Log in
Account
Account
Tools

Can machine learning pick better stocks?

Not in our tests. We tried 244 machine-learning variants on NSE stocks, XGBoost included. They picked barely better than a coin flip, and none passed a luck-adjusted test.

By Shashank · Updated

What did we find?

The idea is natural: give a model many inputs and let it find patterns. Machine learning, XGBoost included, picked barely better than a coin flip. None of 244 variants passed the luck-adjusted test, so we rejected it.

Why does trying many versions matter?

With 244 variants, some will look good by chance alone. So a luck-adjusted test matters more here than one good result. A headline figure from one variant tells you little unless you also know how many others were tried.

Does this mean machine learning can't work on stocks?

No. This is one result, on our data and our inputs. It shows that these versions gave no evidence beyond luck, so plain rules stayed in place.

Questions people ask

Does machine learning work for picking NSE stocks?

In our tests, no. 244 variants, XGBoost included, picked barely better than a coin flip, and none passed a luck-adjusted test.

Does XGBoost work for stock selection?

Not in our tests on NSE stocks. XGBoost was among the 244 variants we tried, and none passed the luck-adjusted test.

Why use a luck-adjusted test for machine learning?

With 244 variants, some will look good by chance alone. A headline figure from one variant tells you little unless you know how many others were tried.