Аңдатпа
The analysis of cross-country spillover effects is becoming increasingly relevant with the rapid growth of the sovereign green bond market and its increasing integration into the global financial system. Traditional econometric models based on quantitative data are unable to fully account for the influence of qualitative factors such as investor sentiment, information flows and narrative risks. This study offers a new methodology based on the specially trained large language model Qwen3-8B FinSpill to identify, measure and interpret high-quality spillover channels. A two-step instruction tuning process is presented, which first adapts the basic Qwen3-8B model to the financial sector, and then calibrates it to a narrow area of analysis of sovereign green bonds and spillover effects. The use of LoRA and QLoRA techniques allows you to achieve high efficiency in further education. The integration of qualitative indicators extracted by the model, including sentiment indices, thematic narratives and greenwashing events, into the TVP VAR and GARCH MIDAS econometric models demonstrates a significant increase in their explanatory power. The results show that narrative shocks explain up to 35% of yield volatility during periods of market instability, confirming the critical importance of considering information channels for investors, regulators and issuers.