中国房地产下行周期已进入第四个年头，但目前尚无明确的触底回升迹象。由于房价企稳对于消费者信心恢复和提振市场情绪来说至关重要，所以高盛利用国家统计局发布的70城二手房价格数据，将城市分成四个不同的集群来进行聚类分析，其目的是为了证实各城市房地产市场间的差异是否显著，进而推出哪一类城市的房价会率先完成触底。

集群1主要由一线城市和强二线城市组成。

集群2主要由普通二线城市组成。

集群3主要由普通三线城市组成。

集群4主要由人口外流的弱三线城市组成。

以下是正文：

原文：The ongoing downturn in the property sector has now extended into its third consecutive year. In the September Politburo meeting, Chinese policymakers pledged to “stem the decline and facilitate the stabilization” of the property market, which we interpret as stabilizing existing home prices.1 While secondary home prices – which better reflect market conditions than primary home prices that tend to be heavily regulated by local governments – have recently started to show narrower price declines in top-tier cities, a clear nationwide stabilization remains uncertain (图表 1, left chart). Given the strong correlation between house prices and consumer confidence, we believe stabilizing house prices remains crucial for supporting household consumption and broader market sentiment (图表 1, right chart).

Against this backdrop, we look at the National Bureau of Statistics (NBS) 70-city existing property prices data to examine regional patterns, investigate the drivers behind the house price changes, and draw lessons for potential future house price movements.

译文：房地产行业的低迷现已持续三年，9月底的政治局会议明确了房地产市场止跌回稳的目标，高盛将其解读为稳定现有房价，二手房价格比受到地方政府严格监管的新房价格更能反映市场状况，最近一线城市的二手房价跌幅有所收窄，但全国范围内的房价仍难言企稳（图1，左图）。鉴于房价与消费者信心之间的强相关性，我们认为稳定房价对于支持家庭消费和提振市场情绪来说至关重要（图1，右图）。

在此背景下，我们研究了国家统计局的70城房价数据，以分区域研究模式来调查房价变化背后的驱动因素，并为未来的房价变动提供判断依据。

图表1：二手房价格尚未触底，房价稳定对经济复苏仍然至关重要

左图为二手房价格预测，数据包括统计局70城房价（深蓝），中原地产6城房价（红色），诸葛找房100城房价（灰色），贝壳25城房价（淡蓝），国信达房产价格（绿色）。右图为房价与消费者信心的走势对比，蓝线表示中国消费者信心，红线表示预期房价上涨的百分比。

分析：从左图可以看出，几乎所有房价指数在2024年都是下跌的，包括926新政后的第四季度房价依然下跌，所以24年第四季度只能算是一个不是很成功的反弹，市场主要还是以价换量为主。再看24到25年的房价预测，诸葛找房、统计局、中原房产分别预测25年到底房价继续下跌6%、7.5%、12.5%，可见高盛对于现有政策强度下的25年房价比较悲观。

从右图可以看出，消费者信心和预期房价呈很强的趋同性，这也解释了目前这么多促销费政策下去但效果却差强人意的原因，和目前国家不遗余力要稳住房地产市场的原因，因为目前房地产市场不但拖累了整体经济，还是通缩治理的核心所在。

原文：For background, there are two main sources of property indices for major cities: official data from the NBS and data from private providers such as Centaline and Zhuge. The NBS 70-city secondary home price index, covering mostly large and medium-sized cities, relies on data from real estate agencies, field investigations, and local housing authority registries. Given the shorter time series and limited city coverage of many private data sources, we focus on the NBS house price index in the analysis below. However, it is not without limitations. For example, local governments face pressure to stabilize property prices in both upturns and downturns, which often leads to data that is overly smoothed, understating actual price fluctuations2.

We first group the property price indices in the NBS 70-city sample using K-means clustering analysis, which divides the data into clusters based on the similarity of their property price trends. The K-means algorithm assigns each city to the nearest centroid (the average of data points in a cluster) and then iteratively adjusts the centroids to minimize the variance within each cluster. To determine the optimal number of clusters, we use the elbow method. This involves plotting the within-cluster variance for different numbers of clusters (K) and choosing the point where adding more clusters no longer improves the result. Using this approach3, we identify four distinctive city clusters based on their house price trends from 2011 to the present (图表 2).

