In green credit markets, information asymmetry and corporate greenwashing increasingly undermine the efficiency of resource allocation, while traditional assessment models relying on static, self-reported environmental data fail to impose effective constraints. To address this limitation, this paper develops a dynamic corporate carbon credit evaluation framework by integrating multiple sources of physical (hard) signals and embeds it into commercial banks’ credit management systems. Anchored in multi-source power-carbon signals (e.g., carbon intensity and compliance records), the framework integrates verifiable physical metrics with ESG disclosures via a Bayesian AHP–CRITIC weighting scheme to construct a dual-dimensional classification scheme (“Credit Rating–Green Label”). It further embeds carbon credit scores into dynamic adjustments to credit limits and differentiated interest rate pricing, forming an integrated risk management mechanism. Empirically, a stratified validation strategy is adopted. Analysis based on a sample of 3327 firms shows that the proposed framework achieves a classification consistency of 81.3%, significantly outperforming both a financial-only baseline model (46.8%) and models based on voluntary carbon disclosure (61.4%). Ablation studies further confirm that physical (hard) signal indicators contribute substantially to ranking stability. Moreover, panel regression analysis, based on 36,185 firm-year observations from 3327 firms over the period 2000–2023, demonstrates that carbon credit scores have robust predictive power for future financial distress. Overall, the proposed framework offers a sustainable, data-driven approach to green credit risk management.

Open set domain adaptation (OSDA) faces two critical challenges: the emergence of unknown classes in the target domain and changes in observed distributions across domains. Although numerous studies have proposed advanced algorithms, recent experimental results demonstrate that the classical empirical risk minimization (ERM) approach still delivers state-of-the-art performance. However, few theories can effectively explain this disputed phenomenon. To address the theoretical gap, we focus on constructing a causal theoretical framework for OSDA. We formulate the novel concepts of the fully informative causal invariance model (FICIM) and the partially informative causal invariance model (PICIM). Subsequently, we derive an OSDA theoretical bound to prove that the ERM performs well when the source domain follows FICIM, while it performs poorly when the source domain follows PICIM. The different results may be attributed to the varying amounts of available information when bounding the target domain's stable expected risk. Finally, across different datasets, we conduct extensive experiments on the FICIM and PICIM source domains to validate the effectiveness of our theoretical results. Moreover, our findings can also support the training and fine-tuning of large language models (LLMs).

Options hedging in real-world markets is challenged by strong asset correlations, partial observability, and nonstationary price dynamics. These conditions limit the effectiveness of traditional strategies and pose challenges for reinforcement learning (RL) methods, which often overfit or become unstable in high-dimensional environments. Existing RL hedging methods suffer from static risk measures and inflexible policies, leading to issues in stability, adaptability, and generalization. To address these challenges, we propose Deep Diffusion Reinforcement Learning (DDRL), a new RL framework that integrates the Soft Actor-Critic (SAC) algorithm with diffusion-based generative policy networks for dynamic hedging. By modeling policy distributions through a denoising diffusion process, DDRL effectively captures nonlinear dependencies and generates more robust actions. To improve stability, DDRL also incorporates double critic networks, entropy regularization, soft target updates, and other technical enhancements. We implement DDRL in a simulated trading environment based on historical market data. Empirical results show that DDRL reduces hedging losses and transaction costs compared to baseline RL methods, while maintaining stable performance across varied portfolio configurations and market conditions. These results highlight the potential of generative diffusion policies to enhance the robustness and reliability of RL-based financial decision-making.