Get documents about "Crowd Out, Stigma, and the Effect of Place-Based Subsidized Rental
Document Sample
Crowd Out, Stigma, and the Effect
of Place-Based Subsidized Rental Housing
By
Michael D. Eriksen
Department of Economics and Center for Policy Research
Syracuse University
Syracuse, New York 13244-1020
Phone: (315) 443-9054
Fax: (315) 443-1081
Email: meriksen@maxwell.syr.edu
and
Stuart S. Rosenthal
Department of Economics and Center for Policy Research
Syracuse University
Syracuse, New York 13244-1020
Phone: (315) 443-3809
Fax: (315) 443-1081
Email: ssrosent@maxwell.syr.edu
http://faculty.maxwell.syr.edu/rosenthal/
September 12, 2007
Funding for this project from the John D. and Catherine T. MacArthur Foundation, the Ford Foundation
and the Department of Housing and Urban Development is gratefully acknowledged. We thank Denise
DiPasquale, Gary Engelhardt, Jeffrey Kubik, Edgar Olsen, Erika Poethig, Michael Stegman, Bruce
Weinberg, Johnny Yinger, and seminar participants at the January 2007 AREUEA meetings and Ohio
State University for helpful comments. Any remaining errors are ours alone.
Abstract
The Low-Income Housing Tax Credit (LIHTC) program has ballooned into the largest ever source of
subsidized construction of low-income housing in the United States, accounting for one-third of all recent
multi-family rental construction. Recent proposals in Congress have sought to double the size of the
program. This paper examines the crowd out and stigma/amenity effects of this increasingly important
source of low-income housing. To do so, we apply a unique geographic approach to the data as well as
instrumental variable methods that facilitate identification.
Results indicate that within one-half mile, LIHTC development likely has a positive amenity effect in
low-income areas but a negative stigma effect in high-income areas. These effects attenuate with
distance. Expanding the geographic scope of the analysis to ten mile areas, one-third of LIHTC
development is offset by crowd out of private unsubsidized rental housing construction. This is
consistent with high levels of crowd out found in other markets (e.g. Berry and Waldfogel, 1999) and also
for earlier forms of place-based subsidized housing (e.g. Murray, 1999; Sinai and Waldfogel, 2005).
These patterns raise questions about the viability of the LIHTC program and should be taken into account
when considering its expansion.
Key Words: Crowd Out, Stigma, Subsidized Housing
JEL Codes: H7, H42, R21, R31
“I rise today to introduce the Affordable Housing Tax Credit Enhancement Act of 2005. … the
bill would double the current LIHTC [annual allocations], which would yield twice the number of
affordable units annually. … Today, the LIHTC program is widely regarded as the nation’s most
successful housing production program resulting in the construction and rehabilitation of more
than 1.3 million housing units for lower income households. …”
Statements Submitted to Congressional Record: May 26, 2005
By Rep. William Jefferson (D-LA)
1. Introduction
The manner in which housing assistance to the poor is provided remains subject to considerable
and even heated debate: should government invest in people through demand side voucher type programs
(e.g. Section 8 vouchers) or in places through supply side construction subsidies such as public and Low
Income Housing Tax Credit (LIHTC) housing? Against that backdrop, this paper examines the rapidly
growing LIHTC program and highlights two features that have received little attention: what are the
neighborhood-level stigma (or amenity) effects of LIHTC housing – a shifter of demand – and to what
extent does the construction of LIHTC housing crowd out unsubsidized private development – a supply-
side effect. To anticipate, our findings suggest that crowd out rates associated with the LIHTC program
are sufficiently high as to raise questions about the viability of the program, let alone recent proposals to
double its size. Some further context will help to put the LIHTC program in perspective.
Between the late 1930s and the mid-1980s, the federal government built over one million low-
income units through the public housing program. During the 1980s, construction of public housing units
ended and was replaced by the Low-Income Housing Tax Credit (LIHTC) program. Under different
variants of the LIHTC program between 30 to 91 percent of project construction costs are subsidized.
Those subsidies are financed by the Federal government which allocates tax credits to individual states
based on state population. State housing authorities then reallocate the credits to developers who agree to
adhere to rent ceilings and tenant income limits. From 1987 to 2005, over 1.3 million LIHTC units were
built accounting for roughly one-third of all recent multi-family rental housing constructed. This makes
the LIHTC program the largest supply-side housing subsidy program in the nation’s history. Figure 1
illustrates this point. The figure displays the level of public housing and LIHTC development by decade
over the last 60 years.1 The recent boom in LIHTC development is evident.2
To assess the crowd out effect of LIHTC housing on unsubsidized rental housing construction,
we combine census tract data from 1970 to 2000 with information on LIHTC units developed since 1987.
Identification of crowd out is based on a simple argument that relies on the geographic scope of the
analysis. Specifically, at the neighborhood level, LIHTC development is likely endogenous because site
selection is sensitive to anticipated capital gains (e.g. Baum-Snow and Marion, 2006) and sharp financial
penalties if a given project is unable to house sufficient numbers of low-income families (Eriksen, 2007).3
However, as the geographic scope of the analysis expands, substitution of new housing capital nets out
across neighborhoods, and LIHTC development becomes exogenous. This implies that non-IV models
should be consistent when the geographic scope of the analysis is sufficiently large. A further implication
is that OLS and IV models should yield different estimates of crowd out for narrow levels of geography,
but similar estimates for broader levels of geography. These and other patterns will be tested.
To implement this strategy, we estimate the impact of LIHTC development on new, unsubsidized
rental housing development. We do this separately for different levels of geographic scope, from one-half
mile to 10-mile radius circle areas. In all cases we control for MSA fixed effects that enable us to strip
away unobserved MSA-level attributes that might otherwise affect rental housing construction. In
addition, for each level of geography, OLS and GMM estimates are computed. For the latter, we
1
The public housing data used to create Figure 1 were obtained from analysts at HUD and with assistance from
MacArthur and Abt Associates. These data differ from the publicly available series found at www.huduser.org in
two ways. First, our data contain the year each “project” was placed-in-service from 1937 to 2000. This allows us
to represent new construction of public housing in each decade. Additionally, our data also includes information on
public housing demolitions in the 1990’s. The LIHTC data were obtained from HUD at http://lihtc.huduser.org.
2
The cost to the federal government in lost tax revenue from the LIHTC program totaled $4.9 billion in 2006. That
loss is expected to increase sharply in the next few years in response to the 40 percent increase in credits allocated
beginning in 2001. Those increases will occur even if current proposals seeking to expand the LIHTC program are
not born out. Additional details on the cost of various forms of low-income housing support are provided in a report
by the U.S. Congress, Joint Committee on Taxation (2005).
3
LIHTC developers are obliged to fill 40 percent or more of their units with low-income families and are required to
charge rents below a specified ceiling. Non-compliance with these terms results in forfeiture of future tax credits
and also repayment of one-third of previously received allocations plus interest.
2
instrument for LIHTC housing between 1987 and 2000 using 1970 housing stock structure types (i.e.
multi-family, single family) and the distribution of bedrooms for both owner- and renter-occupied units.
The intuition behind these instruments is that a past willingness to allow high-density development may
affect the current willingness of the local community to allow further multi-family housing development,
including LIHTC projects.4 An F-test for the joint significance of excluded instruments reported later in
the paper confirm that these instruments are strongly correlated with the level of LIHTC development.
Moreover, in contrast to growth controls, zoning ordinances that limit high-density development need not
affect the level of private rental housing construction provided sufficient open land is available for low-
density development. If such conditions hold, the instruments can be excluded from the structural model.
To test that assumption, Hansen-J (1982) statistics are computed for each level of geography.5 Results
fail to reject the overidentifying restrictions, and especially so for the ten-mile level of geography. This
lends further support for the validity of the instruments.
Results indicate that for one-half mile radius circle areas OLS estimates of crowd out are
significantly larger than GMM estimates. However, those differences diminish monotonically as the
geographic scope of the analysis is increased. This implies that LIHTC developers tend to locate their
projects in neighborhoods that would ordinarily receive little rental housing construction. In certain
respects, that is consistent with the goals of the LIHTC program. But as noted above, it is important to
also recognize that developers may substitute new housing capital from one neighborhood to another
when building LIHTC projects. As we expand the geographic scope of the analysis to ten mile circle
areas, OLS and GMM estimates become nearly identical. This is consistent with the idea that for
sufficiently large geographic areas LIHTC development is approximately exogenous. Importantly, for the
ten-mile circle area models, our estimates indicate that roughly one-third of LIHTC development is offset
4
This could be manifested in local zoning laws, for example, that either explicitly allow or preclude high-density
development.
5
Hansen-J tests evaluate the joint null that the instruments are both exogenous and do not belong in the structural
model.
3
by crowd out of unsubsidized rental housing construction, an amount large enough to warrant attention
when forming policy
In certain respects, the high degree of crowd out should not be surprising. Numerous studies in
the literature indicate that housing demand is inelastic while new housing supply is quite elastic.6 A
simple model outlined later in the paper demonstrates that under such market conditions, high rates of
crowd out will occur. On balance, therefore, advocates of LIHTC development need to look beyond
expansion of the total stock of rental housing to justify the LIHTC program. In that regard, it is important
to also recognize that LIHTC development is not exclusively a low-income neighborhood event, and our
paper has something to offer on this point as well.
As shown in Figure 2a, as of 2000, 16 percent of LIHTC units were located in high-income
census tracts, defined here as tracts in the upper third of their MSA income distribution. Another 28
percent were situated in middle-income communities, with the remaining 56 percent in lower-third
income neighborhoods. The extension of LIHTC housing into middle- and higher-income neighborhoods
differs markedly from public housing. As shown in Figure 2b, 77 percent of public housing units are in
low-income locations with most of the rest in middle-income communities. These summary measures are
strongly suggestive that LIHTC development has expanded low-income housing opportunities in higher
income communities.7
To shed further light on this feature of the program, we consider the stigma/amenity effects
associated with nearby LIHTC development in high- and low-income communities. Because
stigma/amenity effects likely attenuate sharply with distance, our focus in this portion of the analysis is
primarily on activity in close proximity to LIHTC developments, as within one-half mile. In considering
6
Using data from the rental housing allowance demand experiment from the 1970s, Hanushek and Quigley (1980)
estimate rental housing price elasticities for Pittsburgh and Phoenix of -0.36 and -0.41, respectively. Other estimates
of the price elasticity of housing demand based on owner-occupied homes range from -0.3 to -1.0. Estimates of the
elasticity of supply of new housing construction are nearly always far above 1. For rental housing construction, the
most reliable supply elasticity estimates are in excess of 3.
