TL;DR
This paper introduces a novel "pseudo repeat sales" (ps-RS) model to construct more reliable home price indices in China, addressing limitations of traditional hedonic and repeat-sales methods. By leveraging the unique characteristics of China's new housing market, the ps-RS model offers a more robust and transparent approach to tracking price dynamics.
What Problem Does This Paper Address?
Accurate housing price indices are crucial for understanding market dynamics, informing policy, and guiding investment. However, constructing these indices in China presents significant challenges. Traditional methods like the hedonic model struggle with the rapid evolution of urban spatial structures and improving housing quality, leading to potential omitted variable bias. The classic repeat-sales model, which relies on properties selling multiple times, is largely ineffective in China where new homes dominate the market and typically transact only once. This leaves a critical gap in reliable, quality-controlled price indices for China's burgeoning new housing market.
Methods
The authors propose a "pseudo repeat sales" (ps-RS) model. This approach adapts the repeat-sales methodology by creating "pseudo-pairs" of sales. Instead of matching identical units, they match two very similar new homes within the same building, phase, or complex. This "within-building" or "within-phase" matching allows them to control for many unobservable attributes that are common to units in close proximity. The model then regresses the price differential between these pseudo-pairs onto time dummies and differences in easily observable physical attributes, effectively cancelling out more problematic location and community variables. This hybrid approach combines the strengths of repeat-sales and hedonic models while mitigating their respective weaknesses. They also implement a weighting adjustment to account for changes in sample size distributions over time.
Datasets
The study utilizes a large-scale micro transaction dataset of new home sales in Chengdu, Sichuan Province, spanning from January 2006 to December 2011. This dataset includes 2152 complexes and 444,596 housing units, with detailed information on purchase price, physical attributes (size, floor, rooms, etc.), and location attributes (distance to city center, zone ID).
Key Insights and Ideas
- Leveraging Homogeneity: China's new housing market, characterized by large-scale, homogeneous residential complexes, provides a unique opportunity for a modified repeat-sales approach.
- "Pseudo-Pairs" as a Solution: The concept of matching similar units within a building or phase effectively creates "repeat sales" where none would traditionally exist, allowing for quality-controlled price indexing.
- Mitigating Omitted Variables: By focusing on within-building or within-phase comparisons, the ps-RS model effectively cancels out many of the problematic location and community variables that plague hedonic models in rapidly developing urban environments.
- Improved Robustness and Transparency: The ps-RS model offers a more robust and transparent alternative to existing methods, making it easier to understand and communicate to a wider audience.
Key Results
- The ps-RS indices (building-based, phase-based, and complex-based) and the hedonic index show similar overall trends and turning points, validating the general market movements captured.
- The building-based ps-RS index demonstrated the lowest volatility and highest first-order autocorrelation, suggesting it has less random noise and is a more accurate representation of market returns.
- Out-of-sample prediction tests showed a high correlation (0.999993) between indices derived from different random sub-samples, indicating strong robustness.
- The ps-RS indices generally exhibited better performance in terms of random error and estimation reliability compared to the hedonic index.
- The building-based ps-RS index showed the lowest price growth trend among the ps-RS versions, suggesting that improvements in omitted physical quality attributes might impart an upward bias in the hedonic index.
Limitations and Future Work
The authors acknowledge that while the complex-based ps-RS index closely parallels the hedonic index in Chengdu, implying location variables might not be a major issue there, this might not hold true for all cities. They suggest that omitted or poorly measured physical quality attributes can cause an upward bias in the hedonic index. The ps-RS method is presented as a significant contribution, particularly for rapidly urbanizing countries like China, and warrants serious consideration by statistical agencies. Future work could explore the application of this methodology in other similar contexts.
Summary updated