Assessment of Gold Price Predictability and Comparison of Predictions Made by Linear and Nonlinear Methods

Document Type : Research Paper



This article aims to study the predictability of daily return of global gold price, using data during the period 2011/07/25 to 2012/12/17. For this purpose, using Scheinnkman, Dechert & Brock Test (BDS), we try to model the linearity, nonlinearity and chaos of the series under consideration. The results is indicative of the rejection of the hypothesis that the gold price daily return is stochastic, which in turn confirms the predictability of the series. The results of the linear model confirm the nonlinearity behavior of the series. Then, an ANFIS neuro-fuzzy model was designed to predict gold price daily return. To carry out this research, we use ARIMA and GARCH models to eliminate linear and nonlinear impacts of the gold price daily return, respectively. Next, using the several criterions, we compare the results of two models- ARIMA as linear model and GARCH as nonlinear model. As expected, ANFIS linear model presents a better prediction than ARIMA and GARCH models. Finally, using Morgan-Granger-Newbold statistics (MGN), we investigate the statistical significance of the divergence between predictions of the models. The results indicates a significant difference between the predictions produced by nonlinear and linear ARMA models.