7
In principle, this need not be the case if construction of LIHTC housing in higher-income communities is fully
offset by 100 percent crowd out. However, estimates later in the paper suggest that partial but not full crowd-out is
more likely.
4
these issues, it is important to recognize that LIHTC housing is of high-quality relative to most lower-
income housing.8 As a result, LIHTC housing may have a positive amenity effect on nearby low-income
neighborhoods with a corresponding increase in demand for those locations. That would cause
unsubsidized construction of rental housing to increase in the local area, mitigating the impact of crowd
out. On the other hand, it seems likely that LIHTC housing may reduce the appeal of nearby higher
income communities. This could occur either because of fear of publicly supported housing projects in
general, or possibly because of local opposition to the presence of lower income families who occupy
LIHTC units. To the extent that such conditions prevail, NIMBY-type reactions in higher-income
communities would imply that LIHTC development has a negative stigma effect with a corresponding
decrease in demand for the local community. That would cause unsubsidized construction of rental
housing to decrease, exacerbating the impact of crowd out.
Together, these ideas suggest that LIHTC development may have a much more negative impact
on unsubsidized rental housing construction in nearby high-income communities as compared to nearby
low-income neighborhoods. Moreover, those differences should attenuate as the geographic scope of the
analysis expands and stigma/amenity effects diminish. Additional empirical results reported later in the
paper are consistent with these priors. This suggests that whereas LIHTC development will often be
welcomed in low-income areas, political opposition should be anticipated in higher-income communities.
The paper proceeds as follows. In the following section, we provide a brief overview of the
LIHTC program as well as previous assessments of the crowd out effects of place-based subsidized rental
housing. Section 3 examines crowd out effects of the LIHTC program, both at a conceptual level and
empirically. Section 4 stratifies the sample into low-, middle-, and high-income areas, and examines
evidence of amenity and stigma effects of LIHTC development. Section 5 concludes.
8
Cummings and DiPasquale (1999) and DiPasquale, Fricke, and Garcia-Diaz (2003) document this feature of
LIHTC housing.
5
2. Institutional Context and Prior Estimates of Crowd Out
This section provides a brief overview of some additional institutional details of the LIHTC
housing program. Also included in this section is a review of previous research on crowd out associated
with place-based subsidized housing.9
2.1 Institutional Context
Between 1937 and the late 1970’s, low-income housing support was provided almost exclusively
through public ownership and operation of housing built exclusively for low-income families. We refer
here to such housing as “traditional public housing.”10 Typically, occupancy in these projects has been
limited to families near or below poverty levels with tenants paying 30 percent of their gross income
towards rent (Olsen, 2003).
By the 1980s, two concerns about the public housing program had gained attention. The first
was simply that the government builds, owns, and operates these projects: there are basic questions as to
whether at least some portion of such activity is best left to the private market. In addition, because
public housing projects were spatially concentrated and restricted occupancy to very low-income
residents, they created dense clusters of poverty. Facilitated in part by several horrifying newspapers
accounts, the public came to associate public housing with neighborhood decline, violent crime, and drug-
use (Currie and Yelowitz, 2000). Concerns about adverse peer and neighborhood effects on children
growing up in the projects also gained attention, both in academic and policy circles (see, for example,
Currie and Yelowitz (2000) or Jencks and Mayer (1990)). Primarily for these reasons, the government
stopped virtually all construction of traditional public housing in the early 1980’s and began to demolish
9
An exhaustive review of the LIHTC and other related housing programs are in Quigley (2000) and Olsen (2003).
10
Olsen (2003) points out that there are at least 29 different public housing programs.
6
the worst performing projects during the 1990s.11 Figure 1 illustrates the construction of traditional
public housing units between 1935 and 2000 along with demolitions in the 1990s.
The LIHTC program was created as a part of the Tax Reform Act of 1986, both as an alternative
to public housing and to offset the reform’s removal of other tax benefits for owners of rental housing
(U.S. Joint Committee on Taxation, 1987). Different from other low-income housing programs, the
LIHTC program is federally administered by the Internal Revenue Service under Section 42 of the U.S.
Tax code. Figure 1 displays the number of LIHTC units placed in service from 1987 to 2000.
The premise of the LIHTC program is to create a public-private partnership where the Federal
Government subsidizes between 30 to 91 percent of non-land construction costs for private developers.
In all cases, the subsidy is provided to developers through a 10-year stream of annual nonrefundable
federal tax credits – dollar-for-dollar reductions in federal tax liability (Eriksen, 2007). In exchange for
the subsidy, developers agree to set rents below specified ceilings and to lease a minimum specified share
of the project’s units to low-income families for at least 30 years. In practice, most projects contain 100
percent low-income occupancy owing to how the subsidy is structured and administered (see Eriksen
(2007) for additional details).
It is also important to note that rent ceilings imposed on LIHTC units occupied by low-income
families are effectively set at 18 percent of MSA median household income as determined by HUD.12
That amount is sufficiently high and the tax credits are sufficiently generous (30 to 91 percent), that
LIHTC units are generally of high quality relative to most other low-income housing (Cummings and
DiPasquale, 1999). It is also important to note that LIHTC occupants typically have a higher income than
residents of traditional public housing (Wallace, 1995).13
11
In some instances, such as under the HOPE VI program, the structures were remodeled, but in most cases tenants
were usually issued housing vouchers and told to seek housing privately (Jacob, 2004).
12
The 18 percent limit is derived from two conditions imposed by the program. First, low-income families must earn
incomes below 60 percent of MSA median income, and second, rent must not exceed 30 percent of that value.
13
For example, Wallace (1995) estimates that only 28 percent of LIHTC residents earn below 50 percent of an area’s
median income compared to 81 percent of those who reside in traditional public housing.
7
These features of the public and LIHTC programs bear importantly on the analysis to follow.
First, traditional public housing has had a negative reputation in the media, and it seems likely that this
could color perceptions of LIHTC housing, especially in higher income communities. Second, LIHTC
housing is relatively high quality in comparison to other low-income housing, and this could foster
positive amenity effects in lower income communities. Third, LIHTC subsidies are so generous that
development in higher income communities may be profitable. Indeed, as shown in Figure 2b, whereas
the majority of public housing is in low-income communities, almost half of LIHTC development is
found in neighborhoods in the upper two-thirds of the income distribution.
2.2 Previous estimates of crowd out from place-based subsidized housing
The possibility of crowd out arises any time government provides goods and services that are
also offered through the private sector. This has been examined in a variety of markets, including health
insurance, radio, and charitable giving (Culter and Gruber, 1996; Berry and Waldfogel, 1999; Andreoni
and Payne, 2003). Several recent studies have also begun to examine crowd out arising from place-based
subsidized housing, although most do not consider the LIHTC program. The first of these by Murray
(1983, 1999) utilizes national-level aggregate time series data from 1935 to the mid-1980s. These data
pre-date the LIHTC program and are used to assess the impact of public and other earlier forms of
subsidized rental housing construction on unsubsidized housing construction (Murray, 1983) and the
equilibrium stock of housing (Murray, 1999). The general strategy of each paper is to examine whether
construction of subsidized rental housing increases the total stocks of housing on less than a one-for-one
basis. 14 Evidence of such effects would be indicative of crowd out. Murray (1999) finds that subsidized
rental housing programs that target very low income families generate only a small amount of crowd out.
This is consistent with stylized facts that private market developers build little unsubsidized housing for
very low-income families: for crowd out to occur, private markets must first be willing to provide the
14
More precisely, Murray (1999) estimates the cointegrating relationship between the equilibrium stock of
subsidized and unsubsidized housing stocks over the 1935 to 1987 period.
8
product. In contrast, Murray (1999) also estimates that between one-third to 100 percent of subsidized
“moderate-income” place-based housing is offset by crowd out of unsubsidized construction. This is
consistent with the idea that in the absence of construction subsidies, the private market would build at
least some moderate income housing.
Given that LIHTC housing is approximately a moderate income housing program, this is
important. But it should also be emphasized that Murray’s work differs from ours in two very important
respects. First, our empirical design is based on cross-section data and relies on geographic and
instrumental variable methods for identification. Murray’s work is based on aggregate time series data.
Moreover, as noted above, Murray’s data reflect the impact of housing programs that pre-date the LIHTC
program. Because those earlier programs have very different institutional arrangements than LIHTC
development, there is no guarantee that Murray’s results would carry over to the LIHTC program.
More recently, and closer in structure to this paper, Sinai and Waldfogel (2005) examine crowd
out effects of place-based subsidized rental housing programs on per capita occupied housing units in
1990. When the data are measured at the census place level the OLS estimate of crowd out associated
with place-based subsidized rental housing is roughly 70 percent. However, the size of a census place
varies widely, from rather small to quite large. Especially for smaller census places, it is possible that
OLS estimates may suffer from endogenous location of subsidized housing units for reasons described in
the Introduction.15 When the data are aggregated to the MSA level – at which point concerns about
endogenous location of subsidized housing are greatly diminished – their point estimate of crowd-out
falls to roughly 30 percent. For both levels of geography, LIHTC housing is lumped in with other forms
of place-based subsidized rental housing. This is important because their data on subsidized housing
units (obtained from HUD’s 1996 Picture of Subsidized Housing file) includes roughly 2.8 million place-
based subsidized units, of which only 332,085 are LIHTC units, most of which were not present in 1990,
15
Sinai and Waldfogel (2005) attempt to instrument for subsidized housing construction using the number of housing
units per capita built before 1940 and also occupied public housing units per capita in 1980. However, results from
the IV models yield quite different estimates of crowd out depending on which of the instruments are included.
Partly for that reason, Sinai and Waldfogel tend to emphasize their non-IV estimates of crowd out.
9
the period associated with their dependent variable.16 As with Murray (1983, 1999), therefore, Sinai and
Waldfogel (2005) examine primarily crowd out effects associated with place-based programs that pre-
date the LIHTC program.
Malpezzi and Vandell (2002) do consider directly the crowd out effects of LIHTC development.
