literature review
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=rhpd20 Download by: [University of Utah] Date: 21 July 2017, At: 14:08 Housing Policy Debate ISSN: 1051-1482 (Print) 2152-050X (Online) Journal homepage: http://www.tandfonline.com/loi/rhpd20 The Low ‐Income Housing Tax Credit An Analysis of the First Ten Years Jean L. Cummings & Denise DiPasquale To cite this article: Jean L. Cummings & Denise DiPasquale (1999) The Low ‐Income Housing Tax Credit An Analysis of the First Ten Years, Housing Policy Debate, 10:2, 251-307, DOI:
10.1080/10511482.1999.9521332 To link to this article: http://dx.doi.org/10.1080/10511482.1999.9521332 Published online: 31 Mar 2010.Submit your article to this journal Article views: 333View related articles Citing articles: 46 View citing articles Housing Policy Debate · Volume 10, Issue 2251 qFannie Mae Foundation 1999. All Rights Reserved. The Low-Income Housing Tax Credit:
An Analysis of the First Ten Years Jean L. Cummings and Denise DiPasquale City Research Abstract The Low-Income Housing Tax Credit (LIHTC) has been thede factofederal rental housing production program since its creation in the Tax Reform Act of 1986. In this article, using a detailed database on 2,554 LIHTC projects, we analyze the costs of building these projects, where they are built, their financial viability, whom they serve, who finances them, and the size of the subsidies provided to them.
The LIHTC is a flexible program that has built different types of housing in vari- ous markets. While LIHTC projects serve low- and moderate-income households, their rents are beyond the reach of many poor households without additional sub- sidy. Revenues just cover costs for many LIHTC projects. Over time, considerably more of each tax-credit dollar has ended up in the projects, and returns to equity investors have dropped significantly, perhaps reflecting an increased understand- ing of project risks. We estimate that LIHTC projects developed by nonprofits are 20.3 percent more expensive than those developed by for-profits.
Keywords:Low-income housing; Multifamily; Tax policy Introduction The Low-Income Housing Tax Credit (LIHTC) has been the major federal program for producing affordable rental housing since its creation as part of the Tax Reform Act of 1986 (TRA). 1The LIHTC represents a partnership among a variety of public and private sec- tor actors. The basic premise of the LIHTC is to offer federal tax credits to private investors in return for their providing equity for the development of affordable rental housing. The program is ad- ministered by state (or, in a few cases, local) housing policy makers who set goals for the program, review projects proposed by for-profit 1A major feature of the TRA was the elimination of much of the favorable tax treatment of real estate. The LIHTC was created in the final hours of the TRA de- bate when lawmakers realized that, at a time when other federal production pro- grams were being suspended, there were no tax incentives left for affordable rental housing. The LIHTC was viewed as a very targeted tax incentive for building rental housing for low- and moderate-income households. See Case (1991).
252Jean L. Cummings and Denise DiPasquale and nonprofit developers, monitor the reasonableness of project costs, and take responsibility for ensuring that projects stay in com- pliance and that approved projects receive only the tax credits nec- essary to make the project work. The Internal Revenue Service (IRS) is responsible for monitoring compliance and state perfor- mance.
By bringing these various actors together, the LIHTC program is designed to bring the efficiency and discipline of the private market to the building of affordable rental housing. Investor participation is expected to add further oversight to the program, since return to the investors is dependent on the project's staying in compliance.
By allocating the tax credits through the states, the program pro- vides the flexibility to build housing that meets local market needs.
While the LIHTC program may not have been designed to reach the poorest renter households, it is effectively the only federal produc- tion program. As a result, there may be considerable pressure from housing advocates and state and local policy makers to use the pro- gram to serve lower-income tenants.
While all participants share the goal of producing financially viable projects with revenues covering costs, government officials and ad- vocates for the poor often have policy goals that may be viewed by private participants as bringing too much additional risk, a view presented by many of the investors we interviewed (Cummings and DiPasquale 1998a). For example, states may target populations with special needs or may favor the provision of social services as part of a project, both of which can increase management risks.
They may want to serve lower-income tenants who pay lower rents; target underserved areas, which may increase development costs; or favor nonprofit developers (who may bring less experience and less financial capital to a project than larger for-profit developers) to in- crease community participation or to achieve broader community development goals. Clearly, there can be real tension between the policy goals for LIHTC projects and financial viability. To succeed, the program and its participants must meet policy goals while en- suring the investment quality demanded by private market partici- pants.
Despite the critical role of the LIHTC in providing affordable rental housing over the past decade, there is very little detailed historical information on the characteristics and performance of the program's rental housing developments. In this article, we provide an analysis based on detailed data on 2,554 LIHTC projects, covering the entire history of the program, which we collected from four syndicators of tax credit equity. This article builds on our initial report on these data (Cummings and DiPasquale 1998a), in which we provided an in-depth description of the LIHTC projects in this sample.
The Low-Income Housing Tax Credit: The First Ten Years253 Our data provide a unique opportunity to shed light on five key ar- eas: total development costs (TDC), sources of financing for TDC, operating income and expenses, returns to equity and debt inves- tors, and total subsidies provided. TDC per unit varies widely across projects. Location, in terms of region of the country and whether within or outside metropolitan areas, has a significant im- pact on TDC per unit. Controlling for location and a variety of proj- ect characteristics, we find that LIHTC projects developed by non- profits are 20.3 percent more expensive than those developed by for-profit developers. LIHTC projects are often tightly run, with rev- enues just covering costs for many projects; 22.5 percent of the proj- ects had negative cash flows in 1995.
We find convincing evidence that, over time, considerably more of each tax credit dollar provided by the federal government ends up being used for housing rather than for fees and administrative ex- penses. Controlling for a variety of project characteristics, returns to equity investors are higher for rehabilitation projects and proj- ects developed by nonprofits, perhaps reflecting the greater risks associated with rehabilitation versus new construction and with de- velopers who generally have less of a track record than for-profit developers. In addition, returns to equity investors have dropped significantly over the history of the program, reflecting, at least in part, the decline in the perception of risks associated with afford- able rental housing.
The involvement of state and local governments as well as nonprofit developers has a significant impact on the structure of financing and the extent of the subsidies provided. In some local markets, pri- vate banks provide the majority of first mortgages on LIHTC trans- actions, while in others, state and local governments provide virtu- ally all first mortgages. We find that these projects serve low- income households but remain out of reach for very poor renters.
Serving lower-income households requires rental subsidies in addi- tion to the LIHTC capital subsidy. In this article, we estimate the total subsidy provided for LIHTC projects and find that these proj- ects are very heavily subsidized, with substantial subsidies coming from sources other than the LIHTC. On average, subsidies from government and private sources account for 68 percent of TDC; 66 percent of the subsidies come from the flow of tax credits. Our subsidy estimates are underestimates because they do not include tenant-based rental subsidies, such as Section 8, or some project- based subsidies. Even so, these estimates suggest that the total subsidies are quite substantial. This evidence suggests that while the tax-credit program has become more efficient over time, with more of every federal tax credit dollar going to build housing, the overall program design can be expensive in terms of the total amount of subsidy required.
254Jean L. Cummings and Denise DiPasquale In the next section, we provide a brief overview of data used in this article. We then consider six questions that are essential to analyz- ing the extent to which the LIHTC program meets desired housing policy goals:
1. How much do the ªbricks and mortarº cost?
2. What local housing opportunities does the program provide?
3. Are the projects built under the program financially viable over the long term?
4. What income groups can the program serve?
5. Who is providing financing for LIHTC projects?
6. What is the total cost to society of the LIHTC program?
Overview of the City Research data Our sample contains 2,554 LIHTC projects, representing 150,570 units, acquired by four syndicators (Boston Capital Partners, Inc.; Boston Financial; Enterprise Social Investment Corporation [ESIC]; and the National Equity Fund, Inc. [NEF]) from 1987 to 1996, or about 25 to 27 percent of the units generated during this period, al- though there is no accurate account of how many tax-credit projects and units have been built. 2Each of the four syndicators has a na- tional portfolio and has been active in the tax credit market throughout the LIHTC program's history.
Our study complements two other recent LIHTC studies. In 1996, the U.S. Department of Housing and Urban Development (HUD) funded an Abt Associates inventory of all LIHTC projects built be- tween 1992 and 1994; the firm also collected data on a substantial number of projects built prior to 1992. The data collected include basic project characteristics and project location by census tract but 2The calculation is based on estimates by Abt Associates (1996) and Wallace (1998). Abt Associates (1996) estimates that 500,000 units were completed from 1987 to 1994, which suggests about 600,000 finished through 1996. Wallace (1998) estimates that 553,000 units were built through 1996. The National Council of State Housing Agencies (NCSHA), which represents the agencies that administer the allocation of tax credits, estimates that tax credits were allocated for more than 800,000 units through 1995 (NCSHA 1996, 61). Ernst & Young (1997, I) reports that the credit ªhas created almost 900,000 apartments.º The Abt and Wallace esti- mates suggest that NCSHA's estimates are overstated by about 45 percent.
The Low-Income Housing Tax Credit: The First Ten Years255 provide no financial data. 3The U.S. General Accounting Office (GAO) released a study of the LIHTC in 1997 that focused on state oversight and IRS compliance. 4The GAO collected data similar to ours for a sample of 423 projects built between 1992 and 1994; its data include information on tenants, which we do not have. It also collected limited information on all projects built between 1992 and 1994. The data presented in this article are unique in that we have detailed data on a large sample covering the entire history of the program. These data include the only information collected to date on income and expenses and provide many details on funding sources.
Table 1 summarizes some basic characteristics of the projects and units in our data sample and compares our sample to the inventory constructed by Abt Associates (1996). Cummings and DiPasquale (1998a) provide more detailed comparisons between our sample and the Abt and GAO inventories. These comparisons suggest that our sample is quite representative of the overall program. For 1992 to 1994, our sample represents 24 percent of the projects and 34 per- cent of the units in the Abt inventory. 5The Northeast seems to be overrepresented and the Midwest seems to be underrepresented in our sample when compared with the Abt inventory. In both our sample and the Abt inventory, the South has the largest share of projects and units. One reason for the predominance of the South is the significance in the LIHTC program of the Section 515 Rural Rental Housing program, which originally was administered by the Farmers Home Administration (FmHA) and is now administered by the Rural Housing Service (RHS). The program usually provided 1-percent mortgages on LIHTC projects for 50-year terms. Section 515 financing was used extensively in the early years of the pro- gram but has since dropped off sharply. Since 1995, the Section 515 new construction program has been dramatically reduced.
Just under one-third of the projects in our sample were developed by nonprofits (that is, either by nonprofit developers or by for-profit 3The HUD/Abt data are available on the Internet at www.huduser.org.
4Other studies of the LIHTC include Ernst & Young (1997), prepared for the NCSHA and designed as a response to the GAO study; ICF (1991); and Abt Associ- ates (1993), a HUD-sponsored study of 12 tax-credit projects.
5Because there is no accurate accounting of projects and units built since the program began, it is difficult to know with certainty how representative our sample is. For the 1992±94 period, Abt Associates (1996) found 3,987 projects with 168,046 units completed; U.S. GAO (1997) counted 4,121 projects comprised of 172,151 units. Because of data limitations in our sample, we identify the date the syndicators acquired the projects, not the date the project was placed in service, which is the date used by the GAO and by Abt Associates.
