Economic theory has failed to provide sufficient explanation of the dynamic path of price movement over time. Therefore, the use of any linear or non-linear functional form to model the gold price movement is bound to be arbitrary in nature. Neural Networks equipped with genetic algorithm have the advantage of simulating the non-linear models when little a priori knowledge of the structure of problem domains exist. Studies suggest that such a system provides better predictions when compared with traditional econometric models. The NeuroGenetic Optimizer software is applied to the NYMEX database of daily gold cash price covering 12/31/1974--12/31/1998 period. Among different methods, back-propagation neural networks with genetic algorithms is used to predict gold price movement. The results indicate that prices in the past, up to 36 days, strongly affect the gold prices of the future. This confirms the fact that there is short-term time dependence in gold price movements.
