Anyone using GPT for a trading edge? Create an LLM, or LTM (“large trading model ” trained on equities data, generate a future market state, and trading based on this prediction/expectation. I know it’s been done at the institutional level, but now that GPT is becoming available to the Retail investor, perhaps I can now afford it. Anyone playing this game? Thoughts?
Won't work and no it has not been done on any institutional level. Why do you think hedge funds pay hundreds of millions for alternative data. I am not saying that deep learning is not being employed but not in the way how you picture it. And a language model generates and interprets text nothing more nothing less. Has nothing to do with trading and financial models. A language model only helps to map linguistics to tokens an ai model understands. The source data still are current or even realtime textual and numerical data that by majority retail does not have access to. Even if you had access to the entire Twitter feed a mere mortal would be incapable of differentiating between noise and valuable content. There is a reason data science and modeling positions in this space fetch mid to upper 6 figures.
There are no thoughts on this subject just experience, the markets have already discounted simplicity to zero, if it's not so complex you basically don't understand how you got there then it has no utility, ChatGPT produces a 'point in time' view but can never create iterations, it's nearly impossible to stitch point in time together to create anything coherent, and iterations are what make things complex as long as you build a previous generation that works.
Don't forget the (hypothetical ) power of the (hypothetical ) quantum computers, used in finance... by some ups... quants...
I think you misunderstand... I'm not talking about a _language_ model, I'm talking about a generative model "pre-trained" with market data. GPT is far more than language. It can be used to generate images, sound.. and even a market snapshot. In other words, instead of generating language, it generates a market, with prices; say, five minutes into the future. If this prediction of correct say 4/7 times... we're in the money. That's why I called it a "Trading" or "Market" GPT... not a language model. Instead of a Large Language Model (LLM), the GPT is trained on market data, for a Large Market Model (LMM). And yes, predictive models of this type have been institutionally in use for years. Even beyond this proposal, I'm exploring ideas of how to use GPT to gain an edge.
the money is in taking advantage of the ignorant and gullible. Especially those who think they know but are clueless. FOMO
That's not how generative models work, but good luck trying things out, no harm in that. You still don't have access to meaningful data.
I doubt you'll find anything of value using out-of-the-box software on raw data. Would be a bit too easy, wouldn't it?
People are going to get rich tricking public facing AI into recommending stocks to lazy stock pickers.