Article Preview
TopIntroduction
Credit risk warning is an integral aspect of financial activity that helps financial institutions to earn profits, predict financial risks, and decrease the probability of default (Tang et al., 2021). In an effort to curb the severe impact of default problems on the economic market, the Chinese government has developed several systems to restrain defaults. For example, to stimulate the debtor to consciously fulfil the obligations determined by legal documents in force and to promote the construction of the social credit system, the Supreme People's Court of China has formulated relevant regulations for announcing the dishonest civil debtor to society, the social activities of a dishonest civil debtor are restricted, and the purpose of credit discipline is achieved.
According to statistics, by the end of March 2022, there were nearly 7.27 million dishonest civil debtors in China, including defaulted individuals and companies. It is worth noting that the development of listed companies is related to the healthy operation of the capital market and the quality of economic transformation. By the end of March 2022, 4,782 companies were listed in China, with a total market capitalisation of 80.7 trillion yuan, which ranks second in the world in terms of market capitalisation, and the amount of taxes paid by these companies accounted for one-quarter of the national tax revenue. Listed companies are a key factor in developing China's national economy, and any default of listed companies impacts the capital market. Against this background, historical data of listed companies were matched with government credit data to establish an effective early warning model of listed company credit risk. This model not only provides a basis for commercial banks' lending decisions but also has practical significance for the development of listed companies and the regulation of the financial industry.
The above discussion clearly shows that corporate credit is an important influencing factor in financial activities, which has led to the emergence of a considerable body of research on corporate credit risk. Early researchers used statistical methods to evaluate credit risk, and the most representative research was presented by Altman (1968), who constructed a linear discriminant model based on financial indicators and proposed an effective corporate bankruptcy discriminant tool, the Z-SCORE model. In addition, statistical methods, such as logistic regression (Costa e Silva et al., 2020) and probit regression (Chi et al., 2016) are often used to evaluate credit risk, but statistically based credit evaluation methods have the problem of low-discriminatory accuracy. With the application of computer technology in multiple fields and disciplines, scholars have widely used credit risk evaluation methods based on machine learning to construct more accurate credit risk early warning models. Prominent examples are machine learning algorithms, such as K-nearest neighbour (Ata & Hazim, 2020), decision trees (Teles et al., 2020), support vector machines (Shi et al., 2013), neural networks (Zhao et al., 2015; Bekhet & Eletter, 2014) and integrated learning algorithms. Of these, integrated learning algorithms have become a popular avenue of credit risk research in recent years. For example, Wang & Ma (2011) proposed an integrated RS-boosting algorithm combining boosting and random subspaces to predict corporate credit risk. The empirical results showed that RS-boosting achieved the highest prediction accuracy compared to seven kinds of logistic regression analyses, decision trees, ANN, bagging, boosting and random subspaces. Zhu et al. (2017) used data from Chinese-listed companies and found that integrated machine learning offers significant advantages in predicting credit risk for small- and medium-sized enterprises (SMEs). Machine learning methods have greatly improved the accuracy of credit risk assessment; however, the method is prone to problems of insufficient interpretation and weak causality (Lei et al., 2022).