Automated Trading ASystem With Reinforcement Learning

• Autor(a) principal: Neves, José Luís Simões
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/160228
• topic: Machine learning; Deep Learning; Automatic trading systems; Deep Reinforcement Learning; Domínio/Área Científica::Ciências Naturais::Matemáticas

This thesis focuses on using deep reinforcement learning to obtain agents capable of making valid portfolio management decisions that result in profit in increasingly more complex and realistic market simulator trading environments. The goal is to use an incremental evolution of the environment features and price patterns to have more control over the training of an agent capable of performing in more challenging and realistic episodes. The model used for this task is the Proximal PolicyOptimization(PPO) algorithm. The thesis intends to present an environment-building-focused training process as a possible tool to answer the inherent challenges of the deep reinforcement learning model’s training and a newperspective on the creation of automatic portfolio management systems with reinforcement learning.