Driven by sustainability goals, China’s green financial system has rapidly expanded, with green bonds emerging as a key policy and market instrument. However, their associated credit risk has grown increasingly complex, marked by high uncertainty and multidimensional heterogeneity. This study proposes a dynamic monitoring framework for credit risk in China’s green bond market, offering insights into its structural drivers and temporal evolution. A four-dimensional risk indicator system, spanning bond characteristics, financial metrics, green benefits, and social performances, is developed. A static Bayesian network (BN) is constructed to reveal conditional dependencies and infer statistical associations among key risk variables, while scenario-based inference is used to trace critical paths of risk transmission. To capture dynamic patterns, a temporal extension via dynamic Bayesian network (DBN) is introduced. Empirical analysis identifies three conditional dependence pathways demonstrating that both internal financial performance and green-specific factors jointly shape credit risk outcomes. The DBN model further outperforms traditional approaches in predictive accuracy and robustness, validating its effectiveness for dynamic credit risk monitoring. This study advances green bond credit risk analysis by integrating correlation-based reasoning with temporal modeling, providing a unified and interpretable framework. By combining a comprehensive indicator system with Bayesian inference, it offers a novel quantitative approach to green credit risk assessment and contributes methodological insights for sustainable finance development.
Heavy-tailed claim distributions, complex dependence structures, and model uncertainty continue to challenge actuarial risk measurement and capital evaluation. Classical tools such as Value-at-Risk, Expected Shortfall, or distortion risk measures often exhibit instability under tail misspecification, while their structure provides limited flexibility for smoothing or tempering extreme losses. This paper develops a unified kernel-based framework for transforming loss distributions through measurable integral operators. Within this framework, we reinterpret the Incomplete Sadefo Transform (IST) as a specific truncated, distribution-weighted kernel operator, correcting earlier misconceptions regarding coherence and spectrality. We derive general structural properties of kernel-induced risk functionals, identifying precise conditions under which monotonicity, homogeneity, convexity, or coherence may hold, and show that spectral representations arise only in degenerate kernel configurations. We establish robustness results under Wasserstein and Kullback–Leibler perturbations, characterize heavy-tail attenuation under polynomially decaying kernels, and analyze how monotone versus nonmonotone kernels affect copula-based dependence. A stylized stress-testing exercise illustrates the role of bounded or sublinear kernel transformations as preprocessing devices before standard capital assessment.
Artificial intelligence (AI) is reshaping the global economic landscape, prompting China to initiate a new wave of AI application pilot programs. Using the AI Application Pilot Zones policy as a quasi-natural experiment, this study examines the impact of an AI-driven transformation on corporate capacity utilization rates and its underlying mechanisms. Based on 44,537 firm-year observations from publicly listed Chinese companies between 2007 and 2023, and employing two-way fixed effects (TWFE) and staggered difference-in-differences (DID) methodologies, we find that the AI policy did not enhance corporate capacity utilization. On average, this led to a 0.516 percentage point decline. This negative effect is particularly pronounced for firms with longer operational histories, higher asset specificity, and greater market concentration. Mechanistic analysis revealed that short-term impacts operate through resource reallocation and demand shock effects. This study further discusses other unintended consequences of the policy, offering insights for refining AI-related policies to promote broader AI adoption.
Supply Chain Risk Management (SCRM) seeks to identify, assess, and mitigate risks such as fraud, operational delays, and financial irregularities to enhance supply chain resilience. This study introduces an Adversarial Autoencoder-based Heterogeneous Graph Neural Network (AAHGNN) that addresses the complexities of supply-chain fraud detection in networks with intricate supplier–customer relationships. The proposed approach integrates graph convolution, neighborhood aggregation, adversarial representation learning, and graph attention. Financial metrics and supplier–customer interactions were utilized to construct a supply-chain knowledge graph. The framework was evaluated against baseline models, including SVM, CNN, GNN, RNN-LSTM, and GCNN. AAHGNN achieved an accuracy of 87.62%, a precision of 90.34%, a recall of 89.57%, and an [Formula: see text]-score of 90.17%, while the baseline GCNN achieved 91.39% accuracy and 96.87% AUC. Furthermore, the attention mechanism in AAHGNN highlights critical supplier–customer relationships associated with fraud, underscoring its effectiveness for explainable supply chain risk management.
Accurately forecasting cryptocurrency price movements following market overreactions is crucial for traders, investors, and risk managers operating in highly volatile environments. This study presents a novel multi-source, multi-output deep learning framework designed to predict the direction of price changes in four major cryptocurrencies — Bitcoin (BTC), Ethereum (ETH), Binance Coin (BNB), and Ripple (XRP) — immediately after overreaction events. By jointly modeling these assets, our approach captures their interconnected market dynamics, enhancing predictive accuracy. We compile an extensive dataset with over 656 features from diverse sources, including historical trading data, on-chain metrics, technical indicators, and social sentiment data from Google Trends, collected at both daily and intraday frequencies. To improve model interpretability and performance, we introduce two engineered features — price change magnitude and price variation speed — that effectively represent intraday volatility. Feature selection using a Random Forest approach reduces the feature set to 30 key variables, ensuring robustness and avoiding overfitting. Using three advanced deep learning architectures — LSTM, RNN, and CNN — we train models to classify the next-day price movement as upward or downward. Empirical results demonstrate that the multi-output LSTM achieves an F1-score of 73.42%, outperforming both single-asset models (62.95–68.25%) and alternative architectures. These findings highlight the benefits of joint modeling, leading to more reliable forecasts during turbulent market conditions. Our framework offers a practical tool for algorithmic trading, portfolio management, and risk mitigation in the dynamic cryptocurrency landscape.
