Hedonic Indices

Overview

The hedonic-regression is a method that attempts to overcome the issue of compositional bias associated with median price measures. The premise for this lies in hedonic theory which suggests that the value of a composite good – such as a house – is the sum of its components. Thus, by decomposing the sample of houses into their various structural and location attributes, the differences in these qualitative factors across houses can be controlled.

The term hedonic was first applied to this concept by Court (1939) who employed a similar multivariate regression technique to evaluate automobile pricing. Colwell and Dilmore (1999) identify even earlier applications of the technique. Waugh (1928), for example, models price movements in agricultural produce via a multiple regression model. The first academic applications of the hedonic technique to housing were made by Griliches (1971) and Rosen (1974).

Education / Methodology

Two forms of hedonic indices are calculated by RP Data-Rismark. These are: the pooled hedonic index; and the adjacent-period hedonic index.

The pooled hedonic is a time-invariant model since the estimated coefficients are constant through time. That is, the model uses dummy-variables to represent time periods and collated all the data through space and time in a pooled regression. The quality-controlled cumulative growth rate attributable to the time period is then captured by the coefficients of the time dummy variables. Specifically, the dummy variable regression technique controls for the various characteristics and qualitative features of a property that determine price, using a multivariate pooled regression, in which dummy variables are used to represent the time period of each sale. For this reason the pooled hedonic is also referred to as the dummy variable approach.

The adjacent-period approach is conceptually very similar to the pooled model, but pools data only from consecutive time periods. For each of these “adjacent-period” subsets of data, a hedonic function similar to that used in the dummy-variable model is estimated, with one significant modification: only one time dummy-variable is included. This model consequently allows for the implicit value of attributes and characteristics - as represented by their respective estimated regression coefficients - to change through time. This is noted as a major advantage of the adjacent-period hedonic index (Triplett 2004), in addition to it being an index methodology which does not suffer index revision.