Forecasting of Bitcoin Daily Returns with EEMD-ELMAN based Model

Abstract

The present study investigates the application of EEMD-ELMAN model to forecast the daily returns of the Bitcoin. More than seven years data were collected online from 18th July 2010 to 17th January 2018. Then the data signal was decomposed into several sub-signals using EEMD method. After, sub-signals were captured by different ELMAN models, and their output results were combined to generate the final forecast. Besides, the results of this study were compared against ELMAN and ARGARCH models. Hence, the statistical metrics revealed that the used model outperforms ELMAN network, and has approximately the same estimation error as ARGARCH, although the later model is prone to bad generalization due to the high gap between its approximation and generalization errors. Therefore, we can confirm that EEMD can be considered as a promising preprocessing technique, which enables to bring up the forecasting performance of ELMAN network with respect to highly volatile time series.