1 Introduction

Major cities around the world face acute housing supply shortages, creating a need to stimulate residential capacity. As a countermeasure, several cities have implemented property tax exemptions to create incentives for new housing construction. By lowering this tax, these programs are expected to increase the return to residential investment and stimulate development. American cities that have implemented credit programs include New York City (NYC), Seattle, Austin, and San Francisco. Surprisingly, there has been little empirical work on the impact of property tax exemptions on residential investment.

Empirically, whether these tax incentives increase the construction of residential units is not obvious. Theoretically, the response of residential investment to the implementation of tax incentives depends on the size of the incentive and the returns to investment. In land-starved cities with high costs of investment, the returns to investment are limited, and the tax incentives may fail to stimulate the construction of new units. There are two challenges in estimating these impacts. First, developers’ decision to invest in a location depends on its characteristics. Second, the timing of the investment is endogenous to future returns. Well-identified estimates of the causal impacts of property tax reforms on construction is crucial for policymaking in dense cities.

We estimate the short- and long-run impacts of a property tax reform in NYC on new residential investment in the form of new rental units. In NYC, in 2013, the rental housing segment represented 40% of the total housing stock. We focus on the 421a property tax exemption, which protected property owners from tax increases due to new construction-related property value appreciation for a period of 10–15 years (short exemption) or 20–25 years (long exemption). Throughout the manuscript, we consider as property both land and any structures built on it. The duration of the exemption, short term vs. long term, depended on the neighborhood where a building was constructed and whether the developer chose to satisfy the requirements to obtain long exemptions.Footnote 1 This 421a tax exemption was available to projects that converted vacant or underutilized land into new rental units. In some cases, in exchange for the tax exemption, investors had to provide affordable units (on-site or off-site) that were subject to rent stabilization. In other cases, investors could claim the exemption without affordability requirements.Footnote 2

The reform, which was announced in 2006 and implemented in 2008, consisted of four main changes. The first was to expand the regions that had strict affordability requirements to claim the tax exemption (henceforth exclusion regions). With the establishment of the program in 1971, the city was divided into two areas: exclusion regions (blue areas in Fig. 1) and non-exclusion regions (yellow areas). At the start of the 421a program, most regions could access the tax exemption without any requirements, but some Manhattan regions were “excluded” from accessing the program without requirements, leading to the term exclusion regions. In these regions, the provision of affordable units was required to access the tax exemption. Prior to the reform, exclusion regions included only parts of Manhattan, while the reform added neighborhoods across all boroughs.

The second main change introduced by the reform was the imposition of new strict requirements in exclusion regions. Before the reform, the provision of affordable units could be off-site, but after the reform, developers had to provide affordable units on-site to claim the tax exemption. Since exclusion regions were located in high-income neighborhoods, providing affordable units on-site represented a high opportunity cost for developers and a disamenity for high-income renters. Thus, this new requirement substantially increased the costs of claiming the tax exemption in exclusion regions. The third main change brought by the reform was the introduction of new light requirements in non-exclusion regions, where the tax exemption could be claimed as of right, i.e., without any conditions other than the standard statutory requirements. That is, the reform mainly imposed a cap on the size of the tax exemption. As the fourth and last main change created by the reform, new modest requirements on both exclusion and non-exclusion regions were enacted: buildings had to have at least four units, all affordable units were stabilized for 35 years, and affordable units had to have the same number of bedrooms as market-rate units.

We use geocoded administrative data at the parcel level for the 2003–2015 period. The dataset is used to measure residential investment in the form of rental housing buildings, and it comes from the NYC Department of Buildings and records the issuance of permits for new construction. We combine these data with information from the NYC Department of Finance and the list of 421a beneficiaries. The final dataset is used to measure new residential construction, whether a building uses the 421a exemption or not, the type of exemption (short or long), the number of units, the assessed value, and the census tracts where new buildings were constructed.

To estimate the short-run impacts of the tax reform on construction, we use a bunching method. The unit of analysis is the number of new housing units or new buildings per quarter. The delay between the announcement of the reform in Q2-2006 and its implementation in July 2008 (henceforth, the time notch) created incentives for developers to move investment forward in time to avoid losing the old benefits of the tax exemption, a strategy known as grandfathering. The resulting difference in tax levels created a time notch, which we exploit using a bunching methodology.

In our bunching exercise, the treatment group consists of housing units in exclusion regions that claimed the tax exemption. The empirical counterfactual or comparison group is composed of non-tax-exempt units that were also located in exclusion regions, which were unaffected by the policy reform, as in Best and Kleven (2018) and Kopczuk and Munroe (2015). Non-tax-exempt buildings could include but were not limited to new units built due to upzoning or units built by developers who do not claim the tax exemption due to its conditions. The source of variation comes from the exogenous difference in tax levels before and after July 2008 in exclusion regions for units whose developers claimed the tax exemption. Notably, if a developer constructed a non-tax-exempt unit in an exclusion region, the developer’s tax levels did not change due to the reform.

To estimate the long-run impacts of the reform on new residential investment, we use a difference-in-differences (DiD) methodology. This approach allows us to consider both the boom and bust in construction caused by the grace period of the reform and to estimate the net effects in the long run. We leverage the fact that the reform made the requirements to obtain the tax exemption very stringent in exclusion regions, while the new requirements in non-exclusion regions were very lenient—only a cap on the size of the exemption. We consider tracts in non-exclusion regions as our comparison units and tracts located in exclusion regions as our treated units. The source of variation comes from two features of the tax reform. First, census tracts in exclusion regions experienced a major reduction in developers’ rate of return to investment caused by the tax policy reform, mainly due to the requirement to build affordable housing on-site. Second, census tracts in non-exclusion regions experienced only a minor reduction in the rate of return since the reform only capped the exemption and did not impose a requirement to build affordable units on-site. Our DiD approach considers exclusion regions as strongly treated and non-exclusion regions as lightly treated. Thus, our DiD estimates are likely to be a lower bound since the ideal comparison units should be tracts that experienced no change in the conditions to use the tax exemption.

Using the bunching methodology, we find that the reform significantly increased new rental residential construction in the short run. The excess bunching of housing starts in the time notch is 10,735 housing units (273 buildings), and this estimate corresponds to 1% of NYC’s total stock of rental units in 2007. Our results suggest that residential investment is responsive to the property tax reform in the short run. Additionally, our estimate of excess housing starts remains similar when we use as our empirical counterfactual alternative distributions of investment that were not impacted by the tax reform, e.g., housing starts in non-exclusion regions, ineligible units and non-housing construction starts. According to our bunching estimates, around 80% of the excess new units are concentrated in Manhattan and Brooklyn; however, relative to their existing housing stock, these boroughs show the smallest increases in housing due to their limited land availability. Differently, Queens accounts for only a quarter of the excess units, but it experienced the largest increase in units relative to its existing housing stock, largely due to its greater land availability.

Using the DiD methodology, we find that the reform also increased residential investment in the long run. Specifically, our DiD estimates show that tracts in exclusion regions have 0.51 to 0.81 (21.25 to 22.75) more new buildings (new rental units) relative to non-exclusion regions. These estimates represent 0.3x to 0.4x (2.4x to 2.5x) the median number of new buildings (new rental units) per census tract per year in the sample period. When we consider estimates for census tracts that are close to each other, that are divided by a boundary in the borough of Brooklyn, and that are thus exposed to similar time-variant local shocks, our estimates show that census tracts in exclusion regions have 2.25 to 2.78 (34.55 to 48.95) more buildings (rental units) than do tracts in non-exclusion regions. Notably, these estimates are sizable since they are 1.1x to 1.5x (3.8x to 5.4x) the median number of new buildings (rental units) per census tract per year in the sample period.

