Explainable Artificial Intelligence Modeling for Bankruptcy Prediction under Climate Change Risks

Extreme changes in the weather conditions can lead to operational inefficiencies and physical degradation of a firm's assets. As a consequence, firms with a weak capacity to adapt to such changes can be confronted with financial difficulties. Using explainable artificial intelligence modelling (XAI), we examine the performance of variables assessing climate change risks and the magnitude of the climate change phenomenon in predicting corporate failure. The experimental findings, which are supported by real-world datasets from France, show that climate change risk (CCR) and climate change magnitude (CCM) variables combined with accounting ratios can better predict the firm's outcome, which is either survival or liquidation. The findings based on the XGBoost model, which incorporates the CCR and CCM variables, demonstrate high predictive performance, with an area under the receiver operating characteristic curve (AUC) of 0.995. By relying upon such variables, the Extremely Randomized Trees (ERT) model also reveals a strong bankruptcy prediction capacity, achieving an AUC of 0.999 for the transport sector and 0.996 for the industrial sector. Additionally, the effects of CCM variables are heterogeneous across sectors. The liquidation risk of trade firms tends to be higher in French counties exposed to a low degree of humidity, higher temperatures, and strong winds. Transport and industrial firms benefit from a low likelihood of failure in counties with more precipitations that can prevent the risk of droughts or heat waves.