CAJ | 학술논문

上市公司财务风险预警在财务管理中一直都扮演着关键的角色。来自财务数据的特征是上市公司的各种财务指标，代表着上市公司不同方面的财务现状。这些异构特征给财务风险预警带来了巨大的挑战。当前多核学习方法已经显示出了优于支持向量机方法的预测性能。首先，一个利用预定义的基核进行无监督学习。在此过程中，一个稀疏性约束条件被引入，用于限制基核的线性组合，进而改进学习性能和预测模型的可解译性。通过无监督学习可以得到一个“最优”的线性组合核。最终，这个“最优”组合核用于支持向量机中，进而得到具备稀疏学习能力的多核预测机。通过利用214对 ST和非 ST 公司财务数据进行了仿真实验。实验结果证实，稀疏多核学习方法在全部数据集和不同产业数据集上均优于现有的主流预测方法。
상시공사재무풍험예경재재무관리중일직도분연착관건적각색。래자재무수거적특정시상시공사적각충재무지표，대표착상시공사불동방면적재무현상。저사이구특정급재무풍험예경대래료거대적도전。당전다핵학습방법이경현시출료우우지지향량궤방법적예측성능。수선，일개이용예정의적기핵진행무감독학습。재차과정중，일개희소성약속조건피인입，용우한제기핵적선성조합，진이개진학습성능화예측모형적가해역성。통과무감독학습가이득도일개“최우”적선성조합핵。최종，저개“최우”조합핵용우지지향량궤중，진이득도구비희소학습능력적다핵예측궤。통과이용214대 ST화비 ST 공사재무수거진행료방진실험。실험결과증실，희소다핵학습방법재전부수거집화불동산업수거집상균우우현유적주류예측방법。
Financial distress prediction (FDP)is of great importance for managers,creditors and investors to take correct measures to reduce loss.Many quantitative methods have been employed to develop empirical models for FDP.In this paper,a sparse multiple kernel learning (MKL)method is proposed for the task of FDP.In the proposed sparse MKL method,a log-based group selection terms (GST)is introduced to control the sparsity of the learning machine.The log-based GST meth-od instead of the kernels in MKL using"kernel trick",and as a concave penalty,it brings a sparse so-lution for ratios selection and has an ability to find some locally optimal solutions.In experiments,the proposed method is compared with other FDP methods on 214 couples of normal and ST Chinese listed companies during the period 2006-2012 to demonstrate the accuracy.The performance of the pro-posed method is superior to other classifiers and can get a sparse solution indeed.