A NOVEL SPATIAL MIXED FREQUENCY FORECASTING MODEL WITH APPLICATION TO CHINESE REGIONAL GDP

Direct use of economic indicators for different frequencies is important to improve the regional forecast performance, and is the quantitative basis for improving the awareness on regional economic cycle changes, growth drivers and regional differences Considering the spatial mixed frequency data using a high frequency variable to predict a low frequency one in regional prediction problems, this paper proposes a novel spatial mixed frequency forecasting model. Firstly, it analyzes the commonly used spatial forecasting models and the most classical MIDAS. Secondly, it adopts the soft spatial weights to describe the spatial correlation of economic variables to amend the polynomial weighting method of MIDAS. Thirdly, it analyzes the main characteristics and puts forward the prediction error or precision index to test the validity of the model. Finally, it applies the new method to forecast the real GDP growth rate in 30 provinces and autonomous regions in China and compares the different weightings, which show a great feasibility. Futher, it discusses and gets some help findings about the characteristics of parameters such as weighs polynomials, prediction weighs, and lag period in different regions. Finally, it implements the Diebold Mariano tests for RMSEs between different model settings and obtains meaningful conclusions.