256Jean L. Cummings and Denise DiPasquale Table 1.Project Characteristics: Comparing the City Research Sample with the Abt Inventory City Research (1987±96) Abt (1992±94) Projects Units Projects Units Number of observations 2,554 150,570 3,987 168,046 Region Northeast 22.3% 19.8% 13.7% 12.9% Midwest 22.8% 20.4% 32.5% 27.0% South 39.3% 43.4% 39.1% 41.6% West 15.6% 16.4% 14.7% 18.7% Section 515±financed 38.3% 21.2% 34.5% 25.7% Nonprofit developer 31.2% 27.4% 20.3% 23.2% New construction 68.0% 64.5% 65.9% 60.7% Project size (units) 5 to 36 47.9% 20.0% NA NA 37 to 50 18.8% 14.0% NA NA 51 to 99 17.1% 20.3% 12.6% NA 100`16.2% 45.8% 9.8% NA Number of buildings 1 25.7% 21.5% NA NA 2±9 54.0% 40.6% NA NA 10±49 19.0% 34.7% NA NA 50`1.3% 3.2% NA NA Number of bedrooms Efficiency NA 7.4% NA 5.5% 1 NA 32.9% NA 39.8% 2 NA 40.0% NA 38.5% 3`NA 19.8% NA 16.1% Location Central city 42.9% 48.2% 49.1% 54.4% Suburban 24.4% 31.6% 21.0% 26.1% Nonmetropolitan 32.7% 20.2% 29.9% 19.5% Source:Abt figures from Abt (1996).
Note:NA4Not available or not applicable.
developers working with nonprofit partners). In each state, 10 per- cent of annual tax credit allocations must be set aside for nonprofit developers (states may choose to have higher set-aside minimums, and many do). In our data, before 1991, 20 percent of all projects were developed by nonprofits; since 1991, that figure has risen to 36 percent. About two-thirds of the projects and units are new con- struction. A larger fraction of projects in our sample were developed by nonprofits and are new construction than in the Abt inventory.
The Low-Income Housing Tax Credit: The First Ten Years257 The projects in our data range in size from 5 to 589 units; the mean for our sample is 59 units per project. 6Most of the units have one or two bedrooms, although 7 percent are efficiencies (versus 3 percent in the nation's rental stock). Efficiency units typically serve special- needs populations and the elderly and are often an important part of local housing policies. Clearly, however, many of the units in our sample are serving families, with one-fifth of the units having three or more bedrooms.
Using addresses for our projects, we were able to identify the cen- sus tracts for many of them. 7About two-thirds of the projects are in metropolitan areas. The percentage of projects built in central cities has been increasing, from 32 percent in 1987 to 56 percent in 1996, in part due to the decline of the largely rural Section 515 program. 8 As shown in table 1, our sample has fewer central city projects and more suburban and nonmetropolitan projects than the Abt inven- tory. Suburban projects can provide unique opportunities for lower- income households to live in suburban communities. Alternatively, central-city projects can provide important new investment as part of a redevelopment strategy for declining inner-city neighborhoods. 9 Costs of bricks and mortar As with any housing production program, policy makers are con- cerned that development costs for LIHTC projects be reasonable.
For the LIHTC program, states are charged with monitoring devel- opment costs. Our data reinforce the GAO's findings of wide varia- tions in per-unit development costs.
In our data, we have no explicit measure of total development costs (TDC), but we can calculate TDC as the sum of all permanent 6Section 515 projects are considerably smaller; on average, Section 515 projects have 33 units and non±Section 515 projects have 75 units.
7We geocoded 83 percent of the projects in our sample, representing 78 percent of the units.
8ªCentral cityº is defined as the main city or cities of a metropolitan area (e.g., the cities of Minneapolis and St. Paul, MN). ªSuburbanº is within a metropolitan area but not in a central city. ªNonmetropolitanº is located outside of a metropolitan area and is largely but not exclusively rural.
9Keyes et al. (1996) posit that the vast majority of central-city low-income housing is produced by nonprofits. However, our data do not support this claim. While non- profits in our data definitely concentrate in central cities (with 86 percent of both their projects and units in central cities), for-profit developers are also very active in central cities. For-profits develop one-third of the projects and nearly one-half of the units in central cities in our sample; this is fairly consistent over time.
258Jean L. Cummings and Denise DiPasquale financing, equity, and grants. Specifically, we calculate TDC as the sum of the following:
1. Thenet equitypaid directly to the project by the investor in ex- change for tax credits 2. Thefirst mortgage, including all mortgages that have the first lien on the property, whether market-rate conventional mort- gages or subsidized mortgages from the public sector 3. Thegap financing, which we define as any remaining financing, comprising all other mortgages, soft financing, and grants In-kind donations such as land are not included in these data. We use net equity rather than gross equity to calculate TDC because we are interested in the total costs of the development, excluding the costs of raising the equity. Later in this article we consider the difference between gross and net equity as a measure of the costs of raising tax-credit equity; these costs include syndication fees and le- gal and accounting expenses associated with pooling projects into an equity fund. While we believe that our calculation of TDC is quite accurate, a disadvantage of our approach is that it does not provide a breakdown of the uses of funds (e.g., acquisition, hard costs, soft costs, amenities). Unfortunately, our data do not provide sufficient information on the uses of funds. 10 The average TDC for the 2,365 projects for which we have financing data is $65,307 per unit. 11 When we eliminate the 968 Section 515 projects, average TDC overall for the 1,397 projects for which we have financing data rises to $70,226 per unit. 12 On average, for non±Section 515 units, 46 percent of TDC is covered by LIHTC eq- uity, 38 percent by first mortgages, and 16 percent by gap financing.
Figure 1 shows the TDC per unit over time for non±Section 515 10Our TDC figures also may include costs and units used for commercial space.
Commercial space is not always accurately identified in our data.
11The TDC per-unit calculation is unit based rather than project based. That is, we calculate the average cost per unit across the 138,591 units for which we have ade- quate financial data, rather than the average cost per unit across the 2,365 proj- ects containing these units. The unit-based cost measures are 5 percent lower on average than the project-based cost measures, a result of the variations in project size because per-unit costs tend to decline as project size increases.
12We often eliminate Section 515 projects from TDC analysis because these proj- ects have cost structures that are significantly different from other projects, and because Section 515 financing has been cut considerably and is no longer a signifi- cant source of financing for the LIHTC program. For more information on Section 515 projects, see Cummings and DiPasquale (1998a).
The Low-Income Housing Tax Credit: The First Ten Years259 units by sources of financing. While there is no clear pattern in TDC over time, the share of TDC paid by the first mortgage has de- clined from 57 percent in 1987 to 39 percent in 1995. By contrast, the percentage of TDC covered by tax-credit equity has increased every year since the program's inception, nearly doubling from 25 percent in 1987 to 43 percent in 1995. (For most Section 515 units, first mortgages and tax-credit equity alone were sufficient to cover TDC: On average, for Section 515, the first mortgage covers 79 percent of TDC, LIHTC equity covers 19 percent, and gap financing covers only 3 percent.) There is wide variation in per-unit TDC. As shown in figure 2, TDC per unit is between $40,000 and $60,000 for 40 percent of the units, but per-unit costs range from close to $10,000 to more than $250,000 (for 243 units in six projects), and exceed $100,000 for 11 percent of the units. Rehabilitated units can be relatively inex- pensive or quite expensive. Three-quarters of the non±Section 515 units costing less than $40,000 and 38 percent of those costing more than $120,000 are rehabs. 13 Projects developed by nonprofit developers tend to have higher costs per unit on average. For central-city projects, average TDC per unit is $90,268 for units developed by nonprofits compared with $63,778 13Our data do not permit us to accurately distinguish between moderate and sub- stantial rehabilitation. Figure 1.Section 515 TDC per Unit by Year Notes:Data on components for 1996 are incomplete. N41,397 projects; 107,068 units. All dollar figures are in 1996 dollars.
260Jean L. Cummings and Denise DiPasquale Figure 2.Distribution of Units by TDC per Unit for units developed by for-profit developers. For projects with 36 or fewer units, average TDC per unit for for-profit developers is about 50 percent of the average TDC for nonprofit developers ($53,854 versus $107,839). For projects with more than 100 units, TDC per unit for for-profit developers is 88 percent of the average TDC per unit for nonprofit developers. However, in our data, only 9.9 percent of projects developed by nonprofits have more than 100 units; 46.1 percent of nonprofit projects have 36 or fewer units.
The IRS regulations require that states award the minimum tax credits necessary to make a project feasible and mandate that states consider the ªreasonablenessº of development costs. Projects with high per-unit costs raise questions about how well states are monitoring costs. Asking states to ensure that development costs are ªreasonableº is a vague charge. U.S. GAO (1997, 5±10) indicates that there are few meaningful standards for costs and that states vary widely in their efforts to control costs. The GAO found that some allocation decisions were made without complete information and/or certification of key data, and that several states did not fulfill compliance monitoring requirements in 1995 (U.S. GAO 1997, 5).
What accounts for the wide variation in costs? We expect project lo- cation, project size, type of construction, type of developer, unit size, and project amenities to be important determinants of TDC. Our data permit exploration of most of these factors. However, a major weakness in our data is its lack of information on project amenities such as community centers, parking structures, or swimming pools, The Low-Income Housing Tax Credit: The First Ten Years261 all of which can add considerably to TDC. 14 Even with this defi- ciency, the data shed considerable light on the determinants of TDC.
We expect that TDC will vary across regions of the country because of variations in construction costs due to differences in the costs of land, labor, and materials. Costs also may vary across regions be- cause of differences in availability and terms for private financing and public subsidies. We expect that TDC will vary within a region as well. Land and labor costs may be considerably lower for rural projects than for suburban or central-city projects within the same region. Central-city projects may be more costly than suburban ones because of higher land costs or because central-city sites may be lo- cated in more densely developed areas, increasing development costs. Within a city, there may be variations in TDC from neighbor- hood to neighborhood. For example, in poorer neighborhoods with high crime rates development sites may require more security, which adds to the cost of development. Rehabilitation of existing structures in poor neighborhoods also may be more expensive be- cause the buildings are often older and may be more likely to have fallen into disrepair as a result of underinvestment. Finally, be- cause projects in our data were developed over a 10-year period, and financial costs (e.g., interest rates) vary over time, TDC may be expected to vary over time. 15 Regression 1 in table 2 shows that, together, project year, region of the country, location in or outside of a metropolitan area, and neigh- borhood characteristics explain 33.7 percent of the differences in TDC in our sample. All the variables are statistically significant.
There are significant regional differences in project costs. The Northeast is the most expensive, while the South is by far the least expensive. The negative coefficients for suburban and nonmetropoli- tan locations indicate that central-city projects (the omitted cate- gory) are the most expensive. Evaluating regression 1 at the means indicates that moving a project from the central city to the suburbs decreases cost by 7.9 percent. Moving from the central city to a ru- ral area decreases cost by 22.9 percent.
In regression 1, we control for two neighborhood characteristics.
