Analysis of Nonstationary Nonlinear Economic Time Series of Gold Price A Comparative Study

Modeling nonstationary-nonlinear time series has become a major challenge in all fields of scientific research. Some of the popularly used models include ARIMA and GARCH. The limitations of ARIMA model-ing was reported by Harvey [1] and GARCH model by Daniel [2] when applied to model financial time series. Wavelet Networks, a new ap-proach to analyze such time series is seen to be efficient for forecasting. In this paper we have used GARCH model, Wavelet Neural Network (WNN), T-TAR model and WNN with T-TAR model for forecasting a nonstationary-nonlinear time series and these methods are applied for forecasting a nonstationary-nonlinear time series of gold price (price in US Dollar against 1 ounce of gold), an economic time series of current interest. Here it is established that WNN as well as WNN using T-TAR model give accuracy prediction of future values. In the Wavelet Neu-ral Network (WNN) using T-TAR model for preprocessing have further advantages as it includes a parametric model