This paper examines the role of sustainability in shaping sovereign borrowing costs from a macro-financial perspective. Using a multidimensional sustainability index within a cross-country panel framework with country fixed effects, the study highlights two opposing channels through which sustainability influences sovereign bond spreads. On one hand, stronger environmental performance is associated with lower spreads, reflecting reduced ecological risk and improved macroeconomic resilience. On the other hand, the transition toward renewable and sustainable energy is linked to higher borrowing costs, as markets price in short-term fiscal burdens, policy uncertainty, and adjustment costs. The findings suggest that sustainability affects sovereign risk through both risk-reducing and risk-enhancing mechanisms, offering a nuanced view of how green transitions interact with financial markets.

This paper develops an epidemic modeling framework to study financial contagion across the Artificial Intelligence (AI) and energy sectors, two domains that are increasingly interconnected through technological innovation and the global transition toward sustainable energy systems. Drawing on compartmental models such as SIR/SIRS, we reinterpret financial distress as an infectious process, where shocks propagate across sectors through dynamic spillover networks constructed from tail-risk and dependence measures. The analysis captures how periods of heightened uncertainty-driven by rapid AI adoption, energy price volatility, and transition risk- can amplify systemic vulnerability and accelerate the transmission of financial stress. By integrating network-based spillover structures with nonlinear epidemic dynamics, the model provides a tractable yet flexible framework to characterize persistence, recovery, and reinfection of financial distress across sectors. Empirically, the study identifies asymmetries in contagion patterns, with the energy sector often acting as a primary transmitter during transition shocks, while AI-related assets exhibit both amplification and absorption dynamics depending on market conditions. The framework also allows for policy-relevant insights by quantifying how intervention mechanisms-such as regulatory support or technological stabilization-can mitigate systemic risk. Overall, the paper offers a novel interdisciplinary approach to understanding financial contagion in an era shaped by technological disruption and energy transition.

This paper develops a novel framework that leverages large language models (LLMs) for keyword generation and benchmarking in the construction of search-based indices, with direct applications to Bitcoin trading strategies. A key contribution of the study is addressing the memorization and overfitting problem inherent in LLM-generated outputs, ensuring that the extracted keywords reflect genuine, time-relevant information rather than artifacts of pre-trained data. To this end, we design a benchmarking pipeline that evaluates keyword stability, novelty, and out-of-sample relevance using rolling-window validation and perturbation-based tests. Using the filtered and validated keyword sets, we construct a dynamic search intensity index from online query data, capturing shifts in investor attention, sentiment, and information demand in cryptocurrency markets. This index is then incorporated into predictive models to generate trading signals for Bitcoin, with performance evaluated across return predictability, volatility forecasting, and regime detection. Empirical results indicate that controlling for LLM memorization significantly enhances the robustness and predictive power of the constructed index. Compared to traditional keyword selection methods, the LLM-based approach delivers more adaptive and context-sensitive signals while maintaining strong out-of-sample performance. The paper contributes to the growing intersection of natural language processing and financial econometrics by proposing a principled approach to mitigating memorization bias in LLM-driven financial applications.