• Login
• Admin

Zhang, Xinting (2021) Trade Bitcoin with neural networks. [Dissertation (University of Nottingham only)]

PDF - Registered users only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader Download (1MB)

Abstract

Bitcoin price prediction has been researched with various ideas. Challenges especially the great magnitude change of the price exist in current research. Research related to bitcoin price prediction concentrated on the time-series models like ARIMA and machine learning methods. Among them, deep learning models have a comparable performance. Past research related to deep learning seldom focuses on trading, most of them did not give a clear method to transfer model output into holding positions. In this paper, we model the price of Bitcoin with MLP, RNN, and LSTM. Propose an explicit method to transfer the model output into holding positions including classification networks and regression networks. We also propose a novel way to train a model where we call DOL (direct optimizing loss). DOL allows the model to output the next day’s holding position directly and may utilize trading metrics like Sharp Ratio and Total Return to be the loss function. Our results show that LSTM performs better than MLP and RNN, also the proposed DOL may enhance the performance of models at a considerable level.

Item Type:	Dissertation (University of Nottingham only)
Depositing User:	Zhang, Xinting
Date Deposited:	19 Apr 2023 15:30
Last Modified:	19 Apr 2023 15:30
URI:	https://eprints.nottingham.ac.uk/id/eprint/66350

Actions (Archive Staff Only)

Edit View