Academic journal article Journal of Real Estate Literature

Repeat Sales House Price Index Methodology

Academic journal article Journal of Real Estate Literature

Repeat Sales House Price Index Methodology

Article excerpt


We compare four traditional repeat sales indices to a recently developed autoregressive index that makes use of the repeat sales methodology but incorporates single sales and a location effect. Qualitative comparisons on statistical issues including the effect of gap time on sales, use of hedonic information, and treatment of single and repeat sales are addressed. Furthermore, predictive ability is used as a quantitative metric in the analysis using data from home sales in 20 metropolitan areas in the United States. The indices tend to track each other over time; however, the differences are substantial enough to be of interest, and we find that the autoregressive index performs best overall.

(ProQuest: ... denotes formulae omitted.)

Housing is an important part of a nation's economy and house price indices help us understand how such markets operate by tracking changes over time. These indices can be useful for a variety of purposes: as macroeconomic indicators, as input into other indicators, by individuals looking to sell or purchase a home, or for appraising homes.

We conduct a comparative analysis of the autoregressive index introduced by Nagaraja, Brown, and Zhao (2011) with four traditional repeat sales indices: the Bailey, Muth, and Nourse (1963) index, the original Case and Shiller (1987) method, the Home Price Index produced by the Federal Housing Finance Agency, and the S&P /Case-Shiller Home Price Index published by Standard and Poor's. For comparative purposes, we also include the median price index. We do not compare results for hedonic indices or the hybrid index from Case and Quigley (1991) because, other than location information, no hedonic information is available in our data.

We evaluate these five indices using a two-pronged approach: (1) analyzing the components of each index along with the statistical structure and (2) comparing estimates of individual house prices from each index. Using data from home sales for 20 cities in the United States, from 1985 to 2004, we find that the autoregressive model produces the best predictions.

The second approach requires some justification. Generally, to determine how well a model works for its prescribed purpose, we check with the "truth" either using real data or through simulation. We cannot apply either of those techniques here. There is no true index and as each index is constructed under differing data-generating processes, simulation is not an effective tool for comparison. A third option is to examine predictions of individual house prices as a way to determine the efficacy of an index. All of the indices studied here can be applied on a microeconomic level. Therefore, individual price prediction is a practical way of evaluating an index. This type of quantitative metric allows for comparisons on an objective, measurable scale. We assert that methods used to generate better predictions involve higher quality models and thus lead to more accurate indices. Therefore, prediction, combined with qualitative comparisons, provides a more complete analysis of a housing index.

We begin with a literature review. We then describe the data and models and make qualitative comparisons. Finally, we use the data to compare the indices and predictions produced from each method. Finally, we close with concluding remarks.


A major hurdle in constructing house price indices is that homes are heterogeneous goods. Furthermore, the market composition of homes sold changes throughout the year causing even more difficulties. One way to control for differences in the quality of housing stock over time is to use a hedonic index. Characteristics of a house such as floor area or location are considered hedonic variables. Such indices, such as those proposed in Yeats (1965) and Noland (1979), are constructed by regressing the hedonic characteristics against sale prices. Pure hedonic models have been largely abandoned in favor of alternative methods, primarily due to problems with the availability, accuracy, and stability of relevant variables and the difficulty in describing the model. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.