Empirical Study On The Forecasting Of Chinese Housing Price Based On Model Averaging

In recent years,China's real estate market has continued to develop and housing price has been continuously grown,housing price has become an important indicator of the health and stability of the real estate market.The demand for housing price forecasting is increasing,and the requirement is getting higher and higher.The issue of housing price has become the focus of the people in China,a number of studies on the issue of housing price has been published in Chinese and also in international journals.There are many existing forecasting methods on housing price,the most which are based on multiple linear regression and time series analysis.These methods are typically based on a single selected model.The model selection method selects an appropriate model and the model selection uncertainty is imposed into post-model-selection forecasting.As a natural extension of model selection,model averaging method is deemed to be more accurate in prediction.Therefore,this thesis will employ the model averaging method to forecast China's housing price dataFirst,this thesis analyzes China's real estate market,describes the research background,research significance and purpose.Review the literature,summarizing previous research results and the corresponding research method.Then,I introduce the theoretical knowledge involved,namely the spatial econometrics,model selection method and model averaging method.In the empirical analysis part of this thesis,I consider the average sales prices of commodity houses and 7 macroeconomic factors of China's 31 provinces and cities from 2003 to 2018 for analysis.In specific,the Moran index was used to test the spatial autocorrelation.The test results show that there is a spatial autocorrelation between the average sales price of commercial housing in 31 provinces and cities in China.Based on the theory of spatial autoregressive model,model selection and model averaging,I compose the prediction procedure based on model averaging and model selection about different criterions The original data are partitioned into training set and test set according to three scenarios(1:3,1:land 3:1).The mean square error(MSE)is employed to assess the performance of different methods.The results show that the model averaging method dominates the usually used model selection method in terms of MSE,and the MSE of extended Mallows CP criterion is smaller compared with the other two criterions,this finding further support the use of model averaging method in practical situations.