time：2023-01-01
Yifan Chen, Jianhua Gang, Zongxin Qian, Jinfan Zhang
This paper examines whether investors rationally respond to rent news when pricing houses. Previous studies typically estimate rent news using aggregated house price and rent data, but this approach introduces bias in rationality tests. The bias arises because aggregated data capture only the cross-project average relationship between housing returns, rent growth, and macroeconomic dynamics, ignoring the heterogeneity among projects. Even within the same city, housing projects can vary significantly in terms of quality, facilities, and neighborhoods, leading to different dynamic relationships. Thus, investors should rely on project-level dynamics rather than cross-project averages when forecasting housing returns and rent growth.
Rent news reflects investors' revisions of their expectations regarding future rental income based on forecasting errors of housing returns, rent growth, and macroeconomic conditions. If forecasting errors and dynamic relationships are accurately estimated, rent news can be precisely measured. Assuming investors use a vector autoregressive (VAR) model, unbiased estimates can be obtained from project-level VAR models. However, due to data constraints, prior studies often estimate rent news at a more aggregated level, such as metropolitan or national levels. This approach introduces two biases: first, the cross-project average dynamic relationship estimated from aggregated data is itself biased; second, even if it were unbiased, it would still be a poor proxy for project-level dynamics, leading to distorted rent news measurements.
By comparing proprietary project-level data with aggregated city-level data from China’s megacities, this paper finds that using aggregated data for rationality tests results in significant bias. For instance, city-level data might suggest that investors slightly overreact to rent news, while project-level data reveal that investors actually underreact significantly. This bias stems from ignoring heterogeneity in project-level dynamic interactions rather than simple aggregation bias. Moreover, even if a panel VAR model is used at the project level to estimate unbiased cross-project average coefficients, applying these coefficients to rent news calculations yields results similar to those obtained from aggregated data, reinforcing the necessity of capturing project-level heterogeneity.
This paper not only theoretically demonstrates these biases but also empirically shows their substantial economic impact. Unlike prior studies that rely on numerical simulations, this study provides both analytical and empirical evidence of aggregation bias. Additionally, the paper discusses the complementarity between model-based and survey-based approaches in studying investor rationality. While survey-based methods avoid model specification errors, model-based approaches mitigate concerns regarding survey data quality. Both approaches have their advantages and can be used together to gain a more comprehensive understanding.