Comparative Study On Models Of Financial Distress Prediction Based On Neural Networks

Financial distress is also known as the financial crisis, whose the worst condition is bankruptcy. When a company is unable to perform contracts, and repay interest and principal to creditors on time, it is faced with the financial distress. Empirical evidence shows that it is a gradual process for most of companies to get in the financial distress. That is, they first behave the abnormal financial situation. Then, their financial situation gradually deteriorates. Financially, they go into bankruptcy. Therefore, not only does the companies'financial distress have the precursor, but also can be forecast. With the improvement of China's market economy and the integration of international economy, the correct and effective forecast of financial distress takes on important immediate significance to guard against the risk for operators, protect rights and interest of investors and creditors, and strengthen the economic supervision for government administration sections.At present, models widely investigated and used to forecast financial distress mainly include parametric and non-parametric statistical approaches. With the development of information technology, neural network (NN) was introduced into the research area of financial distress. NN doesn't strictly limit the sample data. It can effectively deal with non-linear questions and overcomes the limitation of traditional parametric statistical models. Financial distress forecast models based on NN has become a new research focus. Models for application are mostly based on back-propagation neural network (BPNN), and gradually expanded to other networks. This paper attempts to use Elman neural network and probabilistic neural network (PNN) to construct models.The study uses 90 Chinese listed companies as samples that were randomly selected from all of companies. The 90 samples contain 45 companies that got into financial distress and 45 companies that were in fine financial situation. The principle of companies got into financial distress is whether they are specially treated due to financial abnormity. Then, the study takes financial indicators, as well as corporate governance indicators, into account when selecting financial distress forecast indicators. It is expected to more completely reflect the indicators that will cause companies'financial distress. 14 indicators are selected from 29 indicators by using the statistical analysis. Three forecast models based on BPNN, Elman, and PNN respectively are constructed by using training samples. The data of training samples contain two-year-ahead financial data and corporate governance data. Then, the three models are used to predict the possibility of financial distress of the test samples.Empirical analysis prove that financial distress forecast models based on NN have higher prediction accuracy and are an effective approach to forecast financial distress. The prediction accuracy rates of financial distress forecast models based on BPNN, Elman neural network, and PNN are 90%, 86.67% and 73.33%, respectively.