Here is a fun story about Point72 Asset Management, Steven A. Cohen’s almost-hedge-fund, and its efforts to transform from a classic hedge fund run by warring cliques of smart humans with ulcers into a quantitative hedge fund run by calm quiet rational computers. I feel like I read stories like that a lot: Point72, which is huge and famous and oddly institutionalized considering that it is technically a family office, seems to be at the forefront of the trend of human-driven hedge funds reinventing themselves as computer-driven ones (or at least talking about it). But what is new to me in this story is the description of Point72’s plan for “parsing troves of data from its portfolio managers and testing models that mimic their trades”:
Can you teach a computer to befriend a doctor and get illegal early access to drug trial results, har har har?
A dumb simplistic description of how machine learning works is that a computer shows a person a picture of a thing, and the person clicks “this thing is a dog” or “this thing is not a dog,” and after the person has done that a few hundred times the computer will have figured out how to recognize pictures of dogs. Good job, computer.
A classic dumb simplistic description of machine learning in investing is something like: A computer looks at a state of the market last Monday, and then it looks at whether soybean prices went up on Tuesday, and after it has done that a few million times it will have figured out how to recognize opportunities to buy soybeans.
But I guess you could do it the other way? The computer shows human analysts pictures of states of the market, and the analysts click “this is a trade” or “this is not a trade,” and after the analysts have done that a few hundred times the computer will have figured out how to recognize trades. That is: It will recognize good historical trades not from the fact that they made money, but from the fact that Steve Cohen liked them. That is a subtle but important difference. (Presumably all trades that Steve Cohen liked made money, but not all trades that made money pleased Steve Cohen.) Perhaps that way the computer will not only learn when prices go up, but will also acquire some of the gut instinct or market magic or je ne sais quoi of actual fundamental equity investors. Of course the real trick is to get computers who are better than the humans, not to replicate them exactly.
This, to me, is the core interesting question in debates about corporate governance. Say an activist shareholder wants a company to do a stock buyback, while the company’s chief executive officer wants to instead invest more in the company’s, I don’t know, edible drone initiative. The activist will say: The CEO is just entrenching himself, wasting shareholder money on low-return vanity projects just so he can justify a higher salary for himself. The CEO will say: The activist is just focused on the short term, pushing us to abandon high-return long-term projects just for a momentary pop in the stock price. This is a real tension! Sometimes managers want to hang on to shareholder money for dumb self-aggrandizement reasons; other times they want to hang on to it for long-term value creation. How do you decide which is which?
I am not sure there is an easy answer, but Cremers et al. suggest a plausible one, which is: Shareholders can decide. That is, they can decide in advance to trust managers, or not, and their decision provides some information. If they vote to allow a staggered board, it’s because they trust the managers and don’t want them to be easily displaced by a proxy fight. If they don’t vote for a staggered board, but the board unilaterally puts in place a poison pill to fend off an activist shareholder, then that might suggest that the board is not acting in the long-term interests of the shareholders.
The extreme case is a company — like Snap Inc. — whose public shareholders don’t get to vote. Of course, they chose that fate when they bought the stock. It is an extreme commitment device: If you are investing in Snap, you have to trust that its founders will “pursue long-term strategies that maximize firm value,” because you never get to change your mind.
Super hedge fund!
Elsewhere in governance, there are some well-known issues with hedge fund activism as a corporate governance tool. Activists may not have the same aims and incentives as long-term institutional shareholders. And activists buy a chunk of a public company, agitate for changes, push up the value of the company — and then have to share that value with the other freeloading public shareholders. Here is a blog post (and related article) by law professor Sharon Hannes, titled “Super Hedge Fund,” that proposes a bizarre solution to those problems: