Date of Defense
25-4-2025 4:30 PM
Location
F3-132
Document Type
Thesis Defense
Degree Name
Master of Science in Mathematics
College
COS
Department
Mathematical Sciences
First Advisor
Dr. Ho Hon Leung
Keywords
Long Short-Term Memory Networks (LSTM), Random Forest, gradient descent, recurrent neural networks, diversification, portfolio.
Abstract
We review the theoretical foundations and empirical workings of Long Short-Term Memory Networks (LSTM) and Random Forest for stock market prediction. To be precise, our work initiates with a deep-dive exploration in the mathematics of these algorithms which covers topics like gradient descent, automatic differentiation, and a bit on recurrent neural networks as well.
An application creates stock portfolios through LSTM and Random Forest rules- based strategies, using either technical or fundamental indicators as input variables. The process involves predicting returns and ranking stocks accordingly to build portfolios with a certain level of diversification and annual rebalancing over a defined back-testing period. The paper then benchmarks the performance of such portfolios against market benchmarks like the S&P 500, with CAGR as the key metric to be calculated.
The results of our study measure the ability of Machine Learning based strategies to outperform the traditional market indices and, in a broader sense, to provide more information on whether or not Machine Learning has the potential to aid in financial decision-making. Such results will, thus, have possible extensions for future research.
CONSTRUCTION OF STOCK PORTFOLIOS BY MACHINE LEARNING METHODS
F3-132
We review the theoretical foundations and empirical workings of Long Short-Term Memory Networks (LSTM) and Random Forest for stock market prediction. To be precise, our work initiates with a deep-dive exploration in the mathematics of these algorithms which covers topics like gradient descent, automatic differentiation, and a bit on recurrent neural networks as well.
An application creates stock portfolios through LSTM and Random Forest rules- based strategies, using either technical or fundamental indicators as input variables. The process involves predicting returns and ranking stocks accordingly to build portfolios with a certain level of diversification and annual rebalancing over a defined back-testing period. The paper then benchmarks the performance of such portfolios against market benchmarks like the S&P 500, with CAGR as the key metric to be calculated.
The results of our study measure the ability of Machine Learning based strategies to outperform the traditional market indices and, in a broader sense, to provide more information on whether or not Machine Learning has the potential to aid in financial decision-making. Such results will, thus, have possible extensions for future research.