译文：主要城市的房地产数据包括两种：一种是来自统计局的官方数据，另一种是来自中原房产等第三方数据。统计局70城的二手房价指数主要覆盖了大中型城市，数据源于房地产机构、实地调查、当地住房管理局登记处。鉴于统计局数据城市覆盖范围大且包含第三方数据源，我们在下面的分析中重点关注统计局房价指数。但统计局的数据也并非没有缺陷，比如地方政府在经济上行和不景气时都面临稳定房地产价格的压力，这通常会导致数据过于平滑，从而低估了实际价格的波动。

这段是统计学建模的描述，不用过于关注，所以这里只做简单翻译：我们首先使用K-means聚类分析对统计局70城样本中的房价指数进行分组，根据其房价趋势的相似性将城市划分为多个集群。使用这种方法，我们根据2011年至今的房价趋势确定了四个独特的城市群（图2）。

图表 2:中国住房市场中四个独特的城市群

分析：以上是根据统计局70城房价数据，利用K值聚类分析做出来的四个组，图中红线表示该组的房价均值，在这里高盛只是介绍所使用的统计学模型，对于普通投资者来说不需要关注，只需要参考其结果即可。

另外高盛所说的70城数据由于稳房价压力而过于平缓，说的已经很含蓄了，各位可以自行体会，一般来说我的处理方法是将统计局的数据×2左右以得到接近真实的波动数据。

原文：First, Cluster 1 includes mostly tier-1 and large tier-2 cities, where house prices have shown larger price appreciation and greater resilience (e.g., limited price decline in the 2014-15 downturn). The majority of the 70 cities – 77% on a population-weighted basis – fall into Clusters 2 and 3, which consist primarily of medium-sized tier-2 and tier-3 cities (see Appendix for more detailed city composition and geographic distribution, 图表 8 and 图表 9). Cluster 4 consists of cities such as Jinzhou, Mudanjiang (both in the northeast and with population outflows), and Wenzhou which had a housing bust in the early 2010s.

译文：首先，集群1主要包括一线和强二线城市，这些城市的房价表现出更大的升值空间和更强的弹性（例如在2014到2015年经济低迷期间房价下跌有限）。70城中的大多数（按人口加权计算为77%）属于集群2和集群3，主要由普通二线和普通三线城市组成（更详细的城市构成和地理分布见图8和图9）。集群4由锦州、牡丹江（均位于东北部且人口外流）和温州等城市组成，温州入选的原因是因为在2010年代初经历了房地产泡沫。

分析：以上是高盛对于以统计局70城的分组列表和地理分布图。

原文：Second, as shown in 图表 3, most cities reached their peak house prices before December 2022, with cities that have stronger fundamentals peaking later. Around 57% of the 70 cities peaked between January 2020 and December 2022, while only 6% peaked after December 2022. On average, Cluster 1 cities peaked the latest, around mid-2022, likely supported by stronger demand fundamentals and more stringent supply restrictions. This is followed by Clusters 2 and 3, which peaked around late 2020 to mid-2021. Cluster 4 cities peaked the earliest, with their house prices continuing to decline despite a brief pick up during the 2015-18 shantytown redevelopment program.

译文：其次，如图3所示，大多数城市在2022年12月之前达到了房价峰值，而基本面较强城市的房价达峰较晚。70城市中约有57%的城市在2020年1月至2022年12月期间房价达到峰值，有6%的城市在2022年12月之后房价达到峰值。平均而言，集群1的城市最晚在2022年年中左右达到峰值，这可能是由于更强劲的需求基本面和更严格的限购措施被放开；集群2和集群3在2020年底至2021年中达到峰值；集群4的城市最早达到峰值，尽管房价在2015到2018年棚改期间短暂回升，但最终继续下跌。

图表 3：2020年至2022年期间，超过一半的城市达到了房价峰值

分析：以上是四个集群的城市房价达峰的比例，高盛发现的规律是越是高线的城市房价达峰越晚，越是低线的城市房价达峰越早，这背后的原因如上面所解释的，就是高线城市经济和人口基本面更好，而且限购等措施缓慢放开导致需求源源不断的被释放，所以房价达峰的时间会比较晚，这也符合基本的逻辑和经济规律。

原文：Third, while the recent decline in house prices is similar across clusters – ranging from 15% to 20% since their peaks – the cumulative gains since 2011 differ markedly (图表 4 ). Cluster 1 cities have seen house prices appreciate by 75% over the period, Cluster 2 cities have experienced only 20% cumulative increase, and the latest declines have wiped out all prior gains in Cluster 3. Cluster 4 cities stand out as the only group to experience a net price decline over the last decade. These trends highlight that stronger fundamentals not only delay the timing of house price peaks but also preserve some of the earlier price gains during a nationwide housing downturn.

