中国省际人口迁移短期预测分析

人口迁移是一个时空路径依赖过程,同时受迁移存量和周边迁移状况影响。当前人口迁移预测大多建立在时间序列模型之上,重点考虑迁移流在时间维度上的联系,忽视了其中的时空关联。该文将特征向量时空滤波方法与普通泊松模型相结合,考虑迁移流中可能存在的时空滞后和同期两种结构,对1985-2015年不同时段的中国省际人口迁移流数据进行建模和估计,并利用拟合程度较优的模型预测2015-2025年省际人口迁移的发展趋势。结果表明:1)特征向量时空滞后和同期滤波泊松模型均能较好地模拟研究时段省际人口迁移过程,自1985年以来我国省际人口迁移流不仅受迁出地和迁入地经济、社会等因素影响,也与过去迁移存量及周边迁移流密切相关;2)区域人口规模和GDP对迁移流的“推—拉”作用符合预期,地区人口规模较高和经济发展水平较低会促进人口外迁,反之则有利于吸引外来人口;3)与特征向量时空滞后滤波泊松模型相比,时空同期模型更便于捕捉省际人口迁移过程中的时空路径依赖特性,意味着当前人口迁移流的发展更易受到同时期周边迁移流的影响,表现出明显的羊群效应;4)预计2015-2025年我国省际迁移总量持续增加,呈现更集聚的空间模式,高迁入与高迁出区域在空间上相连,形成一条南北贯通的“高密度迁移地带”。将特征向量时空滤波模型拓展到人口迁移这一空间相互作用领域,可为当前构建更加完善的要素市场化配置体制机制等提供科学参考。

Short-Term Forecast of Interprovincial Migration Flows in China

Population migration is a space-time path-dependence process,which is influenced by both migration stock and surrounding migration status.Most studies on the population migration forecast based on time series models have focused on the connection of migration flows in temporal dimension,thereby ignoring the spatiotemporal autocorrelation among migrations.By combining the eigenvector spatiotemporal filtering method with the ordinary Poisson model and considering two possible modes of spatiotemporal lag and spatiotemporal synchronization,we estimated the interprovincial migration flows in China in different periods from 1985 to 2015,and extrapolated the development trend of interprovincial migration flows in 2015-2025.The preliminary results are as follows.1)The eigenvector spatiotemporal lag and synchronization filtering Poisson models can simulate the process of interprovincial migration in the study period.The interprovincial migration flows in China since 1985 are not only affected by the socioeconomic factors of the origin and destination places,but also are closely related to the past migration stock and surrounding migration flows.2)The push and pull effects of regional population size and GDP on migration flows are in line with our expectation.Higher population size and lower level of economic development will promote migration flows;on the contrary,it will be attractive for immigration.3)Compared with the spatiotemporal lag filtering Poisson model,the spatiotemporal synchronization filtering Poisson model can better capture the space-time path dependence during the process of interprovincial migration.That is,the development of current migration flows is more affected by the surrounding migration flows,it shows obvious herd effects.4)The total interprovincial migration would continue to increase in 2015-2025,which shows a more concentrated spatial pattern.The total amount of interprovincial migration flows during the periods of 2015-2020 and 2020-2025 might be 67.76 million and 76.61 million on average,being more active than the past decades.Furthermore,Chinese interprovincial migration flows would form a high-density belt along Beijing and Guangdong.Extending the eigenvector spatiotemporal filtering model to the field of population migration can provide a scientific reference for constructing a more complete market-oriented allocation system.