Association for Public Policy Analysis & Management

We all know that housing prices have following a boom-and-bust trajectory over the past fifteen years, but which segments of the population experienced the sharpest rise and fall—and in which parts of the country? Though some research has hinted at the answers to these questions, they remain largely unresolved because the literature has focused on housing prices in aggregate. Using transaction-level data from multiple large urban counties, I analyze the entire distribution, breaking down the change in housing prices into quantiles at both a metropolitan and neighborhood level across the country.

Gyourko and Linneman (1993) and Gyourko and Tracy (1999) were among the first to estimate the distribution in housing prices using the American Housing Survey, published by the Census Bureau. Nationally, they found that house prices and quality increased from 1960 to 1974, but thereafter they diverged. From 1975 to 1995, real prices only increased above the 75th percentile. After adjusting for quality using mean regression and quantile regression methods, they find evidence that housing quality declined for the 10th percentile, while an increase in quality explains much of the price increase at the top of the distribution. They do not examine geographic variation in these measures.

To my knowledge, McMillen (2008) is the first paper to study the change in the distribution of house prices at a local level, analyzing Chicago from 1995 to 2005 with a new decomposition approach proposed by Machado and Mata (2005). He too found that higher prices at the top of the distribution resulted from a move toward bigger, better homes. Nicodemo and Raya (2012) extend this approach across cities in Spain, using a similar method developed by Melly (2005), and finding a widening distribution explained mostly by the change in quality variables, as well as significant geographic variation.

In this paper, I first replicate and update the findings of Gyourko and Tracy (1999) to the current period, an extension of twenty years, with the new transaction-level DataQuick records from CoreLogic. Second, I test whether their results hold using the new methods developed by Machado and Mata (2005) and Melly (2005). Third, I apply this approach to a city-level analysis for several major MSAs in the United States, as Nicodemo and Raya (2012) do in Spain. I measure the change in the distribution in house prices in each city, determine how much of the change can be explained by quality variables, and investigate what differences between the cities might be causing the variation in their housing price distributions—especially during the housing bubble, which some cities experienced more acutely than others. This analysis allows me to identify which segments of the population were most sensitive to the boom-and-bust—and in which cities—with policy implications for the role of the housing market in social equity and financial stability going forward.