基于离散化贝叶斯网络的财务预警研究——以香港上市公司为例

Financial Prediction Study Based on Discrete Bayesian Network——Take Listed Companies in Hong Kong for Instance

将贝叶斯网络应用到香港上市公司财务预警,结合基于线性相关性的启发式方法建立了6个不同相关性阈值对应的模型,将E-SM离散方法应用于财务预警,并将其与已被多次使用的E-PT离散方法从不同角度进行对比。结果表明:单模型角度下,E-SM方法的表现优于E-PT方法;而在模型联系角度下,当阈值不断降低,两种方法总体上均能提升模型预测的准确率。由于提升程度不同,综合考虑两种分析角度的结果后提出猜想并得到证实,与目的结点相关性系数绝对值在0.05以上的指标用E-SM方法离散效果更好;与目的结点相关性系数绝对值在0.05以下的指标用E-PT方法离散效果更好。

Combined with heuristic method based on linear correlation, this paper applies Bayesian network to financial prediction among listed companies in Hong Kong and builds six different models corresponding to different correlation thresholds. We also apply the E-SM discretion method to financial prediction and compare it with the E-PT discretion meth-od which has been adopted widely from various angles. The result shows that from the angle of single model, generally, the performance of the E-SM method is better than the E-PT method; but from the angle of linked models, with correlatio thresholds reduced gradually, generally speaking, the two methods can raise accuracy rate of prediction. Nevertheless, due to the different degree, we suggest a hypothesis after comprehensive deliberation from the above-mentioned angles： the E- SM method is the better choice for variables having the absolute correlation coefficients greater than 0. 05 with purpose node ; the E-PT method is the better choice for variables having the absolute correlation coefficients smaller than 0. 05 with purpose node. Finally, we adopt the mixed discretion method to verify the hypothesis.