The Study And Application Of China’s Regional House Price Forecasting Model Based On Online Search Data

As the influence of Internet on daily life deepening, many people search online for useful information before they make big economic decisions. So online search data which record search behaviors of these potential consumers also reflect their intentions. Based on this theoretical hypothesis, careful study of online search data could help researchers obtain these economic intentions before they actually turn into reality, thus helpingmake useful predictions for prices of certain commodities. Google Trends was the first online search data service, which was launched by Google company in 2004, and since 2009, foreign researchershave started using this online search data to predictmany social and economic indicators. As the leading Internet search company in China, Baidu also launched a similar data service called Baidu Index in 2011, showing the relative search frequency for certain key words. Therefore from 2013, domestic scholars also began using this online search data for predictive study in several fields. However, compared to the researchabroad, domestic study in this area still has a long way to go, from the perspective of both quantity and quality.The purpose of this dissertationis to construct house price forecasting models using online search data for 6 cities (Beijing, Shanghai, Guangzhou, Shenzhen, Nanjing and Hangzhou) and assess their predictive power. This dissertation gave an overall review of relevantresearch work on online search data application and house price prediction first. Then the author explored a set of procedure and methods for choosing relevant key words using the method of time difference correlation analysis and developing anonline search composite index using principal component analysis. Finally the author built house price forecasting models based on online search data and then assessed their predictive power compared to basic house price forecasting models, using the method of K fold cross validation. According to the result of study, the online search composite index correlated well with the house price index and could also make a good prediction for house price. However, the predicting improvement of house price forecasting models based on online search data was quite limited, compared to basic models. In the end of this dissertation, the author gave several possible explanations for this limited improvement of predictive power, and then put forwardcorrespondingsuggestions for future related study, hoping to further explore and make full use of the forecasting ability of online search data.