They analyze the impact of 1987-2001 state-level LIHTC allocations on the per capita stock of housing as
measured in 2000 (based on the year-2000 census). Their point estimate implies full crowd out, although
their sample is limited to just 51 state-level observations (including Washington, D.C.) with controls for
14 indicators of demand and supply. As a result, and as recognized by the authors, the standard error on
their estimate of LIHTC effects is several times larger than their corresponding estimated crowd out
effect. They are, therefore, unable to shed much light on the degree of crowd out associated with the
LIHTC program.
Most recently, Baum-Snow and Marion (2007) examine the impact on LIHTC development given
a census tract’s qualified census tract (QCT) status as defined by HUD. QCT tracts are eligible for
additional subsidies through the LIHTC program making these tracts especially attractive for LIHTC
development, all else equal.17 Baum-Snow and Marion (2007) provide evidence that developers respond
to a tract’s QCT status by shifting development from adjacent tracts to the more heavily subsidized
location. That finding, however, does not necessarily reveal anything about the degree to which LIHTC
development crowds out private unsubsidized construction. On the other hand, their finding underscores
the tendency of developers to substitute capital across neighborhoods in response to competing
development opportunities. As noted in the Introduction, it is for that reason that we believe crowd out
effects are most clearly identified at a relatively broad level of geography for which capital substitution
across neighborhoods largely nets out.
16
Place-based subsidized units in the 1996 Picture files include Public (1,326,224 units) Section 8 Moderate
Rehabilitation (105,845 units), Section 8 New Construction (897,160 units), Section 236 (447,382 units), and other
placed-based subsidized units (292,237 units). Additionally, the 1996 picture files only report roughly half of the
LIHTC units allocated between 1987 and 1996 (Malpezzi and Vandell, 2002).
17
LIHTC development in a QCT is eligible for a 30 percent increase in subsidy as compared to those not located in
such a tract.
10
Relative to these studies, the research in this paper makes three important innovations. First, we
directly address the joint effects of stigma and crowd out. Previous studies of crowd out related to
subsidized housing have virtually ignored possible local demand shifts arising from stigma. Second, we
estimate our models multiple times for a range of different levels of geography, from one-half mile to 10-
mile radius circles. This helps not only to address issues about the endogenous construction of subsidized
housing, but also further highlights the different impacts of stigma and crowd out in a manner that we will
make clear later in the paper. Third, previous studies have been hampered by the limited quality of early
data released on the LIHTC program. The LIHTC data used in this study are more current and benefit
from HUD efforts to provide a more complete and accurate accounting of LIHTC development.
3. Crowd Out
3.1 Conceptual Model
Suppose initially that stigma/amenity effects do not occur in response to LIHTC development.
As argued in the Introduction, this is likely a good approximation for larger geographic areas. This also
implies that housing demand is unaffected by the LIHTC program. Consider now Figure 3a which
portrays the market for newly built rental housing. Abstracting from some of the details of the LIHTC
program, the program subsidizes construction up to an exogenously given state-level allocation. This
implies an outward shift in aggregate supply causing equilibrium rents to fall and housing construction to
increase.18
18
Figure 3a suggests that the LIHTC program causes the entire supply function to shift out. This is a simplification
that captures the dominant impact of the LIHTC program that is relevant to this paper while streamlining the
discussion. An alternative and more accurate portrayal is that the LIHTC program flattens the slope of the lower
most section of the supply curve up to the maximum number of LIHTC units allocated to a given location. Beyond
that level of construction, the supply function steepens because additional investment is unsubsidized. Such a
specification would imply that developers choose to invest first in LIHTC developments before unsubsidized
construction. Moreover, to the extent that LIHTC development draws low-cost factor inputs away from the
unsubsidized sector, input costs in the unsubsidized sector will be higher than in the absence of the LIHTC program.
This would result in a further inward rotation of the unsubsidized segment of the supply function. See Olsen (2007)
for a further discussion of this issue.
11
In Figure 3a, notice that aggregate rental housing construction increases by less than the level of
LIHTC construction. This is a direct result of the downward sloping nature of demand, the assumed
elastic nature of supply, and the corresponding reduction in rents following LIHTC development. In
particular, whereas H3-H1 equals the number of LIHTC units built, aggregate supply expands by only H2-
H1. The difference, H3-H2, represents unsubsidized private construction that is “crowded out” by the
LIHTC program. Only when housing demand is perfectly elastic or supply of newly constructed housing
is perfectly inelastic would H3-H2 equal zero and crowd out not occur.
To put this figure in perspective, Mayer and Somerville (2000) estimate that the supply elasticity
of newly built housing of all types is roughly 6. This is close to DiPasquale and Wheaton (1992) who
estimate a supply elasticity for newly built multi-family rental housing of 6.8. Other estimates of the
elasticity of supply of newly built housing are smaller, but generally well above 1 (e.g. DiPasquale, 1999;
Rosenthal, 1999).19
On the demand side, Hanushek and Quigley (1980) use data from the housing allowance
experiments of the 1970s to estimate the elasticity of demand for rental housing. For Pittsburgh and
Phoenix their estimates are -0.36 and -0.41, respectively. Higher price elasticities of demand are typically
obtained when using owner-occupied units. Rosen (1979), for example, estimates a price elasticity of -
0.99, but Rosenthal, Duca, and Gabriel (1991) obtain an estimate of -0.5 for FHA homebuyers, a group
that is much closer in income to that of the typical renter.
Summarizing, only when housing demand is perfectly elastic or the supply of newly constructed
housing is perfectly inelastic would crowd out not occur. However, estimates in the literature strongly
suggest that the supply of newly built housing is very elastic and demand for housing is relatively
19
As a further perspective, suppose that developable land is plentiful, either as raw land or as previously developed
land suitable for redevelopment. Suppose also that long run costs for building materials and labor are constant in
real terms. Then home building is approximately constant returns to scale and new housing supply should be quite
elastic (e.g. Rosenthal and Helsley, 1994). See also Glaeser and Gyourko (2005) for a discussion of the relative
elasticity of supply with respect to expansion versus contraction of the stock of housing.
12
inelastic. This suggests that on a qualitative basis, advocates and opponents of LIHTC development alike
should anticipate crowd out from the program. The question then is, how much?
3.2 Empirical Model
Before developing our regression model it is useful to first highlight certain key features of the
data. We use decennial census data as the primary data source, aggregated to the census tract level.
These data were obtained from the Geolytics, Inc. Neighborhood Change Database file for 1970, 1990,
and 2000.20 Geolytics re-codes data from each of these years to year-2000 census tract boundaries. These
data were combined with information on LIHTC projects placed into service up through 2000. The
LIHTC data were obtained from HUD over the web.21 Information on the LIHTC database includes the
year placed in service and the year-2000 census tract. In all of the estimation to follow, we restrict our
sample to just those tracts located within MSAs. This leaves us with information on roughly 45,000 tracts
including 17,774 LIHTC projects containing 877,972 individual units. Tract-level summary measures of
the variables used in the regressions are provided in Table 2.
Two points should be noted when working with the census tract data. First, tracts are often rather
small and for reasons outlined earlier, crowd out is more accurately identified over a larger geographic
area. Second, tract shapes and sizes differ from one tract to the next and that can complicate efforts to
assess the geographic scope of both crowd out and stigma/amenity effects. For both reasons, we
reorganize our data geographically into uniform circular units using Geographic Information Systems
(GIS) software (i.e. MapInfo and MapBasic). Specifically, data are organized into circles of radius d
miles drawn around the geographic centroid of the individual census tracts, i (i = 1, …, n). In all cases,
20
See www.geolytics.com.
21
See http://lihtc.huduser.org .
13
this is done based on year-2000 census tract geography.22 The circle-based measures are produced for all
of the dependent, independent, and instrumental variables. Summary statistics of circle-level data with
radii 0.5, 1, and 10 miles can be found in Tables 3a – 3c, respectively.
Configuring the data in this manner allows us to examine the impact of different levels of
geography (d) while ensuring that each observation is measured over an area of the same size. In
addition, our empirical design ensures that the activity reflected in the dependent variable is based on the
same underlying geography as the activity of the independent variables and instruments. Bearing these
features in mind, we estimate the following model,
H id ,1990−rental = aH id ,1987 − 2000 +
Private ,
2000
LIHTC
b1 H id ,1985−1989 + b2 H id ,1980−1984 + b3 H id ,1970−1979 + b4 H id , pre −1970 +
Rental Rental Rental Rental
. (3.1)
b5 H id ,1985−1989 + b6 H id ,1980−1984 + b7 H iOwner −1979 + b8 H id , pre −1970 +
Owner Owner
d ,1970
Owner
l1 DistCBDi + l2 LandAreaid + φ MSAi + ε id
The dependent variable in this regression is the level of unsubsidized rental housing construction
between 1990 and 2000 within a radius of d miles drawn around the geographic centroid of the year-2000
census tract, i.23 This is constructed by subtracting LIHTC units built in the 1990s from the number of
rental units reported as under 10 years in age in the year-2000 census tract data. Control measures in
(3.1) are as follows.
Variables in the last row control for location-specific elasticities of demand and supply for newly
built housing. In part, this is done by including MSA fixed effects for each of the 331 MSAs in the
United States. On the demand side, the fixed effects control for population, income, preferences, and
other broader determinants of demand. On the supply side, the fixed effects control for the MSA’s
distribution of land use restrictions (e.g. zoning) and natural features of the landscape that affect
22
MapInfo and MapBasic were used to manipulate the geographic features of the data. When drawing rings around
the census tract centroids, proportional sum measures were used to calculate the various variables. This was done
for all of the variables used in the analysis including the instruments described later in the paper.
23
Roughly 1,000 Public Housing units were also constructed in the 1990s. These units were also subtracted off from
newly built rental housing when forming the dependent variable. The public housing data were obtained from HUD
as noted earlier.
14
opportunities for development (e.g. mountains, lakes, rivers). Also included in the last row are measures
of the distance from the geographic centroid of a given circle i to the central business district (CBD) and
the amount of land within circle id not covered by water. Both of these measures proxy for the supply of
developable land: in the first case, outlying areas have more open land, while in the second case circle-
areas covered with more land also offer greater opportunities for development.
