Revisions in Granular Repeat Sales Indices

Abstract

Price indices based on repeat sales are the most widely used type of real estate index based on asset transaction prices. But such indices are particularly prone to revision. When a new period of transaction data becomes available and is used to update the repeat sales model, all past index values can potentially be revised. These revisions are especially problematical for commercial real estate (as compared to housing), because commercial transactions are relatively scarce and properties are heterogeneous, reducing estimation precision. From a methodological perspective, the magnitude of expected revisions is a particularly useful measure of the quality of the empirical index, as it directly reflects both the precision of the index and its practical usefulness in economic and business applications, since revisions themselves are problematical in practice. This paper focuses on random revisions for indexes in thin, commercial property markets, the type of market that is most challenging for empirical price indexing. We present multiple specifications of the repeat sales model, seeking to reduce revisions. With the objective of minimizing the expected magnitude of revisions, among the specifications we explore, the best result obtains from an index methodology that specifies the periodic returns as a first order autoregressive process, that also uses the periodic returns of an aggregate index as an explanatory variable for more granular indices, and that allows the variance parameters of the signal and the noise to be time-varying. In our small-sample test cases, this model reduces overall index revisions by more than 50%.

Keywords: Commercial Real Estate, Stochastic Volatility, Revisions, Bayesian Repeat Sales

JEL Classification: N9, R31, C32