Detecting fraudulent financial information of a company using Hidden Markov Model

Abstract

Using Hidden Markov Model (HMM), this paper detects some fraudulent financial information of a company. The research index MBSPM (Main Business Service Profit Margin) ratio of company is used to test the effectiveness of HMM. Due to the shortage of the fraudulent data, based on the MBSPM index data we generate the artificial fraudulent financial information data by Poisson process and Uniform distribution, that is, we use the Poisson process to simulate the arrival times of the fraudulent financial information and the Uniform distribution to simulate their fraudulent sizes. By embedding the artificial fraudulent financial information data into the non-fraudulent MBSPM data, we form the sample data series. Applying Henderson filter method, we derive the research sample data by getting rid of the sample data series trends. We do some experiments to test the validity of HMM for detecting the fraudulent financial information, and further make some comparisons with other detecting techniques: logistic regression and ANNs. The detected results show that the HMM approach can significantly improve the identification accuracy.