Indubitably investing in the gold market has been one of the most captivating investments in the world over the centuries. The gold standard has been played a key role in the international monetary system here before. It is utilized to mint coins and gold bullion as the money reserves. Because of the volatile and uncertain nature of the paper currency market, investors alternatively place gold as a safe asset and as the hedge against inflation in their portfolios to boost the performance and reduce the risk of their investments.
So an accurate gold price forecast is vital for central banks, Hedgers, and speculators.
The main purpose of this paper is to present a novel and precise hybrid model to forecast the world gold price. In the first stage, predictability and nonlinearity assumption and chaoticity of time series data, have been examined by the Largest Lyapunov Exponent (LLE) and BDS test. The results of the tests showed the gold price is chaotic and nonlinear and using nonlinear models recognized appropriate for this kind of data. Considering markets are highly correlated, the lags of world oil price, US dollar index and stock market index has been recognized significant for forecasting the gold price.
The price has been modeled in two methods to compare, the first one is based on the combination of artificial intelligence techniques which structured optimized Adaptive Neuro-Fuzzy Inference System by Genetic Algorithm (GA-ANFIS) that contains multilayer feed forward neural network with backpropagation and Takagi Sugeno fuzzy inference system. Since there is a numerous combination of inputs can be exercise and there is exist a complex and nonlinear relationship between independent variables, in this paper, the idea of a combination of genetic algorithm and a neuro-fuzzy system arises to find the best of the best significant inputs and optimal lags.
The second model, which is recognized appropriate for the gold data, is Logistic Smooth Transition Autoregressive (STAR) concerning long memory effect and the original model changed to FI-LSTAR in order to estimate accurate predictions, and US dollar index has been accepted as a transition factor. In this method, STEPLS technique has been used to find the optimal lags. The results show that the hybrid artificial intelligence model produced more accurate forecasts and generally input selection and choosing an appropriate model based on the nature of the data can play vital roles in forecast models.