An Application and Visualization of Deep Learning Neural Networks and Reinfrocement Learning for Stock Portfolio Management

15 May 2020

---

In this project, I have used the deep learning neural network, LSTM to predict the stock prices. The result shows that the LSTM network captures the memory of the time series data and can predict the stock price return well. I then framed the stock portfolio selection problems as a multi-bandit problem where the agents can select from the results from CNN network, mean-variance portfolio strategy, and uniform buy and sell strategy to gain the most benefit from the stock market. The results show that reinforcement learning can potentially help stock traders better allocate stock portfolios.

However, since deep learning network models are highly sensitive to factors such as hyperparameters and network structures, one could also investigate different network structures and calibrate their hyperparameters, to obtain more satisfactory forecasting results.

The video poster is here: https://youtu.be/SHpspYWX0nE