An update on my ML for MI odyssey:

For the last 3 years I’ve been intermittently plugging away at an evolving system of machine learning for my version of mechanical investing. At this point I no longer have any qualms in saying it is a very sound method of selecting the stocks that have a higher-than-normal probability of outperforming. In fact, all it all boils down to is a large set of tools that allow you to use a lot of historical data to determine which factors work together to make a more logical ranking of predicted outperformance. What complicates the whole process is that no one factor works alone to accurately make a prediction. Each factor changes its effect as the mood or outlook of the market changes.

I’m still in the experimenting, learning and gradually improving the system mode. But the first complete ensemble of models built almost 2 years ago and trained on data till end of 21, has now run for 3 years with no retaining. So far it has a post training data from beginning of 2022 a 28.5% CAGR net gain with a 165% annual Turnover. The same system holding 50 stocks actually does a little better with a 28.7% CAGR and 175% annual turnover. Those results are with simulated trading as at the beginning I was not confident enough in the system to actually fund it.

Interestingly some others have apparently discovered that after observing the ML system they could use the ML selected features to build a conventional ranking system that does as well as the complex ML system.

Unfortunately, at this point in time, it is not a simple task to build a working ML system that outperforms.

Mechanical Investing FAQ

---

Contact Shrewd'm
Contact the developer of these message boards.