Financial trading using learning-based approach

Tan, Li Xue (2022) Financial trading using learning-based approach. Final Year Project, UTAR.

Preview

PDF Download (1960Kb) | Preview

Abstract

Financial trading has been widely studied and many algorithms and approaches have been applied to gain higher profit. In this work, deep reinforcement learning algorithms were applied to automate the trading process. The data used in this work were 1-minute, 5-minute, and 30-minute candlesticks from different asset classes including Foreign Exchange markets (FOREX), equity indexes, and commodities. The proposed framework utilised data from different time intervals to make a trading decision. For each time interval, an autoencoder consisting of InceptionTime and Long Short-Term Memory (LSTM) was trained to perform feature extraction. The reinforcement learning algorithms applied include Advantage Actor-Critic (A2C), Proximal Policy Optimisation (PPO), and Twin Delayed Deep Deterministic Policy Gradient (TD3). Both discrete and continuous action spaces were studied. The performance of the models was evaluated by using expected return and risk-adjusted return such as the Sharpe ratio. Furthermore, the models were trained under different transaction cost settings to identify the effect of transaction cost on the performance of the models. The results showed that the most consistent model is PPO and SAC performs the worst in this setting. Furthermore, the results also showed that the best transaction cost setting should be equal to or higher than the actual transaction cost. Bachelor

Actions (login required)