Employing Recurrent Neural Networks to Forecast the Dollar Exchange Rate in the Parallel Market of Iraq

Abstract

In this study, neural recurrent neural networks (RNN) have been used to forecast the price of dollars in Iraqi dinars, as it is clear that the government's efforts to control prices in the parallel markets, the commercial markets witnessed a decline in the exchange rate, but it rose again. Which indicates an economic problem that is still present in the country. Despite adjusting the exchange rate of the dinar, the dollar crisis in Iraq has not ended yet. Here we want to forecast the daily price of the dollar against Iraqi dinars for the common next 30 days. According to the results the RNN model have been performed for the data under consideration with different numbers of hidden layer and nodes. The best architecture for the RNN model was [1,10,1,1] using soft plus activation function, which gives the performance of 85% for the training dataset and (92% and 90%) for the testing and validation datasets respectively, with Mse (0.018, 0.000417, and 0.000477) for training, testing, and validation respectively at epoch 4. According to the results of the forecasted values which start from 15 May 2023 to 13 June 2023 the price of dollars will be between 1390 to 1435.

Keywords

• Forecasting
• Neural Networks
• Activation Function
• Exchange Rate