Automating Stock Trading Decisions with Machine Learning

For many years, chartists–also known as technical analysts in the investment industry–have scoured the stock charts to look for patterns in the trading and the indicators that the prices are about to change. Today, the stock market produces so much data every hour that we need computers to do most of that work; and this is an area where machine learning shines.

Tucker Balch, Ph.D, co-founder of Lucena Research, developed Georgia Tech’s Machine Learning for Trading course as a way to teach students not only the basics of the stock market but also some common techniques for quantitative analysis.

My final project for the class applied reinforcement learning (specifically, a Q-learning algorithm I wrote in Python with NumPy and pandas) to a stock’s historical chart to “learn” the indicators for when to buy or sell. Applying those indicators to the next two years of trading generated a 184% return!

Of course, those returns are wholly fictional, and I doubt I’d ever trust my investment decisions to my algorithm. But on the balance this class provided a great way to practice and apply some of the machine learning algorithms to a real-world situation while dreaming of big bets and no whammies.