网络新闻、需求者关注与房价——基于时变参数向量自回归模型的研究

文章采用基于Linear-SVM的机器学习方法对百度新闻栏目57988篇的房地产市场新闻进行文本分析,构建出网络媒体净看多情绪变量和网络媒体关注度变量。并采用时变参数向量自回归模型(TVP-VAR)对媒体变量、住宅销售面积和房价之间的互动关系展开分析,得出如下结论:网媒房地产新闻的多空情绪变化基本能反映调控政策的阶段性特征。媒体报道对房地产市场冲击的响应相对稳定,时变效果不明显。媒体关注度对房地产市场的传导作用受到其构成成分和所处市场阶段的影响,其对房价的反向作用在紧缩调控放开后逐渐减小,并逐渐转为正向的提升作用,对住宅销售面积的反向作用在2015年后的市场高涨时期逐渐缩小,持续时间也在缩短。媒体情绪对房地产市场的传导作用受到调控政策和市场运行状况的叠加影响,其对房价的提升作用在2015年至2017年的房地产市场上行周期中有所扩大,对销售面积的提升作用则较为平稳。据此,文章提出应根据市场所处的阶段关注网媒发布信息的频度和情绪,加强舆情监测等建议。

Internet News,Demanders' Attention,and Housing prices: A Study based on TVP-VAR Model

A relatively stable market expectation helps suppressing price fluctuations and ensuring smooth operation of market because of the ‘self-fulfilling’ function of expectation. It is also an important aspect of macro-control policy to implement effective expectation management of real estate market. As real estate market is a typical market with asymmetric and incomplete information, in order to make investment decisions, participants especially demanders need to obtain as much information as they can through various sources which mass media may be one of the most important channels. Nowadays, Internet media has become the main source people would obtain information from. Logically speaking, Internet news will affect demanders' mind and behavior as well as market performance. This paper aims to figure out the impact of Internet news on market performance through demanders' attention, using web crawler technology to collect real estate market news from Internet and constructing media sentiment variable based on machine learning process. We obtain 57988 real estate news from news column of Baidu (Baidu.com), one of the biggest Chinese search engines, covering the period from January 1st, 2010 to March 31st, 2018. Then we study the relationship between news, residential sales area and housing price employing the time-varying parameter vector autoregressive model (TVP-VAR model) based on periodic fluctuation characteristic of real estate market. TVP-VAR model can capture dynamic impulse responses among different variables considering time-varying characteristics of coefficients and error variances. The empirical results show that, real estate news reproduced online can reflect the phased characteristics of regulation policies. The response of the media to the shock of the real estate market is relatively stable. The negative effect of media attention on housing price gradually decreased after the lift of tightening regulation and gradually turned to be positive. The negative effect of media attention on residential sales area was gradually reduced in the post-2015 market upsurge period, and the duration was also shortened. The effects of media attention and media sentiment exerting on the real estate market are influenced by the regulation policies and the market operation conditions, which is time-varying. The effect of media net bullish sentiment on housing prices increased during the property market's upward cycle from 2015 to 2017. The positive effect of media net bullish sentiment on sales is relatively stable. The empirical results confirm that there are some connections between internet media variables and the housing market. This paper contributes to the literature in several ways. Firstly, this paper innovatively constructs media sentiment index by using web crawling massive real estate news and employing machine learning process to analyze textual sentiment based on Linear-SVM algorithm. Secondly, the paper studies the interactive time-varying relationship between internet media, residential sales area and housing price. Thirdly, the paper explores for the first time the impact of internet news sentiment on housing market performance. Fourthly, this paper can provide theoretical foundation and practice guidance for improving macro-control of the real estate market using internet media.