Fitting Exponential Smooth Transition Autoregressive Nonlinear Time-Series Model using Particle Swarm Optimization Technique

摘要

Exponential Smooth Transition Autoregressive (ESTAR) family of parametric nonlinear time-series models is considered. The methodology for estimation of its parameters using a powerful Particle Swarm Optimization (PSO) technique is discussed. Further,simulation study is also carried out to test the validity of the proposed methodology. A heartening feature of ESTAR model is that, as opposed to some other nonlinear models involving regimes switching, the change between the extreme regimes is smooth and is assumed to be defined by a bounded continuous function of a transition variable. Further, it is capable of describing those datasets that depict cyclicity. As an illustration, it is employed for modelling and forecasting of Oil sardine, Mackerel andBombay duck time-series landings data in India. Finally, the performance of fitted ESTAR model is also compared by computing goodness-of-fit statistic and various measures of forecast performance. It is concluded that fitted ESTAR model perform better than ARIMA methodology for the datasets under consideration.