Research Of Financial Distress Prediction Based On Bayesian Network

The prediction of corporate financial distress has always been one significantproblem. In real life, most companies suffering from financial problems experience aperiod from being financially normal, worse to financial distress finally, representingin the financial indices gradually. So financial distress is predictable both in theory andpractice. As the Chinese economy participates more and more in the global competition,the society requires badly a prediction system to deal with corporate financial distress.Bayesian Network is introduced in this paper to predict financial distress. Unlikeregression methods, Bayesian Network has no requirement for the populationdistribution and it can be used to model complex relations between uncomplete middlevariables and interacting variables. Bayesian Network also requires nothing for thesamples, which means uncomplete data or information still can be used to train or testthe model. At the meantime, Bayesian Network is dynamic and interactive: the networkcan be updated through new information and subjective opinions can be involved in thenetwork, too. Compared to other machine learning methods, Bayesian Network is moreexplainable and apprehensible because it can represent the relations between variablesby network graph directly.Special treatment is considered as the symbol of corporate financial distress inthis paper, and the research is carried out based on the financial data two years andthree years before the special treatment. The sample includes almost all the financialdistress companies from1998to2010. The variables are selected based on acorrelation analysis method, and EP-T method is considered most suitable for the valuediscretization. The Bayesian Network model's prediction ability is tested by10-foldcross validation to be accurate and solid, and it's also compared to the predictionability of Logistic regression model, which is widely used in the financial distressprediction before and now. The empirical study shows that as a tool of classificationand prediction, Bayesian Network model has a good prediction ability and promisingfuture for application in the research of financial distress prediction.