REINFORCEMENT LEARNING FOR AUTOMATED TRADING
By: Reid Falconer, Sam MacIntyre, Hector Cano and Maximilian Zebhauser
In this study, we employ a deep Q-learning approach to algorithmic trading. Our goal is to build a deep Q-learning system that determines when to buy, sell or hold based on the current and historical market data. Our experiments on the both Apple (AAPL) and Wawel (WWL) stocks demonstrate that the deep Q-learning system is highly effective and that the deep Q-learning model outperforms benchmarks such as a random decision policy and a buy and hold strategy.
Software and Packages required
- The code was written Python 3.6
Python
- NumPy
- Pandas
- seaborn
- random
- tensorflow
- matplotlib
- altair
- fix_yahoo_finance
- datetime