Taken together, our DiD results suggest that the temporal construction boom before the implementation of the property tax reform was stronger than the subsequent temporal construction bust in exclusion regions. In net terms, the reform led to a positive impact of residential investment in the long run for these regions. This finding implies that even if the grace periods of property tax reforms produce a boom-and-bust scenario, they do not necessarily induce a net negative impact on residential investment. Unfortunately, our DiD estimates cannot speak to the impacts on non-exclusion regions. Notably, our DiD model requires a comparison group within NYC, non-exclusion regions, and therefore, the model cannot shed light on how the reform impacted them.

With its high rents and property taxes, NYC provides an ideal setting for this study. Our findings are relevant for other dense, land-scarce cities with high housing costs, including major urban centers in the United States (San Francisco and Miami), Europe (Paris and London), Asia (Hong Kong and Singapore), and Latin America (Mexico City and Sao Paulo). NYC’s property tax rates on rental buildings are the highest in the United States (Lincoln Institute of Land Policy and Minnesota Center for Fiscal Excellence, 2018); the average property tax rate as a share of assessed valuation is approximately 3.3% on rental buildings, compared with 0.6–1% on owner-occupied units. This high rate creates frictions for investors that translate into high costs for residential investment, a characteristic shared with other cities. Our findings show that even in land-scarce cities, tax policy can stimulate residential investment in both the short run and the long run. Notably, in recent years, urban economics research has focused more on how land regulation impacts housing supply, while tax policies have been relatively understudied.

We study both the short- and long-run impacts of a property tax reform because both horizons are policy relevant. Five motives explain this relevance. First, there is persistent pressure to increase housing supply in cities with high rents. Second, residential investment is a key driver of local business cycles; thus, understanding how property tax policies shape investment under both horizons sheds light on how local tax policy influences those cycles. Third, property tax reforms commonly include a grace period postimplementation, which can produce a time bunching and can have consequences under both horizons—yet the grace periods of property tax reforms remain understudied. Fourth, a poorly designed reform can force developers to accelerate investment, only to contract sharply thereafter, generating a boom-and-bust cycle; studying both horizons is therefore essential for designing reforms that prevent or attenuate such dynamics, if needed. Fifth, a surge in short-run investment can shift the expectations of local business owners and induce growth in business amenities.

Why is the presence of a grace period—which can lead to time bunching—common across American cities when they reform their property taxes? There are two reasons for this. First, in some cities, property tax reforms require the approval of multiple government bodies such as the state legislature and the county board or city council. Moreover, in some cases, citizens vote on these reforms. Therefore, the approval process creates a grace period in practice since the first and last approval steps do not occur simultaneously.Footnote 3 Second, many property tax reforms contain de jure grace periods to allow participants in the housing market to adjust. Moreover, judicial processes or revisions to the reforms can extend de facto this grace period.

Our manuscript faces two limitations. First, we do not address the welfare implications of the policy. Addressing these implications would require a dynamic spatial general equilibrium model. Thus, the welfare impacts of the policy fall beyond the scope of this paper. Second, we do not take a stance on whether the short-run impacts or the long-run impacts matter more from a policy perspective. The answer to this question depends on both the preferences of voters regarding housing supply and the preferences and constraints of policymakers. Addressing this political economy question is also outside the purview of our analysis. Nonetheless, from an exclusively empirical perspective, we believe that it is convenient to estimate both the short- and long-run effects of increases in property taxes on construction to identify whether the investment boom induced by a grace period dominates the investment bust caused by the increase in property taxes.

We abstract from the impacts of eliminating the property tax exemption on affordability. In a different paper, we provide evidence on how this tax policy change raised rents and induced gentrification across NYC neighborhoods (Singh & Baldomero-Quintana, 2026). Moreover, Horton et al. (2024) provide a recent summary of the impact of the levels of property taxes on housing affordability across U.S. states and households by income.

Literature review. Research on the impact of property taxes on residential investment is a classical question in public finance. There are several theoretical studies on this topic (Arnott, 2005b; Arnott & Petrova, 2006; Capozza & Li, 1994; McFarlane, 1999; Turnbull, 1988). Early empirical studies focused on whether a two-rate tax system—where land is taxed at a higher rate than structures are—led to higher levels of construction activity relative to a one-rate tax system. These studies produced contradictory results (Ladd & Bradbury, 1988; Mathis & Zech, 1982; Tideman & Johnson, 1995; Wassmer, 1993). Two later studies used count data methods and concluded that taxing land at higher rates than structures is associated with higher residential investment (Lyytikäinen, 2009; Plassmann & Tideman, 2000).

We make four contributions to this literature. First, we use a quasi-natural experiment to estimate the causal impact of property tax exemptions on residential investment. Second, we study property tax exemptions, which are a popular policy tool in U.S. cities. Third, we analyze both the short- and long-run impacts of changes in property taxes on residential investment. Fourth, we study how the grace period of a property tax reform impacts residential investment, a feature that has been understudied in previous work. While Best and Kleven (2018) is an exception, they study the grace period of a housing transaction tax change. To our knowledge, no previous work has focused on the grace period of property tax reforms.

The paper closest to our manuscript is Lutz (2015). He provides causal estimates of the impact of changes in property taxes on residential investment by studying a New Hampshire education reform that reduced property taxes. The reform increased residential investment in municipalities outside the Boston metro area but had no impact in municipalities within the metro area. We contribute to this strand of work in three ways. First, we focus on new rental units, while he analyzes new single-family homes. Second, our policies differ: he focuses on overall property tax levels across suburban and rural areas, while we focus on changes to the requirements to use tax exemptions that stimulate the construction of new rental homes in a dense city with high rents. Third, we carefully examine the boom-and-bust trend caused by the grace period of a tax reform.

Our work relates to work on how property taxes impact different dimensions of the housing market (Carroll & Yinger, 1994; England, 2016; Oates, 1969; Orr, 1968). Nevertheless, finding exogenous sources of variation in property taxes is challenging. In this regard, some recent work includes Loffler and Siegloch (2021), Horton (2023), and Lutz (2015). We contribute to this literature in two ways. First, we provide evidence that economists can use a time notch when a property tax reform has a grace period, which is a common feature of property tax reforms in American cities. Second, we study property taxes in a highly dense city.

Local tax policies can have impacts on outcomes such as the tax policies of neighboring local governments (Agrawal et al., 2022, 2024, 2025a, 2025b; Lyytikäinen, 2023), workers’ state of residence (Agrawal & Chen, 2026), the volatility of fiscal revenues (Seegert, 2016), the progressivity of states’ tax system (Fleck et al., 2025), spatial misallocation (Fajgelbaum et al., 2019), and the spatial distribution of wages, employment, and capital (Braid, 2002). Moreover, intermunicipal cooperation can also impact residential investment and public goods provision (Tricaud, 2025). We contribute to this literature by providing causal evidence that local tax policies impact residential investment in the short and long run. Thus, local tax policy is a key determinant of housing supply overall. In recent years, urban economists have empirically understudied how local tax policy impacts housing supply since most attention has been focused on land regulation (Gyourko & Molloy, 2014; Quigley & Raphael, 2005) and land availability (Saiz, 2010).

This paper is organized as follows. Section 2 describes the property tax reform and the data. In Sect. 3, we present a model of time bunching in residential investment. Section 4 presents the econometric models. Section 5 presents our results, and Sect. 6 concludes.

