Digital Asset Forecasting based on Recurrent Neural Network

Abstract:
Recent years has witnessed the booming development in digital asset markets. The volatility and instability of the digital asset market makes it challenging to accurately predict the movement of digital assets. Over the past decades, multiple studies have proved the ability of artificial intelligence in stock trading in traditional markets. In this paper, we design a Recurrent Neural Network-based trading system for predicting data in the highly volatile digital asset market, enabling it to successfully manage asset portfolios in a real-world setting. By combining asset value forecasts and conventional trading tools, the trading system decides whether to buy, hold or sell digital currency at a given point in time. The experimental results show that, given the data in an interval t , a prediction with an error of less than 0.5% for the data in the subsequent interval t +1 can be obtained. Evaluation of the system through backtesting shows that our proposed system can not only successfully maintain a stable portfolio of digital assets using real historical transaction data in a simulated real-time environment, but even increase the value of the portfolio over time.