Georgii A. Borisenko Assessing the Predictive Power of Investors in Determining the Key Rate of the Central Bank of the Russian Federation Using LLM

DOI: 10.38050/2078-3809-2026-18-2-43-57

Abstract

The main aim of this study is to forecast key rate decsisions of the Central Bank of Russian Federation by analyzing investor opinions in the Pulse social network (T-Bank). Despite the importance of the Central Bank's key rate, most studies aim to forecast macroeconomic indicators related to the Central Bank rate, but not the rate itself. To fill this gap, the paper proposes an approach that combines crowdsourcing data with processing by large language models (LLM). Based on semantically tagged posts and comments from the social network using YandexGPT-5-Lite-Instruct, forecasts directions of rate changes for 48 meetings of the Central Bank of the Russian Federation (2019–2025). The integration of LLM made it possible to correctly interpret the context and emotional coloring of informal statements, avoiding the errors of dictionary methods. The accuracy of forecasts based on Pulse data was 90%, exceeding the indicators of analytical agencies in the context of emergency meetings.

Keywords: Central Bank of the Russian Federation rate, forecasting, large language models, sentiment analysis.

JEL: C45, C53, E47, E52, G41.

For citation: Borisenko, G.A. (2026) Assessing the Predictive Power of Investors in Determining the Key Rate of the Central Bank of the Russian Federation Using LLM. Scientific Research of Faculty of Economics. Electronic Journal, vol. 18, no. 2, pp. 43-57. DOI: 10.38050/2078-3809-2026-18-2-43-57.

Download