Fig. 1
Fig. 1
Full size image

421a property tax reform, 2006–08. The figure shows the changes in exclusion regions that were part of the 421a property tax exemption 2006–08 reform. The light blue shaded region indicates the original exclusion region, where prior to the reform, developers were required to provide affordable housing, either off-site or on-site, to be eligible for 421a benefits. The dark blue unshaded regions indicate newly added new exclusion regions. After 2008, developers in these newly added regions could obtain only a long exemption, which required the provision of on-site affordable housing. In addition, 421a developers in the old exclusion regions had to provide affordable housing on-site, a more expensive requirement. Together, the light and dark blue regions formed the exclusion regions and were to see a loss of generous tax benefits beginning in 2008. Note that the brown regions in Manhattan and Brooklyn represent the “Hudson Yards” and “Willamsburg” regions, respectively, where special provisions applied

2 Background and data

2.1 Property tax reform in New York City

The 421a program. The key change in the property tax on new residential investment in NYC was caused by a reform to a tax exemption called the 421a tax program. This policy exempted developers of new construction from increases in property taxes. In general, new construction increases the assessed valuation of a property (including land and structures), which leads to higher tax payments even when the tax rates are unchanged. In contrast, the 421a property tax exemption provides tax relief for new investment by restoring the assessment to the preconstruction value.Footnote 4 All new project starts on “underutilized” land with at least three proposed units were generally eligible for the exemption (§6-02 of Title 28 of the Rules of the City of New York). This policy was the most extensive tax expenditure program in NYC. For example, in 2015, foregone taxes on tax-exempt units cost the city $1.1 billion, approximately 15% of the city’s expenditures.

The 421a program originated in 1971, when NYC was suffering from steep economic and physical decline, leading to a flight to the suburbs and a reduction in the quality of housing. In response, the state legislature enacted the 421a Real Property Tax Exemption in 1971 to stimulate new multifamily housing development. Between 1971 and 1984, approximately 30% of construction claimed the 421a exemption (The New School for Public Engagement, 2014). The program remained popular in the 2000s: Fig. A-1.2 shows that more than half of the units built each year between 2001 and 2012 claimed the tax exemption.

Since its origin in the 1970s, the program has divided the city into two types of regions. In the first type, developers could claim the tax exemption as of right, i.e., without any conditions. In the second type, developers were excluded from the tax exemption unless they built affordable units. These regions were called non-exclusion and exclusion regions, respectively, due to the existence of this geographical exclusion rule.

Developers could claim a short exemption (10–15 years) or a long exemption (20–25 years). The choice between short and long exemption depended mainly on the provision of affordable buildings (See Table 1 for specific details by region). The choice of 10 vs. 15 years or 20 vs. 25 years depends on the construction of a building in specific neighborhoods. In most cases, developers could claim more years, for example 15 instead of 10, if they built in neighborhoods experiencing decline and used the 421a program in combination with other programs such as the Neighborhood Preservation Program or the Rehabilitation Mortgage Insurance Corporation.

The property tax is a significant cost for developers in NYC. The statutory property tax rate on a rental building with at least 10 units in NYC is 5.5% of the annual market valuation, which is equivalent to 15% of annual rental income (see online Appendix A-3 for details on this calculation). Thus, an exemption that reduces the tax bill is attractive. Notably, the annual market valuation is assessed by the Department of Finance for taxation purposes, as opposed to the actual sale value of the property on the market. As Watson and Ziv (2026) and Singh and Baldomero-Quintana (2026) argue, there are strong incentives for both the NYC Department of Finance and the owners of rental units to ensure that the annual market valuation is accurate. Both the Department of Finance and landlords can provide evidence to correct an imprecise market valuation.

The property tax exemption stimulates investment by increasing the net-of-tax return, which occurs through two channels. First, the availability of the tax exemption directly lowers the tax liability for a significant period, which increases the return to new investment. Second, developers often face liquidity constraints, as suggested by the high volatility of housing starts: high wait times constrain the start of new projects, as developers must sell their current project to obtain cash for the next project, and they resort to a “fishing for liquidity” strategy (Stein, 1995; Topel & Rosen, 1988). Notably, the increased market valuation of a proposed project upon receipt of the tax exemption can help developers raise capital and build more. There is a similar phenomenon in the case of highly leveraged homeowners, who are more likely to increase the trade in the existing housing stock in the presence of a reduction in transaction taxes (Best & Kleven, 2018).

There are several reasons why a developer may not claim the exemption. First, developers cannot claim the exemption for new residential investment due to upzoning. Second, when developers have to provide affordable units on-site, their costs increase. Third, even when developers are not required to provide affordable units, market-rate units in a project are also subject to rent stabilization and rent caps (Cohen, 2009), thus reducing revenue. Fourth, project starts on land that previously contained an income-producing property are not eligible for the exemption. Fifth, buildings that have rent caps or rent stabilization are subject to city inspections, which represent significant hassle for developers since the NYC government is tenant friendly. Sixth, developers avoid providing on-site affordable units since these units are considered a disamenity by prospective high-income tenants, the developers, and the owners of nearby buildings. Moreover, if developers want to sell their property, buildings with rent stabilization are less attractive to potential buyers (Autor et al., 2014). Finally, projects with at least three or four residential units—depending on whether developers build before or after the reform—are eligible for the tax break, which effectively limits the tax break to condos and rental buildings.

The 2006–2008 reform. We provide a description of the reform in Table 1. The reform can be summarized into four main changes. The first is the geographical expansion of exclusion regions, which have affordability requirements to claim the 421a exemption. Second, there are stricter affordability requirements for all exclusion regions. The main change in this regard was that developers must provide affordable units on-site to claim the tax exemption. Third, the reform introduced a new light requirement in non-exclusion regions. In these regions, developers could still claim the exemption as of right, but the exemption was subject to a cap. Fourth, the reform implemented new modest requirements that were applicable to both exclusion and non-exclusion regions.

The first main change brought by the reform is the geographical expansion of exclusion regions. As shown in Fig. 1, the old exclusion regions included only central Manhattan. The new exclusion regions include the entirety of Manhattan and neighborhoods in all other four boroughs. Notably, after the implementation of the reform, both old and new exclusion regions were subject to the same new requirements. The second main change is the imposition of stricter affordability requirements in all exclusion regions, both old and new. Before the reform, developers in exclusion regions could provide affordable units off-site and claim the tax exemption. This was beneficial for developers since units in exclusion regions were desirable for high-income tenants and providing affordable units on-site represented a high opportunity cost. The reform eliminated the option to provide affordable units off-site.

The first two main changes have different but profound implications for both old and new exclusion regions. For the old exclusion regions, the reform led to a major reduction in the benefits of the tax exemption since developers had to provide affordable units on-site, whereas before the reform, developers could provide affordable units off-site. Notably, the old exclusion regions are in central Manhattan, where high-income neighborhoods are located. Thus, this new requirement reduced the appeal of the tax exemption. According to our calculations in online Appendix A-4, the on-site affordability requirement in the old exclusion regions is equivalent to an effective tax rate of 2.35%. The implications for the new exclusion regions are stronger. Before the reform, these regions could claim a short exemption as of right. The provision of affordable units was optional: it allowed developers to obtain a long exemption instead of a short exemption. However, after the reform, the provision of affordable units was obligatory. As we show in online Appendix A-4, in the new exclusion regions, the reform increased the effective tax rate by between 0% and 5.5%.Footnote 5

The third main change brought by the reform is the introduction of a new mild condition to claim the tax exemption in non-exclusion regions. Before the reform, developers could claim a short exemption without any conditions, or they could claim a long exemption by building affordable units. After the reform, developers could still claim the short exemption as of right, but there was now a cap on the tax exemption of $65,000 per unit. The option to claim the long exemption by providing affordable units remained in place after the reform, but it generated a new extra benefit whereby the cap did not apply.

The fourth main change introduced by the reform was the establishment of new minor criteria to claim the tax exemption for both exclusion and non-exclusion regions. Among them, buildings had to have at least four units; all affordable units were rent stabilized for 35 years, whereas before the reform, affordable units were rent stabilized for only the duration of the exemption; and affordable units were required to have either a total number of bedrooms comparable to market-rate units or a specified mix.

