Building an AI-based algorithmic trading strategy

• Main Author: Mok, Zhe En
• Other Authors: Wong Jia Yiing, Patricia
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/158192
• Institution: Nanyang Technological University

Recently, with the development of Artificial Intelligence in finance, using it in stock market trending analysis and prediction has attracted massive attention from various investors and researchers. It has always been a well-known area of study since the creation of Artificial Intelligence. In the last decade, the progress in machine learning has introduced new prospect with new algorithms and models that can be used in stock price prediction. It is almost impossible to predict future stocks prices precisely without any error but Artificial Intelligence can help us to make better decisions on when to buy, sell or hold a stock so the investment can be profitable in the long run. A decent stock investment strategy normally consists of different stock prediction methodologies such as technical analysis, sentimental analysis and fundamental analysis. Artificial Intelligence is well known to have an outstanding advantage in quantitative finance. However, it is difficult for a trader to develop an agent that automatically execute actions to buy or sell stocks in the market due to strenuous debugging and hard programming. Traders often must specifically choose which stock to trade, at what price and what quantity [54]. It would be great to have an AI agent to help trader in making decisions. This project starts with a detailed study on recent applications of different deep reinforcement learning models and long short-term memory model in financial tasks. For deep reinforcement learning model, actor-critics approach is the most popular one and it fits well 3 with the volatile and complex nature of financial market. Six state of the art models such as LSTM, DDPG, A2C, PPO, TD3 and SAC are studied in detail. First, we choose a stock and then collect indicators of the stock data. Next, we preprocessed the historical price and feature data then feed into models for training, the trading strategy is learnt from neural networks. A simulated environment is then created to be used for evaluating trading decisions output by model. Lastly, we set an initial capital and backtest the model on test data and see the amount of profit gained from trading the stocks. Furthermore, the transaction cost is incorporated.