Joint Mid-Term Forecast of Cryptocurrencies in Technics of Inductive Modelling (on Example of XRP, Waves, ETH)

Low accuracy of mid-term forecasts of cryptocurrencies is an open problem that impedes their promotion on the world financial market. One of the reasons is the lack of reliable predictive models including all or the majority of significant objective and subjective factors. In this paper, we propose to partially compensate this deficit by taking into account mutual relations of cryptocurrencies and using the noise-immune algorithms of GMDH technology. Object of consideration is the dynamics of cryptocurrencies XPR, Waves and ETH during 2 full years from March 2019 till February 2021 with the step of 1 week. We consider tasks of forecast with different options: 2 periods of cryptocurrency behaviour (calm, crisis), 2 forecast horizons (week, month), 3 models (autoregression, regression, hybrid), 2 basic algorithms from the GMDH Shell tool (combinatorial, neural network). We also study how preliminary smoothing of cryptocurrencies dynamics affects forecast error. Baseline is presented by the popular Holt-Winters method. The results prove to be promising, which allows us to recommend the proposed simple approach for experts of cryptocurrency marke