Using the Hybrid ARDL–GRU Model in Investigating the Dynamic Relationship between Dinar Deposits and US Dollar Payments at the Central Bank of Iraq

Abstract

This research examines the relationship between dinar deposits and U.S. dollar payments at the Central Bank of Iraq using monthly data for the period January 2016–June 2025. The ARDL model, the GRU neural network, and a hybrid ARDL–GRU model were applied. The results show that U.S. dollar payments are stationary at levels, while dinar deposits become stationary after first differencing, and the two series exhibit a significant positive long-run cointegrating relationship. The linear ARDL model has a limited ability to capture sudden shocks, whereas the hybrid ARDL–GRU model achieves superior forecasting performance both in-sample and out-of-sample. The findings confirm the Central Bank of Iraq's efficiency in managing domestic and foreign liquidity and maintaining market stability and provide three-month-ahead forecasts for U.S. dollar payments. The research recommends adopting hybrid models, incorporating additional economic variables, and using larger sample sizes to improve forecasting accuracy and support monetary policy decisions.