The best of two worlds: Forecasting high frequency volatility for cryptocurrencies and traditional currencies with Support Vector Regression

•Proposal of novel machine learning method to estimate volatility.•Study of new market known as cryptocurrency market.•Comparison between other volatility models.•Evaluation of the models predictive power using statistical tests.•Machine learning model yielded better results for low and high frequen...

Full description

Saved in:
Bibliographic Details
Published in:Expert systems with applications 2018-05, Vol.97, p.177-192
Main Authors: Peng, Yaohao, Albuquerque, Pedro Henrique Melo, Camboim de Sá, Jader Martins, Padula, Ana Julia Akaishi, Montenegro, Mariana Rosa
Format: Article
Language:English
Subjects:
Artificial intelligence
Autoregressive processes
Confidence
Cryptocurrency
Digital currencies
Exchange rates
Forecasting
GARCH
High frequencies
Machine learning
Mathematical models
Model testing
Performance prediction
Regression analysis
Skewed distributions
Statistical analysis
Stochastic models
Support vector machines
Volatility
Volatility forecasting
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Proposal of novel machine learning method to estimate volatility.•Study of new market known as cryptocurrency market.•Comparison between other volatility models.•Evaluation of the models predictive power using statistical tests.•Machine learning model yielded better results for low and high frequencies. This paper provides an evaluation of the predictive performance of the volatility of three cryptocurrencies and three currencies with recognized stores of value using daily and hourly frequency data. We combined the traditional GARCH model with the machine learning approach to volatility estimation, estimating the mean and volatility equations using Support Vector Regression (SVR) and comparing to GARCH family models. Furthermore, the models’ predictive ability was evaluated using Diebold-Mariano test and Hansen’s Model Confidence Set. The analysis was reiterated for both low and high frequency data. Results showed that SVR-GARCH models managed to outperform GARCH, EGARCH and GJR-GARCH models with Normal, Student’s t and Skewed Student’s t distributions. For all variables and both time frequencies, the SVR-GARCH model exhibited statistical significance towards its superiority over GARCH and its extensions.
ISSN:0957-4174
1873-6793
DOI:10.1016/j.eswa.2017.12.004