A Study On Influence Factors And Short Forecast Of Real Estate Price Based On EMD

For nearly 30 years,China real estate industry has experienced several reforms.Involving diverse industries and areas with high relevance,the real estate industry has became a pillar of China's economy.Because of high elasticity of income and demand,the commodity residential house gradually has became hot spots in current consumption.According to the change from macro environment,Governments try to stimulate consumption and investment,which influence the price of real estate,at the same time,the governments also need to predict the future direction in advance based on the price fluctuations of real estate market.This paper combined EMD(Empirical Method Decomposition)with AVM(Support Vector Machine),which are the mature methods in the world,and use for cause and forecast analysis of the price of the real estate market.EMD method,a new adaptive time series method in signal time-frequency analysis,in terms of the characteristics of data itself,extract corresponding intrinsic mode function(IMF),which has the characteristics of different scales step by step according to decomposition,and reflect the physical characteristics of the original time series accurately.At present,the domestic and foreign scholars have applied EMD methods to forecast in economics.Shouyang Wang(2008),they applied neural network learning paradigm model to forecast crude oil price based on EMD.According to list the traditional econometric model such as linear regression,cointegration analysis,GARCH model,random walk,VAR and ECM model,which have been widely applied to predict the price of crude oil.The conclusion shows that the neural network model based on the EMD has more advantages and higher precision than simple ARIMA model.It is an important aspect for the modern intelligent technology and research from observed data to search patterns,and use these data to forecast based on the data of machine learning.The actual time series is not a infinite simple,and it has some disadvantages such as less number of samples,nonlinear and high characteristic dimension.Wapnik and Corinna ms Cortes(1995)proposed a theory which named Support Vector Machine(Support Vector Machine,SVM)based on statistics,which solve the problems from traditional methods,linear inseparable,excessive fitting,dimension disaster and local minimum point,etc.Support Vector Regression(SVR)model is a theory in which SVM apply to Regression.The SVR is a linear learning machine,using linear function for nonlinear regression in regression.It transform data x from a low-dimensional feature space to a high-dimensional feature space according to nonlinear mapping function ?,and then dealt with linear regression in high-dimensional feature space.It adopts the structural risk minimization criterion and provides the global optimal solution for machine learning problem in small sample.It utilize machine learning to predict the future for the nature of the support vector regression(SVR).And the outcome of machine learning(i.e.,the determination of the optimal parameters)is obtained by training sample set for countless times.Therefore,the algorithm selection directly determine the prediction of machine learning.In the field of intelligent Algorithm,Genetic Algorithm(GA)and Particle Swarm Optimization(PSO)is applied more widely.The EMD-SVR-SVR model is a multi-scale prediction model,which is first proposed by Zhang Xun(2008)for the prediction of crude oil,and concluded that the prediction of multi-scale decomposition model is superior to single scale model.This paper first proposed a method combined PSO algorithm with EMD-SVR-SVR based on EMD,and apply this method to real estate market in our country,taking commodity housing prices in Shenzhen as an example.Meanwhile,this paper compare the prediction of reconstruction with the prediction of component sequences based on EMD according to SVR integration.The main content in this paper as follows.(1)By the EMD decomposition of the original time series we can compose of the parts of excessive decomposition for the prices of commodity residential house,and get three parts which have real economic significance.In conclusion,the high frequency series is more closely related to the short-term policies by government control,the money supply(M2),one of which has the highest correlation with the prices of real estate prices;Financial crisis has a significant effects on low frequency series,and the cycle is about 6?7 years;Long-term trend item is consistent with the equilibrium price,which determined by market supply and demand.(2)In the parameter optimization algorithm,This paper first proposed a method combined PSO algorithm with EMD-SVR-SVR based on EMD,and apply this method to real estate market in our country,taking commodity housing prices in Shenzhen as an example.Meanwhile,this paper compare the prediction of reconstruction with the prediction of component sequences based on EMD according to SVR integration.First of all,this paper try to analyse prediction from two dimension,crosswise and lengthwise.From the horizontal point of view,this paper analyse the prediction of the reconstruction for the original series before and after.From the longitudinal point of view,this paper analyse the comparison among algorithms in the same component.The conclusion shows that the prediction of integration series is significantly better than each component,which makes the mean square error(MSE)reduced half,goodness-of-fit raised to 0.97;PSO algorithm is more stable and suitable than GS and GA algorithm.Finally,the integration of original series based on PSO algorithm has best precision than the other,which makes MSE reduced to 563.0599,MAPE reduced to 0.0079.