Please use this identifier to cite or link to this item: https://doi.org/10.21256/zhaw-24408
Publication type: Master thesis
Title: Using financial news for stock price direction prediction : an empirical investigation
Authors: Harder, Pedro
Advisors / Reviewers: Fazlija, Bledar
DOI: 10.21256/zhaw-24408
Extent: 54
Issue Date: 2021
Publisher / Ed. Institution: ZHAW Zürcher Hochschule für Angewandte Wissenschaften
Publisher / Ed. Institution: Winterthur
Language: English
Subject (DDC): 006: Special computer methods
332: Financial economics
Abstract: The semi-strong form of financial market efficiency states that asset prices reflect all publicly available information. Consequently, natural language processing methods can be used to extract the market sentiment from the information such as the news. However, traditional natural language processing methods have the disadvantage that some information such as the context of words or the structure of sentences get lost. The purpose of this master thesis is to extract the sentiment of the financial markets from news articles and to use the estimated sentiment scores to predict the price direction of the stock market index Standard & Poor's 500. To overcome the drawbacks of traditional natural language methods, state-of-the-art natural language processing models based on the Transformer architecture are used in this master thesis. To enable the best possible classification performance of sentiment scores, state-of-the-art bidirectional encoder representations from transformers (BERT) models are used. The pretrained transformer networks are fine-tuned on a labeled financial dataset to be able to estimate the sentiment of the financial markets. After fine-tuning the models, they are applied to news articles from Bloomberg and Reuters to predict the sentiment score of the news. To forecast the price direction of the stock market index, the predicted sentiment scores are fed into a machine learning model. Thereby, the sentiment scores of the titles, the content, and their sentiment scores combined with past time series information of the stock market index are used as input. The results indicate that the use of sentiment scores generated from news content can be used for stock price direction prediction. The use of sentiment scores extracted from the titles or the combination of sentiment scores from the titles and the content does not improve the quality of the prediction. Based on the findings of this master thesis, it can be concluded that the sentiment scores can be used for the prediction of the stock price direction. For further research in this area, the author of this master thesis recommends using recurrent deep learning models. Due to their internal state, these deep learning models have a memory that can be useful for predicting stock price directions. Practical recommendations are that the sentiment scores can be used in a risk-based approach as a complement to the calculation of the value at risk or the expected shortfall.
URI: https://digitalcollection.zhaw.ch/handle/11475/24408
License (according to publishing contract): CC BY-NC-ND 4.0: Attribution - Non commercial - No derivatives 4.0 International
Departement: School of Management and Law
Appears in collections:MSc Banking and Finance

Files in This Item:
File Description SizeFormat 
Master_Thesis_Harder.pdf1.61 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.