Bitcoin price prediction using transfer learning on financial micro-blogs

Abstract

We present a methodology for predicting the price of Bitcoin using Twitter data and historical Bitcoin prices. Bitcoin is the largest cryptocurrency that, in terms of market capitalization, represents over 110 billion dollars. The news volume is rapidly growing, and Twitter is increasingly used as a news source influencing purchase decisions by informing users of the currency and its popularity. Using modern Natural Language Processing models for transfer learning, we analyze tweets’ meaning and calculate sentiment using the NLP transformers. We combine the daily historical Bitcoin price data with the daily sentiment and predict the next day’s price using auto-regressive models for time-series forecasting. The results show that modern approaches for sentiment analysis, time-series forecasting, and transfer-learning are applicable for predicting Bitcoin price when we include sentiment extracted from financial micro-blogs as input. The results show improvement when compared to the old approaches using only historical price data. Additionally, we show that the NLP models based on transfer-learning methodologies improve the efficiency in sentiment extraction in financial micro-blogs compared to standard sentiment extraction methods.