Sentiment Signals in a Crisis Market: A Tri-Source FinBERT Analysis of Reliance Industries and Bharti Airtel During the COVID-19 Pandemic

Abstract

This study examines whether investor sentiment can explain or predict stock price movements for two major Indian firms, Reliance Industries and Bharti Airtel, during the period January to December 2020. The analysis was conducted in two phases using textual data from Reddit archives collected through Pushshift, together with institutional and media sources from ACE Equity and Google News. Sentiment classification was performed using FinBERT, a transformer model specifically trained on financial language. Phase 1 evaluated whether Reddit sentiment alone could act as a predictor of stock returns. 43,061 Reddit posts for Reliance Industries and 10,444 posts for Bharti Airtel were analysed. Across all forward lags, Pearson correlation coefficients remained below 0.10 and p-values exceeded 0.20, indicating that Reddit sentiment did not have statistically significant predictive value for large-cap NSE stock returns. The "Reliance Paradox" was a key finding. Even though Reliance's stock price went up by almost 30% during the study period, retail investor sentiment on Reddit remained persistently negative. The main reason for this difference was institutional capital flows related to the Jio Platforms fundraising cycle, not what retail investors were saying. To get around these problems, Phase 2 added a weighted tri-source pipeline that combined ACE Equity, Google News, and Reddit sentiment. The model gave Reliance Industries an average weighted sentiment score of +0.12 and Bharti Airtel an average weighted sentiment score of +0.19. The findings indicate that sentiment may correlate with stock price fluctuations on a macroeconomic scale, particularly during the COVID-19 market crash and subsequent recovery however, it lacks the requisite accuracy for short-term stock forecasting.