Assessing the Impact of News in a Cryptocurrency Price Forecasting Model

This study investigates the integration of news sentiment analysis into cryp- tocurrency price forecasting field using Long Short-Term Memory (LSTM) model. Focusing on five major cryptocurrencies — Bitcoin, Ethereum, Binance Coin, XRP, and Dogecoin — the research aims to enhance predictive accuracy by incorporating daily sentiment scores derived from news articles. Historical price data and senti- ment scores were collected and processed using the GNews module and TextBlob library in Python, respectively. The results show significant improvements in Mean Absolute Error (MAE) for Bitcoin and XRP when news sentiment is included, with Bitcoin’s MAE decreasing by 23% and XRP’s by 6.3%. However, the impact on Ethereum, Binance Coin, and Doge Coin was minimal, attributed to issues like keyword ambiguity and the nature of the news retrieved. A trading simulation fur- ther demonstrated the practical applicability of the model, particularly for Bitcoin, achieving a 225% increase in net worth over 280 days. Overall, the study provides valuable insights in the field of forecasting cryptocurrency prices. Moreover, it has opportunity for further research, which is able to enhance the robustness and accu- racy of predictive models, providing more reliable tools for traders and investors in the rapidly evolving cryptocurrency market.