CMStatistics 2022: Start Registration
View Submission - CFE
A0433
Title: Forecasting cryptocurrency prices with machine learning: How important is market volatility Authors:  Perry Sadorsky - York University (Canada) [presenting]
Irene Henriques - York University (Canada)
Abstract: Cryptocurrencies are decentralized digital currencies that use blockchain technology to settle transactions. Cryptocurrencies are an emerging asset class that has the potential to increase efficiency in the financial services sector greatly. Forecasting cryptocurrency prices is critical for making well-informed investment decisions concerning this important new asset class. We use machine learning techniques to forecast daily Bitcoin, Ethereum, Cardano, and Ripple price direction. The analysis reveals that random forests, extremely randomized trees, and support vector machines have higher prediction accuracy than Lasso or Nave Bayes. We find that the 10- to 20-day forecasts using random forests, extremely randomized trees, and support vector machines achieve prediction accuracies greater than 85\% with some prediction accuracy reaching 90\%. For a 20-day forecast, Shapley values show that MA50, MA200, and WAD technical indicators are important features for each of the four cryptocurrencies studied. US three-month T bills, ten-year bond yields and inflation expectations tend to be more important features than market volatility. The importance of market volatility varies by cryptocurrency. Ripple is unique in that emerging market stock market volatility is the most important feature. Our results reveal the high prediction accuracy of using machine learning methods in forecasting cryptocurrency price direction and provide valuable information on variable importance.