An overview of stock trading with sentiment analysis

• Main Author: Tang, Yi Qwan
• Other Authors: Erik Cambria
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Trading; Sentiment analysis; Stock; Finance; Natural language processing
• Online Access: https://hdl.handle.net/10356/175215
• Institution: Nanyang Technological University

This study investigates how sentiment analysis, using advanced Natural Language Processing (NLP) techniques, can predict stock market trends, specifically examining Boeing stock (ticker: BA) over three months. Utilising multiple sentiment analysis models like Sentic API, TextBlob, VADER, BERT, and FinBERT, alongside varying preprocessing methods and news formats, the research explores optimal trading signals for high-frequency trading. The findings indicate that FinBERT, tailored for the financial sector, significantly outperforms other models, achieving up to a 63% return on investment. Results show that preprocessing techniques like stop word removal and lemmatisation do not significantly impact performance, while trading signals derived from full news content and summaries yield better results than those from headlines alone. The study challenges the Efficient Market Hypothesis by using random strategies as benchmarks, demonstrating that sentiment-driven strategies can exploit market inefficiencies to generate superior returns.