``Qualified census tract'' (QCT) and ``difficult development area'' (DDA) are two designations that can qualify projects for a 30 per- 14The amount of credits that a project may receive is based on the type of develop- ment, the percentage of low-income units involved, and the ªeligible basis,º or the total development costs meeting approval. Many amenities, particularly in mixed- income developments, can be included in the eligible basis.
15We already have accounted for inflation by converting all amounts into real 1996 dollars.
262Jean L. Cummings and Denise DiPasquale Table 2.Regression Results of Location, Section 515, and Neighborhood Characteristics on Real TDC per Unit Dependent Variable:
Real TDC per Unit Regression 1 Regression 2 Regression 3 Regression 4 Location Suburban15,683.23*13,323.6811,810.7612,217.26 (2,208.46) (2,298.66) (2,191.26) (2,332.95) Nonmetropolitan116,413.04*17,297.59*17,932.24*15,213.66* (2,026.30) (2,434.26) (2,204.91) (2,310.33) Region Midwest127,132.27*127,756.87*129,797.37*126,919.94* (2,855.87) (2,883.14) (2,948.54) (3,252.74) South131,659.19*131,290.71*129,587.97*128,669.57* (2,706.16) (2,714.29) (2,966.45) (3,299.56) West120,480.09*120,804.10*120,261.61*116,196.25* (3,587.47) (3,572.44) (3,513.56) (3,494.88) Qualified census tract 5,709.55* 5,406.13* 4,595.67* 4,683.24* (2,297.77) (2,313.17) (2,279.51) (2,447.89) Difficult development area 15,173.45* 14,797.45* 11,537.38* 8,809.04* (3,359.42) (3,335.38) (3,096.53) (2,840.73) Acquisition year a 677.86* 274.231747.69*11,056.91* (312.52) (331.57) (294.56) (303.71) Section 515112,869.74*128,534.48*123,823.04* (1,593.55) (2,281.32) (2,449.89) New construction 15,662.88* 12,209.24* (2,160.90) (2,083.53) Project size (units) 37±5015,151.45*14,015.91* (1,496.35) (1,493.23) 51±100114,029.07*112,794.73* (2,215.32) (2,040.92) 101`122,340.87*120,532.43* (2,586.53) (2,707.62) Nonprofit developer 12,881.68* 15,044.11* (2,774.30) (3,040.92) Bedrooms 6,190.06* (1,020.40) Constant 83,873.60* 87,591.50* 99,629.72* 88,023.54* (4,033.48) (4,235.41) (4,429.10) (5,203.88) R 2 0.3367 0.3507 0.468 0.4966 Observations 108,267 108,267 100,732 81,689 Note:OLS estimates using Huber corrections for group effects. Standard errors are in parentheses.
aAcquisition year of the project has been recoded so that 198541.
*p,0.05.
The Low-Income Housing Tax Credit: The First Ten Years263 cent increase in the ªeligible basisº used to calculate the tax-credit amount generated for the project. 16 A QCT is any census tract in which at least half of the households have incomes of less than 60 percent of the area median gross income. About 30 percent of the projects in our sample are located in QCTs. HUD defines a DDA as any area having high construction, land, and utility costs relative to the area's median gross income. The signs on QCTs and DDAs in regression 1 are positive, as expected, and statistically significant, indicating that TDC is higher in QCTs and DDAs.
To see if the location and region results are driven by the presence of Section 515 projects, we control for them in regression 2. The re- sults are similar to those in regression 1, although coefficients on suburban location and project year are not statistically significant.
The South is still by far the least expensive area. This may be due in part to the fact that the South tends to have larger projects with more units, and hence per-unit costs are lower, but regional differ- ences in building and labor costs undoubtedly play a role.
Although location clearly has a significant impact on costs, other project characteristics may explain some of the variation in TDC. In regression 3, we add controls for construction type, project size, and developer type. The coefficient on new construction is positive and statistically significant, indicating that, on average, new construc- tion projects are more expensive per unit than rehabilitation proj- ects. As expected, TDC per unit declines as project size increases.
Fixed costs associated with building a new structure such as dig- ging a foundation are spread across more units, bringing down the per-unit TDC.
We expect the track record of the developer to influence TDC. De- velopers with extensive experience and those who build a large number of projects may have long-term relationships with suppliers that provide volume discounts. They also may have invested signifi- cant time and money to understand the local permitting and regula- tory process, resulting in fewer project delays, which can decrease project costs. Many larger developers also have realized efficiency gains by successfully expanding the number of services offered to include those of developer, builder, and property manager.
The only developer characteristic available in our data is whether the developer is a for-profit or nonprofit. Nonprofit developers are often small with little capital, which may mean that they produce fewer projects that tend to be small when compared with the activi- 16These designations were not part of the original TRA but were added as amend- ments in 1989 in an attempt to encourage affordable housing development in areas Congress felt were underserved.
264Jean L. Cummings and Denise DiPasquale ties of for-profit developers. 17 Thin capitalization also may mean that organizations have higher predevelopment costs. Walker (1993) finds that smaller nonprofits need to spend more time and money on fundraising and other predevelopment activities such as market research and marketing. In addition, lacking liquid financial re- sources, they are unable to quickly seize opportunities as properties become available (Walker 1993, 389±90). Because nonprofit develop- ers tend to build fewer units and have less of a track record, we ex- pect that their projects will be more expensive. The results in re- gression 3 confirm this view; the coefficient on nonprofit is positive and statistically significant.
Evaluating regression 3 at the means indicates that, controlling for acquisition year, location, region, Section 515 financing, construc- tion type, project size, QCT, and DDA, units produced by nonprofit developers are 20.3 percent costlier than those produced by for- profit developers. Nonprofit developers whom we interviewed sug- gested that nonprofit TDC was higher because nonprofits build larger units. 18 Our models thus far have not controlled for unit size, which could clearly have an impact on TDC. Our data also do not provide information on square feet per unit. However, we do know the number of bedrooms for many of our projects; in fact, we have number of bedrooms for 75.4 percent of the units in regressions 1 and 2 and for 81.1 percent of the units in regression 3. In regression 4 in table 2, we add number of bedrooms. As expected, increasing the number of bedrooms increases TDC and the result is statisti- cally significant. This model explains 50 percent of the differences in our sample TDCs. Evaluating regression 4 at the means indi- cates that adding a bedroom to the average LIHTC unit increases TDC by almost 10 percent. Our results do not support the claim that nonprofit TDC is higher because nonprofit developers build larger units as measured by number of bedrooms. Including a con- trol for number of bedrooms actually increases TDC for nonprofits to 25.0 percent more than TDC for for-profit developers, holding the other variables in the regression constant.
In our data, the estimated increase in TDC per unit for nonprofit developers is large and statistically significant across a variety of 17Walker (1993) notes that the majority of nonprofit developers are small (e.g., community development corporations typically produce fewer than 10 units per year), with a few large organizations producing a large portion of nonprofit units.
He argues that obstacles facing many nonprofit developers trying to increase pro- duction include undercapitalization, limited technical capacity (in part due to thin operating margins), and time-consuming complex financing (Walker 1993, 406).
18In our data, the portion of large units as measured by number of bedrooms is similar between nonprofit and for-profit developers; 21 percent of nonprofit units and 18 percent of for-profit units have three or more bedrooms.
The Low-Income Housing Tax Credit: The First Ten Years265 specifications. In contrast, a recent GAO report (U.S. GAO 1999) analyzing data on LIHTC projects concludes that there is no statis- tically significant difference between TDC per unit for nonprofit and for-profit developers. The results in U.S. GAO (1999) are based on detailed information on 423 LIHTC projects. These projects are a random sample, based on the size of the project, drawn from an in- ventory conducted by the GAO of all LIHTC projects placed in ser- vice between 1992 and 1994. Conceptually, the specification of the model estimated in U.S. GAO (1999) is quite similar to that esti- mated in regression 4 in table 2. 19 While there are some differences between our data and that used by the GAO that could account for some of the variation in our results, it seems unlikely that these dif- ferences would account for all of the variation, given how robust our nonprofit result is to changes in model specification. 20 Because the GAO sample includes all projects with more than 300 units and thus oversamples large projects, we examined the impact of nonprofit developers on per-unit TDC for large projects. In re- sults not reported here, we estimate regression 4 in table 2 sepa- rately for all projects with 36 or fewer units, 37 to 50 units, 51 to 100 units, and more than 100 units. The size of the coefficient on nonprofit developer decreases as the size of projects in the sample increases and becomes small and statistically insignificant for proj- ects with 100 or more units. These results suggest that the GAO es- 19The model in U.S. GAO (1999, 28) explains TDC per unit using metropolitan lo- cation (urban, suburban, or rural), region of the country, new construction, non- profit developer, building type (e.g., high rise, garden), elderly or family use, eligi- bility for additional credits (located in a QCT or DDA), and located in an economically distressed area (census tract with a poverty rate greater than 20 per- cent, unemployment rate greater than 9.45 percent, and 70 percent or more of households with incomes less than 80 percent of median).
In results not reported here, we have estimated specifications to match more closely the specification in U.S. GAO (1999), given the constraints in our data. For example, we do not have square feet but use number of bedrooms as a proxy; we use our data on QCT and DDA to mimic the GAO use of a dummy variable for eli- gibility for additional tax credits. In other analyses, we also add the census tract poverty rate, redefine our four census region variables to match the GAO's recon- figuration of the nine census regions, eliminate size-of-project variables, and drop the dummy variable for Section 515. In all cases, the nonprofit coefficient remained large and statistically significant.
20Unfortunately, we do not have information on who occupies a unit so our model does not include GAO's family or elderly variable, although GAO did not find this variable statistically significant. Nor do we have information on building type, al- though we found that number of units per building is statistically insignificant.
GAO's data on metropolitan location (central city, suburb, and rural) was reported by the state allocation agency. We found in our data that there were substantial differences between self-reported location and those obtained by using census defi- nitions. Their use of self-reported locations may explain why this location variable is not statistically significant in the GAO results.
266Jean L. Cummings and Denise DiPasquale timates could be driven by the oversampling of large projects. The evidence that per-unit TDCs for nonprofit and for-profit developers are similar for larger projects is consistent with the notion that there are efficiency gains associated with larger, more experienced developers. We expect that larger nonprofit projects are being pro- duced by larger, more experienced nonprofit developers and that these developers thus are more like large for-profit developers.
Without access to the data used by the GAO, it is difficult to do more to explain the divergence between our results and theirs. 21 We cannot control for some factors, given our data, that may ac- count for some of the difference between nonprofit and for-profit TDC per unit. Some nonprofit developers have argued that their re- habilitation of existing housing is more extensive than that of for- profit developers; in our data we cannot measure the extent of the rehabilitation. 22 In some cases, projects receiving government subsi- dies are required to pay prevailing wage rates under the Davis- Bacon Act, which would increase TDC. To the extent that nonprofit projects are more likely to accept subsidies that require prevailing wage rates, this could account for some of the cost differential.
In addition, nonprofit developers may be more likely than for-profit ones to provide support services that may be included in TDC. We do not have good data on whether support services are provided and the extent to which they are included in development costs. (Gener- ally, these are considered operating expenses and therefore not in- cluded in TDC, although space for the provision of these services would be included in TDC.) However, U.S. GAO (1997) reports that for projects completed from 1992 to 1994, only 1 percent served special-needs populations. That rate may well have increased, but it is unlikely that these services could account for all of this cost dif- ference between nonprofit and for-profit developers. Some special- ists have suggested that nonprofits may face stiffer reserve require- ments, adding to their projects' TDC. Unfortunately, we do not have sufficient data on capital reserves to test this view.
