Combination Model Research Based On ARIMA

Prediction plays a very important role in people’s learning, working and life. When it refers to prediction methods, time series analysis must not be ignored. And, when referring to the time series analysis, ARIMA models is an integral part. Predictive accuracy was forecast’s important indices. In order to make the results more accurate, we must try to improve fitting models, and it’s not easy. Study on combination model is to improve the models built to make forecast more accurate.In this paper, we study on the theory of ARIMA model and combination model. At the same time, with China’s GDP data, the combination models based on ARIMA model are also discussed in depth. Firstly, the modeling process of ARIMA is discussed, and solutions to common issues that arise in the process are also given. The forecast results are analyzed based on four evaluations:relative percentage error, mean absolute percentage error, mean absolute error and absolute error. Secondly, the estimated results with exponential smoothing method in the same data are discussed, also based on the same four evaluations. Then, the results of ARIMA model and exponential model are combined to make predict. When solving the weight coefficient, five methods is used, they respectively are simple weighted average method, weighted average method, countdown and inverse square error method, mean square error method and Lagrangian method. Also, the same four evaluations indicators are used to analysis the forecast results. In the end, based on five precisions indications, namely SSE, MSE, MAE, MAPE and MSPE, the results of ARIMA model, index smooth model and the five combination models are compared analysis. It shows that although the composite methods and prediction results of the five combined models are different from each other, compared with the individual models, combination models’predictions are all better than the separate models’. Combination model is worthy to be further studied and applied.