In 2018 Cliff Asness and Aaron Brown published a paper about hockey that I have thought about more than almost anything I read that year. It is called Pulling the Goalie: Hockey and Investment Implications, and it is nominally about when a trailing team should replace its goaltender with an extra skater. It is actually about why capable people knowingly do the wrong thing.
Quick disclaimer before I get into it. The trading I mention in this post is a job I did about a decade ago. It does not reflect my current role or my current employer, and nothing here is about either.
The Math Nobody Follows
The finding is that coaches pull the goalie far too late. Down one goal, the model puts the optimal moment at more than four minutes remaining. Down two, it is around thirteen. Real coaches tend to wait until the last minute or so.
What makes it interesting is that pulling the goalie is obviously, visibly bad on its own terms. It roughly quadruples the chance the opponent scores while less than doubling your own chance. Every instinct says do not do that. And it is still correct, because when you are trailing, losing by one and losing by three are the same outcome. The only thing that counts is tying the game, and the extra skater buys that at a price measured in a currency that does not matter.
| What it changes | Does it matter | |
|---|---|---|
| Chance you score | Roughly doubles | Yes, it is the only path to a point |
| Chance they score | Roughly quadruples | No, a loss is a loss |
| How it looks if it fails | An empty net, on camera | Not to the standings, only to the coach |
The Real Subject Is Career Risk
So why do coaches wait. Because the two outcomes are not symmetric for the person deciding. Lose quietly with the goalie in and you lost a hockey game. Lose to an empty-net goal with six minutes left and you are the coach who did that, in a highlight, repeatedly. The optimal play carries a private cost that never appears in the standings.
Asness and Brown take this straight into money management, and the line that stuck with me is the old Keynes idea underneath it: it is better for a reputation to fail conventionally than to succeed unconventionally. A manager who trails the benchmark doing what everyone does keeps the account. A manager who trails doing something defensible but unusual does not.
Where I Meet This
I do not coach hockey and I no longer trade, but I run into this constantly in product work, and it looks exactly the same. A thing that is scoped for a quarter can often be done in a week if you are willing to work in an unfamiliar way. The organization does not decline because someone ran the numbers and disagreed. It declines because the conventional path fails safely and the unconventional one fails loudly, and everyone can feel which failure gets a name attached.
I have written before about how the org chart lags reality, and this is the mechanism underneath that lag. Structures do not resist better methods out of ignorance. They resist because the person who would have to go first is the person who eats the visible failure, alone, while the upside gets distributed to everyone.
The Useful Part
The paper is not really advice to be reckless. Pulling the goalie is correct only when you are already losing and the clock is real. Most decisions are not that, and the conventional play is conventional for good reasons most of the time.
What it gives you is a question worth asking in any room where the obvious right answer keeps not getting chosen: are we optimizing for the score, or for how the loss will look. Those come apart more often than anyone admits, and once you have seen the shape of it, you cannot stop noticing. The score is the thing. Everything else is an empty net you were afraid of.