Even with these caveats, the evidence presented here seems to indi- cate that TDC per unit is higher for nonprofit developers. In review- ing LIHTC applications, states often give priority to nonprofit de- velopers because they usually bring important community support and commitment to a project. In addition, nonprofits can be impor- 21We have requested access to the data that produced the GAO result but they had not yet been released as this article went to press.
22In our data, rehabilitation of existing structures represents 25 percent of units produced by for-profit developers and 65 percent of the units produced by nonprofit developers.
The Low-Income Housing Tax Credit: The First Ten Years267 tant players in community development strategies and this broader mission often leads them to tackle more difficult projects. These are clearly valuable assets. The evidence presented in this section sug- gests that this value must be compared with the increased develop- ment costs that, on average, come with nonprofit developers.
What is the role of LIHTC housing in local housing markets?
Because housing markets are local in nature, only by examining lo- cal markets can we see the real impact of projects built under the LIHTC program. Our data permit a unique opportunity to examine the role of LIHTC projects in local housing markets. For 19 metro- politan areas, we can examine the diverse roles played by LIHTC housing, how LIHTC projects compare with the overall rental mar- ket in those areas, and how they compete in the marketplace. 23 We address three general issues.
1. Where within local markets is the program providing housing opportunities? We examine neighborhood characteristics within the metropolitan area, including levels of residential construc- tion and neighborhood racial and income characteristics.
2. What is being produced by the program and how viable are the projects? Specifically, we compare unit size, TDCs, vacancies, rents, and cash flow.
3. Is the program serving the poor? While we unfortunately do not have data on tenant incomes, we can explore affordability is- sues by comparing project rents with the income distribution of renters in the local market. Where is the program providing housing opportunities?
Given the variation in local market conditions and the flexibility provided to the states by the LIHTC program, it is not surprising that there is significant variation in LIHTC projects across metro- politan areas. 24 The program's flexibility means that it could be used to provide affordable housing for low- and moderate-income 23In order to ensure that our analysis is not biased by one or two anomalies, we consider metropolitan areas only where the sample has at least 15 LIHTC projects.
Within that list of areas, analyses of individual variables (e.g., number of bed- rooms) are reported only where there are at least 10 LIHTC projects in the metro- politan area with corresponding data.
24Cummings and DiPasquale (1998a) provide detailed information on metropolitan area comparisons.
268Jean L. Cummings and Denise DiPasquale households in higher-income neighborhoods or better quality hous- ing in low-income neighborhoods. 25 Neighborhoods.In many neighborhoods in our data, LIHTC projects represent the only new residential construction in recent years. Ten percent of our LIHTC projects were built in census tracts where there was no new residential construction of any kind in the five years preceding the 1990 census. Some 27 percent of the LIHTC projects in central cities are in tracts that had no new construction ofrentalhousing in the previous five years.
In some central-city tracts, LIHTC units are an especially impor- tant addition because those tracts simply lack rental housing. Over- all, in 13 percent of the 1,820 tracts where LIHTC projects in our sample were built, tax-credit units represented more than 20 per- cent of the rental housing stock in 1990. For many cities, even in neighborhoods where there is residential construction activity, LIHTC projects are a significant addition to a neighborhood's rental housing stock. In Cincinnati; Detroit; Fort Worth, TX; and Kansas City, KS/MO, LIHTC units in our sample represented more than 20 percent of the 1990 rental housing stock in nearly one-third of each city's census tracts.
Race and income.On average, the census tracts that house the LIHTC projects in our sample are 59 percent white, 25 percent black, and 13 percent Hispanic. This average masks the fact that a significant portion of the projects in the sample is located in racially homogeneous neighborhoods. More than 30 percent of the projects in our sample are in neighborhoods where at least 90 percent of the population is white. However, nearly 18 percent of the sample's proj- ects are in neighborhoods with a population that is at least 90 per- cent nonwhite. Another 30 percent of our projects are in racially in- tegrated neighborhoods. 26 In central cities, the racial concentration is even more striking: about 39 percent of the central-city neighbor- hoods in our sample are at least 90 percent nonwhite, and 51 per- cent are at least 80 percent nonwhite.
Newman and Schnare (1997) found that privately owned, subsi- dized rental housing, including LIHTC housing, is concentrated in 25States could set as a goal for the LIHTC program the provision of housing for poor households in suburban jurisdictions that may offer better public schools, lower crime rates, and greater access to suburban employment opportunities. This goal is the centerpiece of HUD's Moving to Opportunity program and the Gau- treaux experiment in Chicago.
26Such neighborhoods are defined as those that have a population that is between 10 percent and 50 percent black (Ellen 1996).
The Low-Income Housing Tax Credit: The First Ten Years269 low-income neighborhoods. Our data support their conclusion. Vir- tually all of the LIHTC projects in our sample were built in low- and moderate-income neighborhoods; about 20 percent were built in neighborhoods with median household incomes below 40 percent of the area median income. The map in figure 3 uses Los Angeles to illustrate the concentration of our LIHTC projects in poorer neigh- borhoods. This map shows the 1990 median household income for tracts in the city of Los Angeles as a percent of the 1990 HUD- adjusted Section 8 median family income (HAMFI) limits, which are the income limits used for Section 8 and related programs. We can see that the lowest-income tracts are concentrated in the downtown area with higher-income tracts at the periphery, a pattern repeated in all of our metropolitan areas. This pattern must be interpreted with caution, however, because in many central cities census tracts are not homogeneous; a single tract may have a wide range of household incomes and neighborhood amenities. The circles on the map represent LIHTC projects, which are clearly concentrated in low-income neighborhoods.
As shown in table 3, nearly half of the Los Angeles projects are lo- cated in neighborhoods with median incomes at or below 40 percent of the HAMFI limits for the metropolitan area. (In 1990 dollars, this represents households with an income below $16,600, which roughly corresponds to the 10th percentile of actual household in- come in the city of Los Angeles.) Another 39.4 percent of the proj- ects are in neighborhoods with median incomes between 40 and 60 percent of the area median. Still, several projects are located in higher-income neighborhoods, mostly near the city borders. In Los Angeles, these projects are scattered across neighborhoods and are less concentrated than the projects in the poorer neighborhoods near the center of the city. (One tract in Los Angeles has nine of our LIHTC projects.) Los Angeles has one project in a census tract in the Westchester area, which has a median income of $63,500.
Table 3 also presents summary statistics of five other cities for com- parison; it is striking how similar these metropolitan areas are in the extent to which the LIHTC projects are serving low-income neighborhoods. The portion of projects in tracts with incomes less than 60 percent of the area median income ranges from 60 percent in Boston to 89.6 percent in Chicago. Very few projects in any of these cities are located in neighborhoods with median incomes at or above the HAMFI limits. It should be noted that some of these cit- ies have very few tracts with median incomes at or above the HAMFI limits. In the city of Boston, for example, only four of the city's 165 census tracts have median incomes at or above the HAMFI limit; in the city of Chicago, only 41 of 867 tracts have me- dian incomes at or above the area median income limit.
270Jean L. Cummings and Denise DiPasquale Figure 3.Los Angeles LIHTC Projects by Neighborhood The Low-Income Housing Tax Credit: The First Ten Years271 Table 3.Distribution of Projects by Neighborhood Income Group 1990 Neighborhood Median Household Income/HUD Section 8 Median Limits Total Number ,40% 40 to 60% 60 to 80% 80 to 100% 100 to 120% 120%`of Projects Boston, MA 30.0 30.0 40.0 0.0 0.0 0.0 20 Chicago, IL 64.6 25.0 6.3 4.2 0.0 0.0 48 Brooklyn, NY 24.0 57.3 12.0 5.3 1.3 0.0 75 Manhattan-Bronx, NY 51.8 36.9 5.0 3.5 1.4 1.4 141 Los Angeles, CA 45.1 39.4 9.9 4.2 0.0 1.4 71 Philadelphia, PA 55.6 33.3 7.4 3.7 0.0 0.0 27 Sources:Neighborhood Median Household Income is census tract income from 1990 Census of Population and Housing Summary Tape File 1C. 1990 HUD- adjusted Section 8 Median Family Income limits are from HUD.
272Jean L. Cummings and Denise DiPasquale The evidence provided in table 3 suggests that in major central cit- ies the LIHTC program is used much more often to provide better housing in poor neighborhoods than to provide affordable housing in higher-income neighborhoods. This may be because of explicit policy goals to site affordable rental housing in poorer neighbor- hoods, or because land in higher-income urban areas is scarce or prohibitively expensive, or a combination of both factors. Suburban projects are much more likely to be located in higher-income neigh- borhoods, providing affordable housing opportunities for low-income households. Suburban projects represent 24 percent of all projects and 36 percent of metropolitan projects in the sample. In our data, 37.8 percent of suburban LIHTC projects are located in neighbor- hoods with incomes at or above the area median; only 9.4 percent of central-city LIHTC projects are located in these higher-income neighborhoods.
Are LIHTC projects financially viable?
Unit type and TDC.Not surprisingly, there is wide variation in the types and costs of units provided across metropolitan areas. 27 The flexibility provided to the states by the LIHTC program allows for a wide range of housing policies and goals, and therefore a wide range of types of projects. For example, in Los Angeles, state and local officials have favored single-room occupancy (SRO) projects, which often serve as transitional housing for the homeless; 50 per- cent of the LIHTC units in the Los Angeles metropolitan area are efficiencies (compared with 16 percent efficiencies for the overall rental stock in the Los Angeles metropolitan area). As we saw on the map, these projects are overwhelmingly built in low-income, central-city neighborhoods.
In other markets, units for larger families are the focus of the LIHTC program. In Cincinnati, 58 percent of LIHTC units have three bedrooms or more (compared with 17 percent of the metropol- itan area's rental housing stock). In metropolitan Philadelphia, 42 percent of the LIHTC units have three or more bedrooms, com- pared with 22 percent in the rental stock. Clearly, these cities are emphasizing housing families.
Figure 4 illustrates how TDC varies widely across metropolitan ar- eas. Average TDC per unit ranges from $36,700 in Fort Worth± Arlington, TX, to $110,000 in Los Angeles. The high-cost areas of California and the Northeast have the highest average TDC per unit, with units in the Boston, New York, Philadelphia, and Los An- geles metropolitan areas all averaging more than $100,000. Figure 4 27See Cummings and DiPasquale (1998a) for more detailed data on unit types across metropolitan areas.
The Low-Income Housing Tax Credit: The First Ten Years273 Figure 5.1995 Vacancy Rates by Metropolitan Area Source:1995 Housing Vacancy Survey, U.S. Bureau of the Census.
Notes:Figures are for rental units only. N4360 projects; 21,539 units.
also illustrates the wide variation in the size of first mortgages and the uses of gap financing.