The reform and time bunching. The time notch arose from the delay in the implementation of the tax change. In February 2006, discussions to reform the exemption policy began with the creation of a task force that identified additional neighborhoods for inclusion in exclusion regions.Footnote 6 This task force was composed of participants in the housing market in NYC: developers (for profit and nonprofit), housing advocates, lenders, City Council representatives, and federal and local public officials. The task force submitted its final report to the City Council in October 2006 containing the areas for which exemptions would be eliminated. However, the report faced political resistance: there was pressure to add more neighborhoods to exclusion regions. As a result, both the city and the state legislature added more new areas to these regions. Figure 2 shows the initial regions picked by the task force (shown in red) and the regions added by the City Council (blue) and the state legislature (yellow). The mayor signed the local law in December 2006, and the legislature expanded/amended the local law.

The final version of the new 421a law was signed by the governor on February 19, 2008, and the law became effective on July 1, 2008. We refer to the period between the last quarter of 2006 and June 2008 as the time notch, where the “notch” arises from the fact that investment in the notch period is associated with a strictly lower property tax compared to investment after the notch period. A developer who obtained a permit and began excavation to install a load-bearing structure strictly before July 1, 2008, was eligible for exemption benefits under the old law. There is a caveat regarding this requirement. In the law, a construction start is defined loosely. Thus, in practice, the mere act of starting excavation for the future installment of a structure is considered a construction start. In addition, construction starts are self-reported and are thus subject to manipulation by developers. As a consequence, the permit date becomes nearly as good as the construction start date in practice, if not in law, since developers around the end of the time bunching period have strong incentives to (a) start basic excavation work the same day they obtain the permit and (b) manipulate the date of the construction start.Footnote 7 Given that NYC inspectors do not know the precise construction start date and that permit records do not contain the self-reported construction start date, the permit issuance date is the best proxy for the construction start date. Figure A-1.3 outlines the steps that a developer must go through before obtaining a permit.

For our time bunching econometric analysis, we focus on excess tax-exempt rental units and buildings in exclusion regions (old and new). As discussed above, the new requirements to claim the tax exemption in these regions, implemented in July 2008, represent an increase in the effective tax rate mainly due to the new requirement of providing affordable units on-site. Our counterfactual is non-tax-exempt rental units or buildings in the same exclusion regions. These units are unaffected by the reform even though they are located in the same regions. Notably, developers of rental buildings in exclusion regions also face some incentives or constraints to avoid claiming the tax reform, as discussed in detail before, such as the construction of new buildings due to upzoning; the opportunity cost of affordable units; the desire to avoid rent stabilization and city inspections; the unsuitability of land for the tax exemption; the disamenity caused by on-site affordable units; the difficulty of selling buildings with rent stabilization units; and the fact that rental buildings are small and thus cannot claim the exemption.

The reform and difference-in-differences. We use the signing and announcement of the reform by the mayor in Q4-2006 as our treatment. The announcement changed the expectations of developers regarding the rate of return to investment in exclusion regions since it became public knowledge that new stringent affordability requirements would be established to claim the tax exemption (Cohen, 2009). These requirements were costly and reduced the appeal of the exemption, as indicated by the second main change of the reform. Thus, starting in Q2-2006, developers expected that the reform would lower the rate of return to residential investment in these regions.

Thus, census tracts in exclusion regions represent our treatment units. Since the new condition for non-exclusion regions was lenient—mainly a cap on the size of the exemption as indicated by the third main change brought by the reform—we consider census tracts in non-exclusion regions as our comparison units. Notably, the new provisions for the entire city are also modest, as indicated by the fourth main change brought by the reform, and these provisions applied to both exclusion and non-exclusion regions.

Table 1 Changes brought by the 2006–08 property tax reform
Full size table
Fig. 2
Fig. 2
Full size image

Political process in the selection of exclusion regions. This figure illustrates the political process through which new exclusion regions (or dark exclusion regions in Fig. 1) were decided by the City Council and by the State Legislature. Data source: Pratt Center for Community Development (2011)

2.2 Data

We collected and organized several datasets from city government agencies. We provide detailed information on the sources of these datasets in online Appendix A-2. Here, we provide a brief summary of the datasets:

  1. 1.

    NYC Department of Finance Records. We obtain data from the Primary Land Use Tax Lot Output - Map (MapPLUTO) and PLUTO website in March 2018. PLUTO provides geocoded locations and details about all buildings on each parcel in the city. Each parcel has a unique identification number, BBL, which identifies the borough, block, and lot. The dataset contains information on the building classification, the number of residential and commercial units, and the actual assessed value (equal to the assessment ratio times the market valuation).

  2. 2.

    NYC Department of Buildings, Permits Dataset. An important measure of the flow of residential investment is the issuance of permits for new construction. The Department of Buildings provides data on the universe of permits issued at a monthly level from 2003. Notably, the use of permits to measure residential investment is common in the literature, both for the case of the U.S. (Lutz, 2015; Plassmann & Tideman, 2000) and Europe (Tricaud, 2025). The type of permit in the dataset depends on the nature of alteration sought. Relevant permit types for new construction are demolition (DM), new building (NB), and major alteration that changes the occupancy of the building (A1). Each permit is linked to a building identification number (BIN). We use the BIN and a crosswalk between the BIN and the parcel identification number, BBL, provided by the property address directory to merge this dataset with MapPLUTO (see Fig. A-28). We exclude Staten Island because it largely consists of family homes and housing starts for rental buildings are very few and noisy.

  3. 3.

    421a Beneficiaries List. Buildings/units that benefit from the 421a tax exemption are identified through their BBL using 421a lists provided by the Department of Finance. Although these lists are exhaustive, they do not identify whether the beneficiary obtains a short or a long exemption. For this reason, we use the 2015 property tax returns of owners residing in these buildings to obtain the type of exemption claimed.

3 A model of time bunching in residential investment

Because the increase in property taxes on new residential investment in 2008 was known in advance by developers, it created incentives to move their investments ahead in time. In this section, we use a simple framework to show that with the uncertainty and irreversibility of investment, an increase in the future property tax stimulates current residential investment by reducing the return to waiting. The forward movement of investment provides exogenous variation in the arrival time of new projects.

Consider the case in which the investment decision for developers is characterized by irreversibility and uncertainty of future return (Dixit & Pindyck, 1994). If things go badly, developers cannot destroy the building and fully recover the initial cost of investment. This impossibility implies that the decision to invest in any period is equivalent to exercising an option. In each period, developers choose to invest or wait, given the uncertainty of future returns. Waiting is profitable when developers expect future prices to be higher. In contrast, a future increase in the property tax stimulates current investment by decreasing the return to waiting. Consider two periods: today and tomorrow \(j\in \{0,1\}\). Developers decide in period 0 whether to invest today or tomorrow. Assume the period 0 capitalized return is known to be \(R_0\). The next-period return is distributed according to the function \(G(\tilde{R})\). Period 1 returns are discounted at the rate \(\beta\).

Developer i has perfect knowledge of investment cost \({C_i}\), which includes land acquisition expenditures and the costs of material and labor. These costs are heterogeneous across developers and are distributed according to a cumulative distribution function \(F(\cdot )\). It is reasonable to assume that land and zoning regulations (Gyourko & Molloy, 2014) and geographical constraints (Saiz, 2010) affect the cost distribution. For instance, we can imagine that zoning regulations make the variance in the probability distribution low. Only a few low-cost developers are able to invest when most developers have high costs.