译文：第三，虽然近期各集群房价的下跌幅度相似，都是自峰值下跌15%到20%不等，但自2011年以来的累计涨幅却截然不同（图4），集群1城市的房价累计上涨了75%，集群2城市的房价累计涨幅仅为20%，近两年的下跌抹去了集群3城市之前的所有涨幅，而集群4城市是过去十年中唯一出现价格净下降的组。这些趋势说明一二线城市强劲的基本面不仅推迟了房价达峰的时间，而且在全国房地产低迷期间的抗跌性也会更强。

图表 4：房价达峰后各集群的房价下跌幅度相似，但自2011年以来的累积涨幅截然不同

分析：上图是四个集群从房价达到峰值后的跌幅（深蓝）和房价从2011年至今的累计涨幅（淡蓝），可以看出集群1、2、3的房价达到峰值后的跌幅类似，集群4房价达到峰值后跌幅比其他三个集群都要大，这是因为集群4的城市经济和人口基本面均不如前三个集群所致。另外从2011年后累计涨幅的角度上看，越是基本面好的高线城市累计涨幅越大，基本面最差的集群4城市从2011年至今的涨幅为负。

高盛想表达的意思是，城市的选择对于房产增值来说至关重要，高线城市由于基本面长期保持强劲而表现为强者恒强，其房价在牛市时涨得多熊市时跌的少，是购房的首选城市。

原文：图表 5 suggests that population flows4 are important for house prices. The left panel shows that every 10% increase in population growth from 2010 to 2020 corresponds to approximately 3% rise in house prices in the 2010s. For example, cities like Chengdu and Zhengzhou, which experienced population inflows of around 45% over the decade, saw house price increases of around 50%. In contrast, some northeastern cities such as Jilin, which faced a 20% population decline, recorded a 20% rise in house prices. The 65% disparity in population growth between these two groups of cities accounts for roughly 30% variation in house price performance during the period. The right panel suggests that cities with strong population inflows in the 2010s also experienced milder price declines in the recent housing downturn.

译文：图表5表明人口流动对房价来说很重要。左图显示，从2010年到2020年，城市人口每增加10%带来的房价涨幅约为3%。例如成都和郑州等城市在过去十年中累积了约45%的人口流入，对应的房价上涨了约50%，作为对照的吉林等一些东北城市人口下降20%，对应的房价上涨了20%，这两组城市之间65%的人口增长差距导致房价表现的差异约为30%。右图表明，在2010年到2020年间人口流入较多的城市在最近的房地产低迷期中房价下跌也比较温和。

图表 5： 人口增长与房价有很强的相关性

分析：上图是高盛根据2010年到2020年历史数据做出的房价与人口关系对比图，得出的结果就是城市人口每增加10%可以贡献约3%的房价涨幅，这个数据可以直接当作结论来用。

原文：Another driver for the recent house price declines could be the elevated housing supply. Longer inventory months – measured as the sellable gross floor area divided by the 12-month rolling gross floor area sold – tend to exert downward pressure on prices. As shown in 图表 6 (left chart), inventory months have risen sharply in lower-tier cities since 2021. Among the seven cities classified in Clusters 2 and 3 that also have inventory months data, there is significant variation in average inventory months since 2023, with Dalian the highest (49 months) and Hangzhou the lowest (10 months) (图表 6, right chart).

译文：近期房价下跌的另一个驱动因素是住房供应量的增加，较长的库存去化周期（以可售总建筑面积除以12个月滚动销售总建筑面积来衡量）会对房价产生下行压力。如图6（左图）所示，2021年以来低线城市的库存去化周期快速上升，在集群2和集群3的7个城市中，自2023年以来的平均库存去化周期存在显著差异，其中大连最高（库存去化周期49个月），杭州最低（库存去化周期10个月）（图6，右图）。

图表 6：低线城市房产库存增加

分析：左上图可以清楚看到三线城市在2023年后的库存增加幅度远大于一二线城市，这也是三线城市房价表现远不如一二线城市的重要原因之一。右上图是集群2和集群3里面挑出来的7个城市的库存去化周期对比，大连的去化周期高达49个月，而杭州去化周期只有10个月，这就解释了杭州的房价为什么远比大连的房价坚挺的原因。

原文：To summarize the two drivers of house price dynamics, we estimate a panel regression model to link annual house price growth with 1) changes in inventory months5 and 2) population growth between 2010 and 2020. We also control for the 1-year lag of house price growth as well as time fixed effects to account for other uncontrolled factors that may drive house prices. The analysis was conducted using two panels, one focusing on cities in Clusters 2 and 3, and another including all available cities in the dataset for increased statistical power. The results are largely consistent across the two panels (图表7）.