The middle two rows include controls that approximate lagged levels of construction. This helps
to control for additional unobserved tract-specific demand and supply factors. This also soaks up serial
correlation in development decisions that would otherwise be embedded in the model’s error term. In
practice, we control for lagged construction by including separate controls for the age distribution of the
rental and owner-occupied housing stocks, where values are measured in numbers of housing units as
reported in the 1990 decennial census. As we do not control for demolitions these variables approximate,
but are not equal to, past levels of construction in the geographic area. An assumption throughout is that
our controls for lagged construction levels successfully remove serial correlation from the model error
term. Under that assumption, the model error term represents contemporaneous shocks and is orthogonal
to lagged housing stocks. As with the dependent variable, each control variable in the middle rows is
subscripted by the years (e.g. 1985 to 1989) of construction its variable represents.
In the first row of (3.1), the number of LIHTC units built between 1987 to 2000 (H3 – H1 in
LIHTC
Figure 3a) is represented by H id ,1987 − 2000 . This is the primary variable of interest. If its coefficient a
equals 0 that would indicate that construction of LIHTC units has no effect on the level of new
unsubsidized private rental construction. Absent stigma/amenity effects, that would imply zero crowd out
and the presence of a perfectly elastic demand function (given strong priors that long-run expansion of
supply is elastic). If instead, a equals -1 that would imply a perfectly inelastic demand function and
complete crowd out.
For reasons described in the Introduction, equation (3.1) is estimated multiple times for different
values of d, where d is the radius of the circles used to measure the dependent and independent variables.
15
Specifically, we estimate the models for ½-mile circles, 1-mile circles, 2-mile circles, …, 5-mile circles,
and 10-mile circles. In addition, each model is estimated twice, first by OLS and then again by two-step
GMM. In the latter case, we instrument for LIHTC construction using 1970 housing stock structure type
(e.g. the distribution of single-family versus various sizes of multi-family) and the distribution of
bedrooms, both measured separately for the renter and owner-occupied housing stock in location id.
Recall that these instruments proxy for a past willingness to allow high-density development and possibly
local zoning ordinances that may restrict or enhance further opportunities to build multi-family LIHTC
housing. For reasons also described in the Introduction, our prior is that as the level of geography
expands, LIHTC development should become increasingly exogenous and crowd out estimates from the
OLS and IV models should become similar. That is because endogenous substitution of capital and
households across neighborhoods in response to LIHTC development nets out.
A final empirical issue concerns the overlapping nature of the circles used to measure the
dependent variables. As noted above, the root data source is tract-level census data and census tracts are
non-uniform in shape and size. The circle measures ensure a common size for each sampled area and are
drawn around the geographic centroids of the underlying tracts. This implies that many circles overlap.
When measuring the independent variables such overlap presents no special problems, but for the
dependent variables underlying census tract information is repeated across observations. Failing to
address that issue would cause the standard errors from the model to be biased downwards but would not
bias the coefficient estimates.
A convenient solution to this problem is available. Specifically, we cluster the standard errors by
MSA (e.g. Woolridge, 2003). Because non-MSA census tracts are excluded from the analysis, any
overlap across dependent variables is fully addressed. At the same time, because many circles within
individual MSAs do not overlap, clustering by MSA overcompensates for the overlapping ring problem.
For that reason, our estimation procedure is inefficient and provides an upper bound on the true standard
errors, or equivalently, a lower bound on the true t-ratios. In most models that is sufficient to identify the
patterns of interest.
16
3.3 Results
Table 4 presents OLS and GMM estimates of (3.1) for various levels of geography from one-half
up to 10-mile radius circle areas. To conserve space and maintain focus on the LIHTC program we report
just the coefficients on the LIHTC variables.24 Values in parentheses below the coefficients are t-ratios
based on standard errors clustered at the MSA level. As indicated in the table and earlier in the
discussion, all of the models include MSA fixed effects.
Consistent with our priors, OLS and GMM coefficient estimates differ markedly when the
geographic scope of the analysis is narrow – as with the one-half mile circle samples – but are very
similar once the circle radius is expanded to ten miles. That pattern is plotted in Figure 4 to facilitate
review. The pattern suggests that sitting of LIHTC projects at the neighborhood level is endogenous, but
that at the ten mile level, LIHTC development is exogenous. This also suggests that our ten-mile circle
sample estimates are especially robust, benefiting both from the IV controls and the broader level of
geography. For narrower geographic levels of analysis (e.g. one-half mile radius circles), we are reliant
on the instrumental variables used in the GMM models for identification.
Observe also that for each level of geography, the OLS estimates of crowd out are always larger
in magnitude than the GMM estimates. This indicates that, on average, developers tend to locate LIHTC
projects in neighborhoods that receive relatively little rental housing construction, resulting in a more
negative coefficient. As noted in the Introduction, that finding is in some respects, consistent with the
spirit of the LIHTC program – to extend housing opportunities for lower income families.
Consider next the diagnostic tests of the validity of the instruments. These are reported at the
bottom of the table. The F-statistic for the joint significance of the excluded instruments clearly support
24
GMM results for all of the estimated coefficients are provided in the appendix for circle radii of 0.5, 1.0, 5.0, and
10.0 miles.
17
the “strength” or relevance of the instruments in all instances.25 Hansen-J (1982) tests also fail to reject
the overidentifying restrictions for each of the models.26 Moreover, the Hansen-J tests appear especially
compelling for the ten-mile level of geography. These diagnostic tests provide some confidence that the
IV estimates for the narrower levels of geography are credible while offering further support for the idea
that our crowd-out estimates based on ten-mile geography are robust.
Finally, consider the magnitude of the ten-mile circle estimates of crowd out. The GMM estimate
implies that roughly 33 percent of LIHTC housing units built are offset by crowd out of private,
unsubsidized rental housing construction; the corresponding OLS estimate is 35 percent. Thus, one-third
of LIHTC development is offset by crowd out. On the surface, this strikes us as a large number, but how
one perceives that estimate depends on perspective. At a minimum, it is important for proponents of the
LIHTC program to be aware that the impact of LIHTC development on the total number of rental housing
units is well below the number of LIHTC units built. However, as noted earlier in the paper, roughly 28
percent of LIHTC units are built in communities in the middle third of their MSA income distribution,
while another 16 percent are built in neighborhoods in the upper third of their MSA income distribution.27
This suggests that the LIHTC program potentially expands low-income housing opportunities in higher
income communities. To explore this feature of the program further, we next stratify the sample to
consider the impact of LIHTC development on low- and high-income areas. This includes consideration
of the degree to which LIHTC housing may be viewed as an amenity or may carry with it a stigma that
could have adverse local spillover effects.
4. Stigma/Amenity Effects
25
This is important because weak instruments bias estimates from IV models. See Murray (2006) for a discussion of
this issue. The F-statistic reported in each table accounts for the clustering of standard errors by MSA.
26
Hansen-J tests evaluate the joint null that the instruments are exogenous and that they are appropriately excluded
from the structural model. We note that these tests are potentially sensitive to model specification and controversial
for that reason (Davidson and McKinnon, 2004; Cameron and Trivedi, 2006).
27
In addition, as noted earlier, other research has emphasized that LIHTC housing is high quality relative to most
other low-income homes.
18
Suppose that local residents do not want to live in close proximity to LIHTC housing because of
either an aversion to the housing structures or to the lower-income occupants of the projects themselves.
Such NIMBY-type responses would reduce demand for the locale and cause market rents and
unsubsidized construction to decline. 28 Stigma, therefore, would exacerbate the negative crowd out
effects of LIHTC development. This is a likely characterization of how LIHTC housing is viewed in
higher-income communities. This situation is portrayed in Figure 3b.29
On the other hand, suppose that LIHTC housing is perceived as a positive amenity that enhances
the appeal of a neighborhood. Cummings and DiPasquale (1999) and DiPasquale, Fricke, and Garcia-
Diaz (1999), for example, report that LIHTC housing is high quality relative to most low-income rental
housing. In addition, Stegman (1991) and Wallace (1995) suggest that the average resident of a LIHTC
unit has a much higher income than the typical resident of traditional public housing and that LIHTC
housing is often welcome in lower income areas. Under these conditions, LIHTC development increases
demand for the local community, causing rents and unsubsidized construction to increase. A positive
perception of LIHTC housing, therefore, would mitigate crowd out effects. This is a likely
characterization of LIHTC development in lower-income communities. This situation is portrayed in
Figure 3c.
Together, Figures 3a to 3c indicate that the influence of LIHTC development on neighborhoods
likely differs with the economic status of that community. To help identify these differences we
reorganize our data geographically. To be precise, for this portion of the analysis, our independent
variables including LIHTC construction and our instruments are measured exactly as in Section 3.
However, the dependent variable is modified in two important ways. First, the estimating sample of
circle observations is stratified into three groups based on the economic status of the core census tract
around which a given circle is drawn. Based on 1990 tract average income, low-income circles are
28
See Jacob (2004) for discussion of the impact of public housing on perceptions of neighborhood quality.
29
For convenience, Figure 3b is drawn such that the crowd out and stigma effects exactly offset resulting in no
change in the level of construction. That need not be the case, of course.
19
defined as those whose core tracts are in the bottom third of their MSA tract income distribution.
Analogous definitions are used to identify middle- and high-income circles for core tracts in the middle
and upper thirds of their MSA income distribution.
It is tempting to estimate our crowd out regressions separately over each of these three samples
with no further adjustments. But such an approach would bias the crowd out estimates to be similar
across the stratified samples. To clarify why, consider the following illustration. Suppose all census
tracts were of the same size and dimension, and all tracts are organized along a line. Further, census
tracts are organized along the line in an alternating sequence of low-, middle-, and high-income tracts.
Then as the circle size increases, the economic distribution of the underlying tracts contained in the circle
would be alike regardless of whether the circle was centered on a low-, middle-, or high-income tract. As
a result, regressions based on such stratified samples would tend to yield similar results even though the
marginal impact of LIHTC development on unsubsidized construction might differ in low-, middle-, and
high-income areas.
