Nonlinear autoregressive with exogeneous input neural network time series model performance: bitcoin price prediction

• Published in: International Journal of Computational Economics and Econometrics
• Main Author: Rashid N.A.; Ismail M.T.
• Format: Article
• Language: English
• Published: Inderscience Publishers 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85198344653&doi=10.1504%2fIJCEE.2024.139764&partnerID=40&md5=1f57baedaf146a3d96ef2ae68d7e68e4

There are over 10,000 listed cryptocurrencies, with bitcoin becoming the most used cryptocurrency at present. This research’s aim is to establish the different dynamic time series architectures of nonlinear autoregressive having exogenous input (NARX) and nonlinear input output (NIO) to forecast the bitcoin price as well as compare their performance. Furthermore, this study attempts to combine the different number of inputs, hidden nodes, and time delay to assess the social media attribute (X) and bitcoin price (Y) past value impact in each model. The results show that all model architectures NARX and NIO with Levenberg-Marquardt backpropagation training algorithm have a significant relationship between inputs and output. This means social dominance, social volume, and weighted social sentiment have a relationship and effect on price except for model 3 with architecture NIO-1-5-1 (d = 1) and NIO 1-10-1 (d = 2). This research is significant because the results of this study will help traders and investors reduce risk and increase returns. Copyright © 2024 Inderscience Enterprises Ltd.