Predicting Stock Market Volatility by Bayesian Treed Gaussian Processes based onGARCH model

摘要

we propose to predict financial volatility by a newtreed Gaussian processes based on GARCH model. Threecorrelation functions, isotropic exponential power, separablepower and Matern families, are applied in the proposed hybridtreed GP models and stationary Gaussian processes. Theempirical results show that the hybrid approaches generatebetter predictive capability than the stationary GARCHmodels; particularly, the treed Gaussian processes withMatern family correlation structure yields superiorperformance among the others.