Forecasting Renminbi Exchange Rates Based on Autocorrelation Shell Representation and Neural Networks

Abstract

Since the implementation of the new mechanism of renminbi exchange rates in 2005, their fluctuation range has become more greater. Therefore, it is very important to control renminbi exchange rates risk via forecasting. This paper describes four alternative renminbi exchange rates forecasting models. These models are based on autocorrelation shell representation and neural networks techniques. An autocorrelation shell representation is used to reconstruct signals after wavelet decomposition. Neural networks are used to infer future renminbi exchange rates from the wavelets feature space. The individual wavelet domain forecasts are recombined by different techniques to form the accurate overall forecast. The proposed models have been tested with the CNY/USD, CNY/EUR, CNY/100JPY and CNY/GBP exchange rates market data. Experimental results show that wavelet prediction scheme has the best forecastingperformance on testing dataset among four models, with regards to the least error values. Therefore, wavelet scheme outperforms the other three models and avoids effectively over-fitting problems.