Essays on Machine Learning in Corporate Finance

Satyajit Saste, PhD
Capital Structure, Financing decisions, Equity Capital Markets, Security Issuance, SEO, Capital Structure, Convertible Issuance, Converts

Abstract :

Machine Learning to Predict Equity Issues: Despite growing interest and practical use of machine learning algorithms within trading, there has been little exploration of these techniques in corporate finance. This paper tries to identify equity issues across the capital structure through a predictive framework built using a Tree-based Machine Learning Technique. The most significant variables predicting future equity issuance are prior equity issuance by the firm, days since the last offering, percentage of peer issuances in the last year, and retained earnings. Although not as significant as these variables, the model shows that features commonly mentioned in the prior literature, such as firm age, firm size, value, and growth, also help predict future equity issuance. Overall, this paper highlights novel variables that are key for predicting future equity issuance and shows the out-of-sample predictability of some of the variables already known to be important in-sample predictors.

Using Machine Learning to Predict Issuance of Convertible: While many theoretical papers have been written explaining why and when firms issue convertible securities, we know very little about what factors predict future convertible issuance. In this paper, I build a framework using tree-based machine learning techniques to predict convertible securities' issuance by a firm in the near future. Using a downsampled random forest approach, I find a few key variables with high feature importance scores for predicting convertible issuance: prior convertible distribution by the firm, days since last offer, retained earnings, percentage of peer issuances in the last year, and short interest in the stock. Interestingly and perhaps surprisingly, except for short interest in the stock, all other features also show up as very important in explaining seasoned equity issuance in Satya (2020). I discuss the relevance of my results for different theories. Interestingly, some of the most important features predicting convertible issuance are consistent with practitioners' insights and thumb rules. My results suggest that firm and industry level features are important, while I do not find evidence supporting the importance of macroeconomic variables in predicting convertible securities' issuance.

Publication date of the thesis

Thesis committee

Supervisor: Hamid Boustanifar, EDHEC Business School 

External reviewer: Diego Garcia, University of Colorado

Other committee members: Nikolaos Tessaromatis, EDHEC Business School