Forecasting Exchange Rates between Malaysian Ringgit and Chinese Yuan with Neural Network

Abstract

Forecasting the exchange rate movement plays a crucial role in international trading since the exchange rates significantly influence investors, traders and countries. This report aims to apply the neural network modelling for forecasting exchange rates of the Malaysian Ringgit (RM) against the Chinese Yuan (CNY). The exchange rate data collected from 1 January 2022 to 24 October 2023 are used for testing, forecasting and validation purposes. The weights and bias parameters in the neural network model are estimated using the gradient method after a loss function is introduced. The forward propagation and backpropagation equations are derived and the value of parameters is updated iteratively until convergence is achieved. Then, the optimal parameter estimates are employed for data validation. Two neural network models demonstrate promising results in forecasting accuracy. The first model achieved satisfactory accuracy with a minimal mean-square error of 2.8289´10^-5, while the second model achieved a minimal loss function value of 2.9308´10^-5. However, the validation resulted in a higher loss function value of 0.1250. The study also shows that the number of input neurons has a more significant influence on the model’s convergence and forecasting performance than the changes in the number of hidden neurons. Lastly, a comparison of forecasting accuracy between the neural network model and time series methods shows that the neural network is the best model for forecasting exchange rates of RM/CNY.