To address this issue, for each stratified sample of core census tracts (i.e., low-, middle-, and
high-income), we measure the dependent variable, construction of unsubsidized rental housing, only in
areas within the circle belonging to underlying census tracts that are of the same economic status as the
core tract. Organizing the data in this fashion allows us to estimate the marginal impact of LIHTC
development on unsubsidized nearby low-, middle-, and high-income areas. It should also be noted that
the estimated crowd-out effects across these stratified samples will typically not add up to the crowd out
estimate from Table 4 based on the full sample. That is because the values of the model covariates differ,
on average, for the low-, middle-, and high-income sub-samples, as illustrated in Tables 3a – 3c. Those
differences affect the weights embodied in the regression algebra used to obtain the crowd out estimates
in a complicated and non-linear fashion.30
30
This is especially relevant for smaller circle sizes (e.g. one-half mile) for which the covariate values differ
markedly across sub-groups (see Tables 3a-3c, for example). For larger circle sizes (e.g. 10 miles), the covariate
values become more similar on average, for the low-, middle-, and high-income circle samples. As a result, and for
reasons outlined above, estimates from large-circle stratified sample models should add up to a value that
20
Tables 5a-5c present results from this exercise for the low-, middle-, and high-income circle
samples. In each case, the structure of the tables is the same as in Table 4. This includes reporting only
the coefficients on the LIHTC variable to maintain focus and conserve space.31 In addition, Figure 5 plots
the GMM LIHTC coefficients for the low- (Table 5a), middle- (Table 5b), and high-income (Table 5c)
circle samples for the different values of d. Notice that the diagnostic tests at the bottom of Tables 5a-5c
once again generally support the validity of the instruments, although in some models they are less
compelling than in Table 4. In particular, the instruments are always very strongly correlated with
LIHTC construction (F-statistic for the significance of excluded intruments). In addition, the Hansen-J
tests generally fail to reject the overidentifying restrictions. The primary exceptions are for the high-
income sample for models based on circle radii two to five miles; for these models the Hansen-J P-values
are low.
Focusing on the GMM estimates in Tables 5a-5c, and also in Figure 5, for the one-half mile
samples these estimates indicate that LIHTC development has a sharp negative impact on unsubsidized
construction in the high-income circle sample, but a small positive effect in the low-income circle
sample.32 To be precise, construction of 100 LIHTC units in a half-mile circle reduces private
unsubsidized construction in the high-income neighborhoods within one-mile by just over 100 rental
units, indicative of full crowd out. Conversely for the low-income sample, the analogous estimate is an
increase of roughly 26 units. Expanding the geographic scope of the analysis beyond one-half mile,
differences in the impact of LIHTC development on rental construction in low- and high-income circle
samples largely disappear. This is anticipated because of the adding up issues note earlier. In that regard,
note that at the 10 mile circle level of geography, our estimates indicate that development of 100 LIHTC
approximates that of the full-sample regression. This is the case for the 10-mile circle samples as can be seen when
comparing results from Tables 5a-5c to Table 4.
31
Complete regression output for the income-stratified samples is provided in the appendix (Tables A-2 to A-4,
respectively) for circle samples of radii 0.5, 1, 5, and 10 miles.
32
In Table 5b, for the middle-income sample, note that the coefficient on LIHTC development within one-half mile
is -0.39, roughly halfway between the corresponding estimates for the low- and high-income samples.
21
units would reduce unsubsidized rental housing construction by 12, 14, and 8 units in low-, middle-, and
high-income neighborhoods, respectively.
To interpret these patterns, it is important to recall two principles that facilitate identification.
First, stigma/amenity effects attenuate with distance and will be most prominent for narrow levels of
geography. Second, crowd out effects are most accurately identified for larger levels of geography.
Bearing these principles in mind, several implications follow from the patterns observed in Figure 5. In
high-income areas, the sharp negative impact of LIHTC development within one-half mile on
unsubsidized construction is indicative of either negative stigma effects, crowd out, or both. On the other
hand, LIHTC effects on high-income areas attenuate sharply with an expansion of the geographic scope
of the analysis; this almost certainly reflects attenuation of stigma effects rather than a dissipation of
crowd out. Overall, the pattern of results is strongly suggestive that LIHTC development has a negative
stigma effect on nearby high-income neighborhoods. This differs markedly from the response to LIHTC
development in low-income communities. For low-income neighborhoods within one-half mile, the
positive impact of LIHTC development on unsubsidized construction can only arise from positive
amenity effects that more than offset crowd out.
5. Conclusion
The recent dramatic growth of the Low-Income Housing Tax Credit (LIHTC) program has
caused an old question to gain new importance. Should government provide low-income housing support
through tenant- or placed-based programs? In that context, the LIHTC program has ballooned since its
inception in 1987, and is now the largest subsidized rental housing construction program in U.S. history.
The program subsidizes 30 to 91 percent of construction costs for eligible projects, 44 percent of which
are located in middle- and higher-income neighborhoods, and has accounted for one-third of all multi-
family rental housing construction in recent years. Moreover, recent proposals in Congress have sought
to double the size of the program. Nevertheless, little is known about the efficacy of this increasingly
important and expensive program. This paper has sought to fill part of that gap.
22
Our most important finding concerns the level of crowd out associated with the LIHTC program.
Over a ten mile area, OLS and GMM estimates both indicate that one-third of LIHTC development is
offset by crowd out of private unsubsidized construction. Moreover, based on conceptual arguments and
also diagnostic tests, we argue this estimate of crowd out is robust. It is clear, therefore, that the impact of
LIHTC development on the total stock of rental housing units is well below the number of LIHTC units
built. This implies a windfall gain for developers and raises questions about the viability of the LIHTC
program. How then might the LIHTC program be justified?
One possibility is the spatial distribution of LIHTC development. Summary measures document
that 16 percent of LIHTC projects have been built in neighborhoods in the upper third of their MSA’s
income distribution, and another 28 percent in communities in the middle third of the income distribution.
This is a radical departure from the historic public housing program in which the great majority of units
were built in low-income communities. The LIHTC program does, therefore, appear to expand housing
opportunities for low-income families in traditionally higher income neighborhoods. To the extent that
mixed-income development at the community level is desirable, the LIHTC program appears successful
in this regard.
On the other hand, it is also important to recognize that not everyone gains from such a
development strategy. Although we find that LIHTC development has a positive amenity effect on
nearby low-income communities, our results are strongly suggestive that LIHTC development has a
negative stigma effect on nearby high-income areas. This latter result implies that LIHTC development
imposes windfall loses on existing residents and landowners in higher income communities. It is beyond
the scope of this paper to offer a full accounting of the costs and benefits of the LIHTC program. But
stigma effects in high-income communities along with crowd out greatly increase the cost of the LIHTC
program. These issues warrant close attention in any further discussion about possible expansion of the
LIHTC program, as well as on-going debates about place- versus tenant-based (e.g. voucher) housing
assistance.
23
References
Andreoni, James and Abigail Payne (2003) “Do Government Grants to Private Charities Crowd Out
Giving or Fundraising?” The American Economic Review, 93(3) June, 792-812.
Baum-Snow, Nathaniel and Justin Marion (2007). “The Effects of Low Income Housing Developments
on Neighborhoods.” Brown University working paper, mimeo.
Berry, Steven T., and Joel Waldfogel (1999). Public Radio in the United States: Does it Correct Market
Failure or Cannibalize Commercial Stations?” Journal of Public Economics, 71, 189-211.
Cameron, Colin and Pravin Trivedi (2006). Microeconometrics: Methods and Applications, Cambridge
University Press, New York.
Culter, David M. and Jonathan Gruber, (1996). “Does Public Insurance Crowd Out Private Insurance.”
Quarterly Journal of Economics, 111(2) May, 391-430.
Currie, Janet, and Aaron Yelowitz (2000). “Are Public Housing Projects Good for Kids?” Journal of
Public Economics 75, 99-124.
Cummings, Jean L., and Denise DiPasquale. (1999) “The Low-Income Housing Tax Credit: The First
Ten Years.” Housing Policy Debate 10 (2), 257-267.
DiPasquale, Denise (1999), “Why Don’t We Know More About Housing Supply?” Journal of Real
Estate Finance and Economics, 18, 9-21.
DiPasquale, Denise, Dennis Fricke, and Daniel Garcia-Diaz. (2003) “Comparing the Costs of Federal
Housing Assistance Programs.” FRBNY Economic Policy Review June, 147-165.
DiPasquale, Denise and William Wheaton (1992), “The Cost of Capital, Tax Reform and the Future of
the Rental Housing Market,” Journal of Urban Economics, 31(3), 337-359.
Eriksen, Michael D. (2007) “Neighborhoods, Risk, and the Value of Low-Income Housing Tax Credits.”
Center for Policy Research Working Paper.
Glaeser, Edward L. and Joesph Gyourko (2005). “Urban Decline and Durable Housing.” Journal of
Political Economy 113, 345-375.
Hansen, Lars P. (1982): "Large Sample Properties of Generalized Method of Moments Estimators,"
Econometrica, 50, 1029- 1054.
Hanushek, Eric A., and Quigley, John M. (1980), ‘‘What Is the Price Elasticity of Housing Demand?’’
Review of Economics and Statistics. 62, August, 449–54.
Jacob, Brian A., (2004). “Public Housing, Housing Vouchers, and Student Achievement: Evidence From
Housing Demolitions in Chicago.” American Economic Review, 94(1), 233-258.
Jencks, C., and S. Mayer, (1990). “The social consequences of growing up in a poor neighborhood.” In:
Lynn, L., McGeary, M. (Eds.), Inner-city Poverty in the United States, National Academy Press,
Washington, DC, 111–186.
24
Malpezzi, Stephen and Kerry Vandell (2002) “Does the low-income housing tax credit increase the
supply of housing?” Journal of Housing Economics, 11 (4), 360-380.
Mayer, Christopher, and Tsur Somerville (2000), “Residential Construction: Using the Urban Growth
Model to Estimate Housing Supply,” Journal of Urban Economics, 48, 85-109.
Murray, Michael. (1983) “Subsidized and Unsubsidized Housing Starts: 1961-1977.” Review of
Economics and Statistics 65 (4), 590-597.
Murray, Michael (1999), “Subsidized and unsubsidized housing stocks 1935 to 1987: Crowding Out and
Cointegration,” Journal of Real Estate Finance and Economics 18, 107–124.
Murray, Michael P. (2006) Avoiding invalid instruments and coping with weak instruments.” Journal of
Economic Perspectives, 20, 111–32.
