Show HN: Ftocks – Everyday I'm asking LLM to give me stock price predictions (opens in new tab)

(ftocks.com)

2 pointsskyfantom11mo ago7 comments

7 comments

ggm11mo ago

Wouldn't put it past somebody to code this adversarial to game their stock picks. You could weaponise misplaced trust here, no different to any other stock recommendation method. The explain part would be Fascinating. "Buy Berkshire Hathaway and apple" aside.

skyfantomOP11mo ago

It's true. I was thinking about also posting requests to LLMs I use to get these recommendations. But again, I can lie about answers. I wonder if there is a way to make it 100% transparent (or reproducible).

ben_w11mo ago

Have you tried it on historical stock data to see how well it would have performed in the past?

skyfantomOP11mo ago

No. But I do have in plans 1. to track results and 2. try different prompts.

Currently I’m working on adding interactive chats. Chats where users add some financial data automatically into messages. Something like:

“Please analyze {historical_data(10, days)} and also check {news}. Give me some recommendations based on these data”

ungreased067511mo ago

Why do you believe an LLM would be good at this task?

skyfantomOP11mo ago

Well. That's a good question. My assumption is that LLMs absorbed enough data and can predict at least how normal people will set these lines (open, close etc).

Next step is to explore how god these forecasts are. Although I don't think they will always green, but if for example 51% of them are good I'll call it a win.

ungreased067511mo ago

When a stock analyst creates a report, they’re doing quite a bit of technical analysis mixed with subjective judgment. An LLM is doing none of that, but guessing what that report might look like, based on no underlying analysis. It’s vibe investing.

I don’t believe this is an appropriate application for an LLM, unless it’s mixed with another program that does some fundamentals analysis based on historical data. That could be interesting.

j / k navigate · click thread line to collapse

7 comments

ggm11mo ago

skyfantomOP11mo ago

ben_w11mo ago

Have you tried it on historical stock data to see how well it would have performed in the past?

skyfantomOP11mo ago

No. But I do have in plans 1. to track results and 2. try different prompts.

Currently I’m working on adding interactive chats. Chats where users add some financial data automatically into messages. Something like:

“Please analyze {historical_data(10, days)} and also check {news}. Give me some recommendations based on these data”

ungreased067511mo ago

Why do you believe an LLM would be good at this task?

skyfantomOP11mo ago

Well. That's a good question. My assumption is that LLMs absorbed enough data and can predict at least how normal people will set these lines (open, close etc).

Next step is to explore how god these forecasts are. Although I don't think they will always green, but if for example 51% of them are good I'll call it a win.

ungreased067511mo ago

I don’t believe this is an appropriate application for an LLM, unless it’s mixed with another program that does some fundamentals analysis based on historical data. That could be interesting.

j / k navigate · click thread line to collapse