Prediction of cryptocurrency returns using random forest

This paper examines the prediction of the closing price of Bitcoin (BTC), Ethereum (ETH), Ripple (XRP), and Cardano (ADA). To achieve this, the Random Forest (RF) regression algorithm is employed, utilizing macroeconomic variables, equity market-related features, technical analysis, and blockchain information features on a daily basis. The features are used both in combination and independently within their respective categories. Each category is tested at four intervals, including two bear cycles, one bull cycle, and the entire period. The findings indicate that technical indicators, particularly Moving Averages (MAs), play a crucial role in predicting the closing prices of all above mentioned tokens. Among the MAs, Weighted Moving Averages (WMA) exhibit the most significant influence, followed by the blockchain information variable of market capitalization. Therefore, momentum features have the highest impact on cryptocurrency prices on a daily basis. The RF algorithm demonstrates high accuracy, with low mean absolute percentage errors (MAPE) of 1.4% for the best interval and category.
Masterarbeit Universität Innsbruck 2023