A Multidimensional Early-Warning Model for Corporate Bankruptcy

To evaluate the proposed framework empirically, we calculate firm specific impact scores, namely F-values, and assess bankruptcy risk for all nonfinancial A share listed firms in Shanghai and Shenzhen from 2010 to 2024. The sample is constructed through the following procedures. First, financial and insurance firms are excluded because their balance sheet structures differ substantially from those of industrial firms. Second, observations with missing values for the variables required to compute the F-value are removed to reduce measurement error. Third, firm year observations with a nonpositive value for the estimated corporate analogue of the cushioning distance d are excluded, because such values have no meaningful mechanical interpretation in Equation (4). Finally, all continuous variables are winsorized at the first and ninety ninth percentiles to mitigate the influence of extreme observations, especially given the squared terms involved in the calculation of F. After these procedures, the final dataset contains 12,279 firm year observations drawn from the CSMAR database.为实证评估本文所提出的研究框架，我们计算了企业特定影响得分（即F值），并对2010至2024年间沪深两市全部非金融A股上市公司的破产风险进行了测算。样本通过以下步骤构建：第一，剔除金融与保险类企业，因其资产负债表结构与工业企业存在显著差异；第二，移除计算F值所需变量存在缺失值的观测样本，以降低测量误差；第三，剔除缓冲距离d的企业模拟估算值非正的年度观测样本，原因是此类取值在公式(4)中不具备合理的机械解释；最后，对所有连续变量在1%与99%分位数处进行缩尾处理（winsorize），以缓解极端观测值的影响——尤其考虑到F值的计算涉及平方项。经上述处理流程后，最终数据集共包含来自CSMAR数据库的12279条企业年度观测样本。