我国省域人口时序预测模型的选择研究

发布时间：2018-05-10 19:30

本文选题：人口预测 + 预测区间　；参考：《东北财经大学》2013年硕士论文

【摘要】：人口预测作为区域规划和政策决策的依据,对于区域经济社会可持续发展有重要的理论和现实意义。从宏观上说,人口预测可提供今后几十年乃至上百年内全国各年龄段的儿童和青少年数量,这对于劳动就业和教育规划是至关重要的；从微观上讲,预测某一地区的人口,则会为区域的基础设施建设、资金投入提供帮助。例如：合并学校、增加电厂、重新设计城市规划、房地产的开发等。然而到目前为止,虽已有不少学者使用时序模型进行了人口预测,但以区域数据作为分析样本,从历史区间长度、预测的临界年和预测区间角度选择最优模型的研究几乎没有。本文第一部分简要介绍了选择此论题的意义与背景、国内外的研究成果、研究的思路与内容、研究方法、和本文的创新与不足之处；第二部分利用一个优良的ARIMA人口时序模型,得出预测误差作为因变量,把有可能影响预测精度的因素作为自变量,做定量的回归研究,并对影响预测精度的因素进行分析；第三部分首先利用多个ARIMA模型对我国部分具有代表性的省域人口进行预测,然后考虑第二部分得出的影响精度的因素,探讨了在不同性质的模型、不同地区、基区和预测区间等条件下人口的时序预测模型选择的一般性规律；第四部分是对前两部分论述的总结,最后附上主要的参考文献以及后记。研究结果发现,人口数目和增长率对人口精度的影响皆呈U型；临界年和地区因素对人口精度也有较大影响,以及对于预测精度来说,各因素的相对影响权重。在模型选择过程中,一些ARIMA模型能够提供相对精确的结果,而另一些则不能；线性模型和非线性模型在省域人口预测精度方面具有较大的差异；历史数据长度不同也可能导致选择不同的模型；从不同角度观测的模型精度有较强的一致性,但也存在一定程度的不一致性,以上结论可以为选择人口时序预测模型提供更深入的参考,并为以后的研究指明方向与建议。本文的创新之处在于,首先没有局限在人口总量数据上进行分析,而是进行了分省份人口数据的建模与分析,从而使分析结果更加细致准确。其次,第二部分在省份数据的基础上通过一个人口时序模型总结出影响人口预测精度的因素,然后把这些因素加入到关于预测精度的回归模型中,得出各因素影响的权重,省份、历史长度,这些因素对预测精度的影响还是比较明显的,这在以往的文献中很少提到。在做变量的回归模型中,不但选取了一次变量,而且选取了变量的二次形式,使模型的拟合优度得到明显提高。再次,在第三部分进行ARIMA模型的评价标准上,运用了从不同基区间,不同预测区间的预测有效性评价标准,然后从不同角度选择时序模型,这点在国内研究中也比较少见。以上从数据的选取,变量的筛选,模型的选择三方面相比于以往的研究都有很大的创新。本文的不足之处在于虽然加入了一些有关的变量因素,但是变量以及变量的形式选取是否具有完全代表性?在从多角度选择模型时,本文得出的结果对于其他省份是否适用?有没有其他的模型更加适合于省域人口的预测?这些问题希望可以在以后的研究中得到解决。
[Abstract]:As the basis of regional planning and policy decision, population forecasting is of great theoretical and practical significance to the sustainable development of regional economy and society. From the macro point of view, population prediction can provide the number of children and adolescents in all ages of the country in the next hundred years, which is vital to employment and education planning. On the microcosmic point of view, the population of a region will be predicted for the infrastructure construction of the region, and the funds will be provided to help. For example, combining schools, increasing power plants, redesigning urban planning, developing real estate, and so on. There is little research on the selection of optimal models from the perspective of the historical interval length, the critical year and the forecast interval.
The first part of this paper briefly introduces the significance and background of choosing the topic, the research results at home and abroad, the thinking and content of the research, the research method, and the innovation and inadequacies of this paper. The second part makes use of a good ARIMA population time series model to get the prediction error as the dependent variable and make the factors that may affect the prediction accuracy. The quantitative regression study is made for the independent variables, and the factors affecting the prediction accuracy are analyzed. The third part first makes use of multiple ARIMA models to predict some of the representative provincial population in China, and then takes into account the factors of the influence accuracy of the second parts, and discusses the different properties of the models, different regions, base areas and prepositions. The general rule of selecting the time series prediction model of population under the condition of measuring interval and so on; the fourth part is the summary of the first two parts, and finally the main references and the postscript. The results show that the population and the growth rate have the U type on the population precision, and the annual and regional factors are also larger for the population precision. In the process of model selection, some ARIMA models can provide relatively accurate results, while others can not. The linear and nonlinear models have great differences in the precision of population prediction in the province, and the difference in the length of historical data may also lead to selection. Different models are chosen; the accuracy of the models observed from different angles has a strong consistency, but there is a certain degree of inconsistency. The above conclusions can provide a more in-depth reference for the selection of the population timing prediction model, and point out the direction and suggestions for the future research.
The innovation of this paper is that, firstly, the analysis of population data is not limited, but the model and analysis of the population data in the provinces are carried out to make the analysis result more detailed and accurate. Secondly, the second part summarizes the factors that affect the accuracy of population prediction on the basis of a personal order model on the basis of the province data. Then, these factors are added to the regression model of prediction accuracy, and the weight of the influence of each factor, the province, the history length, and the influence of these factors on the prediction accuracy are still obvious, which is rarely mentioned in the previous literature. In the regression model of variable variables, not only a variable is selected, but also the two of the variable is selected. In the second form, the goodness of fit of the model is obviously improved. Thirdly, in the third part of the evaluation standard of the ARIMA model, the evaluation criteria of the prediction validity are used from different intervals, different prediction intervals, and then the time series model is selected from different angles. This is also rare in the domestic research. There are great innovations in the selection of the model and the selection of the model in three aspects. The deficiency of this paper is whether the selection of variables and the form of variables is completely representative, although some variables are added, and whether the results obtained in this paper are applicable to other provinces when choosing the model from multiple angles. Is there any other model that is more suitable for the prediction of provincial population? These problems can be solved in future research.

【学位授予单位】：东北财经大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：C81

【参考文献】