| Author: | Kehinde, Temitope Olubanjo |
| Title: | Data envelopment analysis, inverse data envelopment analysis, and machine learning : a novel framework for stock portfolio management |
| Advisors: | Chung, S. H. (ISE) Chan, T. S. Felix (ISE) |
| Degree: | Ph.D. |
| Year: | 2024 |
| Subject: | Data envelopment analysis Stocks Investment analysis -- Data processing Portfolio management -- Data processing Rate of return -- Evaluation -- Data processing Machine learning Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Industrial and Systems Engineering |
| Pages: | xxi, 166 pages : color illustrations |
| Language: | English |
| Abstract: | The stock market is a vital component of the financial sector, facilitating the accumulation of money, spreading risk, setting prices, and communicating information. Given the substantial growth of the global stock market, investors, traders, analysts, and researchers are facing considerable difficulties in accurately selecting stocks and predicting their prices. Traditional statistical and econometric methods have shortcomings when dealing with the inherent uncertainty, noise, nonlinearity, and high dimensionality of stock markets. This thesis presents an innovative framework that combines data envelopment analysis (DEA), inverse DEA (IDEA), and machine learning techniques to improve stock market analysis and decision-making. The first phase of this research presents a novel approach for assessing and ranking equities operating on the Taiwan Stock Exchange (TWSE). This phase addresses the issue of bias and subjectivity in current ranking approaches by combining the strengths of the Shannon Entropy Technique (SET), DEA, and IDEA, which allows a more objective evaluation. The SET method was utilized to identify the most significant financial indicators from a collection of 13 financial ratios. These indicators were selected based on their importance from several financial perspectives, such as liquidity, asset usage, leverage, profitability, and valuation. This approach substantially reduced the complexity of the analysis by lowering the number of dimensions. For each equity under consideration, the DEA conducted further analysis to obtain efficiency scores. The DMUs that demonstrated high efficiency, indicated by a score of 1, were subjected to a ranking process using an IDEA model. The outcome of this phase establishes a standard for the evaluation of DEA models, aiming to improve the level of objectivity and accuracy. In the second phase, a novel DEA methodology framework was proposed, and it deviates from the traditional approach that considers risk as an input and returns as an output. Instead, this concept is reframed by considering return and risk as outputs that arise from a financial production process. However, applying DEA in stock market settings encounters limitations due to its inherent difficulty in efficiently handling negative inputs and outputs. Also, no existing studies have been conducted on applying DEA to reduce potential equity risk using an inverse optimization approach. To effectively address these problems, the directional distance function DEA (DDF-DEA) and IDEA were integrated to estimate inefficiency and potential reductions in non-performing stocks. Using this novel methodology, an analysis was conducted on stocks belonging to the food industry listed on the TWSE as well as consumer staples within the Standard & Poor’s 500 (S&P 500) index. This phase addresses a significant gap in the current body of literature, providing a foundation for enhanced decision-making in managing equity risk. Going forward, a study similar to phase 2 was conducted in phase 3, but in a portfolio-related context. This phase introduces a novel approach to portfolio optimization using IDEA. The study uses a combination of a DDF-DEA and a novel IDEA method to assess the efficiency and volatility reduction of industry-based portfolios, where each industry is a combination of several related firms listed on the TWSE. The empirical analysis shows that only 7 of 20 industry-based portfolios were underperforming. The methodology calculates the potential and maximum reduction in the volatility of all underperforming portfolios. Additionally, for the first time in the literature, the phase proposed a net-zero volatility risk initiative for investors and analysts to keep track of their portfolios efficiently at all times. This work holds potential utility for investors and fund managers seeking to enhance the performance of their investment portfolios. The last phase explores the application of a Transformer model, enhanced with a Tree-structured Parzen Estimator (TPE) for hyperparameter optimization, in predicting stock market indices. The phase focuses on the prediction of three major global stock indices: the S&P 500, Financial Times Stock Exchange 100 (FTSE 100), and Hang Seng Index (HSI), using deep learning techniques. This is a notable advancement from conventional forecasting techniques by incorporating a cutting-edge Transformer model, which is widely recognized for its achievements in natural language processing (NLP), into the field of financial forecasting. The work relies on the innovative use of the Transformer model, which employs self-attention mechanisms to effectively deal with complex and non-linear financial time series features, surpassing traditional recurrent neural network (RNN) and its variants, long short-term memory (LSTM) and gated recurrent units (GRU). The study showcases the model’s proficiency in extracting relevant features from structured financial data, in contrast to prior research that predominantly concentrated on unstructured data such as social media sentiment. An essential aspect of this phase is applying a straightforward trading strategy that relies on the model’s predictions, demonstrating the approach’s practical monetary implications and possible investment gains. To summarize, this thesis introduces a thorough and original framework that integrates DEA, IDEA, and machine learning methods to improve stock portfolio management. The proposed frameworks offer resilient and adaptable strategies for investors and analysts, considering the uncertain and external disturbances encountered by stock markets, such as economic recessions, pandemics, and political instability. This framework provides significant insights for investors, analysts, and decision-makers in navigating the complicated and constantly changing stock market landscape by overcoming the limits of traditional methodologies and utilizing data-driven algorithms. This research holds great value for investors, managers, regulators, and scholars interested in assessing the efficiency of financial markets. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13008