An Empirical Study of Financial BERT Models for Sentiment Analysis and Cryptocurrency Price Correlation

Abstract

Cryptocurrency trading has become a prominent financial domain, with a market capitalization exceeding one trillion USD. As the influence of social media on crypto markets grows, leveraging sentiment analysis (SA) becomes crucial for enhancing trading models. This work showcases the comparative study of advanced sentiment analysis tools, specifically Crypto-BERT, FinBERT, VADER, and SenticNet, and their effectiveness in confirming that Bitcoin is the best cryptocurrency. Also, the correlation between sentiment and prices using a pre-trained transformer DistilBERT is fine-tuned to understand how cryptocurrency prices fluctuate in the market. The study results in a correlation coefficient of 0.88.