Abstract:

The aim of this study is to determine whether successful forecasts for cryptocurrencies such as Bitcoin will be obtained by different methods. The reason for the use of Bitcoin prices (Bitcoin/$) in the study is the idea that this cryptocurrency is still the most commonly used cryptocurrency on the market and that cryptocurrencies will successfully represent the overall situation of the market. The series of financial markets may contain volatilities for some reasons, such as speculations. It also often involves non-linear changes. Such features lead to failures in obtaining forecasts for financial time series. In this study, the GARCH model from the classic time series models and the DVM-ECK method, which is a machine learning method, obtained the quotes of the Bitcoin price series, the model performance was compared. The study used the 1155-day Bitcoin price series ( ) for the period from January 1, 2017 to February 29, 2020. Both models have used the Bitcoin price series and the gameplay of this series, the external variables have not been included in the models. According to both models, the forecasts were obtained for periods of 1 month, 2 months and 3 months. According to MAPE rates for GARCH and DVM - ECK models, the unemployed successful forecast rates are 98,0347 - 95,3423 per month respectively; 97,9544 - 96,1307 per month and 98,1272 - 91,4874 per month respectively. The GARCH model made it possible to more successful results for each three periods. The study finds that the GARCH model can be used in making forecasts for the cryptocurrency price series.

Keywords: