I used AI to analyze every single US stock. Here’s how you can use it to make data-driven investing decisions

In this example, we see that this list of stocks with a low P/E ratio barely beat the market (to a negligible degree). Let’s repeat this exact thing for 2022.

In 2022, what were fundamentally strong stocks (rating 4.5+) that had a positive P/E ratio less than 15?

We get the following list of stocks.

Then, like before, we can create another trading strategy.

Create a strategy to buy and hold this list of stocks

From here, we can either update the backtest settings manually.

Or, we can launch the backtest by typing in the chat the exact period we want to simulate.

backtest this portfolio from Jan 1st 2023 to Jan 1st 2024

Regardless, you’ll get the following result.

In this example, the S&P500 outperformed our list of stocks with low P/E ratios. This suggests that trying to buy “fundamentally strong” stocks with a low P/E ratio may not be a better idea than buying SPY.

This analysis is very informative. If your buddy comes to you next week and says “you should buy this stock; it has a low P/E ratio!”, you now have the practical experience to understand that P/E ratio alone is NOT a convincing argument to buy a certain stock.

And yet, even this analysis is incomplete. We can do better.

How to iterate on this analysis?

This analysis just showed ONE possible test. The results of this analysis were lackluster, but how could we do better?

I’ll leave this as homework for the curious reader. Some ideas include:

• What if we sorted by quarterly P/E ratio instead of TTM P/E ratio?
• What if we used P/B ratio?
• What if we chose the 5 lowest stocks by P/E ratio instead of the full list?
• What if we used market cap instead of P/E ratio?
• What if we did 3-year compound annual growth rate (CAGR) instead of P/E ratio?
• What if we filtered the range for stocks to be stocks with a P/E ratio between 15 and 25?
• What if we started backtesting in 2019? Or tested the robustness in 2025 (so far)?
• Better yet, what if we see if this strategy performs well in bear markets vs in bull markets?

Because of NexusTrade, we can test different trading strategies using pure natural language. We can test different technical and fundamental indicators, and gain insights into how they actually performed.

Pick one change from the list, state your hypothesis, and keep everything else fixed: selection date, universe, rating threshold, weights, rebalance rule, costs and test endpoints. Compare that one variant with the original and SPY. Keep the attempted variants in your notes so a lucky winner is not mistaken for the whole search.

Periods you have repeatedly inspected are development data. Freeze the selected rules and data-timing assumptions, then evaluate a later period you have not used to choose them. If the result remains useful after that check, observe the rules in paper trading and compare actual signals and fills with the intended behavior before considering a funded deployment. Passing a tuned Covid and bull-market test does not establish readiness.

Prepare one controlled fundamental-stock experiment. Review its dates, available inputs and rules before running it.

If you made it this far, then I know exactly which option you’ll choose.

And I can’t wait to hear about what you created in the comme