Time Series Forecasting of Bitcoin Price Based on Autoregressive Integrated Moving Average and Machine Learning Approaches
Majid Khedmati, Farid Seifi, Mohammad Javad Azizi
Abstract
Bitcoin as the current leader in cryptocurrencies is a new asset class receiving significant attention in the financial and investment community and presents an interesting time series prediction problem. In this paper, some forecasting models based on classical like ARIMA and machine learning approaches including Kriging, Artificial Neural Network (ANN), Bayesian method, Support Vector Machine (SVM) and Random Forest (RF) are proposed and analyzed for modelling and forecasting the Bitcoin price. While some of the proposed models are univariate, the other models are multivariate and as a result, the maximum, minimum and the opening daily price of Bitcoin are also used in these models. The proposed models are applied on the Bitcoin price from December 18, 2019 to March 1, 2020 and their performances are compared in terms of the performance measures of RMSE and MAPE by Diebold-Mariano statistical test. Based on RMSE and MAPE measures, the results show that SVM provides the best performance among all the models. In addition, ARIMA and Bayesian approaches outperform other univariate models where they provide smaller values for RMSE and MAPE.