Real Estate Price Index Prediction Based On Internet Search Data

The real estate industry occupies a pivotal position in my country's national economy,among which housing prices are the most important issue that people are most concerned about.Traditional statistical data release has a certain lag,which makes it difficult to make timely and accurate forecasts of real estate price indices.In the Internet era,we are actively or passively receiving various information from the Internet every day.This information includes consumers' purchase intentions and subjective perceptions of real estate development.How to make full use of this information to improve the accuracy of real estate price prediction is a problem that scholars have been exploring.Adding online search information to the real estate price index forecast can overcome the subjectivity of variable selection to a certain extent,make the influencing factors more comprehensive and objective,and the obtained data will not cause errors due to human reasons,and improve the forecasting effect.Therefore,this paper chooses the real estate price index as the research object,taking Shenzhen as the representative,and using Baidu index to make predictions.For the selection of keywords in the Baidu index,this paper mainly refers to the existing literature,determines the initial core keywords according to the main influencing factors of real estate prices,and then forms a keyword library through keyword expansion method.Finally,the Spearman correlation analysis and time difference correlation analysis are used to screen the keywords.Because the selected keywords still have strong correlation,three comprehensive indexes were synthesized by using principal component method.In order to incorporate Baidu index and real estate price index with different data frequencies into the model for regression analysis,univariate and multivariate mixed-frequency data sampling models were established respectively in this paper,and then BP neural network method was used to correct the results of the optimal mixing model,and finally the combined prediction model was constructed.Based on the above research,this article mainly draws three conclusions:(1)People's search behavior is indeed related to the fluctuation of real estate prices,but the two do not change simultaneously.(2)Regardless of whether it is a single-variable or multi-variable mixing model,the prediction accuracy is better than that of the same frequency model,and the combined model modified by the BP neural network model has a higher prediction accuracy.(3)The combined prediction model can realize real-time prediction outside the sample,and the result is very close to the real value.