Olsen, Edgar O. "Housing Programs for Low-Income Households," in Means-Tested Transfer Programs
in the U.S., ed., Robert Moffitt. National Bureau of Economic Research. Chicago: University of
Chicago Press, 2003.
Olsen, Edgar O. (2007), “A Review of A Primer on U.S. Housing Markets and Housing Policies,”
Regional Science and Urban Economics.
Quigley, John M. (200), “A Decent Home: Housing Policy in Perspective.” Brookings-Wharton Papers
on Urban Affairs, 1(1), 53–100.
Rosen, Harvey (1979), “Housing Decisions and the U.S. Income Tax.” Journal of Public Economics,
11, 1-23.
Rosenthal, Stuart S., John V. Duca, and Stuart A. Gabriel (1991), “Credit Rationing and the Demand for
Owner-Occupied Housing.” Journal of Urban Economics, 29, 48-63.
Rosenthal, Stuart S., and Robert W. Helsley (1994), “Redevelopment and the Urban Land Price
Gradient.” Journal of Urban Economics, 35 (March), 182–200.
Rosenthal, Stuart (1999). “Housing Supply: The Other Half of the Market. A Note from the Editor,”
Journal of Real Estate Finance and Economics, 18: 5-8.
Rosenthal, Stuart S. (forthcoming), “Old Homes and Poor Neighborhoods: A Model of Urban Decline
and Renewal,” Journal of Urban Economics.
Sinai, Todd and Joel Waldfogel (2005), “Do low-income housing subsidies increase the occupied housing
stock?” Journal of Public Economics, 89, 2137-2164 .
Stegman, Michael A. (1991), "The Excessive Costs of Creative Finance: Growing Inefficiencies
in the Production of Low-Income Housing." Housing Policy Debate 2, 357-373.
Stock, J., J.Wright, and M. Yogo (2002). “GMM, Weak Instruments, and Weak Identification.” Journal
of Business and Economic Statistics 20, 518–530.
U.S. Congress, Joint Committee on Taxation (1987) “General Explanation of the Tax Reform Act of
1986” 100th Congress, 1st Session, 152.
25
U.S. Congress, Joint Committee on Taxation (2005) Committee Prints, 109th Congress, “Estimates of
Federal Tax Expenditures for Fiscal Years 2005-2009”, JCS-1-05, January 12, 2005.
U.S. Department of Treasury, Internal Revenue Service, Internal Revenue Code, Section 42.
Wallace, James E. (1995) “Financing Affordable Housing in the United States.” Housing Policy
Debate 6, 785-814.
Woolridge, Jeffrey M. (2003), “Cluster-Sample Methods in Applied Econometrics” The American
Economic Review 93(2), 133-138
26
Total Units Constructed (in Thousands)
-75
0
75
150
225
300
375
450
525
1 93 6
-19 4
0
1 94 1
-19 4
5
1 94 6
-19 5
0
1 95 1
-19 5
5
LIHTC Construction
1 95 6
-19 6
0
1 96 1
-19 6
5
1 96 6
-19 7
0
1 97 1
-19 7
5
1 97 6
-19 8
0
1 98 1
-19 8
5
Public Housing Construction (minus demolitions)
1 98 6
-19 9
0
Figure 1. Place-Based Subsidized Construction and Demolitions of Housing Units
1 99 1
-19 9
5
1 99 6
-20 0
0
27
Figure 2a. Location of Low-Income
Housing T ax Credit Units by 2000
Neighborhood Income Status
Middle
Low
196,554
401,861
28%
56%
High
111,724
16%
Figure 2b. Location of T raditional Public
Housing Units by 2000 Neighborhood
Income Status
Middle High
125,994 55,597
Low 16% 7%
618,107
77%
28
Figure 3a: Crowd Out of Rental Housing
Market
House
Rent
Supply without
LIHTC
Demand
LIHTC Construction
Supply with
LIHTC
RNo LIHTC
RLIHTC
H1 H2 H3 Quantity of newly built
rental housing
Increase in new
Crowd out effect on
construction
private stock
Figure 3b: Crowd Out of Rental Housing With Stigma
Market
House
Rent
Supply without
Demand without LIHTC
LIHTC
LIHTC Construction
Supply with
Demand with LIHTC
LIHTC
RNo LIHTC
RLIHTC
RLIHTC & Stigma
H1 H2 H3 Quantity of newly built
rental housing
Stigma effect on
equilibrium stock Crowd out effect on
private stock
29
Figure 3c: Crowd Out of Rental Housing with Amenity Effects
Market
House
Rent
Supply without
Demand without Demand with LIHTC
LIHTC LIHTC
LIHTC Supply with
Construction LIHTC
RNo LIHTC =
RLIHTC & Amenity
RLIHTC
H1 H2 H3 Quantity of newly built
rental housing
Crowd out effect on
equilibrium stock equal
to amenity effect
30
Figure 4: GMM and OLS Estimates of Crowd Out (Table 4)
0.1
0
Coefficient Estimate
-0.1
-0.2
-0.3
-0.4
-0.5
0.5 1 2 3 4 5 10
Circle Radius in Miles
GMM-IV OLS
Figure 5: GMM Estimates of Crowd Out By Neighborhood Income (Tables 5a-5c)
0.4
Coefficient Estimate
0
-0.4
-0.8
-1.2
0.5 1 2 3 4 5 10
Circle Radius in Miles
Low-Income (Table 5a) Middle-Income (Table 5b) High-Income (Table 5c)
31
Table 1. Federal Government Low-Income Housing Expenditures (2005-2006)1
(in Millions of Dollars)
2005 2006
2
Internal Revenue Service
Low-Income Housing Tax Credit 4,700 4,900
Preferential Depreciation Allowance 3,800 4,200
State-issued Tax Exempt Financing for Rental Housing 300 300
Department of Housing and Urban Development3
Housing Choice Vouchers (HCV) 20,064 20,917
Public Housing 5,017 5,734
Other HUD Programs 8,734 4,559
Department of Agriculture3
Rural Housing Administration 1,369 1,029
Total 43,984 41,639
1
LIHTC costs reflect foregone tax revenues associated with the 10-years of tax credit allocations.
Those costs are expected to increase sharply in the next several years given an increase in
LIHTC allocations since 2001. Public housing operating costs are likely to decline in coming
years as increasing numbers of public housing units are demolished.
2
Tax Expenditures as Estimated by Joint Committee on Taxation (2005).
3
Budget of the United States, Office of Management and Budget (2006).
32
Table 2. Sample Means at the Census Tract Level
Values in ___ Income Census Tractsa
Attributes in the Year 2000 All Low Middle High
Private Construction of Rental Housing 1990 - 2000 58.57 46.53 58.05 72.16
LIHTC Construction 1987 - 2000 13.80 22.03 11.52 7.33
# of Rental Units Constructed 1985-1989 56.53 52.73 58.91 58.03
# of Rental Units Constructed 1980-1984 55.73 56.83 58.22 51.91
# of Rental Units Constructed 1970-1979 117.73 131.27 123.42 97.39
# of Rental Units Constructed Prior to 1970 310.40 463.22 270.45 191.07
# of Owner-Occupied Units Constructed 1985-1989 96.50 39.09 103.69 149.59
# of Owner-Occupied Constructed 1980-1984 77.54 37.58 86.15 110.69
# of Owner-Occupied Units Constructed 1970-1979 183.24 94.00 205.99 253.54
# of Owner-Occupied Units Constructed Prior to 1970 539.76 475.19 593.36 551.28
Surface Area not Covered by Water (Square Miles) 14.38 12.61 20.22 10.12
Distance to Central Business District of MSA 12.66 11.01 14.40 12.62
b
Observations 46,930 16,165 15,826 14,939
a
Neighborhood income status is defined by ranking 1990 census tract average incomes by terciles for each MSA.
b
Sample restricted to census tracts located in an MSA as of the year 2000.
33
Table 3. Sample Means in Circles Drawn Around Census Tract Centroids
Values in Circles Drawn Around
Census Tracts That are ____ Incomea
Panel A: 0.5 Mile Circle Radius All Low Middle High
Private Construction of Rental Housing 1990 - 2000 43 53 29 29
LIHTC Construction 1987 - 2000 16 36 7 5
# of Rental Units Constructed 1985-1989 49 70 42 33
# of Rental Units Constructed 1980-1984 49 75 40 30
# of Rental Units Constructed 1970-1979 118 183 100 67
# of Rental Units Constructed Prior to 1970 616 1037 473 311
# of Owner-Occupied Units Constructed 1985-1989 35 21 34 50
# of Owner-Occupied Constructed 1980-1984 31 21 32 40
# of Owner-Occupied Units Constructed 1970-1979 78 54 83 98
# of Owner-Occupied Units Constructed Prior to 1970 529 651 533 391
Surface Area not Covered by Water (square miles) 0.74 0.74 0.75 0.73
Panel B: 1.0 Mile Circle Radius
Private Construction of Rental Housing 1990 - 2000 156 168 82 95
LIHTC Construction 1987 - 2000 59 122 31 21
# of Rental Units Constructed 1985-1989 175 241 155 127
# of Rental Units Constructed 1980-1984 175 256 146 118
# of Rental Units Constructed 1970-1979 423 624 362 270
# of Rental Units Constructed Prior to 1970 2179 3592 1712 1144
# of Owner-Occupied Units Constructed 1985-1989 128 85 126 176
# of Owner-Occupied Constructed 1980-1984 114 81 117 145
# of Owner-Occupied Units Constructed 1970-1979 286 209 303 349
# of Owner-Occupied Units Constructed Prior to 1970 1927 2429 1907 1406
Surface Area not Covered by Water (square miles) 2.94 2.92 2.98 2.91
Panel C: 10 Mile Circle Radius
Private Construction of Rental Housing 1990 - 2000 8,443 4,142 2,418 3,274
LIHTC Construction 1987 - 2000 2,476 3,287 2,144 1,950
# of Rental Units Constructed 1985-1989 9,054 10,402 8,306 8,387
# of Rental Units Constructed 1980-1984 8,444 9,690 7,754 7,826
# of Rental Units Constructed 1970-1979 20,691 24,563 19,002 18,292
# of Rental Units Constructed Prior to 1970 95,493 125,945 87,726 70,771
# of Owner-Occupied Units Constructed 1985-1989 8,352 7,718 8,097 9,308
# of Owner-Occupied Constructed 1980-1984 7,068 6,684 6,821 7,745
# of Owner-Occupied Units Constructed 1970-1979 17,911 17,653 17,322 18,813
# of Owner-Occupied Units Constructed Prior to 1970 102,911 125,936 95,380 85,975
Surface Area not Covered by Water (square miles) 276.25 271.80 278.92 278.22
Distance to Central Business District of MSA (miles) 12.66 11.01 14.40 12.62
Observations 46,930 16,165 15,826 14,939
a
Neighborhood income status is defined by ranking 1990 census tract average incomes by terciles for each MSA.