Vacancies and rents.In 1995, the overall vacancy rate for the 87,623 units in the 1,624 projects for which we have vacancy data was 5.8 percent, well below the national rental vacancy rate of 7.6 percent. Figure 5 shows that in almost all of the metropolitan areas presented, LIHTC vacancy rates are considerably lower than corresponding rates for the metropolitan market. LIHTC projects clearly are faring well in many markets, such as Philadelphia and Kansas City, KS/MO, where LIHTC vacancy rates are quite low de- spite an overall soft rental market. In Cincinnati, however, the av- Figure 4.Total Development Costs per Unit by Metropolitan Area Notes:We consider metropolitan areas only where the sample has at least 15 LIHTC projects.
N442,514 units. All dollar figures are in 1996 dollars.
274Jean L. Cummings and Denise DiPasquale erage LIHTC vacancy rate is a very high 16.4 percent, while the market vacancy rate is only 6.6 percent. Cincinnati focuses on large units. It seems that developers and/or the state LIHTC allocation agency may have misjudged this market.
All participants in LIHTC projects are concerned with whether suf- ficient demand exists for these units at rent levels that cover costs.
The LIHTC program sets maximum rents for units earning tax credits. The maximum rent is 30 percent of the maximum income that can qualify for a tax credit unit. This maximum income is set at 60 or 50 percent of the HAMFI limits for the area. If property owners choose 60 percent of area median as the maximum, then at least 40 percent of the units must be occupied by households with incomes at or below this level; if owners choose 50 percent of area median as the maximum income, then 20 percent of the units must be occupied by households at or below this income level.
To assess the market for their units, developers consider rent levels and vacancy rates in comparable buildings. The median contract rent of the 120,419 units in the sample with rent data was $436 in 1996 dollars. 28 U.S. GAO (1997) estimated rents for tax credit units from 1992 to 1994 at $453, or about $480 in 1996 dollars. On aver- age, rents for units in our sample are 9 percent lower than the aver- age rent for the nation (1995 American Housing Survey). 29 In table 4 we compare LIHTC rents in our sample for two-bedroom units with fair market rents (FMRs) for two-bedroom units, defined by HUD as the rent representing the 40th percentile of the distribu- tion of rents paid by recent movers. 30 The median two-bedroom LIHTC rent in our sample is below the FMR in all cities except Bal- timore and Boston. In Baltimore, Boston, Chicago, and Kansas City, 28LIHTC rents are the most current unit rents available (most from 1992 to 1996, with more than half from 1995), inflated to 1996 dollars using the applicable met- ropolitan or regional Bureau of Labor Statistics rent index.
29The 1995 national median rent for recent movers in 1996 dollars was $480. Re- cent movers are defined as households that moved in the five years prior to the survey. In using census data, we focus on rents paid by recent movers because we assume that recent leases reflect market transactions. As length of tenancy in- creases, rents may not be adjusted to keep up with market conditions. Because renters tend to move frequently, recent movers tend to capture the majority of the rental market; in the 1995 American Housing Survey, 76 percent of renters moved in the previous five years.
30FMRs were changed from the 45th percentile of the rent distribution to the 40th in 1995. We use 1995 FMRs; all rents are inflated to 1996 dollars using metropoli- tan residential rent Consumer Price Indexes (CPIs) for all urban consumers from the Bureau of Labor Statistics. Cummings and DiPasquale (1998a) provide de- tailed comparisons across metropolitan areas of LIHTC rents and average market rents based on 1990 census data.
The Low-Income Housing Tax Credit: The First Ten Years275 Table 4.LIHTC and Fair Market Median Rents for Two-Bedroom Units LIHTC Fair Market LIHTC Rent Metropolitan Area Rent ($) Rent (FMR)($) as a percent of FMR Atlanta, GA 557 622 90 Baltimore, MD 679 603 113 Boston, MA 850 808 105 Chicago, IL 622 709 88 Cincinnati, OH 243 494 49 Cleveland, OH 303 515 59 Detroit, MI 547 568 96 Fort Worth, TX 435 534 81 Kansas City, KS/MO 441 492 90 Los Angeles, CA 530 862 62 Miami, FL 489 766 64 Minneapolis±St. Paul, MN 518 617 84 New York, NY 434 821 53 Orlando, FL 477 633 75 Philadelphia, PA 429 683 63 Portland, OR 542 576 94 Raleigh, NC 396 550 72 Seattle, WA 540 679 79 Washington, DC 598 853 70 Source:FMRs from Office of Policy Development and Research, HUD.
Notes:City Research rents are for two-bedroom units only. N413,337 units. FMRs are 1995 FMRs calculated at 40th percentile. All figures inflated to 1996 dollars using metropolitan residential rent Consumer Price Indexes for all urban consumers from the Bureau of Labor Statistics.
more than 30 percent of LIHTC units in our sample have rents that exceed the area FMR.
What accounts for the rather high rents we see for some units in our sample? First, the LIHTC rents in our sample include some market-rate units, which we could not identify in our data. The data suggest that 83 percent of the projects in our sample are 100 percent affordable (have only ªqualifiedº units) and that only 276Jean L. Cummings and Denise DiPasquale 4 percent of the units overall are market rate. 31 More important, LIHTC units are often much newer than other multifamily units nearby and may come with more amenities than those provided by existing units. As we have shown, in many areas LIHTC projects are the only new construction in the neighborhood. In addition, these high rents may be due to the presence of other subsidies in the development. The rents for units in our sample reflect total rent paid including all subsidies. The LIHTC rent limits are based on the tenant rent payment only; the total contract rent is made up of the tenant payment and rental subsidies and may exceed the maxi- mum allowable tax-credit rent. For a tenant with a Section 8 certifi- cate or voucher, the LIHTC rent limit applies only to the portion of rent paid by the tenant and does not include the certificate or voucher. There is an important distinction between Section 8 certifi- cates and vouchers. If the tenant has a Section 8 certificate, then the total rent paid including the tenant portion and the certificate cannot exceed the FMR under Section 8 rules. However, if the ten- ant has a Section 8 voucher, the tenant can elect to pay more than the FMR by paying more than 30 percent of his or her income in rent. In addition, if there is a project-based Section 8 contract in ar- eas designated as high cost by HUD, such as Boston, rent maxi- mums can be set as high as 120 percent of FMRs. U.S GAO (1997) estimates that for projects placed in service between 1992 and 1994, 39 percent of LIHTC households received additional rental assis- tance. For 25 percent of these households, the total rent paid on the unit exceeded the maximum LIHTC rent (U.S. GAO 1997, 45).
Cash flow.How do rents compare with operating costs? We use op- erating revenue and expense data for 1995, the most recent data available, to examine how projects have performed. 32 As shown in figure 6, 1995 operating revenues exceed operating expenses for 78 percent of the projects. 33 Almost 83 percent of the projects devel- oped by for-profit firms had positive cash flows, while only 60 per- cent of those developed by nonprofits had positive cash flows. While the vast majority of projects had strong cash flows, a significant mi- nority appears to have had cash flow problems. For 10 percent of 31U.S. GAO (1997) estimated that 88 percent of its projects and 95 percent of its units qualified from 1992 to 1994.
32We include in this analysis only those projects that almost certainly have been fully leased by excluding projects acquired by the syndicators after 1993, ensuring at least two years for construction and leasing. In addition, we inspected vacancy and rent history for all projects whose operating expenses exceeded their revenues, to ensure that long delays in construction were not influencing the result.
33Operating expenses are defined as maintenance, taxes, insurance, and interest actually paid on debt, but do not include reserve or principal mortgage payments.
Operating revenues are defined as effective gross rents including subsidies, inter- est income, and other income.
The Low-Income Housing Tax Credit: The First Ten Years277 Figure 7.Percent of Projects with Postive Cash Flow by Metropolitan Area (Operating Revenue Greater than Operating Expenses Plus Interest Payments) Note:N4370 projects. Figure 6.Distribution of Expense-to-Revenue Ratios Note:N41,671 projects.
the projects, operating expenses equaled or exceeded 115 percent of revenues.
Figure 7 shows the percentage of projects in each metropolitan area with positive cash flows. The portion of projects with positive cash flows across the metropolitan areas varies from 91 percent in Kan- sas City to 27 percent in Cincinnati. The mean expense-to-revenue ratio in Kansas City was 0.86 while the mean expense-to-revenue 278Jean L. Cummings and Denise DiPasquale ratio in Cincinnati was 1.19. Although Kansas City's overall va- cancy rate is quite high, as shown in figure 5, indicating a soft rental market, the LIHTC project rents and vacancy rates are rather low. Clearly, LIHTC developers in Kansas City have effec- tively targeted a segment of the rental market. In contrast, in Cin- cinnati, LIHTC rents and vacancy rates are quite high relative to the overall rental markets and only 27 percent of the LIHTC proj- ects are running in the black, suggesting that this market was mis- judged. In Boston, 56 percent of the LIHTC projects had a negative cash flow. Boston's mean expense-to-revenue ratio was 1.11. This is somewhat surprising, given Boston's LIHTC rents, which are higher than comparable city and metropolitan rents, and the rela- tively low vacancy rates for Boston's LIHTC projects.
LIHTC projects often are tightly run, with operating revenues just covering operating expenses. Increasingly, strong incentives exist to keep net income near zero, as many soft loans are structured to re- quire payment only if cash flow is positive. In Los Angeles, for ex- ample, only 60 percent of the LIHTC projects had positive cash flows. This may not be surprising, given Los Angeles's focus on SROs, which may have higher expenses and higher turnover rates than other types of projects. In addition, cuts in local subsidies may have adversely affected these projects. While participation by state and local governments clearly bring important benefits to these proj- ects, relying on these subsidies can be risky if their availability changes with shifts in budgets or political climate.
Despite incentives to keep net income close to zero, no project can continue indefinitely with expenses exceeding revenues. Syndica- tors and investors indicate that as projects increasingly are struc- tured to provide no positive cash flow, funding reserves becomes very important. Because we are only 11 years into the program, there is no evidence on how these projects will fare when they need substantial capital infusions for renovations or systems replace- ment. How well these projects clear such hurdles will be a major de- terminant of long-term viability.
Which income groups does the program serve?
Because the LIHTC program has been thede factofederal housing production program for more than a decade, the program has been used, at times in very creative ways, to meet a variety of goals.
While the legislation creating the LIHTC is not explicit about which households the program is meant to target, it clearly was not de- signed to produce housing that is affordable to the very poor, at least not without considerable additional subsidies. The income lim- its under the LIHTC program of 50 and 60 percent of area median income result in rents that are beyond the means of many poor The Low-Income Housing Tax Credit: The First Ten Years279 households. Wallace (1991, 223) showed that 31 percent of subsi- dized renter households had incomes below 20 percent of area me- dian income. The difficulty in reaching the poorest households is not unique to the LIHTC program. Various federal production pro- grams that preceded the LIHTC and also focused on providing affordable housing by subsidizing development of privately owned rental housing also had difficulty meeting the housing needs of the poorest households without additional subsidy.
While our data do not include information on tenants, rent data can be used to estimate the income levels of households served by units in our sample. Following Stegman (1991), we examine the extent to which the federal government's affordable housing program meets the needs of the nation's households. Assuming that 30 percent of household income is paid in rent, the income for a household paying the median rent of $436 in our sample would be $17,440 (in 1996 dollars). Using the 1990 Public Use Microdata Sample (PUMS) of the census, we find that median household income for the nation is $37,216 (in 1996 dollars).
