How to Use AI for Investment Research
Tips and tricks to compress reading time, find growth markets, and pick stocks
Our AI series continues with how investors can use AI like ChatGPT and Bard for investment research.
The Deload educates growth investors on where the world is going. If you haven’t subscribed, join over 1,400 rational optimists:
A Free Course on Using AI for Investment Research
The best investors over the next decade will not only invest in AI. They will use AI to enhance investment research and creativity.
Early adopters of AI tools like ChatGPT and Bard will gain superpowers, just like early adopters of the Internet. But using AI for investing superpowers is a skill that most users don’t yet have. Not only that, learning how to use AI at a deep level will give you a greater respect for the power of the technology and the potential of the space for investment.
This post will teach you the basics and give you a few tools to make you an AI-superpowered investor.
AI Use Cases
AI grants investors three core superpowers:
Information compression. You can use AI to condense articles, SEC filings, and other digital information into easier to accelerate acquiring knowledge about an investment.
Idea generation. You can use AI to find companies as well as emerging trends and consumer behaviors.
Stock picking. You can use AI to select stocks it thinks are worthy of investment.
I grade AI an A for information compression, a B for idea generation, and a C for stock picking. We’ll explore each of these superpowers in more detail over the next several emails.
While the advanced AI investor can wander into other derivative use cases, the beginner benefits most by knowing which of the three use cases you want to leverage. Poor AI experiences often come from a lack of intent from the user.
Once you know whether you want AI to be a research assistant that helps you with information, a creative brainstorm partner that helps with idea generation, or a portfolio manager that picks stocks, then you can write an effective prompt to achieve your goal.
Before we go into each of the core AI superpowers in more detail, you need to know more about prompting.
Prompts are the commands or questions you submit to an AI to generate some action. “Make me a list of your 10 favorite stocks” is a prompt.
You can think about prompting in four ingredients:
You need each of these ingredients to make a good prompt. Remember, an AI is just a talented improv actor. The more information you give it, the better you extract its talents.
Format clarifies the structure of the output you want from your AI. An investor might tell its AI to, “Write me a 500-word summary about Apple’s June spatial computing event with three key bullet points.” You’re giving the AI instructions about how much output you want and how it should look.
Note: While ChatGPT is arguably a better overall product, I often find myself using Bard for investment-related prompts. Bard is natively connected to the Internet, and that connection just seems to be more flexible than using it with ChatGPT via Bing or Plus. Experiment with both, but just know that if you’re a normal ChatGPT user, you may find some frustration in getting it to do some of the example requests throughout this post. The example posts I’ve used throughout were processed on Bard.
Specificity adds more exact detail about what you’re trying to achieve. “Write me a 500-word summary about Apple’s spatial computing event in June with three key bullet points detailing everything announced about the hardware components.” Now you’re giving the AI more information about your specific aim with the prompt.
Context further information about the purpose of the prompt. “You’re an investment analyst. Write me a 500-word summary about Apple’s spatial computing event in June with three key bullet points detailing everything announced about the hardware components for me as a portfolio manager looking for investment opportunities.” Now the AI knows what role it’s supposed to play.
Examples finally refine what you’re going to get out of the AI. “You’re an investment analyst. Write me a 500-word summary about Apple’s spatial computing event in June with three key bullet points detailing everything announced about the hardware components for me as a portfolio manager looking for investment opportunities. A good bullet might look like this: Apple’s Vision Pro uses vscel laser arrays to understand the external environment.”
You can see our final prompt is several sentences long. It probably looks a lot different than the early prompts you might have used in ChatGPT that were just a short sentence or two. Asking open-ended generic questions is almost guaranteed to get basic responses that you won’t find useful. Just like asking specific questions in real life yields better answers, asking specific questions of an AI to help with investment work will yield better results.
Your prompt checklist should look like this. Did I:
Give a length and format I want?
Specify exactly my goals?
Provide context about the AI’s role and the audience?
Show an example to maximize my odds of getting what I want?
The more you prompt, the more your prompt style will develop. You might find it useful to save your best prompts as they’ll often have structures you can reuse or contain questions you ask frequently.
Prompting 201: The Meta Prompt Trick
My favorite prompt trick is to use ChatGPT or Bard to help generate great prompts for me. What better way to make sure you ask the machine a good question than by asking the machine how to frame what you want into a good question?
This prompt from Jason West works well:
“I want you to become my Prompt Creator. Your goal is to help me craft the best possible prompt for my needs. The prompt will be used by you, ChatGPT. You will follow the following process: 1. Your first response will be to ask me what the prompt should be about. I will provide my answer, but we will need to improve it through continual iterations by going through the next steps. 2. Based on my input, you will generate 3 sections. a) Revised prompt (provide your rewritten prompt. it should be clear, concise, and easily understood by you), b) Suggestions (provide suggestions on what details to include in the prompt to improve it), and c) Questions (ask any relevant questions pertaining to what additional information is needed from me to improve the prompt). 3. We will continue this iterative process with me providing additional information to you and you updating the prompt in the Revised prompt section until it's complete.”
