ARIMA model in forecasting the US dollar exchange rate

Abstract

The article presents a scheme for building an ARIMA model to forecast the US dollar exchange rate against the Polish zloty, starting with an initial analysis of the training time series, then forecasting using the moving average method (7-day and 30-day) and exponential smoothing, and finally forecasting with the ARIMA model. The accuracy of the ARIMA(0,1,0) model on the test data was also assessed using the same function in RStudio. The forecast results were compared with the actual USD/PLN exchange rate using forecast error measures (ME, MAE, RMSE, and MAPE). However, it is necessary to note the model’s limitations, including sensitivity to unpredictable external factors (e.g., geopolitical crises, central bank decisions) that may affect forecast accuracy. The work is based on historical data obtained from the National Bank of Poland, covering the period from 01.01.2024 to 30.04.2025.