Boosting over Deep Learning for Earnings

AAAI 2021

Abstract: Deep learning techniques have become the leading choice for applications in many fields, and finance is no exception. However, the success of such applications in other sciences does not necessarily extend to those in finance. In this paper, we examine the application of machine learning and deep learning techniques to the modeling of corporate earnings. Corporate earnings data are noisy, have limited sample sizes, and require a large amount of disparate inputs for modeling. We illustrate the success of gradient-boosting models in forecasting earnings and examine how standard deep learning techniques fall short in comparison. We also show how encoding is crucial for deep learning to be effective. Overall, our work highlights how deep learning might not be the optimal approach for addressing forecasting corporate earnings. Authors: Stephen J. Choi, Xinyue Cui, Jingran Zhao (LORA Technologies, Hong Kong Polytechnic University)