ties with getting a big contract:
|5-on-5 performance metric||Correlation to cap hit|
|Points per 60 minutes played
|Team shooting percentage with player on ice||0.62|
|Corsi (adjusted for zone starts)||0.46|
|Relative Corsi (adjusted for zone starts)||0.52|
It is clear that teams give bigger rewards for a player helping his team to shoot for a high percentage than for helping his team get a lot of shots. We can illustrate this by dividing our sample into five quintiles of past on-ice shooting percentage and five quintiles of past relative Corsi and looking at the average cap hit of the players in each bin (asterisk indicates 2-5 contracts in that bin; all others had at least 8; a similar result is obtained using Corsi instead of relative Corsi):
|Cap hit||Quintiles of relative Corsi|
|-3% or worse||-3% to -1%||-1% to +1%||+1% to +3%||+3% or better|
|6.6% or worse||$0.9M||$1.0M||$1.2M||$1.7M||$1.3M*|
|6.6% to 7.5%||$1.2M||$1.3M||$1.8M||$2.2M||$3.2M*|
|7.5% to 8.3%||$1.5M||$1.7M||$1.9M||$3.3M||$3.0M|
|8.3% to 9.0%||$2.1M*||$2.4M||$2.8M||$3.3M||$4.2M|
|9.0% or better||$4.5M*||$3.5M||$4.1M||$4.6M||$5.1M|
As you move from the top-left to the bottom-right, the players get better and the cap hits go up quickly. But the more interesting question is what happens when we move from top-right to bottom-left. Going in that direction, we're moving from the good possession guys to the good shooters, and that also comes with an increase in cost. Average shooting and great shot differential costs $3M, but the reverse costs $4.1M.
It's clear that teams are paying more for shooting percentage than shot differential. But that doesn't really prove anything yet -- just because shooting ability is highly valued doesn't mean it is overvalued. The real question is whether a team that targeted the possession guys would outperform a team that targeted the shooters.
For each bin in the table above, we can calculate the plus/minus for those players. But since we know that on-ice save percentage is pretty much completely unrepeatable, we don't want our numbers to be skewed by the performance of the goaltender behind the player. So instead, in the table below, I report what the plus/minus would have been given the player's actual shot differential and shooting percentage but with average goaltending:
|Goal differential per 100 shots||Quintiles of relative Corsi|
|-3% or worse||-3% to -1%||-1% to +1%||+1% to +3%||+3% or better|
|6.6% or worse||-1.9||-1.3||-0.9||-0.6||-0.3*|
|6.6% to 7.5%||-1.3||-0.7||-0.4||-0.1||+0.1*|
|7.5% to 8.3%||-0.8||-0.3||-0.1||+0.3||+0.8|
|8.3% to 9.0%||-0.4*||0.0||+0.3||+0.6||+1.1|
|9.0% or better||+0.1*||+0.5||+1.0||+1.0||+1.7|
Again, moving from top-left to bottom-right has the players clearly getting better. But again, it is the movement from top-right to bottom-left -- from possession skill to shooting skill -- that is interesting. As we go down that line, we get very slightly better performance, but it costs us a lot more money. Teams are paying too much for the marginal gains that the top shooting percentages provide.
It's not hard to see why that might be. A player who drives goal differential by making great passes and shots is very visible, and the connection to positive outcomes is very obvious. He records goals and assists, the strongest driver of salary.
Of course, an NHL team is perfectly capable of recognizing all of the little things a player does to drive possession. But the connection of those little things to positive outcomes won't be quite as obvious -- they'll often be shots that someone else takes, or shots that the opponent doesn't get to take. That means that even if the team knows exactly what contributions he's making, it's easy for them to overlook just how important those two-way contributions are.
And don't forget that a strong shot differential is a much more repeatable talent than a strong on-ice shooting percentage. If we sort the contracts by on-ice shooting percentage in the last three years and take the top 10 deals signed in 2010 or 2011, the average pre-contract on-ice shooting percentage for those top shooters was 10.3%. That figure dropped to 8.8% in the 2-3 years since the contract kicked in, most of the way towards the league-average 8.1%. The teams that paid heavily for an elite shooting percentage often found themselves disappointed.
Shot differential also regresses a bit, but not nearly as heavily. Going through the same exercise, the top ten shot differentials in 2010 or 2011 averaged a zone-start-adjusted relative Corsi of +4.6% before signing their deal and +3.2% after. Because their talent is more persistent, the future goal differential of the possession guys is actually better than that of the shooters -- and remember that they cost much less.
This is what Moneypuck is all about. Nobody is trying to argue that Corsi is a perfect rating system, that shot quality is irrelevant. The point is that the talents that drive Corsi contribute more strongly to winning than people realize.
Lupul is right -- most teams don't give out contracts because of Corsi. But a team that does will get more wins out of their budget than a team that follows the conventional path and overvalues finishing talent.