A Dynamic Framework for Climate Conditional Value-At-Risk Based on Nonlinear Canonical Correlation Forests

Abstract

Climate change poses several challenges for the economic system. The financial and insurance industries are called on to play a pivotal role, in particular in coming up with innovative solutions such as sustainable products. According to the EU’s Sustainable Finance Disclosure Regulation, sustainable products have to meet the requirement of pursuing a sustainable investment objective. Furthermore, they need to have measurable goals in place aligned with the investment objective, which can be reported to clients, in order to achieve an advanced understanding of the potential performance to obtain more effective stress scenario analysis. Accordingly, the design of insurance products requires suitable risk and profitability analytics. In our paper, we provide an innovative risk analytical tool for the design of sustainable insurance products environmentally linked, that is, a Climate Conditional Value at Risk (CCVaR), based on Random Forests with a Canonical Correlation Analysis algorithm, building a VaR of climate-related portfolio volatility that is obtained by assessing how the relationships between volatilities of portfolio assets modify according to the bioclimatic indicators. In particular, we propose a splitting rule for the Random Forest algorithm based on OVERALS, a generalisation of Canonical Correlation Analysis based on similarities and not on correlations, considering a more general perspective of non-linear relationships.