Over my professional career, I have (a) run a small Warren Buffett inspired investment fund, (b) become a Good Judgment Project Superforecaster, and (c) started the Metaculus AI Benchmarking tournament, benchmarking LLM-based forecasters against top human pros on complex, real-world problems.
With these experiences, even just a year ago, I was skeptical that AI could be useful to a serious investor.
I've spent many hours these last few months reading Stockfisher's outputs, finding errors, and improving the research and forecasts. Now, I normally can't point out any significant errors in its analysis, even after I spend an hour reading about a company. I've found it quite valuable for my personal investing, something I can't say about even the highest tier of ChatGPT. (This is not investment advice.)
I’m going to explain how I went from skeptical that any approach like Stockfisher could work, to using it for investing my retirement funds.
My experience is that the best forecasters and investors share a few key traits. They are skeptical by default and they think independently. Fortunately, I don't have to try too hard to convince most people to be skeptical of a tool like Stockfisher. Their default assumption is that Stockfisher produces AI-generated slop. I would think so too!
To be totally clear, I do not blindly buy what Stockfisher says is trading at the biggest discount, nor do I simply accept that Stockfisher’s forecasts are good. I do not expect or encourage anyone else to either!
The coming series of posts will center on the central question: How do I assess the quality of Stockfisher’s forecasts? I don’t have a one sentence answer. My judgments are informed by a wide variety of topics, starting with: How do you assess the accuracy of any forecast, human or AI?
This week, I and my team plan to cover:
- How did we build a model of instrinsic company value that could be forecasted accurately by FutureSearch tech?
- What does a good forecast of long-term business fundamentals look like?
- How do you back-test AI forecasters, when LLMs have memorized vast amounts of information that can’t easily be removed from their memory?
Later on, I plan to write more about:
- Some personal anecdotes from my deep dives into specific companies in order to assess Stockfisher’s quality, where I disagreed, and what I ultimately thought a fair valuation was.
- Estimating the expected returns in my portfolio based on having accurate probabilistic forecasts of future earning (not stock prices) for every company.
In this first short post though, I will simply introduce what Stockfisher is.
In a sentence, Stockfisher is a tool for sorting the stocks in the market by the difference between their stock price and their instrinsic value.
Stockfisher draws from the best stock valuation practices from my investment career, best forecasting practices from my forecasting career, and best LLM agent practices from my highest-performing forecasting bot.
Our methodology is based on the value investing approach made famous by Warren Buffett and his mentor Benjamin Graham. The core task is to estimate the intrinsic value of a company, the discounted value of all the cash that a business will generate for its owners during its existence. With the intrinsic value, a long-term investor can then buy stock in those businesses that are trading at a stock price that is less than the intrinsic value.
Most importantly, with this approach, an independent minded, long-term investor can simply ignore the market. You remain free to take advantage of Mr. Market - buying from him if he offers a low price or selling to him if he offers a high price - but by default you can sit back and simply collect the cash produced by the business, exactly as if you were the owner!
Buffett has said that he spends his time reviewing historic financials and reading annual reports to arrive at his own independent assessment of a company’s intrinsic value. Only then does he check the stock market to see what the price is.
And this is what Stockfisher does as well. It does not anchor on consensus estimates (not there are any reliable forecasts for companies in 2030 or 2035 to anchor to anyway), nor does it pay any attention to the stock price when forecasting a business’s future. The app derives its valuation independently of the stock price, and then, like Buffett, compares its valuation to the stock price in the form of a discount or premium.
The first thing you see are the stocks trading at the biggest discounts to Stockfisher’s valuation. Overall you will see roughly the same number of over- and under-valued companies. Stockfisher does not attempt to time the market in any way, instead it aims to use a consistent, principled methodology for all stocks. You’ll find some companies at big discounts and others at big premiums.
The next post will go deeper into Stockfisher’s intrinsic value methodology.
I encourage everyone to draw their own conclusions about Stockfisher’s quality. A great way to do this is to look up a stock that you know well. If you disagree with Stockfisher’s forecasts and rationales, we want to know!