基于潜在因子响应结构驱动的中国城市二手房价异质性机制研究

Heterogeneous Mechanisms of Second-Hand Housing Prices in Chinese Cities: A Study Driven by Latent Factor Response Structures

摘要: 中国城市房价的异质性波动是宏观政策精准调控面临的核心挑战。本文构建了一个“因子提取–结构分析–驱动聚类”的三阶段分析框架，旨在从城市对共性驱动力的差异化响应模式出发，识别由市场内在传导机制定义的城市类型。首先，运用高维时间序列潜在因子模型，从2015~2023年中国70个大中城市二手房价格收益率中提取三个公共因子：全国金融周期因子、区域发展因子和政策情绪因子。其次，基于先验分类的载荷结构分析表明，城市层级、地理区域和人口规模均能有效区分城市对金融周期的敏感度梯度，但对区域发展和政策情绪因子的捕捉存在局限。最后，以因子载荷向量为特征进行无监督聚类，识别出三类由数据内生驱动的城市集群：金融周期敏感型(11城)、政策情绪敏感型(20城)和增长依赖型(39城)。研究发现，传统分类与数据结构分类存在“同形异质”与“异形同质”的双重关系——层级相同的城市可能分属不同集群，而层级不同的城市可能呈现相似的驱动逻辑。本文的方法论贡献在于将因子模型从降维工具拓展为结构发现引擎，实现了从“外部标签”到“内生特征”的范式跃升；实践意义在于为“因城施策”提供了直指传导机制根源的分类依据。

Abstract: The heterogeneous fluctuations of urban housing prices in China pose a fundamental challenge to precise macro-policy regulation. This paper develops a three-stage analytical framework—“factor extraction, structural analysis, and driver-based clustering”—to identify city types endogenously defined by market transmission mechanisms based on their differentiated response patterns to common drivers. First, employing a high-dimensional time series latent factor model, we extract three common factors from second-hand housing price return series of 70 major Chinese cities from 2015 to 2023: a national macro-cycle factor, a regional development factor, and a policy sentiment factor. Second, prior-classification-based loading structure analysis reveals that administrative hierarchy, geographic region, and population size effectively differentiate cities’ sensitivity gradients to the macro-cycle factor but exhibit limitations in capturing responses to regional development and policy sentiment factors. Finally, unsupervised clustering based on factor loading vectors identifies three data-endogenously driven city clusters: financial-cycle-sensitive (11 cities), policy-sentiment-sensitive (20 cities), and growth-dependent (39 cities). The findings reveal dual relationships between traditional and data-driven classifications: “same form, different nature”—cities with identical administrative tiers may belong to different clusters; and “different form, same nature”—cities across tiers may exhibit similar driving logics. The methodological contribution lies in extending the factor model from a dimension-reduction tool to a structure-discovery engine, achieving a paradigm shift from “external labels” to “endogenous characteristics.” The practical implication provides classification foundations for targeted urban housing policies directly linked to transmission mechanisms.