Sharing my AI stock analysis

Community
,
1

Hi Open AI Community,

TL;DR: I am sharing the tool I built to assist me with my investments

I have been investing for over a decade, and I have always struggled with:

  1. Putting the time to do actual research
  2. Trusting analyst’s recommendations

To * invest * we want to make sure that the price is right and the company will keep improving.

How do we know that “the price is right?”

How can we predict that the company will perform better in the future?

To answer the above, we have to look at the company’s financials and their trends. We have to compare the company with its peers/competitors. We should understand its business model, the sector and geography the company evolves in and the perspective of the economy in general.

We can also look at additional signals like insiders selling or buying.

Just for one investment, this is already a lot of work. And a work that we need to repeat every time there is a significant change - for example, a significant price change from our last analysis or new quarterly results, etc.

To automate all the above, I have built a stock analysis tool and have been using it the past years for my own investments. I have been adding LLMs agents (GPT 4o & Claude 3.5) to perform the qualitative analysis. Recently, I decided to share it and keep on building it in public.

In this initial version, you can get the summary of the stock analysis my model generates.

For now, it covers most of the S&P and Nasdaq stocks.

Here is the link :point_right: https://undervalued.ai

If you are into investing yourself, please feel free to reach out. I would love to get your feedback and know more about your methodology.

3 Likes
2

Cool but…no OCGN or TQQQ?

“Choose from the 5000 NYSE & Nasdaq stocks”

If you’re looking into undervalued, OCGN should be there! :wink:

Would be nice to have ETFs as well.

3
  • TQQQ is an index. I don’t intend to do those.
  • OCGN. For now, I filter out speculative stocks and industries that are hard to understand (healthcare = High R&D costs, binary outcomes, complex regulations, difficult to predict success)

If you have a proven methodology for those, please let me know!

1 Like
4

Sounds like a great way to lose money.

This is not a problem you can solve with LLMs. Whatever you can imagine coming up with, blackrock has something literally a million times more powerful, with humans to monitor it and a couple trillion dollars of buying power to back it up. That doesn’t even get into things your model can’t possibly deal with like how market makers function. The only thing you can really do is try to ride along on their coattails and try not to get left with a bag.

2 Likes
5

Thanks GoldenJoe! I’m busy burying my life savings in a tin can because apparently if we can’t beat Blackrock investing is pointless.

1 Like
6

95% of retail traders fail. Have fun gambling against the house.

7

Thank you for your insight GoldenJoe.

So far it did work for me. The way I like to think about it is the following:

  1. I want to invest in a company (vs Index) => I need to do research.
  2. I already use (and trust) GPT4/Claude3.5 to assist me in synthesizing information and pulling out key data.

Also:

  1. Too many signals do not necessarily translate into better decision-making. I think it degrades performance.
  2. Huge players cannot enter/exit positions like retail. Apart from operational and regulatory (fillings) constraints, any large move carries the risk of impacting the market. In a sense, they are market makers.

If you had bad experiences in stock picking, please share (or DM) what happened / how you made your investment decision.

8

Seems you are referring to day trading based on technical charts. What about retail investors?

9

I’m working on something similar. Realize we are up against the big financial institutions with super computers and algorithms but I dont care. Using AI to help me analyze stocks makes me feel a llittle more confident.

So far my site every evening gets the top 30 stocks in the news and fetches the latest 44 technical indicators per stock, news sentiment, fundmentals, etc. I send that info to a gpt-4mini AI assistant. Get a structured response back that contains the 44 indicators, buy/sell/hold recomendatoion, and price range .

The challenges have been reducing the characters of the data to send to open ai and working on the system instructions for the ai bot. 30 stocks x 18k characers or around 500k characters per day at the gpt4mini price is around $.08 per day.

I’m testing it with my own money through options trading because its just short term analyiss (less than 30 days). Realize I’m in no position to give financial advice but I am having a lot of fun with this. If you want to run ideas by each other let me know.

