Artificial Intelligence-based Detection and Prediction of Corporate Earnings Management

Abstract

In November 2019, Nasdaq announced the application of artificial intelligence (AI) in the US stock market to detect irregular and potentially malicious trading activity. The newly launched initiative aimed to strengthen and revolutionize market surveillance via AI applications such as machine learning (MarketInsite in For the first time, Nasdaq is using artificial intelligence to surveil US stock market. Nasdaq, 2019, [1]). The use of AI in various fields is rapidly becoming ubiquitous. In this vein, various studies argue that AI-based methodologies can help to detect and predict corporate earnings management. Thus, this chapter examines the various AI applications that can be used to detect and predict corporate earnings management. Owing to the remarkable growth in data volume and advancements in computing power, AI applications (e.g., machine learning) have substantially increased over the last decade. Accordingly, previous studies focused on constructing solid methodologies to detect or predict earnings management by utilizing AI-infused methods including supervised and unsupervised machine learning. In general, the level of a model’s accuracy can be enhanced by adopting AI-based methodologies rather than conventional linear regression models. At present, models that utilize supervised learning are more commonly used than unsupervised learning models. However, models for the detection of real earnings activities are scarce. This could be attributed to the characteristics of real earnings management, i.e., real earnings management is more difficult and complicated to uncover than accrual-based earnings management.