Forecasting stock price movements with tweet sentiment, volume and interaction level

• Main Authors: Gew, Grace Jie Yan, Sim, Pei Yi, Tong, De Fang
• Other Authors: Leong Kaiwen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Social sciences::Economic development
• Online Access: https://hdl.handle.net/10356/138789
• Institution: Nanyang Technological University

The desire to understand how stock prices move in the financial markets has led many investors to seek various ways of increasing the quantity and quality of information they obtain. There are many factors affecting the movement of stock prices, but public sentiments from Twitter have been a popular subject of study on its predictive value on stock prices. The aim of the study is to discuss and compare the predictive value of Twitter variables on the short-term price movement of stocks in the Financial and Consumer Discretionary Sector of the SNP500 Index. In this paper, we used ARIMAX to build 3 different predictive models to compare and identify if there is any difference in the predictive accuracy when we involve tweet sentiment, number of tweets and tweet interaction level. The Mean Square Error (MSE) and Mean Absolute Percentage Error (MAPE) measurements are carried out to evaluate the performance of our models. We find that the use of Twitter variables† leads to a better forecast of price movements rather than just using historical data. The use of tweet sentiment, volume and interaction level in the predictive models proved to be more helpful in the Financial Sector as the accuracy increased in two out of three of the models, by between 12.75% to 20.18%.