1 Like
10

The main question is how much better your tool is compared to analysts’ recommendations or your own analysis. Are there any statistics?

11

I guess each analyst has a different performance.
So far, I have been using an evolving version of this tool - in combination with my own assessment - over the past 3 years. The performance was better than the index.

However, I have not let the AI trade on its own (even with paper money) so I cannot say how much better it is as a standalone.

That being said, it is in the roadmap and the performance will be shared transparently. If you are interested, please add your email to the mailing list so that I can:

  1. Keep you updated
  2. Ensure you get lifetime access to whatever perks I add in the future.
1 Like
12

Very cool!

Question: You run it all in a single gpt-4mini call? Isn’t the performance a bit weak (44 elements to analyse might lead to a loss of focus for a weaker model no?)

Realize we are up against the big financial institutions with super computers and algorithms

=> I disagree. We are not doing arbitrage or high-frequency trading. In addition, big players don’t trade like us (their positions are scrutinized and not very liquid).

13

The operative words. I can’t host a sermon on the stock market in a forum post. Take some time away from your project to research how market makers work, how dark pools work, what percent of trading is done by computers, who runs those computers, how they make money, how all of this ignores the order book you’re looking at, etc.

That’s not what a market maker is.

You can research K2s until you are blue in the face. Fundamentals are invalidated by the above. Nothing I say is financial advice. Good luck.

1 Like
14

This isn’t generally true, but is true for the OPs solution.

The rule of thumb though for any post like this is pretty simple however - if it’s any good, why are you advertising it. Anybody who does this kind of thing well knows that just devalues their solution and makes it worthless, throwing out all that work you put into it!

Trying to get returns above index funds is adversarial. Publicizing is very bad strat.

Well, unless it’s scam and you’re trying to front run I guess. Profitable but immoral and illegal.

1 Like
15

So far, I have been using an evolving version of this tool

I am referring to the evolving nature of this tool. You can actually remove “So far”, meaning is the same.

Fundamentals are invalidated by the above.

Disagree. Fundamentals are necessary for value investing - although, in the short term, other drivers have more influence ( In the short run, the stock market is a voting machine . But in the long run, it is a weighing machine)

That’s why Undervalued.ai uses multiple other data like: peer performance comparison, macro-data, technical signals.

Feel free to share valuable insights if you have some. Maybe a blog post and share the link to it?

16

Not really - hedge funds (over 100m) must disclose (K13) their positions in public equities. Albeit with a delay of 45 days. Many outperform the index.

if it’s any good, why are you advertising it.

Because there is value in it.

In this version, you get the “stock assessment made by AI” on Undervalued.ai (alerts coming soon), and it’s free. Eventually, I will let the model trade and share the performance transparently.

Finally, say 20% of the stocks are ranked as buy/strong buy - that could be hundreds of stocks (depending on universe size). Will you buy all of them? Will you do equal weighting?
The next day, one stock takes 10%, and Undervalued.ai downgrades the stock. Will you know? Will you sell? Etc.
=> Knowing if a stock is likely undervalued is not enough, but it is a good start.

17

I like it. Simple, quick, easy tips on each stock you plug in.

But why should I trust it?

1 Like
18

But why should I trust it?

That’s the most important point!
And that’s also the reason why I built it myself rather than relying on analysts or other tools.

As for answering this question. It’s difficult to trust it today because:

  1. My performance was not solely derived from using Undervalued;
  2. AIs / data source / system can have/ make mistakes;
  3. You don’t know me;

And more importantly, trust is earned over time.

That being said, I am building in public and will share backstage content like the methodology.
In addition, I will build a model that will manage a portfolio based on those insights and display its performance.

If you think of other things, or have feedback, would love to hear them (here or DM)

1 Like
19

Perfect. I look forward to following this project!

1 Like
20

This seems useful!

I would use this as a research tool, but maybe not auto buy with it.

1 Like