DEEP REINFORCEMENT LEARNING-BASED STOCK OPTIMIZATION: A STUDY OF DEEP QNETWORK ALGORITHM

• Main Author: Abraar Abhirama, Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87212
• Institution: Institut Teknologi Bandung

The algorithmic trading trend has been gaining significant attention recently. The rapid development of artificial intelligence algorithms and the allure of investing in the stock market have made algorithmic trading appealing, both from an academic perspective, such as algorithm development, and an investment perspective, such as business applications. This thesis aims to implement the Deep Q-Network algorithm, which is part of Deep Reinforcement Learning, and compare the results with conventional ]optimization methods: the DJIA method and the Sharpe ratio method. Unlike supervised and unsupervised learning, reinforcement learning operates by simulating the system into two main components: the interaction between the agent and the environment. The data is sourced from the Dow Jones Industrial Average, sorted by market capital, and divided into three periods: training (January 1, 2009, to December 31, 2019), validation (January 1, 2020, to December 31, 2021), and testing (January 1, 2022, to December 14, 2023). Simulation results show that the portfolio returns achieved outperform the two conventional methods. However, it is important to note that these results are not absolute, as the environment created must be continuously updated to reflect stock market fluctuations.