Forecasting gold prices time series by using joint analysis and separately analysis

In recent years, there are many studies rely on forecasting with artificial neural networks. In this study, artificial neural networks are discussed considerably in demand over the past decade in the world finance literature. In the study, forecasting for the highest and the lowest gold prices with feed forward artificial neural networks are comprehensively studied. Linear or curvilinear functions are used in activation functions of artificial neural networks. Model1 and Model2 are used. Model1 has linear activation function in output layer and Model2 has lojistic activation function in output layer. Initially, we used two separate feed-forward artificial neural networks for analyzing the lowest and the highest gold prices values. Afterwards, lagged variables of time series are joıntly given to tifi i l neu l netwo ks s input. We jointly fo e st the lowest and the highest gold prices. Artificial neural networks gave better results for certain architectures. The forecasting results are discussed according to Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Absolute Deviation (MAD) and Direction Accuracy (DA) criterion. Jointly analysis gave better results