Deep reinforcement learning for investing: A quantamental approach for portfolio management

• Autor(a) principal: Maltêz, Fábio Alexandre Afonso
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/26849
• topic: Deep reinforcement learning; Investments; Portfolio management; Quantitative finance; Quantamental investment strategies; Aprendizagem por reforço profunda; Investimentos; Gestão de portfólio; Finanças quantitativas; Estratégias de investimentos quantamentais

The world of investments affects us all. The way surplus capital is allocated by ourselves or investment funds can determine how we eat, innovate and even educate kids. Portfolio management is an integral albeit challenging process in this task (Leković, 2021). It entails managing a basket of financial assets to maximize the returns per unit of risk, considering all the micro and macro economical, societal, political and environmental complex causal relations. This study aims to evaluate how a machine learning technique called deep reinforcement learning (DRL) can improve the activity of portfolio management. It also has a second goal of understanding if financial fundamental features (i.e., revenue, debt, assets, cash flow) improve the model performance. After conducting a literature review to establish the current state-of-the-art, the CRISP-DM method was followed: 1) Business understanding; 2) Data understanding; 3) Data preparation – two datasets were prepared, one with market only features (i.e., close price, daily volume traded) and another with market plus fundamental features; 4) Modeling – Advantage Actor-Critic (A2C), Deep Deterministic Policy Gradient (DDPG) and Twin-delayed DDPG (TD3) DRL models were optimized on both datasets; 5) Evaluation. On average, models had the same sharpe ratio performance in both datasets – average sharpe ratio of 0.35 vs 0.30 for the baseline, in the test set. DRL models outperformed traditional portfolio optimization techniques and financial fundamental features improved model robustness and consistency. Hence, supporting the use of both DRL models and quantamental investment strategies in portfolio management.