Three essays on the application of quantile regression in real estate economics /

Abstract

This dissertation is composed of three individual essays on the use of conditional quantile regression in real estate economics. The first essay discusses some limitations of the traditional conditional quantile regression methodology. A modification is proposed to improve the interpretability of conditional quantile regression estimates for applications to hedonic price functions in real estate valuation. The second essay provides empirical applications of the methodology suggested in the first essay to analyze the implicit prices of different types of flooring in single-family homes. The third essay suggests that conditional quantile regression can be a viable alternative to duration models to analyze the determinants of the length of time between the market listing of a home and its sale. The essays employ variants of a recently suggested spatial-temporal technique to identify neighborhood effects to avoid spatial autocorrelation and endogeneity problems. The essays rely on data from the Multiple Listing Service (MLS) for single-family homes in Rutherford County, TN, and the county planning commission. The data cover the years 2003 to 2007.

The first essay illustrates that the traditional quantile regression estimates are likely to overestimate the coefficient dispersion across quantiles. As a direct consequence, hedonic price functions in real estate applications may underestimate the prices of homes at the lower-end of the distribution and overestimate prices at the upper end. Unconditional quantile regression is shown to suffer from the same problem, except to a larger degree. An adjustment factor is proposed for the traditional conditional quantile regression estimates to minimize the prediction error.

The second essay applies the methodology proposed in the first one to examine the role of different types of flooring in determining house prices. The results suggest that there are large differences in the implicit values attached to different types of flooring. Almost uniformly across the sales price distribution, finished wood is the most valued flooring type, closely followed by marble and tile. Carpet is the standard flooring type used by almost 96 percent of all homes and, therefore, does not add extra value. The use of vinyl flooring tends to lower the value of a house across all price ranges. The essay identifies large differences in implicit prices across quantiles for different combinations of flooring types. For example, a combination of finished wood, tile, and parquet adds the most value to lower and medium priced homes, while a combination of carpet, wood, and marble has a particularly high implicit value at the upper end of the price distribution.

The third essay shows that conditional quantile regression is a viable technique to study the determinants of the time it takes from the listing of a home to its sale (TOM). Quantile regression has a number of advantages over the traditional methods of duration analysis, such as the Cox proportional hazards model. It allows the determinants of TOM to affect the time on the market differently across the distribution of TOM and the results are much easier to interpret, which is an important consideration for practical applications. For the data set at hand, significant differences are found across the distribution of TOM for a number of variables. For example, a high list price relative to sales price of neighboring properties prolongs TOM perceptively for houses that sell quickly. The same applies to houses which are located in a less desirable school zone.