Deep Hedging with Generative Market Models for Climate-Aware Portfolio Risk and Financial Resilience

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Researchers in climate finance and AI can adopt the G-CADH framework as a template for integrating generative models and deep hedging under climate scenarios.

🏢実務担当者:Asset managers and risk teams can explore the framework for dynamic hedging strategies that incorporate climate-exposure penalties and liquidity constraints.

🏛政策担当者:Regulators may consider the methodology for developing climate stress testing and resilience scoring standards.

Abstract: Climate change alters asset values through acute physical shocks, chronic productivity effects, transition policies, technological substitution and repricing of carbon-intensive activities. Conventional hedging procedures often assume stationary return distributions, limited transaction frictions and historically observed regimes; consequently, they can be fragile when climate information changes the joint distribution of returns, volatility, liquidity and cross-asset dependence. This paper develops a model-based framework that combines deep hedging with climate-conditioned generative market models to support portfolio risk control and financial resilience. The proposed Generative Climate-Aware Deep Hedging (G-CADH) framework couples a conditional time-series generator with a recurrent hedging policy, coherent tail-risk objectives, climate-exposure penalties, liquidity constraints and model-governance tests. Its mathematical formulation integrates factor and matrix representations, stochastic differential dynamics, scenario-conditioned path tensors, distributionally robust optimisation and a resilience score linking hedge effectiveness to capital preservation. 0–10 numerical model demonstrates how scenario fidelity, hedge adaptability, climate integration, liquidity resilience, transaction-cost efficiency and governance quality can be combined with residual tail, model, liquidity and data risks. The demonstration yields an integrated resilience score of 7.11 for the proposed configuration compared with 5.58 for a conventional benchmark; these values are pedagogical rather than empirical. The analysis indicates that generative models are most useful when they expand plausible joint market–climate regimes without substituting for stress design, while deep hedging is most credible when tail losses, turnover, explainability and out-of-distribution controls are jointly evaluated. The paper contributes an interdisciplinary architecture for researchers, asset managers, banks, insurers and supervisors seeking adaptive climate-risk management under non-stationarity and incomplete historical evidence. Keywords: deep hedging; generative market models; climate finance; conditional value-at-risk; climate scenario analysis; financial resilience; time-series GANs; diffusion models; transition risk; physical climate risk; distributional robustness; portfolio optimisation; reinforcement learning; model risk; sustainable finance.

• semanticscholar https://doi.org/10.62311/nesx/rp2jy-30072026first seen 2026-07-26 06:18:46