Reinforcement Learning for Finance

Chapter 6. Algorithmic Trading

Automated stock-trading systems are widely used by major investing houses. While some of these are simply ways of automating the execution of particular buy or sell orders issued by a human fund manager, others pursue complicated trading strategies that adapt to changing market conditions.

Bostrom (2014)

Financial giants such as Goldman Sachs and many of the biggest hedge funds are all switching on AI-driven systems that can foresee market trends and make trades better than humans.

Maney (2017)

In Chapter 3, the deep Q-learning (DQL) agent learns to predict the future direction of the price movement of a financial instrument. We call this a financial prediction game. It is a natural progression to interpret the prediction game as a DQL agent learning to algorithmically trade in financial markets. A prediction of an upward movement can be interpreted as taking on a long position in the financial instrument of interest. Analogously, the prediction of a downward movement is interpreted as taking on a short position. Over time, the predictions might also imply keeping the current position open.

In addition to this reinterpretation of the prediction game as algorithmic trading, the financial side needs to be implemented. Taking on a long or short position in a financial instrument leads to a positive or negative return on such a position. Therefore, to assess the financial performance of the algorithmically trading DQL agent, its positions must be ...