Application of Traditional Machine Learning Models for Quantitative Trading of Bitcoin

Abstract

Bitcoin, the most popular cryptocurrency around the world, has had frequent and dramatic price changes in recent years. The price of bitcoin reached a new peak, nearly $65,000 in July 2021. Then, in the second half of 2022, the bitcoin price begins to decrease gradually and drops below $20,000. Such huge changes in the bitcoin price attract millions of people to invest and earn profits. This research focuses on the predictions of bitcoin price changes and provides a reference for trading bitcoin for investors. In this research, we consider a method in which we first apply several traditional machine learning regression models to predict the Changes of Moving Average in the bitcoin price, and then based on the predicted results, we set labels for bitcoin price changes to get the classification results. This research shows that the method of transforming regression results to the classification analysis can achieve higher accuracy than the corresponding machine learning classification models and the best accuracy is 0.81. Besides, according to this method, this research constructs a Machine Learning Trading Strategy to compare with the traditional Double Moving Average Strategy. In a simulation experiment, the Machine Learning Trading Strategy also has a better performance and earns a 68.73% annualized return.