LLM Sentiment Extraction and RAG for Macroeconomic Time Series Enrichment

AUTOR(I) / AUTHOR(S): Nemanja Jovanović  

  

Download Full Pdf  

DOI: https://doi.org/10.46793/TIE26.190J

SAŽETAK /ABSTRACT:

Official macroeconomic statistics often lack the qualitative context found in institutional reports. This paper presents an automated pipeline combining large language model (LLM) sentiment extraction with retrieval-augmented generation (RAG) for semantic enrichment of monthly macroeconomic time series. The methodology processes multiple report types from the National Bank of Serbia (2006–2024), addressing three engineering challenges: temporal alignment of quarterly and annual reports with monthly data using exponential decay models, quality verification through multiple LLM runs using Krippendorff’s alpha consistency measurement, and cross-document validation via a local RAG module over a ChromaDB vector database. Experimental results demonstrate high extraction consistency (  > 0.75). Correlation analysis between institutional sentiment and monthly GDP growth shows a statistically significant relationship at a one-quarter time lag, with the Financial Stability Report series exhibiting a distinct pattern of strengthening correlation at longer lags, providing preliminary correlational evidence consistent with leading-indicator properties, though formal predictive validation remains a direction for future work. The resulting enriched dataset adds multiple semantic columns to the original monthly records, providing a foundation for integrating qualitative signals into economic forecasting models. The proposed framework offers a scalable, locally executable solution for transforming unstructured institutional narratives into structured features for predictive analytics

KLJUČNE REČI / KEYWORDS: 

large language models; retrieval-augmented generation; sentiment analysis; macroeconomic time series; semantic enrichment

 PROJEKAT / ACKNOWLEDGEMENT:

LITERATURA / REFERENCES:

• Loughran, T., & McDonald, B. (2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance, 66(1), 35–65.
• Tetlock, P. (2007). Giving content to investor sentiment: The role of media in the stock market. Journal of Finance, 62(3), 1139–1168.
• Malo, P., Sinha, A., Korhonen, P., Wallenius, J., & Takala, P. (2014). Good debt or bad debt: Detecting semantic orientations in economic texts. Journal of the American Society for Information Science and Technology, 65(4), 782–796.
• Hansen, S., & McMahon, M. (2016). Shocking language: Understanding the macroeconomic effects of central bank communication. Journal of International Economics, 99, S114–S133.
• Picault, M., & Renault, T. (2017). Words are not all created equal: A new measure of ECB communication. Journal of International Money and Finance, 79, 136–156.
• Apel, M., & Blix Grimaldi, M. (2012). The information content of central bank minutes. Riksbank Working Paper No. 261.
• Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., & Amodei, D. (2020). Language models are few-shot learners. Advances in Neural Information Processing Systems, 33, 1877–1901.
• Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., & Zhou, D. (2022). Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35, 24824–24837.
• Ziems, C., Held, W., Asgari, O., Atanasov, J., Edison, A., Jernite, Y., & Yang, D. (2024). Can large language models transform computational social science? Computational Linguistics, 50(1), 237–291.
• Baker, S., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty. Quarterly Journal of Economics, 131(4), 1593–1636.
• Shapiro, A. H., Sudhof, M., & Wilson, D. J. (2022). Measuring news sentiment. Journal of Econometrics, 228(2), 221–243.
• Bybee, L., Kelly, B., Manela, A., & Xiu, D. (2024). Narrative asset pricing: Interpretable systematic risk factors from news text. Review of Financial Studies, 37(2), 515–570.
• Krippendorff, K. (2013). Content analysis: An introduction to its methodology (3rd ed.). Sage.