Understanding Accruals via Machine Learning

Abstract

Our goal is to better understand the key determinants of accruals and their complex relationships. Using machine learning (ML) techniques, we analyze the accrual process without relying on the linearity assumptions inherent in ordinary least squares regressions. We find that prior-period sales growth, profitability, and previous accruals are the most significant predictors of current accruals. Our ML approach uncovers significant nonlinear relationships and interactions among these variables, revealing dynamics that linear models would struggle to identify. Furthermore, by decomposing accruals into expected and unexpected components using the ML model, we enhance the prediction of future earnings. These findings contribute to the accrual accounting literature by providing a more nuanced framework for accrual estimation and demonstrate the value of integrating advanced analytical techniques in accounting research to improve financial reporting practices.

Keywords: accruals, machine learning, prediction

JEL Classification: C53, C63, M41