基于因子分析的有导师监督型Kohonen神经网络在公司财务预警中的应用研究——以2015年24家上市公司为例

Application of instructor-supervised Kohonen neural network based on factor analysis in corporate financial early warning

针对衡量公司绩效能力的财务指标众多,难以建立高效的公司财务预警系统问题,本文首先通过变异系数和相关系数对最为常用的衡量公司绩效的22项指标进行初步筛选和二次筛选,得到资产收益率、销售净利率、净资产收益率、留存收益与总资产比率、资产现金回收率、营运资本总资产比、经营性现金流量流动负债比、经营性现金流量债务总额比和净资产增长率等9项指标,然后利用因子分析将这9项指标划分为盈利能力、偿债能力和成长能力3个主因子,并以此建立一个按资产收益率层级划分的监督型主元Kohonen神经网络.利用Kohonen神经网络模型对24家上市公司的财务进行实证分析,结果表明该模型相比已有的F分数模型和BPNN模型具有更高的判别准确率,可以有效地规避拒识情形,在实际中具有更高的应用价值.

Due to the large number of financial indicators to measure the performance of the company,this brings many difficulties to the establishment of the company＇s financial early warning system.In this paper,we firstly screened the 22 most commonly used indicators through the variation coefficient and correlation coefficient,and getting the ratio of return on assets,net profit rate,return on net assets,retained earnings to total assets,capital recovery rate,working capital total assets ratio,operating cash flow current liabilities ratio,total operating cash flow debt and net assets growth rate of nine indicators.Then using factor analysis to divide these nine indicators into three main factors of profitability,solvency and growth capacity.The empirical results of 24 listed companies show that this model has a higher discriminant accuracy than the existing F-score model and BPNN model,which can effectively avoid the situation of rejection,so it has a higher application value in practice.