October 2, 2025

Application of Social Media Sentiment Analysis for Developing Trading Models in the Cryptocurrency Market

Key Points

A one-unit increase in lagged social media sentiment can predict a statistically significant 0.24–0.25% rise in next-day Bitcoin returns.
Econometric and machine learning models were used, showcasing the effectiveness of sentiment analysis in predicting market behavior.
Advanced models like random forest and XGBoost outperformed linear baseline models, indicating the complexity of sentiment-driven markets.
The application of sentiment analysis extends to trading bots and risk dashboards, emphasizing its practical relevance in market strategies.

Abstract

This study examines the predictive role of social media sentiment in forecasting short-term Bitcoin price changes using econometric and machine learning models. Based on Twitter and Reddit data (2020–2025), we construct a daily sentiment index and analyse its lagged effect on returns. OLS regression and advanced models (random forest, XGBoost) show that a one-unit increase in lagged sentiment predicts a statistically significant 0.24–0.25% rise in next-day returns. Controls include momentum, volatility, and trading volume, with Granger causality tests and VAR confirming sentiment’s leading role. While volume is insignificant, sentiment and momentum are strong predictors. Machine learning models outperform linear baselines, highlighting nonlinear interactions in sentiment-driven markets. Results validate sentiment as a meaningful input for forecasting, with applications to trading bots, real-time risk dashboards, and supervisory tools. The study contributes to applied economics by showing how quantified investor emotion can serve as a leading indicator in volatile cryptocurrency markets. Future research should consider multilingual sentiment, intraday horizons, and cross-asset extensions.Copyright© 2025 The Author(s). This article is distributed under the terms of the license CC-BY 4.0., which permits any further distribution in any medium, provided the original work is properly cited.Article’s history: Received 5th of August, 2025; Revised 9th of September, 2025; Accepted 16th of September, 2025; Available online: 30th of September, 2025. Published as article in the Volume XX, Fall, Issue 3(89), 2025.

Mark Helpful

Bookmark

Relay