Predicting the debt-equity decision

Abstract

I train the k-nearest neighbors (KNN) and random forests (RF) machine learning models to predict if a firm will issue debt, or equity, in the upcoming quarter. KNN predicts 94% of debt and 80% of equity issues correctly. RF predicts 95% of debt and 86% of equity issues correctly. KNN is 92% correct when predicting debt and 84% correct when predicting equity. RF is 94% correct when predicting debt and 88% correct when predicting equity. The overall prediction accuracy is 90% for KNN and 92% for RF. I conclude that machine learning models can “learn” to predict the debt-equity decision.