With no time notch and no property tax, developer i invests in period 0 if (i) the net present value of investing today is greater than the net present value of investing tomorrow and (ii) the capitalized current return \(R_0\) is greater than costs \(C_i\). If he invests today, his return is \(R_0-C_i\). If he waits until tomorrow, he can expect a return of \(\beta E_G \max (\tilde{R}-C_i, 0)\). Clearly, he invests in period 0 when

$$\begin{aligned} R_0-C_i>\beta E_G \max (\tilde{R}-C_i, 0) \end{aligned}$$
(1)
$$R_0-C_i>0.$$

For developers with a period 0 return lower that the cost, the period 0 vs. 1 investment trade-off is trivial, and their period 0 investment does not respond to any changes in the distribution of period 1 returns. The reason is that when \(R_0-C_i<0\), the investment condition requires \(\beta E_G \max (\tilde{R}-C_i, 0) <R_0-C_i\), which is less than 0. For other developers, the decision to invest today is a function of (i) the investment cost \(C_i\), (ii) the distribution of period 1 returns \(G(\cdot )\), and (iii) the discount factor \(\beta\).

When \(\beta <1\), there exists a cutoff \(\tilde{k}\) that satisfies Eq. 1 such that when \(C_i<\tilde{k}\), developer i invests in period 0.Footnote 8 Intuitively, we can see that \(\tilde{k}\) is a function of the returns in period \(R_0\), the discount factor \(\beta\), and the distribution of period 1 returns \(G(\cdot )\). A higher return in period 0 makes it worthwhile for developers to invest in period 0. When \(R_0\) increases, the cutoff \(\tilde{k}\) decreases. With a higher \(\beta\), developers are patient and willing to wait until period 1 to reap any future rewards. When \(\beta\) increases, the right-hand side increases, and \(\tilde{k}\) falls. Therefore, period 0 investment falls. Finally, a higher mean of period 1 returns lowers period 0 investment. For instance, an increase in the period 1 mean increases the right-hand side of Eq. 1 and lowers \(\tilde{k}\), which reduces period 0 investment. Similarly, a mean-preserving spread that increases the variance in \(\tilde{R}\) reduces period 0 investment by lowering \(\tilde{k}\). The reason is that with a higher variance, larger \({\tilde{R}}\) values are more likely; thus, waiting is worth more. The total period 0 investment is given by

$$I(0)=\sum _{i=1}^{N}\mathbbm {1}(C_i<R_0 \bigcap C_i<\tilde{k})=NF(\min (R_0, \tilde{k})).$$

The time notch introduced a property tax t on the period 1 return, which implies that developers’ net return in period 1 falls to \(\frac{\tilde{R}}{1+t}\) should they choose to wait. This increases \(\tilde{k}\) and, thus, period 0 investment. Investment with a tax is given by

\(I(t)=NF(\min (R_0, \tilde{k_t})),\) where \(\tilde{k_t}>\tilde{k}\) and, therefore, \(I(t)\ge I(0)\). The increase in period 0 investment corresponds to the excess bunching in the time notch.

$$\begin{aligned} B=I(0)-I(t)=NF( \min (R_0, \tilde{k_t}))-NF(\min (R_0, \tilde{k})). \end{aligned}$$
(2)

If we normalize quantity 2 with I(0) and the property tax change t, we obtain the timing elasticity of residential investment with respect to the property tax. The reduced-form estimate is given by

$$\begin{aligned} \epsilon ^t=\frac{\hat{B}}{I(0){ t}} \end{aligned}$$
(3)

where \(\hat{B}\) represents the empirical estimate of excess bunching in the notch period.Footnote 9 It is important to distinguish two kinds of elasticities: the timing elasticity, which is the response of current investment to the future increase in property tax, and the long-run elasticity, which is the response of long-run investment to a permanent property tax increase, given by

$$\begin{aligned} \epsilon ^l=\frac{\frac{\Delta I}{I}}{{ t}}=\frac{F(R_0)-F(\frac{R'_0}{1+t})}{\frac{\Delta t}{t}F(R_0)}, \end{aligned}$$
(4)

where \(R'_0\) is the equilibrium tax-inclusive rent with a nonzero property tax. In general, elasticities 3 and 4 are not equal.

4 Empirical model

In this section, we describe two models that are used to estimate the short- and long-run impacts of the property tax reform on residential investment. First, we use a time bunching method to estimate the short-run impacts. Second, we use a DiD model to estimate the long-run effects.

4.1 Short-run effects: bunching

Econometric model. We estimate excess bunching units by running a simple regression on the count data. The regression is as follows:

$$\begin{aligned} c_{t}=\beta _0+\gamma _t+\beta _1 (Tax\_Exempt)_t \times (Notch_t) + \beta _2 (Tax\_Exempt)_t+ \epsilon _t \end{aligned}$$

where t denotes the quarter, \(c_t\) denotes the aggregate quarterly starts in quarter t, \((Tax\_Exempt)_t\) indicates the total tax-exempt housing starts in quarter t located in exclusion regions, and \(Notch_t\) is a dummy variable that is equal to one for all quarters in the period from Q4-2006 to Q2-2008. The omitted dummy is the indicator for the counterfactual, which in this case includes non-tax-exempt housing starts that are also in exclusion regions. Examples of non-tax-exempt units include new residential investment due to upzoning. The parameter of interest, \(\beta _1\), provides an estimate of the average per-quarter excess bunching during the time notch, which is scaled by 7 to estimate the total bunching observed in the time notch that lasted seven quarters. The main identification assumption is that in the absence of the time notch, the distribution of the treatment group (tax-exempt units in exclusion regions) is similar to the distribution of the comparison group (non-tax-exempt housing starts in exclusion regions).

Where does the source of variation for identifying the parameter of interest come from? The announcement of the policy reform in 2006 changed the future rate of return for developers who were constructing tax-exempt units located in exclusion regions. That is, for these buildings, obtaining a construction permit before July 2008 represents a lower tax level vs. obtaining a permit in or after July 2008. Differently, the counterfactual units consist of non-tax-exempt buildings also located in exclusion regions. Notice that even if the tax policy reform has a strong impact on these regions, it does not affect non-tax-exempt units. Table 2 summarizes the information on the group and comparison units and the source of variation that allows us to use the bunching method. Table 3 displays the characteristics of the comparison and treatment buildings. Our t-tests provide suggestive evidence that both groups have similar characteristics such as the building area, the number of units, rent, the lot area, and the probability of being in a high-income region.

Standard bunching estimation in public finance deals with bunching in housing prices (Best & Kleven, 2018; Kopczuk & Munroe, 2015). Bunching estimates are calculated as the difference between the actual distribution and the counterfactual in the notch region (or time, in the context of our paper). Similarly, the difference between the actual distribution and the counterfactual to the right of the notch region is the missing mass. A key input in the calculation of excess bunching and missing mass estimates is the choice of counterfactual. One choice is a high-degree polynomial fit of the actual distribution that uses data outside the notch region. Estimation using this approach is sensitive to the choice of upper bound and the choice of lower bound, in addition to the choice of the degree of the polynomial (Kleven, 2016). One issue with extending this methodology to the time notch is that this counterfactual does not account for any global time shocks during the time notch. Therefore, similar to Best and Kleven (2018) and Kopczuk and Munroe (2015), we use an empirical counterfactual. Specifically, we pick a comparison group whose housing starts time series serves as the counterfactual for our treatment group, non-tax-exempt housing starts. The lower bound is fixed at the announcement date.

Table 2 Bunching: How the policy reform impacts comparison and treatment units
Full size table

We do not choose housing starts in non-exclusion regions (yellow region in Fig. 1) that are either tax-exempt or not-tax-exempt as our counterfactual for two reasons: (i) we ensure that the actual and counterfactual distributions represent the same geographic area; and (ii) we alleviate the concern that the recession, which started in 2007, had differential effects on housing starts in exclusion and non-exclusion regions. Notably, using housing starts in non-exclusion regions could overstate the bunching observed in exclusion regions if, for example, the recession had a different and larger impact on housing starts in non-exclusion regions. Moreover, there is evidence that foreclosures in 2008 were concentrated in non-exclusion regions (Fig. 3).

