Corporate Financial Distress Early Warning Research

To evaluate the financial statement of the listed companiesobjectively, especially to predict the companies which are prone tofinancial distress are of great importance for the investors to adjust theirinvesting decisions in time, for the supervisors to correctly distinguishblindless financing companies,for the investing banks to dig out thepotential serving customers.However, as financial distress pre-warningissue is a fringe intersection of many subjects includingstatistics,management and computer science,there is no unified standardfor the researching method.The traditional pre-warning methods aremainly based on experts'experience or simplified statistic models whichcan't reflect the nonlinear essence of economic system,and they havestrict requirements on the distribution of pre-warning variables whichcan't meet the practical condition of financial distress prediction.On thebasis of the conclusion of domestic and overseas financial distresspre-warning methods,this paper put rough set and artificial neuralnetwork technology into the practice of financial distress pre-warningissue.Main work and results are as follows:1. Clarify the definition of financial distress, introduce the relatedtheories of financial distress pre-warning issue,on the basis of which itproposes a set of pre-warning index.2. Discuss the basic concept of rough set theory and methods, such asattribute reduction and discretization theory and algorithm.And thenbriefly introduce the basic theory of artificial neural network, mainlydiscuss the structure of BP network, the process of BP algorithm, analyse-its advantages and disadvantages.Adopt beyesian regularization trainingalgorithm to improve the generalization ability of BP network.At last,itproposes RS-ANN pre-warning model.3. After the selection of sample data, descriptive analysis of thepre-warning variables,it puts RS-ANN model into practice study, and thencompares the result of RS-ANN model to those derived from common BP network and traditional statistic model in order to prove theeffectiveness and feasibility of RS-ANN pre-warning model.