基于动态模型平均的中国大中城市房价预测

发布时间：2018-01-11 08:18

本文关键词：基于动态模型平均的中国大中城市房价预测　出处：《西南交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：近二十年来,我国房地产市场经历了较长时期的蓬勃发展,但同时也遭遇了若干次严厉的调控,大中城市的房价出现了一些较大幅度的波动,房价成为媒体、人民群众和政府关注的焦点。因此,如何对未来的房价走势进行科学和有效的预判,也是众多房地产经济学者和业界普遍关心的重要问题。本文率先引入了动态模型平均(DMA)方法及其特例-动态模型选择(DMS),对于全国三十个省会城市和直辖市的房价进行了预测分析。相对于传统模型,DMA方法允许模型变量设置和变量系数的时变性,充分考虑了不同变量、不同时间对于房价影响的大小。同时本文使用等权重平均、自回归、贝恩斯平均、贝恩斯选择以及信息理论平均等多种模型作为对比,充分讨论房价预测的表现。本文不仅使用了传统研究中大量使用的扩展窗口预测模式,同时添加滚动窗口模式作为参照对比,既解决了时间序列中可能存在的结构突变问题,同时也在多种预测模式之下,全面稳健地对于房价进行预测。此外,在使用传统宏观经济变量作为预测变量时,本文也考虑了大数据环境下,互联网搜索指数包含更多需求信息,对于房价的预测会产生新的帮助作用。随后,区别于其他预测研究只采用简单统计指标评价预测表现,本文采用更加高级的模型信度设定方法(MCS),进一步避免了一类统计错误,并且强调在不同标准和统计指标下,多角度全方面检验房价预测模型的精度。实证结果显示,无论是扩展窗口,还是滚动窗口,DMA方法在全国三十个大中城市,在样本内估计精度较高的基础上,样本外预测方面也能够有效地降低全国各个大中城市的房价预测误差,比传统自回归等方法缩小50%以上。此外,本文也发现DMA能够有效筛选变量,降低计算负荷,并且发现搜索指数对于房价影响在近些年逐渐增大,传统变量预测作用式微,表现不及预期。本文尝试提出了需求端和政策不确定性两方面的合理解释。最后,基于稳健性的分析证明,预测三、六期及延长样本外预测区间,与前续结果均一致性地均支持DMA方法的优越预测表现。最后,DMA方法为房价预测提供了新的思路,给予购房者、业界以及政府管理部门更好的房价决策和预判。
[Abstract]:In the past two decades, the real estate market of our country has experienced a long period of vigorous development, but at the same time, it has also encountered a number of strict regulation and control. The housing prices in large and medium-sized cities have experienced some relatively large fluctuations, and the housing prices have become the media. The focus of attention of the people and the government. Therefore, how to predict the future trend of housing prices scientifically and effectively. This paper first introduces the dynamic model averaging (DMA) method and its special case-dynamic model selection (DMS). The housing prices of 30 provincial capitals and municipalities in China are predicted and analyzed. Different variables are fully considered compared with the traditional model / DMA method, which allows the model variables to be set and variable coefficients to be time-varying. At the same time, this paper uses the equal-weight average, autoregressive, Baines average, Baines selection and information theory average as a comparison. This paper not only uses the extended window prediction model, which is widely used in traditional research, but also adds the rolling window model as a reference comparison. It not only solves the problem of structural mutation in time series, but also makes a comprehensive and robust prediction of house prices under various forecasting models. In addition, when using traditional macroeconomic variables as forecasting variables. This article also considers that in big data environment, the Internet search index contains more information on demand, which will help the forecast of house prices. Different from other prediction studies only using simple statistical indicators to evaluate the performance of the prediction, this paper uses a more advanced model reliability setting method to further avoid a class of statistical errors. And emphasizes that under different standards and statistical indicators, multi-angle and all-sided test of the accuracy of the housing price forecasting model. Empirical results show that, whether extended window or rolling window. DMA method can effectively reduce the error of house price prediction in 30 large and medium-sized cities in the whole country on the basis of high estimation accuracy in the sample. In addition, this paper also found that DMA can effectively screen variables, reduce the computational load, and find that the impact of search index on house prices has gradually increased in recent years. The traditional variable forecasting function is declining and the performance is not as expected. This paper tries to put forward two reasonable explanations of demand side and policy uncertainty. Finally, based on robust analysis, forecast three. Six periods and extended prediction interval outside the samples are consistent with the previous results to support the superior performance of the DMA method. Finally, the DMA method provides a new way of thinking for house price prediction, giving home buyers. Industry as well as government management better house price decision making and forecast.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F299.23

【参考文献】