Forecasting China Housing Prices Using Dynamic Model Averaging Approach

Over the past two decades,China's real estate market has experienced a long period of vigorous development,but also suffered numerous severe recession.Housing prices of large and medium-sized cities also experienced great volatility.Therefore,how to predict the future price in the boom and down period is important.Many real estate economists and the industry generally concern about this issue for its importance.This thesis introduces the Dynamic Model Average(DMA)methodology and its special case:Dynamic Model Selection(DMS),and makes the forecast analysis of the housing prices of 30 cities.Compared with the traditional model,the DMA method allows the time varying of the variable coefficients and variables set itself.In the case of this thesis,DMA considers the impact of different variables on the price of housing prices at different time.At the same time,this thesis uses the Equal Weight model,Autoregressive model,Bayesian Model Average,Bayesian Model Selection and Information Theory Model Average to fully discuss the performance of house price forecast.This thesis not only applies the recursive window to forecast,but also adds the rolling window for more benchmark mode.The rolling window can solve the problem of structural mutation that may exist in the time series.In addition,when using traditional macroeconomic variables as predictors,I also consider the Internet search index that contains more demand side information that provide new insight in forecasting housing prices.In this thesis,I also implement the more advanced model reliability setting method(MCS)to avoid Type I error.MCS test can check the accuracy of each forecast model under different loss functions and statistical indicators.The empirical results show that under the recursive window and rolling window,the DMA method can both effectively reduce the forecasting error in 30 cities on the basis of the good in-sample estimation.In terms of the out-sample forecast error,DMA can reduce at least 40%than the traditional autoregressive model.In the advantages of DMA,I find that DMA can effectively filter variables and reduce the computation.From the inclusion possibility aspect,I observe that the possibility of Internet search index on house prices in recent years gradually increases,the traditional macro variables' possibility declines.So,I attempt to propose two reasons,the demand side and uncertainty of policy to explain it.Finally,based on the robustness analysis,it is proved that the superior prediction performance of DMA method in other cases is consistent with my empirical results.This methodology provides a new solution for house price forecasting that afford better decision-making and pre-judgment to home buyers,real estate industry and government bureau.