34
Table 4. LIHTC Crowd Out of Unsubsidized Rental Housing Construction 1990 to 2000a
Circle Radius in miles Around Each Census Tract Centroid
0.5 1.0 2.0 3.0 4.0 5.0 10.0
b
GMM Estimates
0.0354 0.0102 -0.2424 -0.2695 -0.2545 -0.2774 -0.3327
LIHTC construction 1987-2000 (t-statistic)c
(0.65) (0.18) (4.85) (5.53) (5.52) (6.17) (5.39)
OLS Estimates
-0.3495 -0.3737 -0.3962 -0.3946 -0.3762 -0.3692 -0.3502
LIHTC construction 1987-2000 (t-statistic)c
(3.83) (3.98) (3.83) (3.39) (2.94) (2.76) (2.92)
Summary Measures
Observations 46,930 46,930 46,930 46,930 46,930 46,930 46,930
MSA Fixed Effects 331 331 331 331 331 331 331
d
F-stat for Significance of Excluded Instruments 127.93 153.75 75.76 69.04 83.23 144.79 49.56
e 20.37 22.71 17.26 18.65 18.60 18.13 16.55
Hansen (1982) J-Statistic (P-value)
(0.2556) (0.1590) (0.4367) (0.3491) (0.3522) (0.3809) (0.4851)
a
Each of the models includes additional controls for the age distribution of owner-occupied and rental housing stocks as of 1990, distance to the CBD, and the area within
the circle not covered by water. All variables (dependent, independent, and instrumental) were measured over circle areas of radii indicated above. GMM results for all of
the control variables in the model are provided in the appendix for circle radii of 0.5, 1.0, 5.0, and 10.0 miles.
b
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single family and various sizes of multi-family)
and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental housing stock.
c
Corresponding t-ratios are based on MSA-clustered standard errors.
d
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
e
The Hansen (1982) J-Statistic is a test of the overidentifying restrictions with a joint null hypothesis that the excluded instruments are uncorrelated with the error term and
correctly excluded from the estimated equation. Rejection of the null hypothesis calls into question the validity of the instruments (for more details see Davidson and
MacKinnon (2004)).
35
Table 5a. LIHTC Crowd Out of Unsubsidized Rental Housing Construction 1990 to 2000 in Low-Income Areasa
Circle Radius in miles Around Each Census Tract Centroid
0.5 1.0 2.0 3.0 4.0 5.0 10.0
b
GMM Estimates
0.2621 0.1382 -0.0563 -0.0475 -0.0460 0.0577 -0.1162
LIHTC construction 1987-2000 (t-statistic)c
(7.15) (2.48) (1.07) (0.95) (1.13) (1.38) (2.19)
OLS Estimates
-0.3004 -0.2967 -0.2730 -0.1799 -0.0883 -0.0221 -0.0702
LIHTC construction 1987-2000 (t-statistic)c
(2.66) (2.45) (2.00) (1.13) (0.52) (0.13) (0.61)
Summary Measures
Observations 16,165 16,165 16,165 16,165 16,165 16,165 16,165
MSA Fixed Effects 331 331 331 331 331 331 331
d
F-stat for Significance of Excluded Instruments 37.58 113.87 87.81 69.81 67.16 108.27 39.18
e 20.21 21.05 25.93 24.89 19.00 26.60 13.48
Hansen (1982) J-Statistic (P-value)
(0.2632) (0.2241) (0.0758) (0.0971) (0.3287) (0.0643) (0.7032 )
a
Each of the models includes additional controls for the age distribution of owner-occupied and rental housing stocks as of 1990, distance to the CBD, and the area within
the circle not covered by water. All variables (dependent, independent, and instrumental) were measured over circle areas of radii as indicated above. Low-Income Areas
defined by a tract’s average income located in the bottom third of its respective MSA distribution.
b
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single family and various sizes of multi-family)
and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental housing stock.
c
Corresponding t-ratios are based on MSA-clustered standard errors.
d
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
e
The Hansen (1982) J-Statistic is a test of the overidentifying restrictions with a joint null hypothesis that the excluded instruments are uncorrelated with the error term and
correctly excluded from the estimated equation. Rejection of the null hypothesis calls into question the validity of the instruments (for more details see Davidson and
MacKinnon (2004)).
36
Table 5b. LIHTC Crowd Out of Unsubsidized Rental Housing Construction 1990 to 2000 in Middle-Income Areasa
Circle Radius in miles Around Each Census Tract Centroid
0.5 1.0 2.0 3.0 4.0 5.0 10.0
b
GMM Estimates
-0.3999 -0.5703 -0.4192 -0.3586 -0.3192 -0.2590 -0.1360
LIHTC construction 1987-2000 (t-statistic)c
(6.02) (15.88) (18.94) (12.55) (9.82) (8.95) (4.69)
OLS Estimates
-0.4087 -0.3035 -0.2240 -0.2030 -0.2017 -0.1949 -0.1113
LIHTC construction 1987-2000 (t-statistic)c
(9.09) (6.28) (3.32) (3.09) (3.40) (0.34) (1.66)
Summary Measures
Observations 15,826 15,826 15,826 15,826 15,826 15,826 15,826
MSA Fixed Effects 331 331 331 331 331 331 331
d
F-stat for Significance of Excluded Instruments 10.57 27.59 48.86 29.40 30.52 47.26 38.50
e 20.91 24.94 25.08 21.85 20.44 18.82 17.83
Hansen (1982) J-Statistic (P-value)
(0.2305) (0.0960) (0.0929) (0.1907) (0.2525) (0.3389) (0.3998)
a
Each of the models includes additional controls for the age distribution of owner-occupied and rental housing stocks as of 1990, distance to the CBD, and the area within
the circle not covered by water. All variables (dependent, independent, and instrumental) were measured over circle areas of radii as indicated above. Low-Income Areas
defined by a tract’s average income located in the middle third of its respective MSA distribution.
b
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single family and various sizes of multi-family)
and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental housing stock.
c
Corresponding t-ratios are based on MSA-clustered standard errors.
d
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
e
The Hansen (1982) J-Statistic is a test of the overidentifying restrictions with a joint null hypothesis that the excluded instruments are uncorrelated with the error term and
correctly excluded from the estimated equation. Rejection of the null hypothesis calls into question the validity of the instruments (for more details see Davidson and
MacKinnon (2004)).
37
Table 5c. LIHTC Crowd Out of Unsubsidized Rental Housing Construction 1990 to 2000 in High-Income Areasa
Circle Radius in miles Around Each Census Tract Centroid
0.5 1.0 2.0 3.0 4.0 5.0 10.0
b
GMM Estimates
-1.0754 -0.4383 -0.2457 -0.1755 -0.1593 -0.1628 -0.0766
LIHTC construction 1987-2000 (t-statistic)c
(17.30) (8.92) (8.51) (8.72) (6.00) (4.75) (1.43)
OLS Estimates
-0.4037 -0.2382 -0.1471 -0.1585 -0.1582 -0.1544 -0.1293
LIHTC construction 1987-2000 (t-statistic)c
(5.03) (3.63) (2.01) (2.14) (2.14) (2.05) (1.88)
Summary Measures
Observations 14,939 14,939 14,939 14,939 14,939 14,939 14,939
MSA Fixed Effects 331 331 331 331 331 331 331
d
F-stat for Significance of Excluded Instruments 72.20 49.70 63.32 46.78 69.19 136.89 40.84
e 22.95 20.94 28.25 31.00 32.85 34.84 27.45
Hansen (1982) J-Statistic (P-value)
(0.1508) (0.2288) (0.0421) (0.0200) (0.0118) (0.0065) (0.0517)
a
Each of the models includes additional controls for the age distribution of owner-occupied and rental housing stocks as of 1990, distance to the CBD, and the area within
the circle not covered by water. All variables (dependent, independent, and instrumental) were measured over circle areas of radii as indicated above. Low-Income Areas
defined by a tract’s average income located in the upper third of its respective MSA distribution.
b
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single family and various sizes of multi-family)
and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental housing stock.
c
Corresponding t-ratios are based on MSA-clustered standard errors.
d
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
e
The Hansen (1982) J-Statistic is a test of the overidentifying restrictions with a joint null hypothesis that the excluded instruments are uncorrelated with the error term and
correctly excluded from the estimated equation. Rejection of the null hypothesis calls into question the validity of the instruments (for more details see Davidson and
MacKinnon (2004)).
