Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/5F93768D-7EB9-4B2E-82C0-34FB214A7F66 | ||
| Year | 2024 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Christina-Dionysia Panou, "Next-day Bitcoin price forecasting using time series models", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024 https://doi.org/10.26233/heallink.tuc.99131 | ||
| Appears in Collections |
In an era of a thriving cryptocurrency market that disrupts the traditional economic system, new opportunities for capitalization emerge, accompanied by elevated risks. Managing these risks and optimizing investment decisions is imperative, underscoring the need for dependable tools in cryptocurrency market forecasting. This thesis explores the forecasting capabilities of both the Auto-Regressive Integrated Moving Average (ARIMA) and composite Auto-Regressive Integrated Moving Average-Generalized Auto-Regressive Conditional Heteroscedastic (ARIMA-GARCH) time series models for the daily closing prices of the Bitcoin cryptocurrency. The pronounced presence of heteroscedasticity in the Bitcoin time series data renders the ARIMA models unsuitable for accurate modeling and subsequent forecasting of the data. Conversely, the ARIMA-GARCH(0,1) models effectively deal with heteroscedasticity and demonstrate adequacy in capturing the patterns and structure of the time series. This study is conducted using three distinct test time periods and experiments with various training-test splits to evaluate several ARIMA-GARCH(0,1) models. Additionally, it compares their performance to that of some Recurrent Neural Network (RNN) models available in the literature. Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) validation measures are employed for this purpose. Using prominent stock market indices as exogenous variables of ARIMA-GARCH(0,1) models leads to enhanced performance scores for many scenarios, suggesting a possible impact of the stock market on Bitcoin prices. The top-performing candidates among the proposed ARIMA-GARCH(0,1) and ARIMAX-GARCH(0,1) models exhibit similar, and in some cases, superior forecasting performance when compared to the Long Short-Term Memory (LSTM), Bidirectional-Long Short-Term Memory (Bi-LSTM), Gated Recurrent Unit (GRU), and Bidirectional-Gated Recurrent Unit (Bi-GRU) models employed in other research studies.
Copyright © DIAS 2013
