Forecasting the gold prices using time series and novel hybrid deep learning approach

Purpose – This study aims to provide a hybrid deep-learning forecasting model for precise gold price movement direction (increase or decrease) and single-step predictions (a day in advance). Covering an eight-year time span (2011–2019), The proposed model combines the efficiency of Bi-LSTM layers in recognizing short- and long-term relationships, Bi-GRU that typically provide the similar performance of Bi-LSTM with some improvements and lower computation overhead and the capacity of Convolutional layers in extracting meaningful information from time-series data and learning its internal representation. After conducting two groups of the experiment the first one of using standardized data whereas the second applied on normalized data, we found out the model with normalization data without applying the dropout technique outperformed the other models in standardization data with MSE of .0005, RMSE of 0.0244, and MAE of 0.0214. Research limitations/implications The study examines the effects of Crude oil, the Consumer Price Index (CPI), and the inflation rate as macroeconomic factors, and identified technical indicators used by technical analysts to forecast future price changes that enhance the predictive power.