A Nonparametric Bounds Approach for Hedonic Housing Price Indexes With an Application to the Southern California Market 1996 to 2008

I propose a nonparametric bounds approach for hedonic housing price models partially identified due to sample selection of an unknown form. To construct bounds for the identification region, I use a nonparametric hedonic housing price model that allows spatial correlation and propose estimators of extreme conditional quantiles. The estimated bounds are then used to generate price indexes in the form of a time-series of intervals. In contrast to conventional housing price indexes, these interval-valued indexes do not suffer from sample selection bias. The approach is used to construct both metro and zip code level price indexes from a sample of over one million transactions from 1996 to 2008 in Los Angeles and San Diego metropolitan areas collected by a mortgage technology firm, FNC. Although the bounds approach has less identifying power, it provides more reliable results because it is based on more credible assumptions. The metro level indexes show that the housing price in San Diego peaked before Los Angeles and the appreciation rate at peak in Los Angeles was higher while the depreciation rate after peak was lower. The zip code level indexes indicate that the nonparametric hedonic method may underestimate the price of high-value properties and overestimate the appreciation rate of low-value properties