Abstract: Heterogeneity of financial indicators is an important factor affecting the accuracy of financial distress prediction. Existing multiple kernel learning (MKL) methods can be used to solve the problem of heterogeneous data learning. Firstly, this paper introduces the theoretical framework of financial distress prediction based on subspace MKL. On this basis, three subspace MKL methods are proposed according to the maximization variance criterion, the maximization of class separability criterion and the principle of non-linear subspace mapping. They are maximization variance projection subspace MKL, separability maximization subspace MKL, and nonlinear subspace MKL, respectively. Using the collected data of Listed Companies in China, the proposed methods are compared with the existing representative financial distress prediction methods, and the experimental results are analyzed. The experimental results show that the proposed subspace MKL financial distress prediction framework is effective, and the subspace MKL prediction methods constructed under this framework can effectively improve the accuracy of financial distress prediction.

Key words: prediction of financial distress, kernel methods, support vector machine, multiple kernel learning

摘要： 财务指标的异构性是影响企业财务困境预测精度的重要因素,现有多核学习方法能够用于解决异构数据学习问题。本文首先介绍了子空间多核学习财务困境预测理论框架,在此基础上根据子空间学习的最大化方差准则、类别可分性最大化准则、非线性子空间映射原理,提出了三种子空间多核学习方法,分别为最大化方差投影子空间多核学习、类别可分性最大化子空间多核学习、非线性子空间多核学习。利用采集的我国上市公司数据进行实验,对比所提出的方法同现有代表性财务困境预测方法,并对实验结果进行分析。实验结果表明,本文提出的子空间多核学习财务困境预测框架行之有效,该框架下所构造的子空间多核学习预测方法能够有效地提升财务困境预测精度。

关键词: 财务困境预测, 核方法, 支持向量机, 多核学习