Three Essays on the Applications of Machine Learning in Corporate Finance
2023
Hochschulschrift
Zugriff:
This dissertation consists of three chapters that answer fundamental questions about corporate finance and sustainable finance using advanced methods in big data and machine learning.The first chapter investigates whether employees have useful information for assessing firms’ environmental, social, and governance (ESG) practices. In this chapter, I analyze 10.4 million anonymous employee reviews via a word-embedding model to construct an inside view of corporate ESG practices. The inside view has useful information beyond external ratings in predicting a firm’s future misconduct, governance issues, downside risk, growth, and valuation. In addition, the inside view appears robust to greenwashing, both theoretically and empirically. In various settings including a novel exogenous shock, I show that low-cost changes in a firm’s stated ESG policies do not affect the inside view while more expensive changes do.In the second chapter, I investigate how diversity and inclusion (D&I) considerations affect a firm’s operating flexibility and financial policies. To do so, I extrapolate a diversity and inclusion rating introduced in 2020 back to 2008 for over 10,000 companies using a breakthrough model that surpasses humans on reading comprehension. With this extrapolated rating, I show that diverse and inclusive firms (D&I firms) have lower operating flexibility than other firms, especially following adverse economic shocks. With less operating flexibility, D&I firms likely hold more cash and use less debt in case of unexpected shocks, a finding I confirm empirically in various tests, including a quasi-natural experiment. The results suggest that D&I firms have less operating flexibility and thus adopt more conservative financial policies.In the third chapter, co-authored with Xinjie Ma and Sudipta Basu, we show that firms’ investment opportunity sets (IOS) are multidimensional. Analyzing Form 10-K texts, we identify 445 unique keywords that predict firms’ future investments during 1995-2009 and combine them into 43 underlying factors. Industry-specific factors include Bio-Pharma, Banking, Information Technology, Oil & Gas and Retail Stores, while more general factors include Equity Intensity, Debt Intensity, Lease, Going Concern and Acquisition. These factors form our multidimensional measures of IOS. They outperform Tobin’s Q and/or industry fixed effects, in predicting future out-of-sample (2010-15) investments and related corporate policies, and even inform incrementally over lagged dependent variables. We trace the factors’ improved predictive power to their multidimensional nature, which captures IOS-related variation within and between industries, and stability in IOS that allows 10-K texts to be more informative.
Titel: |
Three Essays on the Applications of Machine Learning in Corporate Finance
|
---|---|
Autor/in / Beteiligte Person: | Briscoe-Tran, Hoa |
Link: | |
Veröffentlichung: | 2023 |
Medientyp: | Hochschulschrift |
Schlagwort: |
|
Sonstiges: |
|