Scholarlink

<p>This study examines the Monetary Policy Committee (MPC) communication of the Central Bank of the Republic of Türkiye (CBRT) over the period 2006–2025 using a hybrid framework that combines large language models, fine-tuned transformers, and econometric regime identification. A CBRT-specific RoBERTa-Large classifier is trained on sentences extracted from MPC summary reports to identify “hawkish”, “dovish”, and “neutral” statements. The trained classifier is subsequently applied to MPC decision reports to determine overall communication tone at the sentence and document level. Model evaluation is conducted using a time-order split to preserve chronological structure and reduce temporal leakage. At the sentence level, tone classifications are analyzed across econometrically identified inflation regimes, governor tenures, and economic agent references. The results indicate that hawkish communication becomes more prevalent during high-volatility regimes, while neutral tone dominates in more stable periods. Tone distributions also vary across leadership periods, suggesting that institutional communication adapts to macroeconomic conditions. At the report level, tone composition is evaluated jointly with inflation regimes, governor information, and interest rate decisions. Decision tree analysis shows that policy decisions are systematically associated with communication tone and regime conditions under a time-ordered robustness framework. Overall, the findings suggest that CBRT communication evolves in alignment with macroeconomic regimes and institutional transitions. The study provides a structured and replicable approach to analyzing monetary policy tone in an emerging market context.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fine-Tuning Language Models to Decode Monetary Policy Tone: Evidence from the Central Bank of the Republic of Türkiye Across Inflation Regimes

  • 期刊论文

  • 发表日期：

    2026年7月31日

查看全文
  • Muhammet Rıdvan İnce,
  • Meltem Kurt Pehlivanoğlu,
  • Hikmetcan Özcan

摘要

This study examines the Monetary Policy Committee (MPC) communication of the Central Bank of the Republic of Türkiye (CBRT) over the period 2006–2025 using a hybrid framework that combines large language models, fine-tuned transformers, and econometric regime identification. A CBRT-specific RoBERTa-Large classifier is trained on sentences extracted from MPC summary reports to identify “hawkish”, “dovish”, and “neutral” statements. The trained classifier is subsequently applied to MPC decision reports to determine overall communication tone at the sentence and document level. Model evaluation is conducted using a time-order split to preserve chronological structure and reduce temporal leakage. At the sentence level, tone classifications are analyzed across econometrically identified inflation regimes, governor tenures, and economic agent references. The results indicate that hawkish communication becomes more prevalent during high-volatility regimes, while neutral tone dominates in more stable periods. Tone distributions also vary across leadership periods, suggesting that institutional communication adapts to macroeconomic conditions. At the report level, tone composition is evaluated jointly with inflation regimes, governor information, and interest rate decisions. Decision tree analysis shows that policy decisions are systematically associated with communication tone and regime conditions under a time-ordered robustness framework. Overall, the findings suggest that CBRT communication evolves in alignment with macroeconomic regimes and institutional transitions. The study provides a structured and replicable approach to analyzing monetary policy tone in an emerging market context.