Forecasting volatiletime-series data through Stochasticvolatility model

Abstract

Forecasting of volatile data is generally carried out using Generalized autoregressive conditional heteroscedastic GARCH) model.However, there are some limitations of this methodology, such as its inability to capture empirical properties observed in time-series data. Further, the GARCH assumption that volatility is driven by past observable variables only sometimes becomes a constraint. Accordingly, in this paper, a promising methodology of Stochasticvolatility (SV) model, in which the time-varying variance is not restricted to follow a deterministic process, is considered.The estimation of parameters of this model is carried out using a powerful technique of Kalman filter (KF) in conjunction with Quasi-maximum likelihood (QML) method.As an illustration, volatile dataset of Month-wise total exports of fruits and vegetables seeds from India during the period April 2004 to January 2012 are considered. It is concluded that SV model performs quite well for modelling as well as forecasting of the volatile data under consideration.