An alternative approach to capture cyclical and volatile phenomena in time-series data

Abstract

Exponential autoregressive (EXPAR) and generalized autoregressive conditional heteroscedastic (GARCH) models are usually employed for fitting of cyclical and volatile data respectively. However, in practical situations, there may be data which embodies both this phenomena at the same time. To tackle such situations, a new form of parametric nonlinear time-series model, EXPAR-GARCH is proposed. Methodology for estimation of parameters of this model is developed by using a powerful optimization technique called Genetic Algorithm (GA). Entire data analysis is carried out using SAS and MATLAB software packages. For illustration, monthly price series of edible oils in domestic and international markets is considered. The individual models as well as the proposed model were assessed on their ability to predict the correct change of direction in future values as well as by computing various measures of goodness-of-fit and forecast performance.