ARIMA Based Time Series Forecasting Model

As a data mining tool, time series facilitates better understanding nature of the development process of things and permits forecasting the future values of the process parameters based on the data recorded in a chronological order. ARIMA is one of the general time series models and capable of representing time series which, although not necessary stationary, is homogeneous and in statistical equilibrium. This paper presents the characterization, main methods and problems of the time series by the detailed specific algorithm of software Eviews on analyzing the ARIMA model modeling methods, as well as its specific steps on drafting particular flow chart. Finally, it deals with the Producer Price Index (PPI) collected from the year 1978 to 2013 in China. The statistics related to first 33 years are used to train the models and the 3 past years are used to forecast. This paper constructs two models as ARIMA (1, 1, 1) and AR (1) by using the autocorrelation and partial autocorrelation function of time series, and by comparing with the Akaike information criterion (AIC) and the results of the model test, the ARIMA (1, 1, 1) is chosen as the best model for forecasting. The future value of PPI in the 3 past years shows that ARIMA (1, 1, 1) model has a minor error. It is concluded that a properly performed analysis of time series can be a useful tool for analysis and short-term prediction.