| Abstract [eng] |
This thesis explores the prediction of steep price movements in cryptocurrencies by employing Reddit messages and technical indicators. 'Steep' is defined as daily price changes that exceed 10 percent in either direction. The analysis involves a wide variety of cryptocurrencies traded on Binance, one of the leading cryptocurrency exchanges, covering data from 2020 and 2021. The study explores the impact of sentiments extracted from Reddit messages on predicting these significant price movements. Sentiment analysis was conducted using advanced Natural Language Processing models, including Bidirectional Encoder Representation from Transformers (BERT), XLNet, BigBird-RoBERTa, and OpenAI's Generative Pertained Transformer 3.5 (GPT-3.5). The effectiveness of Reddit-derived sentiments in predicting price movements were conducted using traditional machine learning algorithms, namely Random Forest and Gradient Boosting. This was contrasted with a recurrent neural network model, the Bidirectional Long Short-Term Memory (Bi-LSTM). Results revealed that the Bi-LSTM model outperformed traditional machine learning models. Furthermore, the study demonstrated that sentiments extracted from Reddit messages significantly improved the performance of price movement prediction models, thereby highlighting the potential of integrating social media sentiment analysis into cryptocurrency price forecasting. However, the implementation of a simple trading strategy using these predictions did not yield profitable results when backtested. |
