Comparative Study of Euro-Dollar Exchange Rate Forecasting Based on BP Neural Network and ARIMA

Beijing Institute of Technology, 102401, Beijing, China

^* Corresponding author: lyt060316@ldy.edu.rs

Abstract

Effective economic activity prediction is crucial for financial market stability, as accurate exchange rate forecasting can significantly impact international trade and investment decisions. This study aims to anticipate the EUR/USD exchange rate utilizing the automatically picked BP neural network structure with the Autoregressive Integrated Moving Average(ARIMA) model, assessing their efficacy in capturing market dynamics. Traditional models, like ARIMA, often struggle to account for the complexities and nonlinearities of financial markets, which are influenced by various economic and political factors. Advanced techniques, such as BP neural networks, are designed to overcome these limitations by better capturing complex patterns within the data. With a mean squared error (MSE) of 2.8899e-06 and a lower relative error, the results demonstrate that the BP neural network model performs significantly better than the ARIMA model in terms of prediction accuracy. The ARIMA model's forecasts are presented as a horizontal line, which indicates that it performs less well than the BP neural network model. The study demonstrates that the improved neural network designed by BP has greater predictive capacity in terms of anticipating exchange rates. This provides decision-makers in the market for securities with useful information that can be utilized in their practice.