Forecasting mid-price movement of Bitcoin futures using machine learning

Area of Science:

• Quantitative Finance
• Computational Economics
• Machine Learning Applications

Background:

• Global financial instability and the COVID-19 pandemic create complex asset price dynamics.
• Accurate forecasting of Bitcoin futures prices is crucial for investors amidst market uncertainty.
• Traditional forecasting models may struggle with the high-frequency, volatile nature of cryptocurrency markets.

Purpose of the Study:

• To evaluate the efficacy of various machine learning algorithms (MLAs) in predicting Bitcoin futures mid-price movements.
• To compare the forecasting performance of MLAs against benchmark models like ARIMA and random walk.
• To analyze prediction accuracy across different intraday time frequencies (5-60 minutes).

Main Methods:

• Utilized high-frequency intraday data for Bitcoin futures.
• Implemented and assessed six distinct machine learning algorithms.
• Compared MLA performance against autoregressive integrated moving average (ARIMA) and random walk models.

Main Results:

• Five out of six MLAs demonstrated average classification accuracy exceeding the 50% threshold.
• MLAs consistently outperformed ARIMA and random walk models in forecasting Bitcoin futures prices.
• The predictive power of MLAs was evident across various tested time frequencies.

Conclusions:

• Machine learning algorithms are effective tools for forecasting Bitcoin futures prices.
• MLAs provide superior predictive accuracy compared to traditional econometric models.
• These findings underscore the relevance of MLAs for navigating cryptocurrency market volatility, particularly during periods of economic turmoil like the COVID-19 pandemic.