Using Machine Learning to Predict Corporate Fraud: Evidence Based on the GONE Framework

Bagging or boosting? Empirical evidence from financial statement fraud detection

“…For example, Bao et al (2020) used 1% of the sample size, while Xu et al (2023) used 12% of the sample size. Considering that in practical applications, regulators pay more attention to the inclusion of those samples with the highest suspected risk in the inventory list, this study adopts a fixed value; namely, k = 50, which accounts for approximately 2.3% of the number of samples in the test set, which is between those values of the above studies.…”

“…Cases of financial fraud among Chinese A-share listed companies, such as Kangdexin, Kangmei Pharmaceutical, and Zhangzidao, have attracted the attention of market investors and regulators. In the short term, financial fraud announcements are usually accompanied by stock price declines, and 3-day cumulative abnormal returns (CARs) are significantly negative after such announcements (Xu et al, 2023). Executives and boards also suffer reputational damage after such announcements (Fich & Shivdasani, 2007).…”

“…This work finds that the RF model based on bagging outperforms other ensemble learning models based on the boosting strategy on the test set. Specifically, this study uses the NDCG@k indicator commonly used in ad recommendation ranking (Bao et al, 2020) and refers to the practice of Xu et al (2023) to construct the Precision@k and Recall@k indicators. In this way, model performance is measured and compared by analysing the k = 50 companies with the highest probability of predicting fraud via different models.…”

“…The research in this paper is highly relevant to Xu et al (2023) and Bao et al (2020). However, there are differences in specific research objectives and implementation methods.…”

Why do employees commit fraud? Theory, measurement, and validation

“…While according to Occupational Fraud 2022 , 86% of fraud case types are asset misappropriation and only 9% of financial statement fraud. Second, correspondingly, current research has limited the search for the antecedents of fraud to firm-level factors ( Dechow et al, 2011 ; Perols et al, 2017 ; Bao et al, 2020 ; Xu et al, 2022 ), paying insufficient attention to the individual-level factors, the factors that account for larger percentage in the variance of fraud losses ( Holtfreter, 2008 ; Timofeyev, 2015 ). While in the field of similar research, white-collar crime and unethical behavior, the literature focuses more on influencing factors at the individual level.…”