The median rent paid on our sample of tax credit units therefore implies a household income that is about 48 percent of the national median household income. This may, in fact, overstate the incomes of tenants in our sample because the median rent figure includes some unknown portion of additional tenant subsidies, and some households in our sample may be paying more than 30 percent of their income in rent. Even so, our estimate of 48 percent of national median household income is certainly within the LIHTC program limits.
Further comparisons of the rents in our LIHTC database to the in- come distribution of households in 1990 reinforces the point that LIHTC units do not reach the very poor without substantial addi- tional subsidies. In 1990, based on PUMS data, 53 percent ofrenter households in the nation were eligible for LIHTC units by having incomes at or below 60 percent of the applicable area median in- come, as determined by the program (using 1990 incomes inflated to 1996 dollars). Of those qualifying renters, 33 percent had in- comes that could make it possible for them to pay the median rent of $436 without spending more than 30 percent of their incomes.
The medians do not tell us much about the distribution of either rents or incomes. In our sample, 75 percent of rents are at or below $543, a rent implying an income of $21,707 if the household spends 30 percent of its income on rent. Of qualifying renter households, 17 percent could afford this rent without additional subsidies. At the bottom of our rent distribution, the 1st percentile, the rent is $203 per month. This rent level is quite low, and clearly well below 280Jean L. Cummings and Denise DiPasquale the maximum levels allowed in the LIHTC program. Still, over one- third of the nation's renter households (36 percent) in 1990 could not afford even that rent.
The LIHTC program, like other federal housing programs, is depen- dent on the federal definition of the applicable area median incomes for setting rent and income limits. The maximum incomes in the LIHTC program are based on the Section 8 HAMFI limits, adjusted for household size, not on the actual area median household in- come. 34 Congress has authorized HUD to adjust Section 8 median income limits from actual income figures to reflect, among other things, the relative costs of housing with respect to income. In most areas, this means that HAMFI limits are higher than actual income figures.
The benefit of the adjusted Section 8 figures is that income and rent limits are higher, permitting properties to generate more income.
These higher limits can be essential to the financial viability of a project. A disadvantage is that this adjustment widens the gap be- tween the poorest renter households in the area and the renter households that realistically can be housed by the program. The HAMFI limit for a household of four for the nation is estimated at $41,600 in 1996, 98 percent of actual median income for a four- person household. However, this income reflects all households, both renters and homeowners. Average income for renters is consid- erably lower than that for homeowners. According to the 1990 PUMS, the median income of renter four-person households was $30,874 (in 1996 dollars). The HAMFI limit, therefore, is 135 per- cent of the actual renter median income. Using HAMFI limits rather than actual area incomes permits LIHTC rents to be set higher than what is implied by actual renter income.
In table 5, we present the incomes implied by the median rent in our sample, assuming a household size of four, and assuming that 30 percent of household income is spent on rent. We also present ac- tual area median income for all households (owners and renters) and for renters only from the 1990 census, and HAMFI limits for 19 metropolitan areas, all adjusted for a household size of four per- sons. As shown in the fifth column of the table, HAMFI limits are higher than actual median incomes in many of these metropolitan areas. The divergence between HAMFI limits and renter medians is quite dramatic across all metropolitan areas. In Chicago; Cleveland; Detroit; Miami; Minneapolis/St. Paul, MN; and Washington, DC, the HAMFI limits are more than 200 percent of the actual median 34While Section 8 incomes are applied by metropolitan area or county, for this il- lustration we use HUD's estimate of the national Section 8 median.
The Low-Income Housing Tax Credit: The First Ten Years281 Table 5.Comparison of Median Incomes and Median Income Limits, Adjusted to Households of Four Persons Calculated IncomesArea HAMFI as HAMFI as City Research Incomes from City ResearchMedian Incomes (AMI) HAMFI a Percent of a Percent of as a Percent of Section 8 Rents All Households Renters Limits All AMI Renter AMI Median Incomes Atlanta, GA $23,995 $48,000 $26,400 $52,100 109 197 46 Baltimore, MD 23,763 48,731 30,000 52,400 108 175 45 Boston, MA 34,400 57,770 36,000 56,500 98 157 61 Chicago, IL 24,519 46,772 25,000 54,100 116 216 45 Cincinnati, OH 21,512 42,900 24,468 46,700 109 191 46 Cleveland, OH 12,759 42,348 20,000 44,600 105 223 29 Detroit, MI 24,687 48,085 23,000 50,100 104 218 49 Fort Worth, TX 19,543 42,200 26,000 47,500 113 183 41 Kansas City, KS/MO 18,752 48,500 27,432 43,600 90 159 43 Los Angeles, CA 20,589 43,509 26,400 51,300 118 194 40 Miami, FL 21,418 35,600 21,830 44,600 125 204 48 Minneapolis±St. Paul, MN 22,093 49,836 26,860 54,600 110 203 40 New York, NY 19,362 42,500 29,000 49,000 115 169 40 Orlando, FL 21,432 40,995 27,300 41,900 102 153 51 Philadelphia, PA 19,561 47,430 25,300 49,300 104 195 40 Portland, OR 22,299 42,000 26,460 44,400 106 168 50 Raleigh, NC 17,888 47,500 30,600 50,700 107 166 35 Seattle, WA 22,904 48,990 31,000 52,800 108 170 43 Washington, DC 28,208 42,000 28,000 68,300 163 244 41 Sources:City Research incomes calculated by dividing City Research rents by 30 percent. Area median incomes from Public Use Microdata Sample (PUMS) of the 1990 Census. HUD data provided to authors by HUD.
Notes:HAMFI is HUD-adjusted Section 8 median family income limits. HAMFI was calcuated as two times very low income limit for households of four for 1996. Very low income limit is defined as 50 percent of HAMFI. All dollar figures are in 1996 dollars.
282Jean L. Cummings and Denise DiPasquale income for renter households. The last column shows that the esti- mated incomes for a household occupying an LIHTC unit with the median rent in our sample range from 29 percent of HAMFI limits in Cleveland to 61 percent in Boston.
The illustration of affordability of LIHTC units presented in table 5 clearly shows a tension in setting national housing policy. The HAMFI limits are an important part of the eligibility criteria for privately owned subsidized housing. Basing Section 8 income limits on the income distribution of all households rather than on renter households pushes all programs that depend on these definitions, including the LIHTC program, toward higher-income renters. Per- mitting increases in Section 8 income limits above actual income levels results in rents that make many projects financially viable, particularly in areas with high housing costs, but also results in rents that require higher incomes, greater additional subsidies, or higher rent burdens on households.
Who is financing LIHTC projects?
As discussed in the introduction, the LIHTC program is structured to combine the efficiencies of the private market with the public policy goals of government participants. An important aim of the program is to leverage public financing sources with private dollars.
Even though LIHTC rents are often out of reach of the poorest households, the previous section illustrates that LIHTC projects are often dependent on additional subsidies to be financially viable. To reach lower-income families, the program would need even greater subsidies.
The production of LIHTC housing depends on four key actors: the developer; the lender, who provides market rate and subsidized debt; the ultimate investor, who puts up the equity in exchange for the tax credits; and the tax-credit syndicator, who acts as a broker between the developer and the investors. Each financial partici- pant, in effect, wants its contribution to be the last piece in the transaction. The federal government requires that states allocate the minimum amount of tax credits necessary to make a project via- ble. State and local governments want to use the smallest subsidies possible to make a project viable, so that their subsidy dollars can build more projects. Private lenders want more government subsi- dies in the transaction because, since the first mortgage holder has the first lien, the presence of subsidy decreases the lender's risk of loss if the borrower defaults on the mortgage (DiPasquale and Cum- mings 1992, 91±92).
The Low-Income Housing Tax Credit: The First Ten Years283 As shown below, our data provide some information on syndicators and extensive information on lenders and investors. The only infor- mation our data provide on developers is whether they are for-profit or nonprofit; we use these data throughout this article.
Syndicators Syndicators pool several projects into one tax-credit equity fund and market the credits to investors, who buy a portion of the fund, thus spreading investor risk across the fund's various projects. Syndica- tors also provide underwriting, legal, and accounting services re- quired to syndicate the tax credits; structure investments to meet individual investor needs; monitor projects for the investors; and sometimes fund reserves for fund-level legal and administrative costs.
Our data do not permit us to isolate the fees paid to syndicators.
However, we do know the gross equity raised and the net equity, which is the equity that actually ends up in the project. The differ- ence between net and gross equity represents syndication fees, overhead, fund-level reserve funds, legal and asset management fees, bridge loan interest, and other miscellaneous costs. In our data, the mean ratio of net equity to gross equity is 0.71 and has remained fairly steady over time. While there has not been much change in the ratio of net equity to gross equity, there may have been significant changes in where the difference between the two has gone. For example, our interviews with investors and syndica- tors suggest that, over time, more investors have extended the time over which they pay in their equity, which has increased the amount of bridge loan interest paid. While our data do not permit us to track bridge loan interest, an increase in bridge loan interest, holding the net equity to gross equity ratio constant, would mean a decrease in other fees, which suggests that syndicator fees have de- clined over time.
Lenders LIHTC projects rely heavily on debt financing from private and gov- ernment lenders. Debt financing includes first mortgage loans and gap financing, which includes all other mortgages and soft loans. In addition, we include grants in gap financing.
First mortgage loans. Private-sector lenders are an important source of financing for LIHTC projects. In our sample, private banks provide about 40 percent of non±Section 515 first mort- 284Jean L. Cummings and Denise DiPasquale gages. 35 State governments provide 26 percent, local governments 19 percent, and nonprofit organizations 9 percent. For-profit devel- opers are much more likely to use private lenders for first mort- gages than are nonprofit developers; private lenders provide 51 per- cent of the first mortgages on non±Section 515 projects developed by for-profits but only 30 percent of the first mortgages on non± Section 515 projects developed by nonprofits.
First mortgages from state and local governments can be similar to those from private lenders with respect to term, interest rate, and payment schedules, or they may be subsidized mortgages with below-market interest rates or soft loans that are more like grants.
In some cases, payments on loans from state and local governments may be forgiven if the project is in trouble. However, since grants are excluded from the eligible basis for determining tax credits, it is often important to ensure that soft loans are counted as loans and not grants. 36 Our sample data do not provide enough information in a sufficient number of cases to distinguish between hard and soft first mortgages.
Although private lenders are a primary source for first mortgages, their first mortgages cover, on average, a smaller portion of TDC than government lenders, with the banks' loan-to-TDC ratio 37 aver- aging 40 percent, local governments' 50 percent, and state govern- ments' 43 percent. Loan-to-TDC ratios for nonprofit lenders are lower, averaging 28 percent. In addition, the loan-to-TDC ratios for private-bank first mortgages have declined over time (from 56.1 percent in 1987 to 34 percent by 1996). This is puzzling be- cause our sample offers no evidence that banks are lending on risk- ier (in terms of location) projects over time. During the same time, the average loan-to-TDC ratio for state and local governments has held steady at about 45 percent.
Interest rates on these mortgages vary across lenders as well. Rates on private-lender first mortgages in our data average about 180 ba- sis points (bps) above state government rates and 561 bps above lo- 35Since the RHS provides first mortgages on Section 515 projects, this analysis is limited to non±Section 515 projects.
