Regional eco-efficiency prediction with Support Vector Spatial Dynamic MIDAS

To obtain good eco-efficiency prediction with factors accompanied by spatial relationship, mixed frequency data and nonlinearity, based on the existing spatial panel data forecasting models and MIxed DAta Sampling (MIDAS), we established Support Vector Spatial Dynamic MIDAS to incorporate the spatial interaction, different frequencies of sampling data, and non-linear relationship between the eco-efficiency and various factors. Further to testify the effectiveness, we applied the new model to regional eco-efficiency prediction in China. Prediction Error of the Last Year, Mean Percentage Error, Mean Square of Prediction Error and Standard Deviation of Prediction Error were utilized to measure prediction accuracy. Results showed SVSD-MIDAS effectively considered the mixed frequency factors--Financial Development Level, Foreign Direct Investment, Urbanization Level, Price Index, Fixed Asset Investment and their spatial interaction. Prediction performances of 30 regions are very good, with low prediction error below 1% or smaller. And regional prediction characteristics in the eastern, central, western and northeast regions were compared. The different spatial weights impacted the prediction no matter in individual province or the whole 4 areas. Accurate prediction by SVSD-MIDAS can save costs of collecting and calculating indicators, and guide the formulation of regional sustainable development strategies of residents, business managers, government departments in advance.