Date of Award

7-2022

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Dr. Hesham Ali

Abstract

With the current availability of massive data sets associated with stock markets, we now have opportunities to apply newly developed big data techniques and data-driven methodologies to analyze these complicated markets. As stock market data continues to grow, analyzing the behavior of companies listed on the market becomes a massive task, even for high-performance computing systems. Hence, new big data techniques like network models are very much needed. We conducted this study on data collected from CRSP during the years 2000-2021 inclusively. In this study, we proposed a novel population analysis by constructing a correlation network model based on the monthly data of different companies’ excess returns; additionally, we employed the Louvain clustering algorithm to generate individual clusters/communities. After constructing correlation networks from input data, hidden knowledge was extracted from the network by using community detection and measuring network centralities. The Louvain algorithm was applied to the network as a data analysis shortcut tool and grouped different companies with high correlations or similar financial behavior over the period of study. In each community, different centralities were measured. Centrality measurements came from Closeness, Betweenness, and Eigen centralities for this study. The empirical result of this study showed that the meaning of centrality measurement in network analysis in the stock market has a different meaning compared to social network analysis. In most networks, high central entities are the most important entities; however, in this study, we learned that high centrality is not something that researchers should look for when developing and building a portfolio with low risk. What was discovered was that nodes in the network with lower degrees of centrality led to developing a diverse portfolio with lower risk, with acknowledgment of the Modern Portfolio Theory. Since this new approach was applied on the years 2000-2021, this study revealed behavioral patterns from stock movements depending on different events such as the 9/11 attacks, 2008 economic crashes, and the Covid-19 pandemic. As a result of this study, we would like to suggest a system based on a weighted portfolio to make a proper decision in selecting portfolios that can outperform the benchmark during normal circumstances or crises.

Comments

Copyright held by author.

Download

Share

COinS