Analyzing the Dynamics of Cryptocurrency Rate of Return through Social Media for Financial Trends (Euro vantage journal of artificial intelligence)

Summary

• The study examines Twitter and Telegram data between July 19, 2023, and July 19, 2024, to evaluate social media sentiment impacts on Bitcoin, Ether, and Binance returns.p.1
• By July 2022, the cryptocurrency industry reached a market capitalization of $1.025 trillion, with Bitcoin trading near $21,000 compared to its initial price of $0.0008.p.2
• On September 10, 2023, market-wide downside volatility was observed, with daily returns dropping -11.44% for Bitcoin, -11.50% for Ether, and -10.29% for Binance.p.14
• Predictive evaluations on Twitter data showed the LSTM model achieving an RMSE of 3.1917 and MAPE of 85.28% for Bitcoin return rates.p.20

Key insights

1. Twitter sentiment aligns closely with real-time short-term price fluctuations due to its concise messaging format.p.1
2. Telegram content reflects broader community-driven market discussions suitable for long-term forecasting, despite higher noise levels.p.1
3. Bitcoin exhibits elevated price volatility, measuring approximately five times higher than gold or standard fiat currencies.p.2
4. Deep learning architectures consistently achieve better forecasting performance than traditional statistical techniques by handling non-linear dependencies.p.4
5. Traditional ARIMA and SARIMA time series models yielded higher prediction errors, with SARIMA reaching a MAPE of 185.53% for Bitcoin on Twitter data.p.17
6. ARIMA models slightly outperformed SARIMA on Telegram data, achieving a MAPE of 17.15% for Bitcoin return predictions.p.18
7. Bi-LSTM models outperformed standard LSTM for Ether on Twitter sentiment data, recording an RMSE of 1.5492 and a MAPE of 33.82%.p.20
8. Telegram-based Value at Risk models at a 95% confidence level estimated downside risks of -2.64% for Bitcoin, -3.63% for Ether, and -3.00% for Binance.p.22