This research introduces a methodology that presents a hybrid model that integrates ARIMA with LSTM/GRU architectures to forecast monthly trends in the Turkish gold and foreign exchange markets. The volatility inherent in the economic frameworks of emerging economies such as Turkey has become increasingly evident, especially following the sudden fluctuations in exchange rates and gold prices that started in 2018. In this context, traditional time series models struggle to effectively model complex economic behavior. To address these challenges, this study formulates a hybrid model that combines the advantages of ARIMA with the nonlinear learning capabilities of the LSTM and GRU deep learning frameworks. The dataset used covers a period of 24.5 years from 2000 to 2024. The model includes both Turkey-specific and global macroeconomic factors. To improve the forecasting accuracy of the model a Walk-Forward validation approach was used to continuously improve the model with each successive observation. The findings show that the proposed model achieves high success in both short-term and long-term forecasting of gold and exchange rates and effectively adapts to sudden market changes.