News sentiment index and its predictive power for the Russian stock market

This paper analyzes the informational value and predictive power of retail investor sentiment, formed by news flows and social media publications, for the dynamics of the Russian stock market. The aim of the study is to assess the informational value and predictive power of sentiment indicators, calculated using the FinBERT model, for key stock market indicators: stock returns, trading volume, volatility, and the MOEX Russia Index. The empirical dataset includes data on 925 companies listed on the Moscow Exchange from 2015 to 2024, as well as messages from popular financial Telegram channels. Textual sentiment analysis and user visual reactions (emojis) were employed to measure sentiment, from which proxy variables were constructed. Hypothesis testing was conducted using gradient boosting models and vector autoregression (VAR), controlling for macroeconomic and financial variables. The results indicate that investor sentiment is significantly associated with market dynamics, with the strongest effect on trading volumes and the returns of low-capitalization companies. We find that negative sentiment has a more pronounced effect than positive sentiment. The proposed machine learning models, which incorporate sentiment proxy variables, outperformed classical econometric approaches in forecasting accuracy. Model interpretation using Shapley values helped identify the key sentiment factors. Event study analysis revealed the dual nature of sentiment: despite systematically improving the quality of shortterm forecasts (predictive value), there is no significant post-event market reaction to sentiment shocks, and in cases with borderline effects, a pronounced pre-trend in market indicators is observed. This indicates that news sentiment acts primarily as a reactive indicator, promptly reflecting processes already underway, rather than an independent causal factor in market dynamics. These findings underscore the importance of accounting for behavioral factors in investment analysis and can enhance the accuracy of predictive models.