基于神经网络的金融风险预警模型及其实证研究

Early Warning Model of Financial Risks and Empirical Study Based on Neural Network

本文以金融风险预警单指标区间评价标准为依据,生成足够多用于BP神经网络(BPNN)建模用的训练样本、检验样本和测试样本,在遵循BPNN建模原则和基本步骤的情况下,建立泛化能力较好的金融风险预警BPNN模型。对1994～2010年中国金融风险的实证研究表明:BPNN模型能较好地应用于中国金融风险的预警研究,实证结果与中国金融实际运行情况吻合度高,除2008年和2010年金融风险处于"警惕"状态外,其他年度处于"基本安全"状态;BPNN模型克服了因子分析法及其与BPNN相结合方法的缺陷,且能分析评价指标与金融风险之间存在的非线性关系和评价指标的灵敏度等。

Based on the evaluation criteria of every indicator for early warning of financial risk （EWFR）, this paper produces the efficient number of training data sets, verification data sets and testing data sets for back-propagation neural networks （BPNN） model and, following the principles and basic steps of BPNN model, establishes a BPNN model with a good generalization for early warning of financial risk. The empirical study of China＇s financial risks from 1994 to 2010 shows that BPNN model can be better used in the research on the early warning of China＇ s financial risks and empirical results can better match the actual operation of China＇s finance. Except for the financial risks in 2008 and 2010, which are in Rank III （guard）, the financial risks in other years are in Rank II （subordinate safe）. The BPNN model overcomes the shortages of both factor analysis （FA） and the method of FA combined with BPNN and can analyses the nonlinear relationship between evaluation indicators and financial risks and the sensitivity of evaluation indicators,etc.