A Comparison of Estimators of Capital Asset Indexes

The two basic models used for constructing price indexes for durable assets (such us real estate assets) have been the hedonic and repeated sales models. Case and Quigley (1991)-CQ proposed a generalized least squares (GLS) procedure to estimate a combined (single and repeated sales information) model (CSRS). Hill, Knight and Sirmans (1997)-HKS proposed a maximum likelihood procedure to estimate the CSRS accounting for the autocorrelated error process in the hedonic model. Rambaldi, Hill and Doran (2003)-RHD proposed a methodology based on interpolating incomplete observations. The sale price of a particular asset, a house for instance, can be viewed as an incomplete time series since the price is observed only when a sale occurs. The method proposed by RHD can estimate a CSRS model with autocorrelated errors using a maximum likelihood estimator. It improves on HKS in that it also produces time-space consistent interpolations (model's prediction of repeated sales always equal the actual observations) and explicitly estimates a cross-sectional correlation parameter. The present paper compares RHD to traditional estimators as well as the HKS estimator through a simulation experiment.