|Abstract:||This dissertation studies empirical corporate finance and investment. In Chapter 1, we utilize big data and machine learning tools to examine the information content of executive social media with regard to applications on forming investment strategies. The 2-4 articles focus on the relationship among payout policy, executive compensation and earnings management. To establish a causal effect, we are revisiting and expanding previous research questions and examining them from another viewpoint with identification strategies. Specifically, we are seeking to understand whether Earnings Per Share (EPS) dilution due to options exercises causes firms to manage earnings in Chapter 2, whether having a high level of institutional ownership mitigates the problem of conducting repurchases through Accelerated Share Repurchases (ASR) for the purposes of EPS manipulation in Chapter 3, and whether payout choices affect executive compensation decisions in Chapter 4. Moreover, Chapter 5 explores machine learning algorithms with an effort to improve Fintech lenders in their selection process for online borrowers.
Chapter 1 "The Information Content of CEOs' Personal Social Media: Evidence from Stock Returns and Earnings Surprises" demonstrates a previously unexplored and close connection between firm performance and CEO Tweets. Utilizing big data, we implement a novel dataset about the personal social media of CEOs, which mostly deals with CEOs' personal activities and opinions, and the analysis further employs machine learning algorithms to identify and exclude Tweets that have relevance for firm operations. The results show that a high proportion of positive words in the CEOs' Tweets predicts positive future abnormal returns and firm performance. These results suggest that the content of personal Tweets can elicit the moods of CEOs. Executives tend to embark on pleasurable activities like going on vacation when they are confident and satisfied with their firm's performance. This project provides a framework for future research using big data as a tool to gain additional information about the behaviors of managers and shed additional light on corresponding firm performance.
Chapter 2-4 embark on three innovative projects devoted to understanding the causal relationships among executive compensation, payout policy and earnings management. Chapter 2 entitled "Equity-based Compensation, Dilution, and Earnings Management: Evidence from a Regression Discontinuity" for the first time we show that firms engage in earnings management to counter the dilutive effect on EPS due to option exercises, in lieu of using buybacks as suggested by the literature. To isolate the effect of option exercises on payouts, we explore what firms do in response to plausibly exogenous option exercises using a fuzzy regression discontinuity framework. We show that these quasi-exogenous exercises do not cause additional buybacks, but they cause firms to reduce R&D spending and increase accruals. Therefore, we conclude that instead of buying back shares, firms have alternative ways of countering EPS dilution from option exercises, such as boosting income through real- and accrual-based earnings management. Furthermore, we examine the methods of repurchases in Chapter 3 "The Effects of Institutional Investors on Earnings Per Share Bonuses and Methods of Share Repurchases: Evidence from Accelerated Share Repurchases". Using hand-collected, originative data on Accelerated Share Repurchases (ASR), this article concludes that firms with a high level of institutional investors are less likely to conduct ASR, because of concerns about managing EPS for compensation motives. We further explore the determinants of stock option grants in Chapter 4 "Do Dividend Affect Stock Option Award? Evidence from Job and Growth Tax Relief Reconstruction Act." Employing a difference-in-difference framework to identify an exogenous increase in individual level dividend payments, the paper finds that an expected increase in dividends due to dividend tax decrease causes firms to shift future compensation plans away from option grants. Therefore, we conclude that the exogenous change in shareholders' tax-related payout preferences leads to a new equilibrium in terms of executive compensation and payout choices.
Lastly, we use machine learning to improve Fintech lenders in their selection process for online borrowers in Chapter 5 "Predicting the Performance of Peer-to-Peer Loans using Machine Learning". We show that machine learning algorithms provide a higher accuracy than the benchmark model in prediction loan default, especially for out-of-sample predictions. Moreover, using feature selection mechanism, we identify independent variables of borrower and loan characteristics that are the most important features in predicting default. Therefore, machine learning algorithms are valuable to select the most relevant features and can achieve higher accuracy in predicting the performance of Peer-to-Peer loans.