Forecasting Chinese Interprovincial Migration Flows By Spatio-temporal Filtering Models

Recently,Chinese interprovincial migration flows have been increasingly active,attracting a widespread attention of academia.By the means of building different models,most scholars have attempted to grasp the interprovincial migration patterns and spatio-temporal characteristics,and explain the economic,social,cultural and other factors that affect the migration flows,with a view to forecast the future migration flows scientifically.The interprovincial migration flows not only depend on current social and economic developments of origins and destinations,but also relate to migration flows at different periods over space.However,without the combination of the temporal and spatial dimensions,most of the current studies ignored the temporal dependence or spatial dependence.In addition,most work focused more on modeling the impacts and mechanism of migration flows,but less on the forecasting and the development of migration flows for the future.Thereby,based on five periods’ inter-provincial migration flows(1985-1990,1990-1995,1995-2000,2000-2005,2005-2010)of China,this paper,firstly analyze the temporal and spatial characteristics and autocorrelation effects of Chinese interprovincial migration flows from temporal,spatial and spatio-temporal perspectives.Secondly,we construct the temporal autoregressive models for interprovincial migration flows of China,and establish eigenvector spatial filtering and spatio-temporal filtering models based on eigenvector filtering methods,then make a comparison of the above models.Finally,we forecast the short-range and long-range interprovincial migration flows through the spatio-temporal filtering models,and evaluate the forecast results in comparison with those results of temporal autoregressive models.The results are as follows.(1)In the temporal dimension,there are significant temporal autocorrelation effects in Chinese interprovincial migration flows during the five periods.In the spatial dimension,Chinese interprovincial migration flows show significant positive network autocorrelation effects,which have been enhanced over time,and the spatial focusing of migration flows is more focused.In spatio-temporal dimension,there are significant positive spatio-temporal autocorrelation effects under both the contemporaneous and lagged dependent structure,and the autocorrelation effects of the former are much larger than those of the latter.(2)The temporal autoregressive models can explain the interprovincial migration flows,and the first-order lagged model has the largest effects.The eigenvector spatial filtering method can further improve the fitting capacity of the models,which shows that eigenvector can capture the network autocorrelation among migration flows.The eigenvector spatio-temporal filtering models have the best model fitting,which indicate that spatio-temporal autocorrelation should be accounted for in modeling multi-temporal migration flows.And spatio-temporal contemporaneous filtering model is better than lagged filtering model.(3)Spatio-temporal filtering model can contribute to both short-range and long-range forecasting.The short-range forecasting results show that interprovincial migration flows are more active in 2010-2015,and the migration distribution patterns will do not change a lot.The long-range forecasting results show that the overall tendency of inter-provincial migration flows will change slowly during 2010-2035,which take on a trend of first rising and then falling.Furthermore,in the future,Chinese interprovincial migration flows will form a high density migration belt across the space.