Quantitative cryptocurrency trading: exploring the use of machine learning techniques

Abstract

Machine learning techniques have found application in the study and development of quantitative trading systems. These systems usually exploit supervised models trained on historical data in order to automatically generate buy/sell signals on the financial markets. Although in this context a deep exploration of the Stock, Forex, and Future exchange markets has already been made, a more limited effort has been devoted to the application of machine learning techniques to the emerging cryptocurrency exchange market. This paper explores the potential of the most established classification and time series forecasting models in cryptocurrency trading by backtesting model performance over a eight year period. The results show that, due to the heterogeneity and volatility of the underlying financial instruments, prediction models based on series forecasting perform better than classification techniques. Furthermore, trading multiple cryptocurrencies at the same time significantly increases the overall returns compared to baseline strategies exclusively based on Bitcoin trading.