Cryptocurrency price prediction: A narrative-based approach to sentiment analysis
This paper takes a novel approach to sentiment analysis of cryptocurrency-related Twitter data by applying concepts from narrative economics. The aim is to determine whether tweet engagement and sentiment metrics are predictors of cryptocurrency price returns and trading volumes. The machine-learning algorithm latent Dirichlet Allocation (LDA) was used on a dataset consisting of cryptocurrency-related tweets to unveil the following four narratives: Decentralised Finance (DeFi), Non-fungible tokens (NFTs), Gaming and Memecoins. Empirical analysis consisting of Granger causality testing and OLS regressions revealed a complex relationship between tweet engagement and cryptocurrency prices, where both are predictive of each other. Out of the identified narratives, the Memecoin narrative was found to hold the most predictive power over cryptocurrency prices and trading volumes. A strong association between the S&P500 stock market index and cryptocurrency prices was also revealed, which goes against the common belief that cryptocurrencies are able to act as a hedge against traditional markets.
Faculteit der Managementwetenschappen