Market Pulse: How to get fooled by AI when building a strategy

On Fridays, the focus is on systematic stock strategies, strategies with clear rules for when we buy and sell, and how we manage positions and risk.

Last week we tested a trend-following strategy from the book Automated Stock Trading Systems, first on American stocks where it held strong even six years after the book was published, then on Swedish stocks where the results were significantly weaker. I ended with a cliffhanger: I had a computer optimize a significantly better figure for the Swedish market, over 20 percent in annual returns. The question I left you with was whether that figure could be trusted.

This week I’m giving you the answer, and it’s about more than just a single strategy. We live in an era where AI tools make it faster and easier than ever to test thousands of combinations and find an impressive curve. But a tool that finds patterns for you can’t always tell you whether the pattern is real or just a coincidence that happened to look good in history. That’s exactly the trap we’re walking through today, concretely, with real numbers, not just as a passing warning.

We show what actually happened when we tested the ”winning” optimization on the period that matters most. Then we show the path that actually works, fewer parameters, a genuine separation between the data we optimize on and the data we test on, and why a strategy that looks a little worse on paper is sometimes the one you should trust the most.

In today's episode I'll go over:

1. Conclusion on the cliffhanger, did the optimized strategy hold?
2. The fundamental trap, why more parameters are not better, and what it means for those of you who use AI tools
3. A genuine pattern in the noise, an insight that actually holds true
4. The rugged road, and how it performed when it really mattered

Create an account
to read the full article

Create an account to read the full article

This article is part of our analysis service. Create an account to get full access to analysis and our strategies.

Get 50% off Premium

If you already have an account, you can

log in