Cryptocurrency portfolio allocation using a novel hybrid and predictive big data decision support system

Abstract

Establishing a cryptocurrency portfolio management and allocation decision support is vital in enabling investors to thrive in this market. More than nearly 5000 cryptocurrencies are traded daily at cryptocurrency exchanges, and one of the primary challenges is the optimal selection of the assets. While research studies considered cryptocurrency price prediction as a typical time series forecasting problem, a portfolio allocation problem can be defined as a Multiple Criteria Decision-Making (MCDM) problem. To tackle this problem and generate a response to these dynamic challenges, this study developed an analytical and predictive approach to guide investors through their decisions. Time series forecasting was conducted using the Prophet Forecasting Model (PFM) to predict the prices. For the asset allocation stage, this study extended the Cluster analysis for improving the Multiple Criteria Decision Analysis (CLUS-MCDA) algorithm by adding additional features to this large-scale decision-making technique, including Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and the Višekriterijumsko kompromisno rangiranje (meaning: multi-criteria optimization and compromise solution) (VIKOR) methods. While the concept of Parallel Decision Making (PDM) in massive structured problems was previously introduced in the CLUS-MCDA algorithm, this study also utilized a target-based normalization technique that gives the decision-makers more control over the extended CLUS-MCDA II algorithm. To validate the reliability of the method, a cryptocurrency trade experiment considering more than 70 cryptocurrencies and multiple scenarios based on decision-makers' preferences was analysed. The outcomes of this study showed that the proposed decision support system is a reliable tool for such practical and real-world big data problems.