Hedonic Price Functions: Guidance On Empirical Specification

The hedonic pricing model is widely accepted as a method for estimating the marginal willingness to pay for spatially delineated amenities. Empirical applications typically rely on one of three functional forms—linear, semi-log, and double-log—and rarely involve rigorous specification testing. This phenomenon is largely due to an influential simulation study by Cropper, Deck and McConnell (CDM) (1988) that found, among other things, that simpler linear specifications outperformed more flexible functional forms in the face of omitted variables. In the 20 years that have elapsed since their study, there have been major computational advances and significant changes in the way hedonic price functions can be estimated. The purpose of our paper is to update and extend the CDM (1988) simulations to investigate current issues in hedonic modeling. Three preliminary results obtained from our theoretically consistent Monte Carlo simulation have been highlighted in this paper: (i) we find that adding spatial fixed effects (census tract dummies) to linear models does improve their performance. This is true both when all attributes are observed, and when some attributes are unobserved, (ii) adding the spatial fixed effects to the more flexible specifications such as the quadratic and quadratic box-cox does not improve their performance when all housing attributes are observed. However, when some housing attributes are unobserved, the spatial fixed effects significantly improves the performance of flexible specifications as well, and (iii) increasing the sample size from CDM’s 200 observations to a sample size of 2000 (which is more representative of modern applications) changes the relative performance of different specifications.