AI-based stock market trending analysis

• Main Author: Ko, Johann
• Other Authors: Li Fang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/138093
• Institution: Nanyang Technological University

In this paper, we present two hierarchical neural networks for market trends predictions. These models utilised sentiment analysis of news as well as past information of returns and prices to predict the next day trend (i.e. bullish, stagnant, bearish). The hierarchical models were trained in the context of swing trading of 2 days using price actions of Dow Jones Industrial Average (Ticker: DJI). Experimental studies showed an F1-accuracy of 0.53 on this 3-class problem with Hierarchical LSTM. This was a considerable improvement over the industry-standard model, ARIMA. The Hierarchical LSTM came out as the best performing model.