i was under the impression that people in the major quant companies are broadly not deep learning pilled, do not make full use of frontier models, and do not seem inclined to rapidly change in this regard. i’m not sure what to make of this. maybe they’re just set in their ways, and this opens them up to being disrupted.
I think people at quant companies use frontier models, sometimes spending thousands of dollars a month on tokens
Quants usually do not use deep learning to try and directly predict the market, though they might use it to process data. Even in 2018, everyone I knew in the field was quite excited about deep learning and reinforcement learning
Quants are generally open to rapid change, though they’re not likely to spend a lot of time on new tools until they’ve shown some clear benefit. The strategies you spent years working on can stop working overnight, and the market provides very strong incentives to adapt
I don’t think it matters, but I’m not sure that’s true. I think the major quant firms are already using LLMs in their trading and plan to use more.
In terms of making full use of frontier models, the models aren’t currently good enough to send forth, and trade, and you have the problem of adverse selection (if you start putting bad quotes, they get snapped up). Firms are definitely making use of the models in building their systems. I think quant firms are perhaps the 2nd-most AI-pilled segment of society.
But my point is that I think there are a lot of what you might call “LLM-based trading strategies”. One example is using LLMs to read press wires or earnings reports/filings. Another might just be that LLMs allow you to MM on markets much more easily that you wouldn’t have bothered with before to set up the infrastructure. For example, if there are some 100 markets in Chile that have $10k of volume a day that nobody currently bothers with, all of a sudden, you can throw your LLM at it and make 1% on that volume, maybe. Not worth paying a quant trader to actively MM and get adversely selected against. But worth $20 of tokens,
i was under the impression that people in the major quant companies are broadly not deep learning pilled, do not make full use of frontier models, and do not seem inclined to rapidly change in this regard. i’m not sure what to make of this. maybe they’re just set in their ways, and this opens them up to being disrupted.
Those are a few separate claims:
I think people at quant companies use frontier models, sometimes spending thousands of dollars a month on tokens
Quants usually do not use deep learning to try and directly predict the market, though they might use it to process data. Even in 2018, everyone I knew in the field was quite excited about deep learning and reinforcement learning
Quants are generally open to rapid change, though they’re not likely to spend a lot of time on new tools until they’ve shown some clear benefit. The strategies you spent years working on can stop working overnight, and the market provides very strong incentives to adapt
(I work at a quant fund.)
I agree with all of this and matches my experience in quant trading (though ive been out of the game for a while)
I don’t think it matters, but I’m not sure that’s true. I think the major quant firms are already using LLMs in their trading and plan to use more.
In terms of making full use of frontier models, the models aren’t currently good enough to send forth, and trade, and you have the problem of adverse selection (if you start putting bad quotes, they get snapped up). Firms are definitely making use of the models in building their systems. I think quant firms are perhaps the 2nd-most AI-pilled segment of society.
But my point is that I think there are a lot of what you might call “LLM-based trading strategies”. One example is using LLMs to read press wires or earnings reports/filings. Another might just be that LLMs allow you to MM on markets much more easily that you wouldn’t have bothered with before to set up the infrastructure. For example, if there are some 100 markets in Chile that have $10k of volume a day that nobody currently bothers with, all of a sudden, you can throw your LLM at it and make 1% on that volume, maybe. Not worth paying a quant trader to actively MM and get adversely selected against. But worth $20 of tokens,