Earnings management visualization and prediction using machine learning methods

Abstract

To create new insights and understanding of earnings management, this study attempts to diagnose firms’ financial profiles using machine learning methods and thereby provide a visual representation of the financial profiles that characterize earnings management strategies (upward and downward) and tools (accruals and real activities). By applying a novel machine learning method to detect signs of earnings management, this research reveals diverse financial profiles related to earnings management. Firms that conduct downward manipulation (accruals and real activities) share a sound financial profile. For firms that manipulate earnings upward, different types of financial distress influence the earnings management tool they use: Companies with liquidity constraints undertake accruals earnings management; companies with solvency difficulties are prone to real activities management. Notably, the proposed machine learning method outperforms traditional prediction methods in detecting signals of earnings management.