Fig. 3
Fig. 3
Full size image

Foreclosures in 2008 in New York. Figure from the NYU Furman Center Report 2008

Threats to identification. We have two threats to identification. First, to use the bunching method, the distribution of the comparison group and the distribution of the treatment group in the absence of a notch should be similar. Second, the presence of the housing crisis in 2008 may represent a threat to identification if the crisis impacted the comparison group and the treatment group differently. We discuss these two issues below.

Testing whether the distribution of the comparison group and the distribution of the treatment group, absent a time notch, follow the same distribution is difficult. However, we can provide suggestive evidence that this assumption is satisfied in our context. We implement a two-sample Kolmogorov–Smirnov test for equality of distribution functions. In this test, the null hypothesis is that the distributions are similar. We implement this test on three different cases: first, the distributions of new units for both the comparison and treatment groups before the time notch; second, the distributions of new units for both the comparison and treatment groups after the time notch; and, third, the distributions of new units for both the comparison and treatment groups over the entire sample period but excluding the time notch quarters. The Kolmogorov–Smirnov test results are presented in Table 4. For all cases, we cannot reject the null hypothesis that the treatment and comparison groups share the same distribution. Therefore, the test results suggest that our choice of comparison group is reasonable.

We explain why the 2008 financial crisis is unlikely to impact differently the comparison and treatment groups. There are three reasons for this. First, the housing segment that caused the 2008 financial crisis (single-family homes) is not the same as the housing segment that we analyze in this manuscript (rental buildings). The general agreement among researchers who have studied the 2008 crisis is that the subprime mortgages that originated the financial crisis were associated with single-family homes (Dell’Ariccia et al., 2012; Kochhar et al., 2009; Mayer & Pence, 2008). Moreover, in NYC the segment of single-family homes is small. Between 2000 and 2008, only 7.7% of new units in NYC were single-family homes (Furman Center, 2009). Thus, the impact of the 2008 financial crisis on the comparison and treatment groups occurs via macroeconomic conditions and the the conditions of the U.S. financial system, but it is unlikely that these two factors impact the comparison and treatment groups in an unequal manner.

Second, the regions where housing markets were affected the most by the crisis do not include NYC and are not near the city. The foreclosure rate was the highest and above the national rate in the following regions: the Southeast, the Southwest, Florida, parts of the Midwest, the DC metro area, and Atlanta. Notably, most of the cities where the boom and bust was the most intense are in the Southeast, the Southwest, and Florida.Footnote 10 Third, the Kolmogorov–Smirnov tests suggest that the comparison and treatment groups behave similarly when the period of high financial stress in the banking system started in 2008 due to the collapse of Bear Stearns. Hence, these tests suggest that the financial crisis impacted tax-exempt and non-tax-exempt buildings in similar ways. Notably, in this section, we do not discuss whether the 2008 financial crisis could bias our estimates, but we address this concern in the results section.

Table 3 Summary statistics for non-tax-exempt and tax-exempt parcels located in exclusion regions, 2015
Full size table
Table 4 Kolmogorov–Smirnov tests for equality of distributions: Comparison vs. treatment group outside the notch period
Full size table

4.2 Long-run effects: difference-in-differences

Exclusion vs. non-exclusion regions. We use a DiD model to estimate the impacts of the tax reform on new residential investment in the long run. Using a DiD model allows us to focus on the net impacts of the tax reform since we do observe a large boom and bust in construction before and after Q2-2008, as shown in Fig. 5.

Our DiD model compares census tracts in exclusion regions vs. tracts in non-exclusion regions. Although both regions have new requirements due to the policy reform, the new affordability conditions for obtaining the tax exemption in exclusion regions are quite stringent, whereas the new requirements for obtaining the tax exemption in the non-exclusion regions are relatively lenient—simply a cap on the tax exemption. We estimate the following model at the census tract level:

$$\begin{aligned} RI_{ct} = \theta _{0} + \theta _{1} 1[Exclusion \ region]_{c} 1[Post]_{t} + \omega _{t} + \phi _{c}+ \xi _{ct} \end{aligned}$$
(5)

where \(RI_{ct}\) represents new residential investment in census tract c during year t; \(1[Exclusion \ region]_{c}\) is a dummy variable for whether census tract c is in an exclusion region; \(1[Post]_{t}\) is a dummy variable that takes a value of 1 for those years during and after the announcement of the tax policy change when the mayor signed the reform, that is, during and after 2006; \(\omega _{t}\) represents time fixed effects; and \(\phi _{c}\) is census tract fixed effects. We cluster standard errors at the census tract level. In this model, the comparison group consists of tracts in non-exclusion regions, and the treatment group is composed of tracts in exclusion regions.

The parameter of interest is \(\theta _{1}\) and represents the impact of the reform on census tracts in exclusion regions with respect to tracts in non-exclusion regions. Importantly, exclusion regions experience a strong treatment (provision of affordable units on-site), whereas non-exclusion regions experience a light treatment (can claim a short exemption without providing affordable units but face a cap on the tax exemption). The ideal parameter of interest would compare treated regions with untreated regions, but there are no untreated regions for this tax policy change in NYC since even the non-exclusion regions experienced some minor additional regulations to claim the tax exemption. Nonetheless, \(\theta _{1}\) may represent a lower bound of the treatment effect since the ideal comparison units would not experience any tax policy changes. Unfortunately, our econometric model cannot provide any information on the impacts of the reform on non-exclusion regions.

What is the source of variation for identifying the parameter of interest? The announcement of the tax policy reform changed the expected rate of return for units built in tracts located in exclusion regions. This change was mainly caused by the new requirement to provide affordable units on-site if developers wished to claim the tax exemption. Differently, in non-exclusion regions, developers could still claim a capped short exemption as of right, while the provision of affordable units on-site extended the exemption in time. Notably, the provision of affordable units on-site substantially reduces the rate of return for developers, as shown by our calculations in online Appendix A-4. Moreover, affordable units on-site are considered a disamenity by some renters, they reduce the revenue that the building generates, and they expose the building to inspections by a tenant-friendly local government. Table 5 synthesizes the information on the treatment and comparison groups and describes the source of variation for identifying the parameter of interest.

Table 5 Difference-in-differences: How the reform impacts comparison and treatment units
Full size table

The identification of the parameter of interest requires parallel trends in the pretreatment period, i.e., before 2006. Figure 4 shows the results of an event study, where we graph the estimates for each year. The figure shows that the parallel trends assumption is satisfied: the estimates of the event study during and before 2006 are statistically insignificant. In addition, the estimates for 2004, 2005, and 2006 are almost zero and become positive during and after the announcement of the tax reform.

Fig. 4
Fig. 4
Full size image

Event study. Standard errors are clustered at the census tract level. The model considers both time and census tract fixed effects. The red and green lines indicate the start and end years of the time notch, respectively. The outcome variable is permits for new residential investment in a census tract (levels). We consider only buildings classified as rental walkups and rental buildings with an elevator

5 Results

Descriptive evidence of residential investment across time and space. Figure 7 plots the quarterly number of tax-exempt permits issued in NYC in the raw data, split by exclusion and non-exclusion regions. Dotted red and green lines indicate the quarter-year when changes to the policy were announced and implemented, respectively. The time stamp in the permit data indicates the date on which the permit for construction of a new building was issued. Importantly, these do not include tax exemption applications and reflect only approved permits. Moreover, as discussed above, the permit issuance date is a good proxy for a construction start date since the law considers a construction “start” the initial excavation to install load-bearing structures and developers have incentives to manipulate the reported construction start date.

Figure 6 shows the spatial distribution of tax-exempt noncondo projects. The top panel presents the distribution for seven quarters before the time notch. Each project is weighed by the number of residential units. Notably, tax-exempt projects in Manhattan tend to be fewer but larger; however, there is no specific spatial pattern. In contrast, the bottom panel shows the spatial distribution of tax-exempt noncondo projects during the time notch. We observe that more and larger tax-exempt projects are started in exclusion regions in this period and, in particular, in Manhattan and Brooklyn. This finding is consistent with the concentration of residential investment in exclusion regions during the time notch.

