An intelligent detecting model for financial frauds in Chinese A-share market

Abstract

Financial frauds can cause serious damage to financial markets but are hard to detect manually. In this study, we develop an intelligent detecting model to efficiently identify financial frauds by using XGBoost on raw financial data items in corporation financial statements. With listed companies in Chinese A-share Market taken as samples, empirical results reveal that the proposed model works better than traditional models by a large margin in detecting fraud. Notably, the proposed model exhibits superior performance when used together with raw financial data items than with financial indicators. Moreover, the proposed model remains robust on outperformance in fraud detection when serial fraud cases are recoded, test periods are altered, more raw financial data are input, as well as other machine learning models–the AdaBoost and SVM–are selected as benchmark models. Our study enriches the application of machine learning in finance sector, and highlights the economic significance of raw financial data as the financial system's most fundamental components.