The Selection Rerearch Of Time Series Prediction Of Provineial Population In China

Population prediction has important theoretical and practical significance for it is the basis of regional economic and social development. So far, many researchers have used time series models to construct population prediction, but there is almost no research on the selection of optimal model from the prediction precision, bias and uncertainty. From the macro level, population prediction can provide the number of the age of children and adolescents in the future, which is crucial for employment and education planning. From the micro perspective, predicting regional population will help investment in infrastructure or the use of funds. For example, school consolidation, the power plant is increased, the city planning is redesigned and real estate development will use the regional population prediction. Although, many scholars have been using the time series model of population prediction, few people study the choice of the optimal model from the forecasting interval length, the historical critical years and the prediction interval by the regional data samples.Therefore, the first part briefly introduces the background and significance of the topic, the research results at home and abroad, ideas, content, research method, innovation and deficiency of the paper. The second part,we get the forecast errors by a good ARIMA population time series model, make the forecast errors as the dependent variable, the possible factors affecting the accuracy of prediction as independent variables, and analysis the factors which can influence the prediction accuracy. The third part we firstly forecast the Chinese representative provincial population by multiple ARIMA model,considering the factors affecting the accuracy of the second part, discuss the general law of population time series forecasting model selection under the condition of different models, different area, base area and prediction interval.The fourth part is the conclusion of the first two parts. The last is the main references and postscript. The results show some ARIMA models can provide relatively accurate results while others can not; Linear model and nonlinear models have different population prediction precision, and different length of history data could also lead to different models; The precision of the model from the different angles observation has strong consistency, and some models have not. Finally, this paper points out the direction of future research and suggestion.The following is the innovations of this paper. Firstly, we use the provincial data analysis rather than aggregate data, then we can get the more accurate results. Secondly, we can summarize the factors affecting the accuracy of population prediction by a population time series model which founded on the the province data, and then we put these factors into the regression model about the prediction accuracy. We know provinces and history length have obviously impact on the prediction accuracy, but this point rarely was mentioned in the previous literature. We not only selected the variable prototype, but also selected the variables’ square in the regression model. The goodness of fit of the model is improved obviously. When we conclude the ARIMA model evaluation criteria, We use evaluation of the effectiveness of prediction criteria by the different base interval and different prediction interval, then we select the timing model from different points of view, this is relatively rare in the domestic research. Compared to the previous studies, data selection, variable selection and model selection are my innovations in this paper. The shortfall is whether the variables and the variables’forms are perfectly representative although adding some related variables? In the choice of models from multiple perspectives, whether these results can be adapted other provinces? Is there other model is more suitable for the prediction of the provincial population? Whether different evaluation criteria will get different answers? Whether we can select population forecast model from more angles? Which point of view for the selection model is more important than the other point of view? How to distribute the weight point of view? We hope these problems can be resolved in future research.