Cryptocurrency trading with ensemble machine learning algorithm

Abstract

In this project, we aimed to use ensemble machine learning algorithms to trade ten cryptocurrencies along with attempting to add more external factors. Cryptocurrency included in this project were Cardano (ADA), Binance Coin (BNB), Bitcoin (BTC), DOGE, DOT, Ethereum (ETH), LINK, Polygon (MATIC), Uniswap (UNI), and Ripple (XRP). Furthermore, ten external factors, which are ten major stock indices, were added to the algorithm. All machine learning algorithms in this project are used to trade for four trading circumstacnes, an 1-hour interval, six-hour interval, daily interval and weekly interval. There are six machine learning models in this project which will be separated as based and ensemble models. XGBoost (XGB), Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) are based, models. Next, three based models are used to combine with three ensemble methods which are Equally Weighted Forecast Combinations (EW), Adaptive Regression by Mixing (ARM), and Aggregation of Forecasts Through Exponential Reweight (AFTER). Models are evaluated under two criterias, accuracy and Sharpe Ratio. As a result, XGBoost outperformed other models for all trading invervals while Equally Weighted Forecast Combination ranked second in both accuracy and Sharpe Ratio.