An interpretable decision-support systems for daily cryptocurrency trading. (1st October 2022)

• Record Type: Journal Article
• Authors: Dolatsara, Hamidreza Ahady; Kibis, Eyyub; Caglar, Musa; Simsek, Serhat; Dag, Ali; Dolatsara, Gelareh Ahadi; Delen, Dursun
• Is Part Of: Expert systems with applications. Volume 203(2022)
• Publication Date: 2022-10-01
• Subjects: Cryptocurrency -- Bitcoin -- Price prediction -- Predictive analytics -- Machine learning -- C&RT -- Decision support systems; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.117409
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A short-term Bitcoin price prediction system is developed. A three-level feature selection approach is proposed. A parsimonious, transparent, and interpretable prediction model is constructed. The system is tested for viability on challenging time period like the COVID-19 era. The proposed prediction system provided a high level of accuracy, sensitivity, and specificity. Abstract: Cryptocurrencies, especially Bitcoin (BTC), have become an important commodity for both individual and corporate investors within the last decade. The limited supply, high volatility, and random price fluctuations have increased investors' interest in BTC, especially in daily trading. Although BTC has been yielding a high rate of returns, price fluctuations and constant speculations make the investors wary of unexpected price movements. Predictive modeling suffers from the complexity of the datasets (i.e., the high number of features employed to forecast BTC movements) as well as the black-box nature of most machine learning algorithms (which is especially problematic for corporate investors since they are obligated to disclose their investment decisions to their clients). Therefore, the main goal of the current study is to assist individual and corporate investors in making transparent and interpretable daily BTC trading decisions by developing a predictive analytics framework. To address the complexities posed by the datasets, a comprehensive tri-level feature selection approach is proposed. The selected features are then, fed into the Classification & Regression Tree (C&RT) to build a highly parsimonious, transparent, and interpretable prediction model. The resultant model was not only evaluated on the test (holdout) sample but was also tested on challenging time periods, including the first half of 2020 (the start of the pandemic era) to exhibit the viability and reliability of the proposed framework. Finally, a decision support tool is developed for the practical implementation of the model. The tool can be used by short-term investors not only due to its highly simplistic, transparent, and interpretable structure, but also its higher accuracy, sensitivity, and specificity results when compared to the extant literature.