Revisiting the Autocorrelation of Real Estate Returns

Abstract

Real estate returns often show highly positive autocorrelation. However, with the Hong Kong housing market data, we found two anomalies. For one, the autocorrelations of the district-level submarket returns are mostly negative. For the other, despite the negative or insignificant submarket autocorrelations, the autocorrelation of the aggregate market returns is highly positive. This study explains the two patterns. First, we show analytically how observed autocorrelation, transaction noise, and the speed of return adjustment are related. The model suggests that even if return adjusts instantly to news, the transaction noise in observed prices will lead us to observe a negative autocorrelation. An empirical approach is derived to recovering the adjustment speed of returns from the negatively biased autocorrelation. Second, the autocorrelation of returns of a market is a function of not only the autocorrelations of its submarkets, but also the cross lead-lag relationships between the submarkets. Strong cross lead-lag relationships inflate the autocorrelation of the aggregate market returns. Two competing hypotheses for explaining the cross lead-lag relationships between submarkets, namely spatial information diffusion and transaction costs, are tested. Empirical tests based on Hong Kong housing market data support the transaction cost hypothesis against spatial diffusion.