Theses and Dissertations

Date of Award

7-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Finance

First Advisor

Ahmed Elnahas

Second Advisor

Incheol Kim

Third Advisor

Suin Lee

Abstract

This dissertation consists of two essays. In the first paper, I find that machine learning (ML) models predict the likelihood and magnitude of insider trading significantly better than linear models such as OLS and logistic regression. I use ML models, including LASSO, Random Forest, and eXtreme Gradient Boosting, optimizing model parameters through Bayesian hyperparameter tuning to identify the best configuration. Additionally, I apply SHAP values to better understand the determinants of insider trading. I also use Gaussian Thompson Sampling (GTS) to explore the sources of insiders’ market-timing capabilities. I find that ML models can boost the R² for models predicting the magnitude of insider selling by 150% compared to OLS models. Furthermore, ML models predict the likelihood of insider selling with an accuracy of 82%, recall of 95%, and precision of 82%. Additional tests indicate that the improvement in predictability due to the use of ML models is more pronounced for female than male insiders. Finally, while male insider trading is largely driven by risk-taking incentives, female insider trading appears to be more influenced by possessing private information about future cash flows. This paper provides evidence that informational advantages are a key driver of insider trading's market-timing capabilities.

In the second essay, I examine the relationship between Corporate Social Responsibility (CSR) and market reactions to dividend omission or reduction announcements across 33 countries. I find a significant negative relationship between CSR scores and cumulative abnormal returns following such announcements. This finding supports the information asymmetry hypothesis rather than the insurance hypothesis. The negative effect is more pronounced in countries with weaker shareholder protection and civil law systems, and for firms with higher free cash flow problems, greater information asymmetry, and more financial constraints. My results remain robust after addressing endogeneity using the instrumental variable (IV) approach and Oster’s omitted variable diagnostic test, as well as selection bias using the propensity score matching (PSM) and entropy balancing (EB) approaches.

Comments

Copyright 2025 Solmaz Batebi. All Rights Reserved. https://proquest.com/docview/3246966162

Download

Share

COinS