[Paper Review] Optimizing Portfolio Management and Risk Assessment in Digital Assets Using Deep Learning for Predictive Analysis

Qishuo Cheng, Le Yang+3 more|arXiv (Cornell University)|Feb 25, 2024

Stock Market Forecasting Methods17 citations

TL;DR

The paper introduces a Deep Q-Network approach for portfolio management in digital assets, highlighting improvements over benchmarks and addressing cross-asset Q-value drift with distribution-sharing constraints.

ABSTRACT

Portfolio management issues have been extensively studied in the field of artificial intelligence in recent years, but existing deep learning-based quantitative trading methods have some areas where they could be improved. First of all, the prediction mode of stocks is singular; often, only one trading expert is trained by a model, and the trading decision is solely based on the prediction results of the model. Secondly, the data source used by the model is relatively simple, and only considers the data of the stock itself, ignoring the impact of the whole market risk on the stock. In this paper, the DQN algorithm is introduced into asset management portfolios in a novel and straightforward way, and the performance greatly exceeds the benchmark, which fully proves the effectiveness of the DRL algorithm in portfolio management. This also inspires us to consider the complexity of financial problems, and the use of algorithms should be fully combined with the problems to adapt. Finally, in this paper, the strategy is implemented by selecting the assets and actions with the largest Q value. Since different assets are trained separately as environments, there may be a phenomenon of Q value drift among different assets (different assets have different Q value distribution areas), which may easily lead to incorrect asset selection. Consider adding constraints so that the Q values of different assets share a Q value distribution to improve results.

Motivation & Objective

• Motivate the need for improved deep learning-based portfolio methods beyond single-prediction models.
• Propose integrating market-wide risk factors into asset-level decision making.
• Evaluate the effectiveness of a DRL-based portfolio strategy against benchmarks.
• Identify and mitigate Q-value drift across independently trained asset environments.

Proposed method

• Introduce the Deep Q-Network (DQN) algorithm into asset management portfolios.
• Implement strategy by selecting assets and actions with the largest Q value.
• Treat different assets as separate environments to train independently.
• Investigate Q-value drift due to different Q value distributions across assets.
• Suggest adding constraints so that Q values across assets share a distribution to improve consistency.

Experimental results

Research questions

• RQ1 Can a DRL-based approach using DQN outperform traditional benchmarks in digital-asset portfolio management?
• RQ2 How does treating each asset as a separate environment affect learning dynamics and decision quality?
• RQ3 What causes Q-value drift across assets, and how can distribution-sharing constraints mitigate it?
• RQ4 Does incorporating broader market risk data improve predictive performance beyond isolated asset data?
• RQ5 What constraints on Q-value distributions yield more stable and reliable asset selection?

Key findings

• The DQN-based portfolio management approach substantially exceeds the benchmark according to the authors.
• Training assets as separate environments reveals Q-value drift issues that can affect asset selection.
• Addressing Q-value distribution drift through distribution-sharing constraints can improve results.
• The study highlights the importance of considering market-wide risk factors beyond single-asset data.