Intelligent cryptocurrency trading system using integrated AdaBoost-LSTM with market turbulence knowledge

Abstract

The Bitcoin market is firmly positioned as a global asset market. However, due to its extremely high volatility and the lack of a custom Bitcoin trading system, investors find it difficult to establish an effective investment strategy in this market. In this study, we build an intelligent Bitcoin trading system to maximize profitability by predicting the market. We propose a fusion approach combining technology with economic knowledge to achieve accurate predictions. Firstly, we provide an ensemble prediction framework that integrates the AdaBoost algorithm with the LSTM deep learning model as a form of technological convergence. This dramatically improves the predictive performance due to the high representational capacity of LSTM and the overfitting minimization capability of AdaBoost. Secondly, to account for the repeated structural volatility in the market, we combine the econometrics Markov regime-switching model with the AdaBoost-LSTM model to predict the period of market turbulence caused by structural breaks in the regime. Our prediction model suggests that market turbulence lasting longer than 21 days could be a potentially high-risk investment in the future. Finally, we have developed a Bitcoin-customized trading algorithm that maximizes rewards by predicting upward movements in prices and warning of high-risk investments during a long-lasting turbulent regime. In trading tests, our system yielded a cumulative return of 2.4 times and 4.0 times on 1-day and 5-day forecasts, respectively, compared to other representative trading strategies such as the stochastic oscillator.