A hybrid forecasting model was proposed in this article; seasonal time series ARIMA and neural network back propagation (BP) models were combined together in which is known as SARIMABP. This model was used to improve forecasting of high frequency data with application on exchange rate (Egyptian pound / US dollar). The aim is to combine models to build a complete picture especially if a time series exhibits different patterns. The forecasting performance was compared among SARIMABP and SARIMA models and showed that the mean square error (MSE) and the mean absolute error (MAE) of the SARIMABP model were the lowest. The turning point evaluations also show that the proposed model has the ability to capture the actual direction of turning points of the time series.