Research On Financial Fraud Identification From The Perspective Of Organization Impression Management

In the era of big data,China's financial and securities market is booming,and the financial fraud issues that accompany it have also received increasing attention from investors and regulators.Financial fraud of listed companies has caused huge losses to investors,disrupted the economic order,triggered a crisis of trust,and caused great harm.Therefore,it is of great significance to construct an effective financial fraud identification model for listed companies.Although the research on financial fraud identification has made great breakthroughs with the introduction of machine learning methods,as an important part of the research on financial fraud identification,non-financial characteristics research still has problems such as incomplete system and lack of quantitative standards.Therefore,theoretical guidance has important theoretical and practical significance for the identification of financial fraud.This article looks at corporate fraud from the perspective of organizational impression management,systematically extracts non-financial features under the guidance of theory,and further uses it to construct fraud classification models.This article first extracts three corresponding corporate behaviors based on the classification of organizational impression management strategies.That is,to promote the positive propaganda behavior under the promotion strategy,guide the positive emotional behavior,and the manipulative language expression behavior under the downgrade defense strategy.Second,build features based on these three types of behavior,and use a text analysis framework to obtain data.The framework uses a variety of text analysis methods to quantify the social behavior data and annual report text data of listed companies,and extracts social behavior characteristics,emotional characteristics and readability characteristics,which are the characteristics of impression management in this paper.Further,this study uses cluster analysis to perform variable cluster analysis on three groups of features,and the results confirm that the feature extraction of the impression management in this paper is reasonable and effective.In the construction of the classification prediction model,this article selected 62 companies punished by the Securities and Futures Commission for breach of trust in 2014-2017 as a fraud sample,and selected 62 companies that did not commit fraud in the year in a 1: 1 ratio as a matching sample,the company obtained a research sample with a sample size of 124.At the same time,this paper identified 11 financial ratio variables through literature research,and used random forest feature selection to screen for 5 important variables.Finally,this study uses support vector machines,gradient boosted trees,and artificial neural networks to build models,and uses a confusion matrix and AUC(Area under the Curve of ROC)as evaluation indicators for comparison and verification.The experimental results show that the classification effect of the model combined with impression management features is significantly better than the model using financial features alone.The models built by the three classifiers have significantly improved TPR(the ratio of correctly identifying fraudulent companies as fraudulent companies)after adding organizational impression management features,proving that organizational impression management features play an important role in identifying financial fraud.The research contributions of this paper are:(1)Discuss the problem of financial fraud identification under the theory of organizational impression management.In the study of the characteristics of financial fraud identification models,scholars pay less attention to the characteristics of impression management,and most of them only deal with one aspect of it;the research methods in China are currently focused on qualitative research or verification experimental research.Under the guidance of the text analysis framework,this paper systematically and quantitatively analyzes the impression management behavior in the information released by listed companies,and uses the characteristics of impression management for the detection of financial fraud.Provides a reference frame for the selection of non-financial characteristics.The experimental results show that the model combined with the characteristics of impression management can better identify financial fraud companies.(2)Organizational impression management behavior in social behavior data is studied.Data for financial fraud identification studies usually consists of financial ratios and unstructured data in annual reports.Based on this,from the perspective of organizational impression management,this paper analyzes the behavior data of enterprises on social platforms,extracts social behavior characteristics from them,and applies them to financial fraud detection models to optimize model classification results.Social behavior data has enriched the data sources for financial fraud identification research and provided more options for non-financial characteristics.