Bitcoin, Sentiment Analysis and the Efficient Market Hypothesis: A Machine Learning Approach

Abstract

This paper uses machine learning models to directionally forecast the Bitcoin and implicitly test the Efficient Market Hypothesis. We use technical, asset based and sentiment-based data that are fed to 4 machine learning algorithms. Sentiment analysis is introduced by using data from Google Trends. The frequency of the data used is weekly spanning the period from 23/10/2017 to 02/08/2021 (198 observations). The results of the models that do not include sentiment data support the efficient market hypothesis. Furthermore, an investor following our optimal model was able to generate 297.35% higher returns than the buy and hold strategy.

Keywords: Machine Learning, Cryptocurrency, Bitcoin, Google Trends, Efficient Market Hypothesis