Supply Shock Versus Demand Shock: The Local Effects of New Housing in Low-Income Areas

Abstract

We study the local effects of new market-rate housing in low-income areas using microdata on large apartment buildings, rents, and migration. New buildings decrease nearby rents by 5 to 7 percent relative to locations slightly farther away or developed later, and they increase in-migration from low-income areas. Results are driven by a large supply effect-we show that new buildings absorb many high-income households-that overwhelms any offsetting endogenous amenity effect. The latter may be small because most new buildings go into already-changing areas. Contrary to common concerns, new buildings slow local rent increases rather than initiate or accelerate them. JEL Codes: R21, R23, R31 the regional and local effects of new housing construction. It benefits both.Causal identification in this setting is challenging because developers select the locations of new buildings based in part on unobserved local characteristics and trends. In addition, the size and shape of a new building's amenity or reputation effects is unknown, making it difficult to know where they may shrink or reverse the negative effect of added supply. We attempt to overcome these challenges by leveraging our unique data to construct three related empirical strategies. The first is a difference-in-differences specification that compares the area very close to a new building to the area slightly farther away (our "near-far" specification). 4 The idea is that frictions in the land assembly and development approval processes lead to random variation in building placement and timing at the hyper-local level, making the outer area a good control for the treated inner area. This specification is well suited to detect one way that new buildings could raise rents-through amenity effects that fade out quickly with distance, such as increased retail options or the replacement of a vacant lot.The second exercise is a difference-in-differences that compares listings near buildings completed in 2015 and 2016 to listings near buildings completed in 2019, after the conclusion of our sample (our "near-near" specification). The underlying logic is that developers choose sites in both groups for similar reasons, but one building is completed before the other for idiosyncratic reasons, such as the timing of when sites are available for purchase. Because the treatment and control areas are not necessarily in the same neighborhood, this specification can detect price changes driven by broader effects that may cross the near-far boundary of the first exercise. A variant of this specification also allows us to examine how congestion effects near the building could influence our results. Finally, we combine both sources of variation into a triple-difference specification that effectively compares the near-far difference around 2015-2016 buildings to the near-far difference around 2019 buildings.We first study the effect on rents using listing-level data from Zillow ™ that span 2013 to 2018. All three empirical approaches show that new construction in low-income neighbo...