Document Type

Thesis

Degree Name

Master of Science (MSc)

Department

Mathematics

Faculty/School

Faculty of Science

First Advisor

Roman Makarov

Advisor Role

supervisor

Second Advisor

David Soave

Advisor Role

supervisor

Abstract

This thesis aims to predict Apple's stock return using the sentiment of financial news headlines. There is a tremendous amount of financial news posts daily, and their headlines include crucial information. The financial news headlines may influence the readers and lead them to make certain trading decisions after reading the news. Therefore, financial news headlines could potentially impact the financial market. By understanding the connection between the headlines and the financial market, we can construct profitable trading strategies based on our analysis of daily headlines. Specifically, the sentiments in the headlines can result in positive, negative, or neutral impacts on the financial market. By analyzing the headline sentiments, we can learn the market trend. Before we examined the sentiments, we used natural language processing techniques to clean the texts and select essential features from the headlines. We also implemented the principal component analysis to reduce the dimension of the data set. Then, we constructed statistical models to explore how headline sentiments impact the financial market. Specifically, we focused on detecting how the headlines affect Apple's stock price. The headlines can impact the stock price positively, negatively, or neutrally. Logistic regression was proposed to model stock returns. We constructed one-stage, two-stage, and three-stage classification models. In addition, we used logistic, LASSO, and support vector machine (SVM) to construct regression models and analyze the connection between headline sentiments and stock price. Also, we introduced a new metric to assess the performance scores of multi-stage regression models so that we could select the best approach. Furthermore, we constructed trading strategies consisting of vanilla call and put options based on the predictions made with the best models selected. We created nine trading strategies applied for the period from 2018-05-31 to 2019-05-31. The annual return for the best trading strategy is 169.265%. Even the worst trading strategy we constructed earned 58.39993% per year, while the annual return of the S&P 500 was 1.73% during this period. Therefore, our thesis successfully created multi-stage models that use financial news headlines to predict stock returns and applied predictions to construct trading strategies with high returns.

Convocation Year

2023

Convocation Season

Spring

Download

Share

COinS