译文：为了总结房价波动的两个驱动因素，我们建立了一个面板回归模型将房价增长与库存周期（月数）的变化和2010年至2020年的人口增长联系起来。我们还通过统计学手段去除了可能影响房价的其他不受控因素。分析是用两个面板进行，一个面板侧重于集群2和集群3中的城市，另一个面板包括所有城市，分析的结论是两个面板的结果基本一致（图表7）。

图表 7：在回归模型中，人口变化和住房库存都推动了房价的波动

分析：上图可以不用看懂，只需要明白高盛通过回归模型的分析证明了一件事，就是人口和房产库存这两个因素对房价确实有显著的影响，换句话说，在其他因素不变的前提下，人口增加和房产库存减少的城市房价会显著上涨，这个结果也可以直接当做结论来用。

原文：We find that a one-month increase in housing inventory supply leads to around 20bp decline in house price growth. Meanwhile, a 10% increase in population between 2010 and 2020 results in an average annual house price increase of 0.3%, accumulating to 3% over a decade – consistent with our earlier scatter plot analysis. Combining these estimates, average population increased 23% in our Clusters 2 and 3 sample over the last decade, which accounts for roughly 28% of the total house price appreciation during this period. Moreover, housing inventory has risen by about 17 months since the price peak in 2020-22, contributing to around 18% of the subsequent price decline.

Furthermore, we find that the persistence of house price growth, as measured by the one-year lagged house price growth, is less pronounced (0.2%) compared to the US (around 0.4%). This implies that house prices are less sticky in China, potentially reflecting the larger role of government policies in the economy.

While demographics fundamentals and housing inventory are important drivers of house prices, they don’t explain all of the house price movements, and other variables not captured by our model also play an important role. For example, previous academic literature illustrated the importance of household income in driving local house prices.

Taken together, our analysis suggests that both structural population shifts and cyclical fluctuations in housing inventory supply are important drivers of house prices in China. Therefore, we expect top-tier cities like Shanghai and Shenzhen, bolstered by stronger migration inflows, lower inventory levels and the recent easing of housing purchase restrictions, may find a price bottom sooner than the rest of the nation – consistent with our property team’s view. In contrast, while inventory destocking may provide some relief, the implementation challenges and still limited government support suggest that lower-tier cities may continue to face pressure on housing prices, further exacerbated by migration outflows.

译文：我们发现住房库存的去化周期每增加一个月会导致房价下降约0.2%。2010年至2020年期间人口增长10%，期间房价年均上涨0.3%，十年内累计上涨3%，与我们之前的散点图分析结果一致。利用以上的估算，在过去十年中集群2和集群3城市的平均人口增长了23%，贡献了同期房价总升值的28%。自2020到2022年房价见顶以来，住房库存已经上涨了17个月，贡献了房价总贬值的18%。

另外我们发现中国的房价粘性比美国更低，这反映了中国政府的政策在中国经济中发挥的作用比美国政府的政策在美国经济中发挥的作用更大。

虽然人口基本面和住房库存去化周期是房价的重要驱动因素，但它们并不能解释所有的房价变动，回归模型中不包含的其他变量也起着至关重要的作用，比如家庭收入在推动当地房价方面的重要性。

综上所述，我们的分析表明，结构性人口转移和住房库存去化周期是中国房价波动的重要驱动因素。因此在更强劲的人口流入、更低的房产库存水平、以及最近被放宽的限购等政策的推动下，一线城市的房价会比其他城市的房价更早完成触底。相比之下，低线城市会继续面临房价压力，而人口外流会进一步加剧这种房价压力。

分析：高盛总结了库存去化周期和人口因素对于城市房价的影响，结论就是住房库存去化周期每减少一个月贡献的房价涨幅约为0.2%，城市人口每增加1%贡献的房价涨幅约为0.3%。另外我国的政策对本国经济的影响大于美国的政策对本国经济的影响，换句话说就是同样一个政策，在中国的作用比在美国的大，所以政策是改变我国经济走势非常重要的因素之一。

需要注意的是，模型只对人口和房产库存对房价的影响做了有效性验证，而并没有包括其他影响房价的关键因素，比如最重要的家庭收入因素，这在逻辑上也很直观，家庭没有钱就没法买房。