36Section 42 of the Internal Revenue Code states clearly that federal subsidies must be subtracted from the eligible basis, although there have been recent regula- tions making some exceptions. The LIHTC code is less specific on state and local subsidies, although Guggenheim (1994, 36) specifies that nonfederal grants are de- ducted from the eligible basis while nonfederal loans may be included.
37These ratios are loan-to-TDC, not loan-to-value (LTV), a standard measure that was not available in our data. LTV is based on appraised market value of the building, which can be lower than TDC for affordable rental housing projects.
The Low-Income Housing Tax Credit: The First Ten Years285 Figure 8.First Mortgage Lenders by Metropolitan Area Note:N4471 mortgages.
cal government rates. In 1996, the private-lender rate in our data was 104 bps above the rate on 30-year treasuries. This relationship has changed considerably over time. In the early years, the average rate on mortgages from private banks hovered around the 30-year treasury rate. These low rates may reflect the considerable amount of LIHTC project lending done through the community lending or Community Reinvestment Act (CRA) branches of banks, which sometimes provide concessionary rates. The spread over treasuries has increased in recent years, perhaps due to a decrease in the availability of concessionary financing or the mainstreaming of some LIHTC mortgage lending into commercial lending divisions of banks.
Not surprisingly, the roles of different types of lenders vary widely across local markets, as shown in figure 8. In Cleveland, Atlanta, and Chicago, private lenders play a major role in first mortgage lending for tax-credit projects in our sample, providing 81, 67, and 60 percent of the mortgages, respectively. In New York, private banks provide no first mortgages, and in Boston they provide only 8.3 percent of the mortgages. In New York City, the city provides virtually all LIHTC mortgages, with rates averaging 1.2 percent, a very significant subsidy. In the Boston metropolitan area, two- thirds of the mortgages are provided by the state, at an average rate of 8.7 percent; in many cases, the rates are very close to mar- ket rates.
The portion of TDC covered by the first mortgage likewise varies across local markets, as illustrated earlier in figure 4. Combining the information in figures 4 and 8, however, suggests the differ- ences in mortgage size for different lenders. In Cleveland, 81 per- cent of the LIHTC projects have mortgages from private banks, but those mortgages cover only an average of 13 percent of TDC. In At- 286Jean L. Cummings and Denise DiPasquale Figure 9.First Mortgage Size by Lender Type Notes:Federal Govenment includes Federal Home Loan Bank, Fannie Mae, and other non± Section 515 federal sources. N41,257 mortgages.
lanta, two-thirds of the mortgages come from banks; they cover 54 percent of TDC, on average. First mortgages from the city of New York cover 55 percent of TDC, on average.
It is often unclear whether government lending is filling a void left by private lenders or government and private lenders are competing for the same business. In New York City, for example, local policy makers decided that the city would provide mortgages on LIHTC transactions, eliminating private lenders from the tax-credit mar- ket. In other cities, there is no government institutional or financial capacity to lend on rental housing and hence private lenders domi- nate the market. In the majority of markets we have examined, there appears to be a mix of private and government lending.
Market participants indicate that it is more difficult to obtain con- ventional mortgage debt for smaller projects because lender fees are based on mortgage size and larger loans are more profitable (see Cummings and DiPasquale 1998b). Figure 9, however, illustrates that private banks are leading lenders for mortgages of all sizes.
They are responsible for nearly half of all mortgages of less than $1 million, and 40 percent of the mortgages of $3 million or more.
State and local governments have the largest share of all mortgages over $1 million. Of course, our data provide no information on proj- The Low-Income Housing Tax Credit: The First Ten Years287 Figure 10.Non±Section 515 First Mortgage Lenders by Neighborhood Income Group Sources:Neighborhood income level is the ratio of the 1990 neighborhood median household income, from 1990 Census of Population and Housing Summary Tape File 3A, to the 1990 HUD-adjusted Section 8 median family income (HAMFI) limits for the metropolitan area or nonmetropolitan county.
Notes:Federal government includes Federal Home Loan Bank, Fannie Mae, and other non± Section 515 federal sources. N41,041 mortgages.
ects that were abandoned because conventional mortgages were not available.
There has long been concern that lenders are hesitant to make loans on properties located in low-income neighborhoods. As shown in figure 10, private banks are a significant source of first mort- gages in all neighborhoods, regardless of income. In the poorest neighborhoods (with median income less than 40 percent of HAMFI), banks provide one-quarter of first mortgages, states pro- vide one-quarter, and local governments provide 37 percent. Private banks are clearly the dominant provider of mortgages for LIHTC projects in higher-income neighborhoods. Again, we have no infor- mation on the distribution of projects by neighborhood that were not built because of lack of mortgage financing.
Gap financing. Forty percent of the projects in our sample have gap financing, often critical to a project's economic viability. 38 State and local governments are the source for nearly 70 percent of the loans 38For the 60 percent of projects without gap financing, tax-credit equity and a first mortgage stand alone to provide the financing for the project. Of these projects, 63 percent have Section 515 financing.
288Jean L. Cummings and Denise DiPasquale and grants included in gap financing. Nonprofits provide 11 percent of the loans and grants in gap financing, while private banks fund 5 percent, although, on average, bank loans tend to be about twice the size of loans from nonprofits. Rates on non±first mortgage loans vary considerably, from 0 to 15 percent. Of the 2,554 projects in our data, only 85 projects are identified as having grants. However, 261 projects had 0 percent loans, which in many cases are very much like grants; there is often a presumption that 0 percent loans will not be repaid.
Importance of concessionary financing. We make a distinction be- tween gap financing and concessionary financing. Gap financing is any financing beyond the tax credit equity and the first mortgage.
Concessionary financing is any financing with a below-market in- terest rate. As discussed above, many first mortgages, especially among those provided by state and local governments, have below- market interest rates. Because market rates may vary across local markets, we cannot know with certainty when a mortgage carries a concessionary rate. We use the rate on 30-year constant maturity treasuries (CMT) as a conservative measure of market interest rates. Rates below the CMT rate are assumed concessionary while those above the CMT rate are categorized as market-rate loans.
Using this definition, 64 percent of all the first mortgages in our sample carry a below-market interest rate; 2 percent have a 0 per- cent interest rate. For non±Section 515 first mortgages, 38 percent have below-market interest rates. Of the gap financing, 64 percent carry below-market interest rates and 23 percent have a 0 percent interest rate. Figure 11 revisits TDCs for all of our projects, includ- ing Section 515 projects. This time, we show the portions of TDC financed by market-rate loans, concessionary financing, 0 percent interest rate loans and grants, and LIHTC equity. On average, 26 percent of TDC is covered by market-rate financing; 40 percent by below-market financing, 0 percent interest loans, and grants; and the remaining 34 percent is covered by LIHTC equity. The por- tion of TDC covered by market-rate financing has been growing slowly since the early 1990s. Grants and all concessionary financing combined peaked at 52 percent of TDC in 1991 and have been de- clining steadily since, representing 30 percent of TDC by 1995.
Figure 12 compares the types of financing used for different kinds of projects. The first set of bars compares units with and without Section 515 financing. On average, non±Section 515 units cost 43 percent more than Section 515 units. For Section 515 projects, the low TDC per unit coupled with the deep discount of Section 515 mortgages results in projects funded virtually entirely by net equity (19 percent) and concessionary financing (79 percent).
The Low-Income Housing Tax Credit: The First Ten Years289 Non±Section 515 projects exhibit significant differences in types of financing as well. As shown in the second set of bars, suburban proj- ects use virtually no soft loans or grants and have little concession- ary financing. In contrast, central-city projects use relatively little market-rate financing. In central cities, concessionary financing and grants make up a significant portion of TDCÐabout 43 percent of TDC, compared with 17 percent in the suburbs and 28 percent out- side of metropolitan areas. Figure 12 also illustrates the importance of concessionary loans and 0 percent loans for projects developed by nonprofits and for smaller projects.
Finally, we find considerable variation in types of financing used across metropolitan areas. Figure 13 demonstrates that below- market loans dominate in most cities. In Chicago, concessionary loans and 0 percent loans are very important; and in New York, concessionary loans from the city dominate the financing of LIHTC projects. Figure 8 illustrated that most of the LIHTC projects in Cleveland have first mortgages from private banks, but we noted that the amount of TDC covered by those mortgages is very small.
As a result, market-rate mortgages are a small fraction of TDC in Cleveland, as shown in figure 13. Market-rate financing is clearly very important in some markets, such as Atlanta, Boston, and Washington, DC. These results are consistent with the wide varia- tion that we found in the roles of private and public lenders across local markets. Figure 11.Total Development Costs by Financing Type Notes:All dollar figures are in 1996 dollars. N42,089 projects; 120,125 units.
290Jean L. Cummings and Denise DiPasquale Investors In proposing the LIHTC as part of the Tax Reform Act of 1986, con- gressional members cited the need for ªusing tax incentives to at- tract private capital to low-income housing development and reha- bilitation; [because] . . . absent some incentives, investment in low-income housing is a fundamentally uneconomic activity.º 39 Be- cause of rent limits, LIHTC projects would not be expected to gener- ate market rates of return without the stream of tax credits. Pri- vate investor participation also was expected to add a layer of over- sight. Because the return to investors in tax credits is dependent on the project staying in compliance, investors have an incentive to en- sure that the development continues to meet requirements. 40 39Senator George Mitchell's entire statement can be found in the Congressional Record (1986, pp. 14,918).
40The LIHTC program was structured to include the retroactive recapture of tax credits in the case of noncompliance as an incentive to investor oversight. U.S.
GAO (1997) reports that IRS compliance oversight has been inadequate to date.
IRS admitted that it has not been able to estimate the extent of taxpayer noncom- pliance because it has not completed sufficient tax credit audits. In addition, the GAO found that the IRS tax credit audit program design itself is methodologically flawed (U.S. GAO 1997, 108±11). Improved IRS compliance oversight and enforce- ment are major recommendations of the GAO report.Figure 12.Total Development Costs for Each Financing Type by Project Type Notes:All dollar figures are in 1996 dollars. N42,237 for projects and 130,761 units for Section 515 and Non-Section 515; 1,061 projects and 77,544 units for location; 1,239 projects and 94,533 units for developer type; and 1,269 projects and 99,238 units for project size.
The Low-Income Housing Tax Credit: The First Ten Years291 Figure 13.Total Development Costs for Each Financing Type by Metropolitan Area Notes:All dollar figures are in 1996 dollars. N4594 projects and 39,514 units.
The past decade has seen significant changes in the type of inves- tors participating in the LIHTC market. Before 1992, a substantial portion of project equity was raised through retail funds sold through brokers to individual investors. In part because of changes in passive loss rules for individual investors, and spurred by grow- ing experience with the LIHTC program, corporate involvement be- came more important beginning in 1992 and now dominates the market. In recent years, particularly with the permanence of the LIHTC established in 1993, more investors have entered the mar- ket, significantly increasing the competition for LIHTC transac- tions. While there are no formal statistics on the number of corpora- tions that have participated in this market, interviews with investors and syndicators suggest that 100 to 200 corporations have made LIHTC investments. 41 How efficient is the LIHTC program in raising equity for rental housing? The industry defines the price per tax-credit dollar as the net equity (portion of total equity that ends up in the project) di- vided by the total tax credits (10-year sum). This price indicates the 41Cummings and DiPasquale (1998a) report the results from a written survey of 50 corporate investors and in-depth follow-up interviews with 23 survey respon- dents.
