Imagine a student who memorizes the exact answers to last year’s exam without understanding any of the material. On that specific test, they’d score perfectly. On a new test, they’d be lost. Overfitting is that student, turned into a trading strategy.
A strategy is overfit when it has learned the random noise of the past rather than a genuine, repeatable pattern. It fits history so tightly that it has effectively memorized it — including the parts that were pure coincidence and will never repeat.
The tell-tale sign
Overfit strategies share a signature: spectacular on the data used to build them, useless on anything new. The bigger the gap between past performance and fresh performance, the more of what looked like skill was actually just noise being memorized.
Most of the strategies in our Graveyard didn’t fail because the idea was stupid. They failed because their impressive historical results turned out to be overfit — they collapsed the moment they met data they hadn’t been shaped around.
Why more complexity makes it worse
The more knobs and rules a strategy has, the easier it is to twist it into fitting the past perfectly — and the more likely it is fitting noise. This is why we favor simple strategies with few parameters. Simplicity is harder to fake.
Overfitting is mistaking the random texture of the past for a rule about the future. It looks like genius and behaves like luck.
- 1An overfit strategy memorizes past noise instead of learning a real pattern.
- 2The warning sign is a big gap between historical and fresh-data performance.
- 3More parameters make overfitting easier, which is why simpler is safer.
overfitting — the reason most of our Graveyard entries failed.