Università del Salento · 2017 · 11 citations · 0 references
Open access
BusinessEconomicsFinanceFinancial EconometricsFinancial EconomicsMachine LearningMaximum Likelihood EstimationSupport Vector MachinesTime Series ForecastingTime Series PredictionVolatility AnalysisVolatility ModelingAsymmetric Volatility ModelsConditional Heteroskedasticity ModelsForeign Exchange Volatility ModelingGeneralized Autoregressive Conditional Heteroskedasticity ModelsSupport Vector RegressionTime Series Volatility ModelingVolatility ForecastingVolatility Modelling
In finance, volatility is fundamentally important because it is associated with the risk. A growing body of literature shows that risks associated with volatility are priced in stock, option, bond, and foreign exchange markets. Therefore, an accurate measurement and estimation of the volatility is critical in financial markets. The generalized autoregressive conditional heteroskedasticity (GARCH) has been one of the most popular volatility models and the model is usually estimated from the maximum likelihood estimation (MLE) method. In this paper, we attempt to improve the MLE-based GARCH forecast using the support vector machine (SVM). We also compare with two asymmetric volatility models: exponential GARCH (E-GARCH) and Glosten-Jagannathan-Runkle GARCH (GJR-GARCH). We carry out the analysis through simulations and real datasets. The results show that the GARCH- and SVM-based volatility model provides better predictive potential than the existing volatility models.