I Tried Market Prediction with Machine Learning

It sounds like the validation set was seriously overfit.

If you run 10'000 experiments over thousands of indicators, you're bound to find some that idiosyncratically perform well on the validation set but underperform out-of-sample.

You also can't conclude that "there was no subtle underlying pattern" just because you couldn't find one...

Original for Medium-phobes:

https://shortcircuitsandinfiniteloops.blogspot.com/2019/11/w...

Thanks for this. I still want to try, even if it does fail. Sounds like my idea of a very good time.

I agree! It sounds like a fun experiment.

So many more failures should be published and documented like this, it's immensely valuable.

Thanks to the author for writing this article.

Yeah, those were my thoughts. The idea of making money from projects like this is alluring but probably unrealistic, and it's good to have an honest assessment like this.