Research And Implementation Of Financial Crisis Early Warning For SMEs Listed Companies In China

In recent years, with the development of market economy, the Small Medium Enterprises (SMEs) have been growing rapidly and gradually become the main force in the market economy system. But the unreasonable capital structure, incomprehensive finance information and lack of management cause weak anti-risk ability of SMEs in China. When the deterioration of the external environment happens, SMEs are very hard to resist the consequent risk. Therefore, the SMEs also need to continue to strengthen the building of the financial crisis early warning mechanism to detect potential crisis as far as strengthen their self-construction, to ensure the normal operation of the enterprises. Based on this, the SMEs need to establish a set of financial crisis identification, prevention and treatment methods, effective financial crisis early warning mechanism to help enterprises with financial management and decision-making.One of the reasons for SMEs in lack of financial crisis warning is that it’s difficult to obtain data. With the establishment and development of China’s SME board, all of the market capacity, the trading system and institution-building step onto a new level, which provides a good foundation and platform for the related academic research. The other issue is the majority of the previous studies are based on annual data granularity rather than quarterly data, possibly resulting in poor effectiveness of the model.The thesis makes use of the data of Small and Medium-sized Enterprises Listed Companies to establish financial crisis early warning system. The main ideas of the thesis include three aspects:Firstly, according to the characteristics and the causes of financial crisis of the SMEs listed companies, the thesis chooses20financial indicators. In the process of determining the financial warning interval, it uses the Principal Components Analysis (PCA) method to evaluate the financial situation, and the improved clustering method named K-mean++to divide the score into three categories, which are defined as good financial condition, general financial condition and financial failure, individually, based on the distribution of Special Treatment enterprises in the score and clustering evaluation method Silhouette. Weka software is used to complete Feature Selection, removing the correlation between indicators as well as redundant indicators. Through majority voting, there are12indicators remained.Secondly, the thesis proposes an ensemble algorithm Weighting Based on Maximum-between-Cluster-Distance (WBMCD) to generate balanced weights through test set and regularization item, in order to avoid overfitting on train data which directly causes ensemble algorithm failure on prediction classification. We choose quarterly data as the warning interval granularity, and use the data of the previous period to predict the classification of the current one. Support Vector Machine composes the weak learners. Particle Swarm Optimization (PSO) method is used to search for the best initial parameters for SVM. Then we specify a factor under a given range as a disturbance to generate a certain number of SVM weak learners, and integrate them with the weights generated by the WBMCD algorithm model, gaining a better and more accurate classifier.Lastly, the thesis designs and implements the SMEs Listed Companies Financial Crisis Early Warning System, using the data from CFinance data warehouse of the Platform of the Listed Companies in China, with the matlab API for Java and web development technologies such as JSP, Ajax, structure Struts, Hibernate, which is integrated into the Platform of the Listed Companies in China, promoting the decision support system to complete.