关键词: 违约预测, 指标组合遴选, 对数似然函数, 惩罚系数, 中小企业

Abstract:

Corporate default prediction is to infer the default probability in the future, which is related to the economy and society. The contribution of this paper is three folds. First, it constructs a multi-objective function that considers the performance of default prediction and the number of indicators. It transforms the mayfly algorithm into a binary mayfly algorithm using a sigmoid function, which is introduced into the financial risk field to select the optimal indicator combination. Then, in the log-likelihood function of the logistic regression, it adds a penalty coefficient to defaulted enterprises to maximize the F-measure and compute the optimal penalty coefficient to improve the accuracy of defaulted enterprises under the premise of ensuring the overall discriminatory accuracy. It makes the objective function of the logistic regression more closely match the actual situation and ensures that the estimated weight vector more accurately reflects the functional relationship between the indicator data and its default status. It is  found that internal non-financial factors such as “executive salary disclosure”, “whether the top ten shareholders are related” and “supervisory board shareholding” and the macroeconomic factors such as “per capita GDP”, “benchmark interest rate for medium and long-term loans” and “money and quasi-money supply growth rate” have a key influence on default prediction of small and medium-sized enterprises (SMEs). The proposed methodology can improve the identification of corporate credit risk and reduce the non-performing loan rate of commercial banks.

Key words: default prediction, indicator combination selection, log-likelihood function, penalty coefficient, small and medium-sized enterprises (SMEs)