The predictive power of the Ethereum transaction network
Today, there are more than 1,500 traded cryptocurrencies with a combined market capitalization exceeding 800$ billions. These numbers are still growing as well as the interest of some investor in this market. Cryptocurrencies were rst devised as a payment method, but the large uctuations in the pric...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2020 |
| País: | España |
| Institución: | Universitat Oberta de Catalunya (UOC) |
| Repositorio: | O2, repositorio institucional de la UOC |
| OAI Identifier: | oai:openaccess.uoc.edu:10609/118926 |
| Acceso en línea: | http://hdl.handle.net/10609/118926 |
| Access Level: | acceso abierto |
| Palabra clave: | complex networks machine learning cryptocurrency redes complejas aprendizaje automático criptomoneda xarxes complexes aprenentatge automàtic Machine learning -- TFM Aprenentatge automàtic -- TFM Aprendizaje automático -- TFM |
| Sumario: | Today, there are more than 1,500 traded cryptocurrencies with a combined market capitalization exceeding 800$ billions. These numbers are still growing as well as the interest of some investor in this market. Cryptocurrencies were rst devised as a payment method, but the large uctuations in the price led to their use as an alternative to the traditional stock market. As with other nancial assets, cryptocurrencies have also captured the attention of the research community. In current literature we can nd many works facing the problem of cryptocurrencies price prediction. Most of them are regression models, market simulations to calculate ROI, or classi cation models to predict the sign of future change. However, anticipating this market is not an easy task and the results obtained still have a high degree of uncertainty. This study contributes to the current literature by analyzing the Ethereum transactions as a complex network. Next, we evaluate its predictive power on Ethereum future price by deploying two predictive models. The rst one considers a set of features that performed well in current literature, such as price of the previous days, technical indicators and volume of tweets mentioning Ethereum. The second model considers the features of the rst model together with the properties computed from the transaction network. We found an increment in the accuracy when considering the properties of the transaction network. In addition, three network's properties appear in the top 11 of most important features in the nal model. We conclude that the properties computed from the transaction network provide additional information that does not exist in the current literature variables. |
|---|