by Aynur İncekırık
ABSTRACT
Gold is an important precious metal widely traded in global financial markets, and its price dynamics interact with various commodity and financial indicators. The current study investigates the performance of unidirectional and bidirectional deep learning architectures in gold price forecasting through comparative analyses. The empirical application consists of a dataset containing daily closing prices for Gold (ALT), Silver (GMS), Platinum (PLT), Copper (BKR), Crude Oil (BTL), Exchange Rate (USDTL) and the Dollar Index (DXY). Daily closing prices for the period between June 8, 2018 and February 3, 2026 were obtained and the relationships between the variables were evaluated using Pearson correlation analysis. The findings showed that precious metals exhibit strong positive correlations among themselves, while crude oil clearly diverges from this group. Specifically, there is a strong positive correlation between ALT and GMS, and between GMS and PLT, while there is a very weak negative correlation between precious metals and BTL. Relationships with the dollar index generally exhibit a limited and heterogeneous structure. During the prediction phase, the data set was split into 80% training and 20% testing using LSTM, GRU, Bi-LSTM and Bi-GRU models. Model performance was evaluated using RMSE, MAE and MAPE metrics. The results revealed that GRU and Bi-GRU architectures have lower error values for some variables, while the Bi-LSTM model does not provide a general performance advantage. Overall, the results demonstrate that deep learning-based approaches offer an effective tool for modeling the price behavior of commodities and foreign exchange markets affected by gold and constitute a strategic decision support mechanism for sustainable finance.
