August 4, 2025

A Comparison of Machine Learning Models for Predicting the Direction of Bitcoin Using Technical and Macroeconomic Indicators

HSHussain SabborChung Yuan Christian University DTDramane Thiombiano JCJo-Hui ChenChung Yuan Christian University

Key Points

Machine learning models outperform traditional statistical methods in predicting bitcoin returns.
Gradient boosting and bagging exhibit the highest predictive performance across all evaluation metrics.
The study uses a dataset of daily bitcoin returns along with technical and macroeconomic indicators.
Feature selection effectively reduces multicollinearity, enhancing model accuracy and robustness.

Abstract

Abstract This paper investigates the predictive power of technical indicators and macroeconomic variables for forecasting the future direction of Bitcoin returns over a 30-day horizon. A range of machine learning models including classification and regression algorithms are applied to assess forecasting performance. The dataset comprises daily Bitcoin returns, technical indicators (such as RSI, SMA, and MACD), and macroeconomic variables (including the VIX, US 10-Year Treasury Yield, Gold Price, and USD Index) spanning from April 2014 to 2024. Model performance is evaluated using three key metrics: accuracy, the area under the ROC curve (AUC), and the F-measure. The findings show that machine learning models significantly outperform traditional statistical approaches, particularly after applying feature selection to reduce multicollinearity. Among the evaluated models, gradient boosting and bagging achieve the strongest predictive performance across all metrics, demonstrating the effectiveness of ensemble methods for capturing complex patterns in volatile cryptocurrency markets. JEL Classification Codes: C45, C53, G17, E44, G15

Ask AI

Helpful

Bookmark

View Full Paper