Bitcoin Price Direction Forecasting and Market Variables

Abstract

This paper aims to improve Bitcoin price direction prediction using a CNN-LSTM model that incorporates various relevant indicators, such as stock market indices, commodity indices, and interest rates. Separate models are trained for predicting price up and down direction and combined to enhance prediction accuracy. We utilize binary classification models to independently analyze the impact of different features, verified through explainable artificial intelligence techniques. Additionally, an investment strategy based on our model is proposed and compared with traditional strategies, specifically focusing on maximum drawdown relative to the S&P500 buy-and-hold strategy. Results suggest that our strategy offers potential for stable investment in Bitcoin, showcasing its value as a financial asset. This study demonstrates the role of deep learning in Bitcoin price direction prediction and investment strategy development and contributes to future research on cryptocurrency forecasting and investment approaches.