38
Appendix: Supplemental Tables
Table A-1. Impact on Private Rental Construction in All Neighborhoods 1990 to 2000
in Response to LIHTC Construction from 1987 to 2000
Radius of Circles in Miles
GMM Estimatesa (t-statistics)b
0.5 1.0 5.0 10.0
LIHTC Construction 1987-2000 0.0354 0.0102 -0.2545 -0.3327
(0.65) (0.18) (5.52) (5.39)
# of Rental Units Constructed 1985-1989 0.3461 0.4341 0.5397 0.3777
(25.67) (16.81) (25.89) (10.55)
# of Rental Units Constructed 1980-1984 0.1547 0.1837 0.2168 0.3943
(12.72) (9.74) (5.56) (8.24)
# of Rental Units Constructed 1970-1979 0.0135 0.0041 0.0205 0.0428
(2.31) (0.54) (1.69) (2.03)
# of Rental Units Constructed 1970 or earlier 0.0083 0.0095 0.0106 0.0138
(13.06) (15.75) (10.40) (10.28)
# of Owner-Occupied Units Constructed 1985-1989 0.1317 0.1316 0.2705 0.2353
(7.86) (5.47) (6.22) (3.66)
# of Owner-Occupied Units Constructed 1980-1984 -0.0518 -0.0173 -0.0810 -0.0556
(2.53) (0.54) (1.40) (0.62)
# of Owner-Occupied Units Constructed 1970-1979 -0.0153 -0.0158 -0.0416 -0.0432
(2.83) (1.53) (2.57) (2.19)
# of Owner-Occupied Units Constructed 1970 or earlier -0.0167 -0.0133 -0.0089 -0.0078
(8.59) (8.81) (7.48) (7.17)
Distance to Central Business District of MSA -0.4818 -0.7859 -10.8006 -11.9175
(4.61) (2.58) (3.18) (1.09)
Surface Area not Covered by Water (Square Miles) 23.1592 13.2437 1.5941 0.2646
(6.14) (4.63) (1.03) (0.26)
Observations 46,930 46,930 46,930 46,930
MSA Fixed Effects 331 331 331 331
F-stat for Significance of Excluded Instrumentsc 127.93 153.75 144.79 49.56
d
Hansen (1982) J-Statistic (P-value) 20.37 22.71 18.13 16.55
(0.2556) (0.1590) (0.3809) (0.4851)
a
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single
family and various sizes of multi-family) and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental
housing stock. All variables (dependent, independent, and instrumental) were measured over circle areas of radii indicated above.
b
Corresponding t-ratios are based on MSA-clustered standard errors as described in the text.
c
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
d
Hansen (1982) J-test is a test of overidentifying restrictions with a joint null hypothesis that the excluded instruments are
uncorrelated with the error term and correctly excluded from the estimated equation. Rejection of the null hypothesis calls into
question the validity of the instruments (for more details see Davidson and MacKinnon (2004).
39
Table A-2. Impact on Private Rental Construction in Low-Income Neighborhoods 1990 to 2000
in Response to LIHTC Construction from 1987 to 2000
Radius of Circles in Miles
GMM Estimatesa (t-statistics)b
0.5 1.0 5.0 10.0
LIHTC Construction 1987-2000 0.2621 0.1382 -0.0460 -0.1162
(7.15) (2.48) (1.13) (2.19)
# of Rental Units Constructed 1985-1989 0.2751 0.3100 0.2582 0.3074
(18.22) (17.03) (14.01) (22.79)
# of Rental Units Constructed 1980-1984 0.1138 0.1660 0.1880 0.1009
(6.10) (6.08) (5.05) (2.35)
# of Rental Units Constructed 1970-1979 -0.0185 -0.0068 0.0108 -0.0052
(1.87) (0.75) (0.66) (0.30)
# of Rental Units Constructed 1970 or earlier 0.0008 0.0034 0.0160 0.0154
(0.96) (2.89) (13.65) (13.13)
# of Owner-Occupied Units Constructed 1985-1989 0.1268 0.1408 0.0271 0.0364
(3.26) (3.92) (0.84) (1.07)
# of Owner-Occupied Units Constructed 1980-1984 -0.0043 -0.0291 -0.1482 -0.2215
(0.07) (0.45) (2.71) (2.98)
# of Owner-Occupied Units Constructed 1970-1979 -0.0164 -0.0481 -0.0345 0.0238
(1.05) (2.61) (2.44) (1.48)
# of Owner-Occupied Units Constructed 1970 or earlier -0.0166 -0.0171 -0.0157 -0.0088
(4.49) (4.78) (8.94) (6.72)
Distance to Central Business District of MSA -0.6042 -1.3503 -5.7270 -8.7532
(5.40) (4.13) (2.37) (1.56)
Surface Area not Covered by Water (Square Miles) 34.4016 18.8986 11.1282 1.9005
(2.37) (2.08) (3.69) (1.77)
Observations 16,165 16,165 16,165 16,165
MSA Fixed Effects 331 331 331 331
F-stat for Significance of Excluded Instrumentsc 37.58 113.87 108.27 39.18
Hansen (1982) J-Statistic (P-value)d 20.22 21.05 26.60 13.48
(0.2632) (0.2241) (0.0643) (0.7032)
a
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single
family and various sizes of multi-family) and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental
housing stock. All variables (dependent, independent, and instrumental) were measured over circle areas of radii indicated above.
b
Corresponding t-ratios are based on MSA-clustered standard errors as described in the text.
c
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
d
Hansen (1982) J-test is a test of overidentifying restrictions with a joint null hypothesis that the excluded instruments are
uncorrelated with the error term and correctly excluded from the estimated equation. Rejection of the null hypothesis calls into
question the validity of the instruments (for more details see Davidson and MacKinnon (2004).
40
Table A-3. Impact on Private Rental Construction in Middle-Income Neighborhoods 1990 to 2000
in Response to LIHTC Construction from 1987 to 2000
Radius of Circles in Miles
GMM Estimatesa (t-statistics)b
0.5 1.0 5.0 10.0
LIHTC Construction 1987-2000 -0.3999 -0.5703 -0.3192 -0.1360
(6.02) (15.88) (9.82) (4.69)
# of Rental Units Constructed 1985-1989 0.3714 0.3156 0.1936 0.1313
(13.84) (14.81) (16.07) (9.27)
# of Rental Units Constructed 1980-1984 0.0986 0.0866 0.0383 0.0428
(7.22) (6.38) (1.94) (1.50)
# of Rental Units Constructed 1970-1979 0.0103 -0.0020 -0.0118 0.0035
(1.99) (0.40) (1.80) (0.28)
# of Rental Units Constructed 1970 or earlier -0.0014 0.0013 0.0008 0.0031
(4.02) (5.60) (2.07) (4.18)
# of Owner-Occupied Units Constructed 1985-1989 0.0989 0.1223 0.0647 0.0352
(7.61) (6.37) (3.59) (2.34)
# of Owner-Occupied Units Constructed 1980-1984 -0.0283 -0.0290 0.0332 0.0793
(1.75) (1.85) (1.53) (2.56)
# of Owner-Occupied Units Constructed 1970-1979 -0.0004 0.0086 -0.0031 -0.0073
(0.05) (1.36) (0.38) (0.97)
# of Owner-Occupied Units Constructed 1970 or earlier -0.0028 -0.0008 0.0033 0.0006
(4.45) (1.55) (6.46) (1.11)
Distance to Central Business District of MSA -0.3448 -0.7870 -2.9349 -0.8227
(5.41) (4.31) (2.24) (0.33)
Surface Area not Covered by Water (Square Miles) 14.3984 5.6086 3.1669 0.5299
(5.02) (3.16) (3.15) (1.10)
Observations 15,826 15,826 15,826 15,826
MSA Fixed Effects 331 331 331 331
F-stat for Significance of Excluded Instrumentsc 10.57 27.59 47.26 38.50
Hansen (1982) J-Statistic (P-value)d 20.91 24.94 18.82 17.83
(0.2305) (0.0960) (0.3389) (0.3998)
a
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single
family and various sizes of multi-family) and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental
housing stock. All variables (dependent, independent, and instrumental) were measured over circle areas of radii indicated above.
b
Corresponding t-ratios are based on MSA-clustered standard errors as described in the text.
c
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
d
Hansen (1982) J-test is a test of overidentifying restrictions with a joint null hypothesis that the excluded instruments are
uncorrelated with the error term and correctly excluded from the estimated equation. Rejection of the null hypothesis calls into
question the validity of the instruments (for more details see Davidson and MacKinnon (2004).
41
Table A-4. Impact on Private Rental Construction in High-Income Neighborhoods 1990 to 2000
in Response to LIHTC Construction from 1987 to 2000
Radius of Circles in Miles
GMM Estimatesa (t-statistics)b
0.5 1.0 5.0 10.0
LIHTC Construction 1987-2000 -1.0754 -0.4383 -0.1593 -0.0766
(17.30) (8.92) (6.00) (1.43)
# of Rental Units Constructed 1985-1989 0.4363 0.4309 0.1681 0.0807
(15.52) (16.41) (8.09) (4.56)
# of Rental Units Constructed 1980-1984 0.1531 0.1200 0.0906 0.2725
(7.03) (6.45) (2.79) (7.42)
# of Rental Units Constructed 1970-1979 0.0479 0.0119 0.0378 0.0049
(6.25) (1.95) (3.45) (0.19)
# of Rental Units Constructed 1970 or earlier 0.0012 -0.0024 -0.0045 -0.0061
(2.47) (4.94) (5.88) (3.14)
# of Owner-Occupied Units Constructed 1985-1989 0.0773 0.1123 0.1239 0.1855
(3.88) (5.02) (4.01) (4.93)
# of Owner-Occupied Units Constructed 1980-1984 -0.0553 -0.0272 0.1366 0.0720
(3.07) (1.74) (3.46) (1.13)
# of Owner-Occupied Units Constructed 1970-1979 -0.0182 -0.0227 -0.0538 -0.0443
(2.16) (2.71) (4.18) (3.10)
# of Owner-Occupied Units Constructed 1970 or earlier -0.0089 -0.0068 -0.0055 -0.0041
(7.54) (8.32) (5.76) (3.85)
Distance to Central Business District of MSA -0.4549 -1.1769 -8.6142 -14.4049
(5.34) (4.36) (3.33) (1.93)
Surface Area not Covered by Water (Square Miles) 19.5199 9.4554 -1.1215 -2.4112
(5.61) (4.41) (0.56) (2.22)
Observations 14,939 14,939 14,939 14,939
MSA Fixed Effects 331 331 331 331
F-stat for Significance of Excluded Instrumentsc 72.20 49.70 136.89 40.84
Hansen (1982) J-Statistic (P-value)d 22.95 20.94 34.84 27.46
(0.1508) (0.2288) (0.0065) (0.0517)
a
LIHTC Construction 1987 – 2000 treated endogenously. Instruments include 1970 distribution of housing structure type (single
family and various sizes of multi-family) and bedroom counts (0, 1, 2, 3, and 4+ bedrooms), separately for owner-occupied and rental
housing stock. All variables (dependent, independent, and instrumental) were measured over circle areas of radii indicated above.
b
Corresponding t-ratios are based on MSA-clustered standard errors as described in the text.
c
The F-statistic for the joint significance of excluded instruments accounts the clustering of standard errors by MSA.
d
Hansen (1982) J-test is a test of overidentifying restrictions with a joint null hypothesis that the excluded instruments are
uncorrelated with the error term and correctly excluded from the estimated equation. Rejection of the null hypothesis calls into
question the validity of the instruments (for more details see Davidson and MacKinnon (2004)
42
