Search Queries to Internet as a Tool for Mid-term Forecast of Cryptocurrencies (XRP, Waves, ETH) on GMDH-based models

• Published in: International Scientific and Technical Conference "Computer Sciences and Information Technologies" (Online) Vol. 1; pp. 375 - 379
• Main Authors: Boldyreva, Anna, Bayburin, Bulat, Alexandrov, Mikhail, Koshulko, Olexiy, Mogilev, Pavel
• Format: Conference Proceeding
• Language: English
• Published: IEEE 22.09.2021
• Subjects: barometers; cryptocurrency; ETH; GMDH; GMDH Shell; Prediction algorithms; Predictive models; Smoothing methods; Sociology; Software; Software algorithms; time series; Vocabulary; Waves; XRP
• ISSN: 2766-3639
• DOI: 10.1109/CSIT52700.2021.9648740

In spite of optimistic predictions of experts in recent years, cryptocurrencies still hold an inferior position on financial market due to the inability to provide a satisfactory forecast of their dynamics. The latter essentially depends on various subjective factors: attitude of authorities, interest of business community and trust of population. Our hypothesis is that subjective factors are well reflected in Internet queries, and in this study we confirm it experimentally. In the paper, we describe formation of descriptors' vocabulary, creation of barometers, and building of predictive models. The source information covers the period 2019-2021 (2 years) with the step of 1 week. In the paper we consider 3 cryptocurrencies (XRP, Waves, ETH), 2 forecast horizons (week, month), 2 forecast periods (calm, crisis), 3 models (autoregression, regression, hybrid), 2 basic algorithms from the GMDH Shell software (combinatorial, neural network). We show that preliminary smoothing of cryptocurrencies dynamics essentially reduces forecast error. Baseline is presented by the popular Holt-Winters method. The results allow us to recommend the proposed approach for experts of cryptocurrency market.