L1正则化Logistic回归在财务预警中的应用

L1-regularized Logistic Regression Modeling for Financial Distress Prediction

线性模型和广义线性模型已广泛地用于社会经济、生产实践和科学研究中的数据分析和数据挖掘等领域,如公司财务预警,引入L1范数惩罚技术的模型在估计模型系数的同时能实现变量选择的功能.本文将L1范数正则化Logistic回归模型用于上市公司财务危机预报,结合沪深股市制造业ST公司和正常公司的T-2年财务数据开展实证研究,对比Logistic回归和L2正则化Logistic回归模型进行对比分析.实验结果表明L1正则化Logistic回归模型的有效性,其在保证模型预测精度的同时提高模型的解释性.

The linear model and the generalized linear model are widely employed in data analysis and data mining in social economic and scientific research,such as Financial Distress Prediction. If L1 norm penalty is added with model parameters, It can achieve feature selection at the same time when the model coefficients are estimated. L1 norm penalized logistic regression model is proposed for financial distress prediction with listed companies in this paper. Together with normal logistic regression and L2 norm penalized logistic regression model, three logistic regression models are built and tested on the two-years-before data from ST companies and normal counterparts in China security market . The results demonstrate the performance of L1 norm penalized logistic regression model. The model can achieve better prediction accuracy and explanation ability.