The AI will take you through several iterations of clarifying questions where it will add detail to its proposed prompt. It will eventually run out of questions and give you a final prompt, or you can stop it at any time.
The goal of information compression is to help you consume information about an investment in a more efficient way. There are several ways to achieve this.
Feed AI a Link
You can give your AI analyst a link to have it synthesize an article or document. A document synthesis prompt might be:
“You’re my investment analyst. From this document https://investor.okta.com/static-files/8a461359-83f8-4332-9f41-eb7ee8c65c3b, summarize how Okta makes money in 1,000 words. Provide detail on the company’s major business lines, how much revenue it generates from each business line, and how it recognizes revenue. Make each different business line a section of the writeup.
Section 1: Customer Identity Cloud
Okta’s Customer Identity Cloud helps enterprise companies XXXX. The company makes XX% of its total revenue from Customer Identity Cloud. Revenue is recognized XXX.”
Feed AI Your Own Documents
There are plugins like AskYourPDF that allow you to upload PDFs and other files to ChatGPT for the same type of analysis as above; however, you need to have a Plus account to use plugins ($20/month). Another solution is to use Bearly, a paid service built for both analyzing docs provided by links or files.
If you’re an AI pro, you can take the uploading technique even further to build a second brain.
I have thousands of docs in my Apple Notes on books, podcasts, and articles I read as well as random ideas I have from observing the world. Unfortunately, those notes are only occasionally revisited and not in any efficient manner. It can take some money , you can have conversations with your notes and have AI systems automatically unearth prior insights for remembrance.
Ask AI for a Chart
You can ask ChatGPT via the diagr.am plugin to make charts of information from a link to help you visualize information. The chart request doesn’t work well on Bard, although Bard will give you tables that you can export to Google Sheets.
The charts and tables usually aren’t the most beautiful things you’ve seen, but if you’re a visual information consumer, it can give you a format that agrees with your learning format.
I use AI for idea generation in two main ways: company discovery and trend finding.
You can use AI as a simple screening tool to find companies by feeding it parameters about what you’re looking for. “List EU-based growth companies that have a unique, hard-to-replicate product, high customer affinity, and a strong brand-driven moat. They should be valued at a maximum of $5 billion with an annual revenue exceeding $100 million.”
That prompt yielded Pipedrive, a private SAAS company I had previously not heard of.
I’ve always appreciated the Tim Ferriss idea that many of the best early opportunities come from what niche groups are doing — rich people, geeks, experimenters, super hip influencers, bodybuilders, etc. These groups are often ahead of the curve. Whatever they’re doing today, a lot more people will be doing in a few years. AI can help us discover esoterica that may be where the world is going.
With Bard or ChatGPT on Bing since they’re connected to the Internet, you can ask the AI to tell you about trends that are emerging on services like Google Trends or Reddit. Pick a narrow category or subreddit and get deep into it. “You are my research assistant. We're looking for problems to solve for the longevity community. Analyze the Longevity subreddit here (https://www.reddit.com/r/longevity/) and give me five frequently recurring problems that users discuss in posts and commons over the past three months. Please only select problems that you see recur at least 5 times in the three-month period.”
That search yielded an idea around senolytics — a class of drugs that eliminate senescent cells which may help with aging. Many Redditors complain about the cost of the drugs. Perhaps something rich longevity experimenters do today that many people will do in the future.
Investors have used ChatGPT to beat markets already.
However, while you can use AI to “pick stocks;” there is a challenge inherent to the say the systems work to get useful output. Prompts dictate the quality of an AI’s response, but the more detailed your prompt, the more your question turns into a stock screen given a bunch of criteria you care about rather than the AI picking stocks of its own volition. If you give Bard a lot of criteria like high FCF yield, ROIC over 20%, revenue growth of 10%, etc. you end up with a list of stocks that meet that criteria more than thoughtful stock picks.
You could ask Bard something very generic like pick five stocks it thinks will beat the S&P 500, but you get a generic response that tells you to buy AAPL, MSFT, AMZN, GOOGL, and TSLA.
Thanks for telling me to buy megacap tech, Bard. Although if you followed that advice at the beginning of the year, you’d probably be pretty happy.
AI also be useful for you if you’re interested in understanding momentum, although that’s different from investing on fundamentals.
One way to get data on sentiment is via a query like this: “Categorize the sentiment of tweets about Apple stock over the past month from this link https://twitter.com/search?q=%24aapl&src=typed_query as positive, neutral, or negative. Count the number of total tweets in the search, how many each are categorized as positive, neutral, or negative, and the percentage of tweets that are positive, neutral, or negative.”
Humans + AI
We like to debate whether AI will replace human jobs or enhance us. As with so many debates about technology, time scale matters. For the next decade plus, maybe several decades, AI will be an augmentation tool. Humans who learn to embrace it and learn how to use it will have effective superpowers. Those who don’t will fall behind.
Investing is the same. The tips and tricks I shared here give you 80% of the power of AI for investing. As you experiment with ChatGPT and Bard, you’ll learn new creative ways to build them into your process for better results.
But remember. While AI can help improve your investment process, the returns are still on you. At least for now.
Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.