The Research On Financial Early-warning Of Listed Companies Base On NRS-PSOBP Network

In2011, because of the weak capital market performance, the capital requirements and thedifficulty of financing increased greatly than before in many listed companies. Enterprises arefacing high risk in finance. Therefore, it is very necessary to make further research in thetheory and practice of financial warning and to construct a financial crisis warning model thatis suitable for Chinese national conditions for listed enterprises. Financial statements and thechange of financial indicators can reflect the management of enterprise is normal or not.Meanwhile, more and more non-financial factors also play an important role in thedevelopment of enterprise in the21st century. So we need to monitor these indexes everymoment and use feasible technology to predict the running state of enterprise according to thechange of these indexes, and all these measures have very important significance for thoselisted enterprises themselves, investors, capital market and other stakeholders.Based on above, this paper mainly focuses on three aspects: firstly, the enterprise financialwarning index system constructed in this paper includes not only traditional financial indexes,but also some non-financial indexes. Secondly, in order to solve the classical rough set (CRS)theory in processing numeric data insufficiency, this paper will introduce neighborhood roughset (NRS) theory into financial research area for attribute reduction. Thirdly, this paper alsouse particle swarm optimization (PSO) algorithm to make up the BP neural network algorithmwhich has some defects and then construct a model named PSOBP. We make NRS andPSOBP combined to construct a NRS-PSOBP financial crisis warning model so as to realizeearly-warning for listed enterprise financial status.The main research process is described as follows:First of all, we study and analysis the financial warning research situation in domestic andoversea, and then summarize various methods of index selection and model research.Second, we systematically introduce the NRS theory, PSO algorithm and BP neuralnetwork algorithm, and make a comparative analysis between the classical rough set and NRStheory, construct a model based on PSOBP. Third, according to many selection principles we get initial indicators, and then do attributereduction for indicators in using CRS and NRS respectively, construct the listed enterprisefinancial early-warning index system.Finally, according to the new listed enterprise financial early-warning index system, we usethe training sample data to train the PSOBP model, and then use the test sample data toexamine the early-warning performance of the model. We compared this forecast results withother results that are got from the CRSBP model and the NRSBP model, and then find that itis feasible and effective to make NRS replace CRS and using PSOBP. The combination modelprovides a new method for enterprise financial early-warning research.