Real estate returns typically have a positive autocorrelation. However, we discovered two oddities in our Hong Kong housing market data. One is that the autocorrelations of submarket returns at the district level are generally negative. In contrast, even with minor or no submarket autocorrelations, the autocorrelation of total market returns is highly positive. This study provides a clear explanation for the two different patterns. We first show analytically how the observed autocorrelation transaction noise and rate of return adjustment are interrelated. This model suggests that even if the return can adjust immediately in response to changes, fluctuations in prices observed will cause us to see a negative autocorrelation. An empirical method is proposed to determine the speed of adjustment of returns based on the negatively biased autocorrelation. In addition, the autocorrelation returns from an individual market result from more than just the autocorrelations between its submarkets and the cross-lead-lag relations between the submarkets. The strong cross-lead-lag relationships boost the autocorrelation of aggregate market returns. Two hypotheses are examined to explain the cross-leap-lag relationship between submarkets, including spatial diffusion of information and transaction costs. Tests based on empirical data derived from Hong Kong housing market data confirm the hypothesis of transaction costs against the spatial diffusion hypothesis.

This is a sneak peek of the subscription content you can access through the institution you work for.

Data Availability

The information used in the article is proprietary data. The authors have not been granted permission to share the data.

Code Availability

Not suitable.

Notes

1. Hong Kong is divided into 18 council districts. Four districts lack trade volume, leading to estimated NAs in the index series. They are then discarded.
2. Equation (2) can be expanded in the appendix to accommodate higher-order return-generating processes, like the AR(2) method.
3. The returns from C&S monthly indices show more significant autocorrelation coefficients than those of FHFA indexes. The reason could be that the C&S indexes are calculated using the moving average algorithm over the three-month time horizon, which could have created positive biases in the autocorrelation coefficients of the returns. However, suppose you consider the algorithm identical across all indices of the same series of indexes. In that case, the proportions of autocorrelation coefficients in different submarkets will still be dependable.

This study provides a clear explanation for the two different patterns. We first show analytically how the observed autocorrelation transaction noise and rate of return adjustment are interrelated. This model suggests that even if the return can adjust immediately in response to changes, fluctuations in prices observed will cause us to see a negative autocorrelation.

An empirical method is proposed to determine the speed of adjustment of returns based on the negatively biased autocorrelation. In addition, the autocorrelation returns from an individual market result from more than just the autocorrelations between its submarkets and the cross-lead-lag relations between the submarkets.