The Outlier Analysis In Transfer Function Model

Time series is a sequence that observed by time, which exists in various fields extensively, such as finance, science and engineering fields. Time series analysis is an important method for analyzing and processing dynamic data. It is established based on statistical methods to predict and control behavior with historical and present observations.This paper mainly discusses outlier analysis in transfer function model, builds a new model and predicts future sales data.1. This paper builds the ARMA and transfer function model, using observed sales and the dynamic relationship between the sales and leading indicator to predict the future sales. Comparing forecasts average errors, the accuracy of the transfer function model is better than that of ARM A model.2. Combining outlier analysis with transfer function model to predict future sales. Through analyzing the outlier effects, determining the new transfer function model, we can get the prediction sales by former observations of leading indicator and sales. Comparing forecasts average errors, if the outlier of output series (sales) is caused by the input series (leading indicator), the accuracy of the model without outliers does not change much; if the outlier is only because of the output series (sales), the model without outliers has a better accuracy than before; if the outlier is only because of the input series (leading indicator), the model without outliers has a better accuracy than before.