Volume 12, Number 1, June 2022
| Bitcoin price prediction via LSTM and CNN models |
Fu-Ming Lai 1, Cheng-Shiain Lin 2*, Wan-Rung Lin 3, Yi-Hsien Wang 3
Abstract
Bitcoin is an emerging peer-to-peer electronic currency that can be used to make payments directly around the world without third parties, and as a new transaction channel and cryptocurrency. Investors consider Bitcoin as a profitable investment object that attracts a lot of money from all over the world, and thus, rush to enter and exit the market. Hence, as Bitcoin is considered by investors, companies, and governments as having the effects of real currencies, Bitcoin’s price prediction can help investors arbitrage in the future.
According to literature, many scholars have studied how to use traditional statistics and econometrics to predict Bitcoin, while few studies have used convolutional neural networks to predict Bitcoin. However, traditional models have very limited predictive effects, or their data are highly volatile and non-linear due to no control by a third party. Hence, with the sample period from January 1, 2017 to February 28, 2020, this paper predicts Bitcoin with the neural networks of long short-term memory (LSTM) and convolutional neural networks (CNNs).
Keywords: Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNNs); Bitcoin
1 Ph. D. Candidates. College of Management, National Taipei University of Technology.
2 Department of Computer Science and Information Engineering, Tamkang University
( E-mail: 157446@ mail.tku.edu.tw)
3 Department of Banking & Finance, Chinese Culture University