In some time series, especially financial time series, the high/low frequency sometimes occurs in short time scale, or for a specific period, the phenomena happen such that not observed in the rest of the time series; in the situation, decomposition of the time series into constitutive series can be very useful ,because  the wavelet transform has the ability to transform the time series into low-and-high-frequency information, which allows detecting trends, breakdown points, and discontinuities in the data that may it is lost when other analysis methods are used. to cope with non-stationary time series forecasting and with the aim  of  improving the forecasting accuracy of the volatility pattern in exchange rates , we use the wavelet transform with hybridization methodology to build a hybrid model combining Maximal Overlap Discrete Wavelet Transform (MODWT), Autoregressive Integrated Moving Average (ARIMA), GJR- GARCH model and Radial Basis Function Neural Network(RBFNN) model, so that it is  capable of capturing the volatility in financial time series, by applying it to the Chinese Yuan exchange rate for the period ( 2015/1/5 to 2022/11/11)  , where the wavelet transform technique provides a useful feature based on data analysis, which improves the performance of the model.  The time series of exchange rates were analyzed into their (approximation and detailed) coefficients at three levels using the Maximal Overlap Discrete Wavelet Transform (MODWT). The experimental results of this research demonstrate that the proposed model has a higher prediction accuracy than the single models (SVR and RBFNN) and other hybrid models (MODWT-ARIMA-GJR-GARCH-SVR and ARIMA-GJR-GARCH).

Paper type Research paper.