You’re absolutely right, past performance of a particular strategy is not an indication that the strategy will perform similarly in the future. However, what else do we have to go on other than past performance? Don’t millions of people invest in US equity indexes with the expectation that future performance will match or come close to historical performance of that index?
Is that such a good assumption? What’s to say the US market doesn’t go the direction of Japan?
To combat the threat of overfitting, Portfolio123 offers numerous way to test robustness, including interchangeable universes and ranking systems, to ensure a strategy works across a wide range of assumptions. They’ve also published educational materials, including a strategy design class, that teach best practices in avoiding common pitfalls such as overfitting.
In defense of overfitting in general, O’Shaughnessy published several strategies that have returned above 15% annually since 1964 (limits of his data), and those strategies have continued to work since being published (out of sample).