Case Study Based On False Discovery Rate

With the explosive growth of the multidimensional data,multiple hypothesis testing as a major tool for large-scale statistical inference becomes a hot spot of research.In the multiple comparisons,control error rate of the entire test is the key to success.The concept of false discovery rate was first proposed by Benjamini and Hochberg in 1995,they also carried out the original method to control this error rate.Comparing to the family-wise error rate,false discovery rate not only provides a new theory for control falsely rejected hypotheses in multiple significance test,but also enhances the power of the whole test and finds a new balance between error control and the test power.The development of false discovery rate can be divided into two different directions: First,improve the procedure of controlling false discovery rate;second,expand the application field of false discovery rate.Through two cases of the same industry stock selection and the automobile sales enterprise tax evasion identification,this paper apply the false discovery rate to the field of stock selection and the field of data mining,which extend the application field of false discovery rate.In the case of the same industry stock selection,this paper proposes a short-term stock selection method based on the false discovery rate.By constructing the hypothesis test,the stock selection problem is transformed into multiple comparison,so the false discovery rate method can be applied to the stock selection problem.This method can simultaneously examine the performance of all stock returns in a specific industry and improve the efficiency of stock selection in a specific industry.The results of case study show that the short-term gains of stocks selected by the false discovery rate method are higher than the industry average returns.Further study shows that compare to the stocks selected by the t-test method,the short-term gains of the stock selected by the false discovery rate method also have an advantage.In the case of the automobile sales enterprise tax evasion identification,people tend to find as many indicators as possible which have an impact on the classification results,and all of the acquired data is imported into the data mining model,the following problems may arise: the complexity of the model reduces its comprehensibility,import redundant data causes the model to be less effective,and massive amount of data preprocessing work.Therefore,in this paper,the false discovery rate method is applied to the field of data mining,be used for input variable selection in the data preprocessing stage to improve the model efficiency,reduce the complexity of the model and reduce the data preprocessing work.Through the case study,it is shown that use the false discovery rate method to select input variable can improve the accuracy of model and reduce the complexity of the model.Through the study of above two cases,this paper successfully applies the false discovery rate method to the field of stock selection and the field of data mining,and the feasibility and validity of this method using in these two fields have been verified.