David Rosenthal: prediction

Transcript context

…And frankly, that computers got good enough to actually do it, too. That’s another big piece of this. I don’t know that Straus could have done his data engineering too much earlier in time. So they’ve got this grand new plan and vision with the Medallion fund. Unfortunately, right out of the gate, the fund stumbles a bit, and Ax ends up getting burned out. Berlekamp though is like, no, no, no, no. This is an anomaly. Like we’re going to fix this. I really, really believe that what we’re doing with these models is going to be extremely profitable. He buys out most of Ax’s stake in the summer of 1989, and he moves the offices up to Berkeley. There he comes up with the idea that, hey, we should trade more frequently, a lot more frequently, because if what we’re trying to do is understand the state of the market from the data we have and then predict the future state of the market, and then combine that with figuring out the right bet sizing to make. We actually want to make a lot more trades to get a lot more data points, and learn a lot more about the bets we’re making so that we can then size them up or size them down. It’s that, and it’s two other things. One is the further into the future you look, the less certain you can be about it. If you know something is worth $10 right now, what you know five minutes from now is it’s probably going to be worth about $10. The most likely situation is it’s within 5% of that. If you ask me three years from now, I have almost no intuition about that. A state machine is the same way. If you flash forward a whole bunch of states, you lose predictability as you continue down that chain. The second thing is, if your models are showing that you’re going to be right, call it something like 50.25% of the time, then the amount of money you can make is gated by the number of bets you can make at a quarter percent edge. If I walk up to the casino and I think I’m right about this particular roulette wheel, which of course you’re not 50.25% of the time, and I decide to play once or play twice or play five times, there’s a chance I could lose all my money. Or if I have tiny little bet sizes, then I’m just not going to make that much money. But if I walk up to said game with a little bit of edge, and I use small bet sizes, and I play 10,000 times, I’m going to walk out with a lot of money.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.