I know of one machine learning approach which is currently in use by at least one hedge fund. numer.ai is using an ensemble of user-provided machine learning algorithms to direct the actions of the fund.
In other words:
A hedge fund provides open access to an encrypted version of data on a couple of hundred investment vehicles, most likely stocks. Thousands of data scientists and the like train all sorts of machine learning algorithms against that data and upload the results to a scoreboard. The highest scorers get a small amount of money depending on the accuracy of their results and how long their result has been available online.
The best predictions are supposedly made by ensembles of algorithms.
So you have a lot of scientists providing trained guesses, some of which are themselves ensembles of guesses and the hedge fund uses the ensemble of all provided guesses to direct their investments.
This rather interesting hedge fund's results taught me two things:
- Ensembles are often viewed as a good way of making predictions on the stock market.
- Good predictions require more ensembles than I'm willing to build myself...
If you want to have a go, visit: https://numer.ai/
No, I'm NOT affiliated with them, I'd most likely not spend my days online were I connected to a hedge fund that employs thousands of people, but paying only those that provide measurable results :)
The numer.ai community has a forum where they discuss their approach so you CAN learn from others who are trying to do the same.
Personally I think anyone with a good algorithm is going to keep it very, very secret.