Forecasting gold based on ensemble empirical mode decomposition and Elman Recurrent Neural Network

Abstract

Gold is an attractive form of investment for investors because it is considered the safest investment compared to other invesment. Forecasting gold price in the future is an importans aspect because of uncertain gold price fluctuations. In this paper, a forecasting model based on EEMD and Elman Recurrent Neural Network (ERNN) is used to predict world gold price. The data used is a daily period of the world gold price data obtained through the investing.com website. First, use the EEMD to decompose the world gold price time series data into several Intrinsic Mode Functions (IMF) and residuals. Then, each component of the IMF and residuals is then modeled and forecasted using the ERNN method. The final forecast result for the gold price time series is the sum of the forecast results for each IMF and the residual. The EEMD-ERNN model was adopted to make modeling easier and to increase forecast accurracy. In forecasting using two methods, namely the EEMD-ERNN and ERNN methods, it is concluded that the EEMD-ERNN hybrid model on gold price data gives better results than the ERNN model without EEMD pre-processing because the EEMD-ERNN has a smaller value of MAPE and RMSE.