- First is to actually evaluate whether these LLMs have any intelligence around investing. If you actually give them all the data, can they do well? Can they beat the market? I'm not sure, we're testing that.
- My thesis is that they will actually beat the market (I know a lot of you will disagree). If that's the case, how can we invest a lot of resources in building the best harness, tool calling, etc to enable these models to invest.
What does "all the data" mean here? I see you mentioned SEC posts. What about news articles, twitter / blog / other posts, general info on the industries, etc?
I assume these are simulated trades, not real trades being executed. How accurately do you take into account trading fees, time from order-decision to order-placement, and things like this?
I would be interested to see the same test run on some prediction market (kalshi / polymarket / etc). In the stock market, a rising tide lifts all boats, so it's easy to deceive yourself about how well you've done, vs how important initial timing was. I suspect that prediction markets will eliminate that source of noise, since it's truly a 0 sum game. That said, it also adds lots of complication, insider trading will eat into your performance more, etc.
- We've built a local vector database with every SEC filing over the last few years. And we've built a tool call on top of that to allow these LLMs to read and query sec filings.
- Have done the same for a lot of other data sources, just giving the LLM access to them and allowing it to spend some time to actually research.
I actually think it's doing better now. It was just too stubborn to exit its position for the first few months. It did that, and put some money into MSFT/JPM recently.
This is an experiment to see how well can LLMs invest in the market through a lot of research. We give them tool calls to access every financial dataset that exists online, and also some money to manage. And we then see how well they do.
SPY is up at least as much as your Claude bot since the Nov 25 2024 start date, but you show it down 3%. If AI is both doing the trading and reporting the results, you...may have a problem.
Show HN is meant for people that build a thing. It's not meant for incremental features. However, I'd say this counts as a 'new thing', not an incremental feature, so I think it should be allowed. At any rate, I'm happy to see it, and I think it meets the bar as compared to most Show HN's I've seen in the past.
Welp, sorry it got taken down in the end. At any rate I feel it's significant enough effort to earn a space on HN one way or another -- maybe worth submitting through the 'main channel' if people really don't feel it belongs in Show HN.
ML driven is. LLM driven is still nascent, especially the idea that as large language models get more advanced, can they research and invest like a fund manager.
Founder here: YC actually had an "AI hedge fund" idea in one of their recent "request for startups" post. We've been working on evaluating the capabilities of frontier models in investing money in the stock market. Results are encouraging and we're not doubling down on it.
> Results are encouraging and we're not doubling down on it.
Personally I believe LLM-assisted trading is destined to underperform passive indices, so I also would have moved on from this. But you say results were promising, so I'm interested to hear why you're not pursuing it further. Is it just that you have other things to focus on? Is there something else that's making you move on?
Working on building an investment assistant backed by real time data. ChatGPT and Perplexity finance are amazing, but all of them are based on web search data only, which is a big limitation in finance since realtime data is important.
We have an agent that has access to almost every data point you can think of in the stock market (as much as we can get), which gets leveraged before answering.
And we also figured out ways to build amazing charts in between answer snippets, which looks very cool. Investors are usually very visual.
- First is to actually evaluate whether these LLMs have any intelligence around investing. If you actually give them all the data, can they do well? Can they beat the market? I'm not sure, we're testing that.
- My thesis is that they will actually beat the market (I know a lot of you will disagree). If that's the case, how can we invest a lot of resources in building the best harness, tool calling, etc to enable these models to invest.