Identifying new earnings management components: a machine learning approach

Abstract

Purpose This study aims to apply machine learning (ML) to identify new financial elements managers might use for earnings management (EM), assessing their impact on the Standard Jones Model and modified Jones model for EM detection and examining managerial motives for using these components. Design/methodology/approach Using eXtreme gradient boosting on 23,310 the US firm-year observations from 2012 to2021, the study pinpoints nine financial variables potentially used for earnings manipulation, not covered by traditional accruals models. Findings Cost of goods sold and earnings before interest, taxes, depreciation and amortization are identified as the most significant for EM, with relative importances of 40.2% and 11.5%, respectively. Research limitations/implications The study’s scope, limited to a specific data set and timeframe, and the exclusion of some financial variables may impact the findings’ broader applicability. Practical implications The results are crucial for researchers, practitioners, regulators and investors, offering strategies for detecting and addressing EM. Social implications Insights from the study advocate for greater financial transparency and integrity in businesses. Originality/value By incorporating ML in EM detection and spotlighting overlooked financial variables, the research brings fresh perspectives and opens new avenues for further exploration in the field.