Machine Learning and its application in the Portfolio Management industry
Loading...
Files
Date
2024
Authors
Murphy, Conor C
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Abstract
Machine Learning (ML) is a subdivision of Artificial Intelligence (AI). AI is a term used for technology that “enables machines to mimic human thoughts and behaviour” (Xu, 2021). Recently, there has been a significant increase in the use of ML techniques by finance professionals, primarily by the Portfolio Management industry (Perrin, 2019). This thesis reflects on ML’s application in the Portfolio Management industry. To further understand this relationship a literature review is carried out in Chapter 2 to identify the intersection of ML and Portfolio Management and to highlight key criteria required for ML’s application in the Portfolio Management industry. A key finding from Chapter 2 found that the lack of quality data is a critical barrier to ML’s application in the Portfolio Management industry.
Chapter 2 examines the use of sentiment from unstructured data, in the main tweets, for stock market prediction in the Portfolio Management industry as alternative data source for ML. There are a variety of ML algorithms that can be applied for Twitter sentiment stock market prediction, Chapter 3 employs the Multilayer-Perceptron (MLP). MLP has been employed successfully in other studies for Twitter sentiment stock market prediction (Livingston, 2019; Turchenko, 2011). Before running the MLP prediction algorithm, EmoLex was employed to identify the underlying sentiment that may be apparent in the tweets. Subsequently, a prediction algorithm MLP Classifier was run to ascertain daily sentiment stock price predictions for the data. For the prediction analysis section the Vader Sentiment Analyser from the Natural Language Tool Kit (NLTK) in python was employed to split the Twitter sentiment into three categories (positive/neutral/negative), while the MLP classifier was run through SickIt-learn package in python. Interestingly, the performance of the MLP in Chapter 3 is not as accurate as other studies including, Kolasani (2020), Usmani (2016) & Khan (2020).
Following the same guidelines as Pagolu (2016) and Mittal (2012) incorporating a new algorithm (the Random Forest algorithm) using the same methodology outlined in Chapter 3 may outperform the MLP. Studies such as Bollen (2011) indicate that positive Twitter sentiment will be reflected in the stock market by a positive increase in stock price and negative Twitter sentiment will result in a fall in stock price. The use of the Random Forest regressor in this study found that in the case of Black Swan events, Twitter sentiment has the same predictive power as a chance model. These results are integral to the Portfolio Management industry as it clearly indicates that Twitter sentiment cannot be used to gauge the severity of Black Swan events nor can it be used as a method to predict it. Only one of the Tweets in the database collected referenced that China allowed one of its banks to go into liquidation, indicating that the speed of stock price drop was ahead of the Twitter sentiment.
Description
Controlled Access
Keywords
ML , Machine Learning , Artifical intelligence , Finance
Citation
Murphy, C. C. 2024. Machine Learning and its application in the Portfolio Management industry. MRes Thesis, University College Cork.