Dr. Elizabeth Delmelle of the University of Pennsylvania and the Department of City & Regional Planning will be presenting a talk, "Exploring micro-geographic housing dynamics and predicting neighborhood racial and income changes using real estate listing and mortgage lending data" on Friday, April 28th at 3:00pm

Abstract:

Emerging non-traditional data sources and advancements in machine learning are
advancing the predictive potential for understanding urban changes in near real-time. In this
talk, I discuss the use of annual point-level real estate listing data for understanding housing
dynamics at a micro-geographic scale and show how these listings can serve as a bellwether for
subsequent changes in the racial and income makeup of mortgage applicants at the
neighborhood scale. I will discuss a case study using property listing data from the Multiple
Listing Service from 2001-2020 for the Charlotte, North Carolina MSA combined with annually
collected mortgage lending from the Home Mortgage Disclosure Act (HMDA). The analysis
begins with a semi-supervised, embedding-based classification of the textual description of
properties into 5 types of dwellings along a housing lifecycle continuum of investment to
disinvestment: disinvestment; opportunity; new suburban; expensive investment; renewed. The
first analysis explores the spatial temporal dynamics of properties aggregated to a small,
gridded tessellation of the study area. Next, I demonstrate how the marketing of properties
explains changes in the racial and income composition of mortgage applicants in the following
year. To do so, we estimate a model that explains the share of mortgage applicants by different
income and racial groups in time t, as a function of the share of properties in a neighborhood
listed in each of our five categories, in time t-1. This framework enables us to explore threshold
effects needed to predict significant racial or income changes.