# Application of Deep Q-Network in Portfolio Management

**Authors:** Ziming Gao, Yuan Gao, Yi Hu, Zhengyong Jiang, Jionglong Su  
**arXiv:** [2003.06365](https://arxiv.org/abs/2003.06365)  
**Date:** 2020-03-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2003.06365v1/

## Key points

- Various applications of machine learning and neural networks explored:
- Stock market prediction
- Face recognition
- Population analysis
- Application of Deep Q-Network (DQN) for portfolio management in the stock market:
- DQN is a deep reinforcement learning algorithm optimized by Q Learning
- Discretization of action space to define weight of portfolios in different assets
- Combination of Convolutional Neural Network and dueling Q-net techniques to enhance recognition capabilities
- Experimental results showcasing superiority of DQN-based strategy over traditional strategies:
- Profit generated by DQN surpasses other strategies by 30%
- Lower risk levels indicated by metrics like Sharpe ratio and Max Drawdown
- Study authored by Ziming Gao, Yuan Gao, Yi Hu, Zhengyong Jiang, and Jionglong Su highlighting the potential of advanced deep reinforcement learning algorithms in revolutionizing portfolio management practices within financial markets.

## Summary

In the realm of machine learning and neural networks, various applications have been explored. These range from stock market prediction to face recognition and population analysis. This paper delves into the application of a classic Deep Reinforcement Learning algorithm known as Deep Q-Network (DQN) for portfolio management in the stock market. DQN is a type of deep neural network optimized by Q Learning, making it suitable for tackling complex financial problems. To adapt DQN to the financial market environment, the authors discretize the action space and define it as the weight of portfolios in different assets. This transformation enables DQN to effectively address portfolio management challenges. Furthermore, the authors enhance the algorithm's recognition capabilities by combining Convolutional Neural Network and dueling Q-net techniques. In an experimental setting involving five low-relevant American stocks, the results showcase the superiority of the DQN-based strategy over ten traditional strategies. The profit generated by the DQN algorithm surpasses that of other strategies by 30%, indicating its efficacy in maximizing returns. Additionally, when considering risk assessment through metrics like Sharpe ratio associated with Max Drawdown, it is evident that policies formulated using DQN exhibit lower risk levels compared to alternative strategies. Authored by Ziming Gao, Yuan Gao, Yi Hu, Zhengyong Jiang and Jionglong Su, this study sheds light on how advanced deep reinforcement learning algorithms can revolutionize portfolio management practices within financial markets. By leveraging cutting-edge technologies and methodologies such as DQN and neural networks' integration with financial data analysis, this research paves the way for more efficient and effective investment strategies in today's dynamic economic landscape.