Random Forest and Gradient Boosting Applications in Forecasting Corporate Capital Structure Adjustments

Abstract

Capital structure adjustments are among the most impactful decisions that company management can make. These decisions are naturally driven in part by the assets supported by the capital structure and the capital markets, in addition to the discretion of the managers themselves. In turn, capital structure can support or inhibit realization of value for equity investors. Understanding patterns in capital structure changes can provide useful insights for investors ranging from hedge funds to private equity managers as they orient their expectations regarding firm value and future cash flows. Using 50+ years of annual financial data, dating from 1965-2018, for US-based public companies, we apply modern quantitative techniques, including regression and ensemble predictors, and find patterns in corporate financial data and macroeconomic factors which allow investors to better understand and predict active leverage adjustments in companies. We find multiple models with significant predictive power. Most notably, random forests reach a r2 value of 0.35 for predicting the magnitude of active leverage change and an accuracy of 64.4% for predicting the direction of active leverage change. This result is robust across different economic regimes. We also find that three main variables explain a large part of predicted active variation in leverage ratio - difference between firm leverage ratio and median industry leverage ratio, profitability, and the firm’s previous year active leverage ratio change.