To understand what exemption type developers sought before the reform, Fig. 5 provides descriptive evidence across time by exemption type separately for exclusion and non-exclusion regions. We notice three patterns. First, as expected, the largest spike occurs for short exemptions in exclusion regions, and the size of the spike is close to 6,000 buildings. Notably, prior to the reform, developers in new exclusion regions could claim short exemptions as of right. Second, there is a spike in short exemptions in non-exclusion regions, but it is much smaller, approximately 900 buildings. Third, there are several increases in long exemptions during the time notch in exclusion regions, driven by the removal of the off-site long exemption in old exclusion regions (see Figs. 6 and 7).

Fig. 5
Fig. 5
Full size image

Quarterly tax-exempt housing starts by exemption type. This graph plots the number of permits issued for a building weighed by its units in each year for 421a properties in the region. The red and green lines denote the start and end quarter-years of the time bunching, respectively. The data include permits for new building (NB) construction. Source: Department of Buildings and PLUTO dataset

Fig. 6
Fig. 6
Full size image

Spatial distribution of tax-exempt projects in the time notch. This figure presents the spatial distribution of tax-exempt project starts before and in the time notch. Each project is weighed by the total number of residential units built. The top panel includes permits issued between Q2-2005 and Q4-2006. The bottom panel includes permits issued between Q4-2006 and Q2-2008 (the time notch). Data: Department of Buildings permits, 2001–15, and Primary Land Use Tax Output, 2002–16. Note: Blue regions represent old and new exclusion regions

Bunching results. Table 7 reports the bunching estimates for NYC and different boroughs, while Fig. 8 reports the excess bunching estimates for exclusion regions in NYC using non-tax-exempt housing starts as the counterfactual. We observe that while the two distributions are fairly parallel in the prereform period, there is a large spike at the notch for both tax-exempt units (left panel) and tax-exempt buildings (right panel). Reassuringly, there is no spike in non-tax-exempt starts, which supports the assumption of an absence of regional shocks that affect only non-tax-exempt units in exclusion regions in this period. Overall, in exclusion regions, there was excess construction of approximately 10,735 tax-exempt units and 270 buildings in the time notch. As shown in Table 7, these findings correspond to 1% of the total stock of rental units in NYC for 2007.

Figure 9 plots tax-exempt housing starts across four boroughs: Manhattan, Brooklyn, Queens, and the Bronx. Each graph displays the corresponding bunching estimate and standard errors in black. We exclude Staten Island since most new residential investment consists of single-family homes and new rental buildings are rare. We find that Manhattan, Brooklyn, and Queens see sharp increases in housing starts, despite the start of the recession in 2008. The magnitude of bunching varies across boroughs. Finally, our estimate is higher by 80% if we include the response of tax-exempt condo starts (Fig. 10a).

Fig. 7
Fig. 7
Full size image

Aggregate quarterly tax-exempt housing starts in New York City, exclusion and non-exclusion regions. This graph plots the number of permits for units in a quarter-year for tax-exempt properties in each region. The red and green lines denote the start and end quarter-years of the time bunching, respectively. The data include permits for new building (NB) construction. Source: Department of Buildings

Fig. 8
Fig. 8
Full size image

Bunching estimation of tax-exempt housing unit starts. Diamonds \(\blacklozenge\) denote the empirical distribution of tax-exempt housing starts. Black squares \(\blacksquare\) denote the counterfactual: the empirical distribution of non-tax-exempt housing starts in the blue region. The red and green lines denote the start and the end quarter-years of the time bunching, respectively. The top number \(B = value\) is the bunching estimate, and the lower number in parentheses is the standard error. Data: Department of Buildings permits, 2001–15

Fig. 9
Fig. 9
Full size image

Bunching of tax-exempt housing starts in exclusion regions by borough. Diamonds \(\blacklozenge\) denote the empirical distribution of tax-exempt housing starts. Squares \(\blacksquare\) denote the counterfactual: the empirical distribution of non-tax-exempt housing starts in the exclusion region. The red and green lines denote the start and the end quarter-years of the time bunching, respectively. The top number is the time bunching estimate, and the lower number in parentheses is the standard error. Data: Department of Buildings permits, 2001–15

Fig. 10
Fig. 10
Full size image

Bunching estimates: Robustness checks. Diamonds \(\blacklozenge\) denote the empirical distribution of tax-exempt housing starts. Squares \(\blacksquare\) denote the counterfactual: the empirical distribution of non-tax-exempt housing starts in the exclusion region. The red and green lines denote the start and the end quarter-years of the time bunching, respectively. The top number is the time bunching estimate, and the lower number in parentheses is the standard error. Data: Department of Buildings permits, 2001–15

There is substantial variation in relative housing starts across boroughs. Our estimates in Table 6 suggest that Manhattan had approximately 3,000 excess rental unit starts, Brooklyn 5,000, the Bronx 200, and Queens 2,500. These findings correspond to approximately 0.4%, 2.1%, 0.8% and 6.8% of the existing housing stock in each borough, respectively. The differential excess housing starts across boroughs could reflect borough-wide differences in the net benefits of the tax exemption after the reform since the new requirements for claiming the tax exemption made it mandatory to provide affordable housing on-site, the cost of which depends on rents that vary across neighborhoods. In addition, the bunching estimates seem to be negatively correlated with the density of the boroughs, as we document in Fig. A-1.7. Notably, the smallest relative increase in housing stock occurs in Manhattan, the densest borough and, thus, the borough with the greatest scarcity of land. The largest relative increase in housing stock occurs in Queens, the borough with the lowest density among the four examined and, thus, with the greatest availability of land.

Table 6 Summary statistics: New residential investment across census tracts
Full size table

We address three potential concerns that might impact the validity of our bunching estimates. The first concern in the excess starts estimation is the possibility that developers substituted non-tax-exempt housing with tax-exempt housing in the time notch. In such a case, using non-tax-exempt starts as a counterfactual leads to an overestimation of the true excess bunching. To alleviate this concern, we check robustness with respect to alternative distributions as the counterfactual. The estimates are fairly robust when we use either housing starts in non-exclusion regions (Fig. 10b) or ineligible units and nonhousing construction starts—family homes and commercial construction—in exclusion regions as the counterfactual (Fig. 10c). In either case, the excess bunching estimate is close to 10,500 tax-exempt rental units.

The second concern relates to the time to completion of the proposed projects. While the bunching estimates include only units that were finished by 2015, it is possible that tax-exempt units were not completed in a timely manner. This possibility is even more relevant because the recession started in 2008. In fact, a rule change in 2013 extended the "undue delay" period from 36 months to 72 months for projects that were subject to mortgage foreclosure or other lien enforcement litigation before May 14, 2012 (§11–245.1 of Title 2, New York City Administrative Code). To address this concern, Fig. 11 uses PLUTO data to identify the completion year of buildings whose construction began in the time notch. While a large share (approximately 92%) of the buildings were finished by 2010, a very small share of projects took more than three years to complete. Moreover, by 2013, almost all projects (approximately 99.1%) were completed.

The third concern is that the housing crisis could significantly affect the size of our bunching estimates if the crisis impacted the comparison and treatment groups differently. Specifically, if the housing crisis led to a stronger housing boom during the national housing cycle upturn in 2000–05 and a larger bust after the financial crisis started in non-tax-exempt buildings compared to tax-exempt buildings, this would bias our bunching estimates upward. Five considerations suggest that this concern is overstated. First, we do not observe heterogeneous trends for rental housing or buildings before 2006 or after 2008 for either tax-exempt or non-tax-exempt units, as shown in Fig. 8. Recall that the peak of housing construction is 2005, while the stress in the financial system starts in 2008. In other words, we do not observe a larger housing boom or a larger bust in the comparison group relative to the treatment group. The results of the Kolgomorov–Smirnov test before and after the time notch confirm this observation, as shown in Table 4. Second, as discussed above, the housing market segment that produced the 2008 financial crisis was single-family homes in cities in the South, the Southwest, and Florida, while we study rental buildings in NYC. Third, our estimates barely change when we use investment in commercial construction, which is unrelated to housing construction, as a comparison group.

