A Comparative Study Based on Time Series and Machine Learning Approaches to Forecast National Currency Exchange Rate of Pakistan

The exchange rate is crucial because it can influence a country's economy. It helps brokers and traders make informed operational decisions that reduce risk and maximize profits. Several methods exist for forecasting currency exchange rates. The present study focused on various methodologies, including Box-Jenkins, Holt's method, artificial neural networks, the Facebook Prophet Model, and the Multi-layer Perceptron, for predicting exchange rates. The performance of these techniques is evaluated based on small mean squared error, mean absolute error, and mean absolute percentage error. The results revealed that MLP outperformed all the models. It is a promising method for forecasting the exchange rate of Pakistan, as it yields a relatively minor forecast error. Additionally, the predicted values obtained using MLP are very close to the actual values. The experimental results and time series plot revealed that the exchange rate of Pakistan is expected to increase in the upcoming years. It is concluded that the present study will help to determine the aggregate demand for domestic currency in the coming years. It is also helpful for the government and policymakers. However, understanding exchange rates is essential for anyone involved in international business and finance.