The combining prediction of the RMB exchange rate series based on diverse architectural artificial neural network ensemble methodology

Abstract

Motivated by the neural network ensemble approach, this paper puts forward a diverse architectural artificial neural network (ANN) ensemble method to optimize the combining prediction of the RMB exchange rates. On the one hand, four types of architectures are adopted here including multilayer perceptron (MLP), recurrent neural networks (RNNs) to diversify the learning mechanism. On the other hand, the nonparametric kernel smoothing technique is applied to make combining forecasts, which can overcome the drawbacks of traditional methods. The empirical results show that the proposed method has significantly improved the forecasting performance of the optimal single ANNs and random walk model, especially in RMB exchange rate series forecasting.