Research On Financial Fraud Identification Of Listed Companies Based On Granular Computing

The rapid development of China's securities market makes accounting information disclosure particularly important.However,whether it is domestic or foreign,financial fraud is not uncommon.Many listed companies have been repeatedly exposed by financial fraud scandals,which has adversely affected investors,enterprises,and regulatory agencies,and threatened the stability of the securities market.How to identify the financial frauds of listed companies,and then take effective measures to curb financial frauds,is an important issue that needs to be resolved urgently in China's securities market.Therefore,we should accurately identify financial frauds of listed companies,establish a set of accurate financial fraud identification models,continuously improve audit countermeasures,reduce financial frauds,and ensure the effective operation of the securities market.This article mainly studies the financial fraud identification model of listed companies.Based on previous research and the data available at this stage,this article selects the 2014-2018 China Securities Regulatory Commission,Shenzhen Stock Exchange,Shanghai Stock Exchange,listed companies,A sample of 335 A-share fraudulent listed companies in 17 industries with financial violations released by the Ministry of Finance and other agencies,and 1,395 non-fraud listed companies without financial fraud in five years were modeled and analyzed as a paired sample as the original sample.With regard to the selection of indicators,this article selects as many and comprehensively as possible from the four aspects of listed companies' solvency,operating ability,profitability,and development ability,39 indicators that may affect the identification of financial fraud of listed companies.This article first uses Lasso feature selection to filter out 9 relevant indicators of financial fraud identification,and verifies the identification characteristics of financial fraud of listed companies.Then,use 4 integrated learning algorithms: Ada Boost algorithm,Gradient Lifting Tree(GBDT)algorithm,XGBoost algorithm and the latest Light GBM algorithm for fraud recognition,the recognition effect is not very good,the AUC value is up to 76%;meanwhile,with three Single classifier: Logictic regression,decision tree,and Naive Bayes' comparison of recognition results proves that the integrated learning algorithm is superior to a single classifier in identifying financial fraud.Finally,a financial fraud recognition model based on granular computing is proposed.Based on the original data,the five fiscal year data are integrated into the time series information through the ratio method,and then the principal component analysis is used to improve the granularity to calculate the debt solvency and operation of the listed company.The four coarse-grained indicators of capability,profitability and development ability,and finally the use of integrated algorithms on the coarse-grained indicators to identify financial fraud.The AUC values of the four models are all above 91%,which is much greater than the identification of financial fraud without granular calculation.The model proves that the effect of financial fraud recognition based on granular computing is significantly higher than that of financial fraud recognition based on integrated learning.