VALUING PRODUCT ATTRIBUTES USING SINGLE MARKET DATA - A COMPARISON OF HEDONIC AND DISCRETE CHOICE APPROACHES

This paper compares, via simulation, the performance of the multinomial logit and hedonic models in estimating consumer preferences for product attributes. We ascribe preferences over the attributes of houses to a population of consumers, and, by having them bid for a set of houses calculate equilibrium prices. The resulting data are used to estimate the two models. We find that the gradient of a linear Box-Cox hedonic price function estimates marginal attribute bids at least as well as a linear logit model, although the difference between the two is small when some variables are not observed or are replaced by proxies. The logit model, however, outperforms the hedonic model in valuing non-marginal attribute changes. This is true when the researcher knows the true form of consumers' utility functions and when the utility function must be approximated.