The role of earnings components and machine learning on the revelation of deteriorating firm performance

Abstract

This study explores the proficiency of earnings components for detecting earnings and cash flows distress. The authors examine the deterioration of these two performance indicators for two aggregate and two disaggregate earnings models, each of which is subject to examination through different machine learning, non-parametric, and parametric methods. The results, obtained from firms in 22 countries, reveal that the current information content of earnings not only has explanatory power for future earnings and cash flows but also can support advance classifications of the two performance indicators as negative or positive. Each aggregate and disaggregate model offers distress classification ability, the disaggregation of earnings generates better, robust detection accuracies for cash flow distress, while aggregate earnings model provides improved classification for prospective earnings distress. The findings also suggest that machine learning estimation methods provide superior distress detection compared to a parametric method, despite its still decent performance.