Research On Whitewashing Detection Of Financial Indicators Based On Machine Learning

Financial whitewashing detection refers to the analysis and identification of enterprise financial data to determine whether the enterprise has inflated,reduced or concealed financial data.The detection of financial whitewashing can help investors,analysts,regulators,etc.to evaluate the true financial status of the company,and avoid investment risks caused by misjudging the financial status of the company.As the economy evolves,so does the financial whitewashing situation.In some developing countries and emerging markets,financial whitewashing may be more common due to lax supervision and unsound systems.In developed countries,although the regulation is relatively strict,companies may still use legal loopholes or innovative financial tools to conduct financial whitewashing.In the past,financial researchers mainly used traditional financial indicators and ratios to judge the authenticity of financial statements.These indicators and ratios can only reflect the surface phenomenon of financial data,but cannot accurately reflect the actual financial status of the enterprise.With the development of computer technology and data mining technology,more and more researchers have begun to use machine learning,deep learning and other methods to identify financial whitewashing.These methods are able to mine the underlying patterns and regularities behind financial data to more accurately identify financial whitewashing.This thesis conducts research based on domestic financial data,and uses machine learning methods to conduct research on the identification of financial whitewashing at the indicator level.This thesis collects public corporate financial statement data based on the authoritative CSMAR database,builds a data set of whitewashing indicators,and proposes an LSTMM model.The specific work of this thesis is as follows.(1)Aiming at the lack of data set for the current financial indicator whitening task,a financial indicator whitening data set for the financial field is constructed.The data set is composed of whitewashed samples and related data screened out according to the CSRC’s punishment clauses for corporate financial violations.Specifically,first of all,this thesis summarizes the financial violation information of the violation cases,and screens out the whitewashed samples in the financial summary table according to the enterprise code and year in the violation information;secondly,filter out the non-whitewashed samples in the financial summary statement through the attributes of the whitewashed samples,then construct a dataset of the whitewashed samples and non-whitewashed samples with their historical data and industry data;finally,construct a label for each whitewashed indicator according to the violation information.A large number of experimental results prove that the screened whitewashing samples are reasonable and effective and the constructed dataset brings new challenges to the existing financial whitewashing detection tasks.(2)This thesis proposes an LSTMM model to solve the problem of financial indicator whitewashing detection,which is a fusion of LSTM and MLP models.On the proposed data set,the proposed method is compared with Decision Tree,XGBoost,LSTM and other methods.The experimental results show that the proposed method is superior to other comparison methods,which proves the effectiveness of the proposed method.