Fourth, the bunching estimates from Queens—the borough hit hardest by the crisis—do not seem to be contaminated by the housing crisis. Notably, Queens is the borough where most foreclosures occurred. The sample from this borough could bias the bunching estimates upward if we observed major differences in the trends between tax-exempt and non-tax-exempt buildings during the time notch. Nonetheless, as shown in Fig. 9, residential investment in Queens during the time notch is mostly flat for both tax-exempt and non-tax-exempt buildings, except in Q2-2008, right before the time notch ends. That is, in the borough where the bunching estimate could be biased upward due to the high levels of foreclosures, the variation that produces the estimate comes from the increase in residential investment right before the reform implementation. Fifth, the 2008 financial crisis was a systemic event that impacted the entire financial system (Financial Crisis Inquiry Commission, 2011), unlike other financial events such as the savings and loan (S&L) crisis, the junk bond collapse, and the Long-Term Capital Management (LTMC) crisis, which impacted specific banks or financial market segments. That is, it is likely that the 2008 crisis impacted all developers alike, i.e., those who build tax-exempt buildings and those who build non-tax-exempt buildings. Moreover, these two types of buildings belong to the same market segment, building rentals in a dense city; thus, they are likely to have similar funding sources.

Fig. 11
Fig. 11
Full size image

Completion year of notch projects. The graph plots the distribution of the year of completion of tax-exempt projects that started in the time-notch. Data: Permit data matched with PLUTO data, 2002–16

DiD Main Results. Table 7 shows the estimates of the parameter of interest from Eq. 5, which represent the long-run impacts of the tax policy change on residential investment at the census tract level. When we measure residential investment by the number of buildings, our results suggest that census tracts in exclusion regions—which experience a “strong treatment”— increased residential investment by between 0.58 and 0.81 buildings. When we consider apartment units, we find that tracts in exclusion regions increased residential investment by between 21.25 to 22.75 housing units. How large are these estimates? Table 8 shows that in a census tract, the median number of new buildings (units) per year is 2 (9), for the period 2003–2015. Therefore, the estimates that we obtain represent 0.3x to 0.4x (2.4x to 2.5x) the median number of new buildings (units) per census tract per year in the sample period.

Table 7 Excess housing starts in the time notch
Full size table
Table 8 Long-term impacts of the tax policy reform: Difference-in-differences for all census tracts in NYC
Full size table

Notably, the estimates have large standard errors. This finding can be explained by the fact that most census tracts receive very few new buildings. Table 6 shows that in the period 2003–2015, a quarter of the census tracts receive one new residential building per year. Furthermore, the results of our event study in Fig. 4 show that even though we observe an increase in residential investment in the treatment group relative to the comparison group, after Q2-2008, there is considerable variation among census tracts.

Our DiD results yield two conclusions. First, although the property tax reform increased the requirements and reduced the net benefits of the property tax exemption, it produced a net positive impact on construction in the long run. Therefore, the boom during the time bunching dominated the bust after the implementation of the reform. Second, the grace period was key for the positive impact of the reform on residential investment, as it induced a large spike in construction since developers moved their investment forward in time. There is a caveat regarding our DiD results: we are unable to document what were the long-run impacts of the property tax reform on non-exclusion regions since we use them as the comparison group in our analysis.

Importantly, the estimates that we obtain are likely to be a lower bound of the true treatment effect. Notably, all census tracts in NYC are treated by the reform, but some census tracts experience a stronger treatment than others do because they are located in an exclusion region. Our econometric model compares census tracts in exclusion regions and tracts in non-exclusion regions. Therefore, the estimates of this model would likely be larger if the comparison group units were not exposed to any treatment.

DiD: exclusion vs. non-exclusion regions exposed to similar local shocks. We take advantage of a unique feature in tax policy: there is a boundary between exclusion and non-exclusion regions within the borough of Brooklyn. This policy features deliver two advantages. First, Brooklyn is the quintessential example of neighborhood change among American cities due to the large increase in residential investment in the borough during the 2000 s and the 2010s. Before these years, this borough was associated with minority and immigrant neighborhoods (Gonzalez, 2022, 2023). Second, since we are comparing neighborhoods that are close to each other, our regression accounts for unobserved time-variant local economic shocks impacting both the treatment and comparison groups. In a sense, this robustness check is similar to using a difference-in-discontinuities approach with the caveat that the discontinuity is constant across time.

We build the sample as follows. We choose treatment and comparison census tracts that are located d meters away from the boundary separating exclusion and non-exclusion regions, where \(d \in \{750, 1,000, 1,100, 1,300\}\). We drop census tracts that are intersected by the boundary. To illustrate this approach, in Fig. 12, we show a map of Brooklyn that includes the boundary, the treatment census tracts, and the comparison census tracts. In online Appendix Figs. A-1.4, A-1.5 and A-1.6, we include maps for treatment and comparison tracts that are located 1,000, 1,100, and 1,300 meters away from the boundary.

We display the results for this robustness check in Table 9 for new buildings and in Table 10 for housing units. Our results suggest that the tax policy change increased residential investment in census tracts that belong to exclusion regions near the boundary by between 2.25 to 2.78 (35 to 49) buildings (apartment units) per tract relative to census tracts near the boundary that are in the non-exclusion regions. This impact is 1.1x to 1.4x (3.9x to 5.4x) the median number of new buildings (apartment units) per census tract per year in NYC during the sample period, as shown in Table 6.

Interestingly, the estimates using the Brooklyn census tracts (Tables 9 and 10) are much larger than the main estimates (Table 8). Specifically, for the regressions using the number of buildings (apartment units) as the dependent variable, the estimates using the Brooklyn tracts near the boundary are 2.8x to 4.8x (1.5x to 2.3x) the size of the main estimates. This means that when we compare census tracts that are exposed to the same time-variant local shocks, the effect of the tax policy reform on residential investment increases in size.

Fig. 12
Fig. 12
Full size image

Exclusion and non-exclusion regions in Brooklyn: Treatment and comparison tracts located 750 meters away from the boundary. The green line indicates the boundary between exclusion and non-exclusion regions in the borough of Brooklyn

Table 9 Long-term impacts of the tax policy reform on new buidings per census tract in Brooklyn areas d meters away from the boundary
Full size table
Table 10 Long-term impacts of the tax policy reform on new housing units per census tract in Brooklyn areas d meters away from the boundary
Full size table

6 Conclusion

We estimate the short- and long-run impacts of a tax policy reform in NYC that eliminated tax exemptions for rental apartment buildings. In the short run, we find that the announcement of the future elimination of the property tax exemption leads to excess housing starts that correspond to 1% of the entire rental housing stock in 2007. In the long run, the policy results in sizable increases in the number of buildings and rental units.

Our findings yield two conclusions about property tax policies. First, local tax policy is a key determinant of housing supply, even in dense cities with land scarcity. Second, when a property tax is effectively increased, grace periods can induce a large short-term increase in housing supply by moving forward the planned investment of developers. This increase can be sufficiently large to dominate the construction bust after the property tax effectively increases.

There are two limitations of our study. First, we cannot estimate the impacts on nonrental apartments. Second, we do not measure the welfare impacts of the tax policy across different types of households. Future empirical research is needed to understand how changes in property taxes impact (i) nonrental housing units in dense cities and (ii) the welfare impacts on households across different income brackets.