292Jean L. Cummings and Denise DiPasquale Figure 14.Price of Tax Credits by Year Note:N42,272 projects.
portion of each tax-credit dollar spent by the federal government that actually ends up in housing, which is one measure of efficiency.
Based on 2,272 projects, the price per tax credit in our sample aver- aged $0.52. 42 This standard industry calculation of price does not discount either the net equity pay-in or the stream of tax-credit dol- lars. (The average ªpriceº rises to more than $0.70 if we use the present value of the flow of the tax credits. 43) As shown in figure 14, price has risen from $0.47 in 1987 to $0.62 in 1996, with a dramatic increase after 1993. U.S. GAO (1997) found prices to be $0.45 in 1987 and $0.60 in 1996. In our survey of LIHTC investors, many indicated that 1997 prices approached $0.70. These higher prices likely reflect, at least in part, a significant increase in competition among investors for tax credits. The rise in prices does seem to indi- cate a substantial increase in program efficiency over time. Some investors we interviewed were cautious about concluding that this price increase primarily represents efficiency gains, because it also may simply reflect longer pay-in times for equity, and that those differences in pay-in time are reflected in the total net equity price.
While our data do not permit evaluation of trends in these pay-in schedules, the price increase seems too large to reflect only differ- ences in pay-in schedules.
Internal rates of return (IRR).Investors need information on risks and returns to weigh participation in the market. Virtually no infor- mation is publicly available on risks and returns to equity investors 42We have excluded projects with historic tax credits through the federal program designed to encourage rehabilitation of historic structures. Historic tax credits may be combined with LIHTC, but the result is a complicated financing structure that is difficult to compare with other LIHTC transactions.
43We assume that total net equity is paid in the first year and calculate price by dividing net equity by the present value of the 10-year stream of tax credits, using a discount rate of 6.7 percent, the average interest rate on U.S. Treasury securities with a 10-year constant maturity for the period of 1992 to 1994.
The Low-Income Housing Tax Credit: The First Ten Years293 in LIHTC projects or to investors in other rental housing develop- ments. A unique feature of our database is the substantial informa- tion it contains on project performance, including income state- ments and balance sheets. While it is still too early to fully assess the performance of LIHTC investments, because tax-credit benefits flow across 10 years and the project compliance period may extend to 30 to 40 years, these data provide a useful base for evaluating project performance to date.
The IRR is the standard industry measure of the return to inves- tors. Our sample permits calculation of IRRs and examination of IRR trends over the history of the program. It is important to note that while syndicators and investors calculate IRRs at the fund level (IRRs to investors in a fund representing a group of projects), IRRs in this study are calculated at the project level to maintain the project as the unit of analysis throughout. 44 Our calculations are based on a series of assumptions outlined below and do not re- flect the actual IRRs to investors generated by these projects, nor do they exactly duplicate the various ways individual investors cal- culate potential IRRs when considering projects. While our inter- views with syndicators and investors have indicated that our IRR calculations and findings are reasonable, there are no published data on LIHTC IRRs to compare with our results, and actual IRR calculations are closely guarded by syndicators and investors.
We consider two pay-in schedules: 100 percent of investment made initially, and an eight-year pay-in made in eight equal installments.
For the latter schedule, payments are discounted to the initial pe- riod using an 8 percent discount rate (our assumed rate on bridge financing), and the IRR then is calculated based on the discounted equity investment and the discounted flow of benefits to the inves- tor. Interviews with investors and syndicators indicated that longer pay-in periods were becoming very common in the mid-1990s. Using these two pay-in schedules generates good benchmarks for a rea- sonable range of returns, given standard practices in the market- place. A shorter pay-in of equity will significantly decrease IRR, and a longer pay-in will significantly increase IRR. We assume an an- nual stream of benefits for 15 years, with the flow of tax credits to the investor for only the first 10 years (a longer stream of benefits will, of course, increase the IRR). The annual stream of benefits is calculated using the equation [Losses2Tax rate]`Flow of tax credits.(1) 44The distinction between project-level and fund-level IRRs is important. Investors in funds care only about the fund-level IRR. The syndicators can provide a blended IRR by grouping projects with varying IRRs into one poolÐoffsetting a project of- fering high returns with a more complicated project (perhaps providing community development or social service benefits) with a lower IRR.
294Jean L. Cummings and Denise DiPasquale Figure 15.IRRs by Year Note:N41,577 projects.
Lossesequaltotal operating incomeminustotal operating expense minustotal financial expensesminusdepreciation. 45 A 40 percent tax rateis assumed. Operating income and expenses are based on actual 1995 data. Operating expenses are assumed to grow at 3 per- cent per year and operating revenue at 2 percent per year, the stan- dard industry assumptions. We assume no residual value at the end of the investment, which is a standard assumption used in the in- dustry; if there is a residual value, these return calculations under- state actual returns.
As shown in figure 15, assuming 100 percent of the investment is paid initially, the average IRR peaked at 20.5 percent in 1987, and declined steadily through 1994 to an average of 11.8 percent. With an eight-year pay-in, the average IRR fell from 28.7 percent in 1987 to 18.2 percent in 1994. 46 These declines appear to reflect the pro- 45If the investor reports a loss on the K-1 tax form, then the benefits calculation inside the brackets is equal to the loss times the tax rate. If the investor shows a profit on the K-1 form, then we calculated the benefits amount inside the brackets as equal to the profit times one minus the tax rate to account for the after-tax profit.
46As a comparison, we looked at returns to investors in equity REITs (real estate investment trusts). An equity REIT is a real estate company or trust in which shareholders own equity shares in a pool of properties. The National Association of Real Estate Investment Trusts (NAREIT) publishes annual data on total returns to investors in equity REITs. These returns have varied widely in the 1990s, ranging from ±15 percent in 1990 to 35 percent in 1996. In 1994, NAREIT began tracking returns on equity REITs by property type. Total returns on apartment REITs were 2.2 percent in 1994, 12.3 percent in 1995, and 28.9 percent in 1996 (authors' calcu- lations based on data supplied to the authors by NAREIT).
The Low-Income Housing Tax Credit: The First Ten Years295 Figure 16.IRRs Using a 100 Percent Pay-In Note:N41,342 projects.
gram's increasing efficiency over time. As more tax-credit projects have been developed, investors know more about these investments, which is likely to have decreased the perception of risks associated with them. In addition, important changes have taken place over the course of the program. One change has been the shift from pub- lic offerings of tax syndications to individual investors toward pri- vate placements with corporations. Additionally, over time, banks have become major investors in the LIHTC; these investors often are motivated more by CRA requirements than by the returns on the investments. These changes also have helped to reduce the cost of raising the capital. 47 As shown in figure 16, in the program's early years, IRRs on central-city projects were considerably higher than those for all projects in the sample. From 1987 to 1991, the gap narrowed from 12.9 to 2.1 percentage points, on average. From 1992 through 1994, the average IRR on central-city projects is quite close to the average for all projects. This declining gap may reflect a decreased percep- tion of risk associated with central-city projects relative to other LIHTC projects. Similarly, in 1988, IRRs on nonprofit projects aver- 47U.S. GAO (1997) reports that major investment syndicators and allocating agency officials attribute increased program efficiency in part to these changing in- vestor characteristics and syndication structures. In addition, they point to equity funds developed by states and localities that add to the competition; the fact that tax-credit properties are relatively more attractive to corporations than to individu- als because corporations are often exempt from passive investment loss rules that limit individual investors' deductions; and ªgrowth in the economy and in corporate profitability [that] has increased the taxable income that could be sheltered by tax creditsº (U.S. GAO 1997, 90±91).
296Jean L. Cummings and Denise DiPasquale aged 33 percent, or 17 percentage points above those for for-profit projects. By 1992, this gap in IRRs virtually vanished, suggesting a decrease in the perception of risk associated with nonprofit develop- ments or perhaps a change in the mix of projects done by for-profit and nonprofit developers. We further explore this finding in the dis- cussion that follows.
In order to identify the determinants of differences in IRRs across projects, we estimated regressions with IRR as the dependent vari- able and various project characteristics, location, and the year the project was acquired by the syndicator as independent variables. As shown in the first regression in table 6, the acquisition year has a negative and statistically significant impact on IRR, meaning the newer the project the lower the IRR, suggesting that the perception of risk has declined over time. The estimates suggest that IRRs have declined 1.4 percentage points per year, which is consistent with the trend shown in figure 15. New construction projects have lower IRRs, perhaps reflecting the increased risk associated with rehabilitating an existing structure. Project size has no impact on IRRs. Section 515 projects generate higher returns to equity inves- tors and the impact is statistically significant. Controlling for acqui- sition year and other project characteristics, projects developed by nonprofits generate higher returns to equity investors, suggesting that for comparable projects nonprofit developers are perceived as riskier than for-profit ones. There is little difference in IRRs among projects in the Northeast, Midwest, and South; however, projects in the West generate IRRs that are 2.8 percentage points lower than projects in the Northeast. Locations in the suburbs and outside metropolitan areas have IRRs that are lower than those for central- city projects, but the sizes of the coefficients are small and are not statistically significant. Similarly, the coefficients on QCT and DDA are small and statistically insignificant.
The results on location are somewhat surprising, given that we would expect location to be an important determinant of risk. To ex- plore location further, in regression 2 we added the portion of the population of the census tract that is below poverty. Poverty has a positive and statistically significant impact on IRRs; a one standard deviation increase in neighborhood poverty increases IRRs by one percentage point. This result suggests that neighborhood character- istics may be more important in determining risk than region or central city versus non±central city locations.
What are the total costs in subsidies for LIHTC housing?
All levels of governmentÐfederal, state, and localÐactively partici- pate in the LIHTC program in many areas of the country. The fed- The Low-Income Housing Tax Credit: The First Ten Years297 Table 6.IRR Regressions Dependent Variable: IRR Regression 1 Regression 2 Location Suburban10.013510.0093 (0.0083) (0.0084) Nonmetropolitan10.010410.0107 (0.0098) (0.0097) Region Midwest10.004610.0035 (0.0076) (0.0075) South10.005610.0080 (0.0070) (0.0071) West10.0284*10.0304* (0.0085) (0.0085) Qualified census tract 0.009110.0058 (0.0069) (0.0089) Difficult development area 0.0044 0.0052 (0.0077) (0.0077) Poverty 0.0669* (0.0251) Acquisition year10.0146*10.0149* (0.0012) (0.0012) New construction10.0264*10.0244* (0.0065) (0.0065) Nonprofit developer 0.0464* 0.0442* (0.0080) (0.0081) Section 515 0.0230* 0.0202* (0.0087) (0.0087) Project size (units) 37±5010.007110.0060 (0.0060) (0.0060) 51±100 0.0074 0.0079 (0.0073) (0.0072) 101`10.007910.0068 (0.0087) (0.0087) Constant 29.3675* 29.9402* (2.3068) (2.3111) Adjusted R 2 0.2300 0.2338 Observations (projects) 1,243 1,243 Note:Standard errors are in parentheses.
*p,0.05.
298Jean L. Cummings and Denise DiPasquale eral government provides substantial subsidy through the tax credit. In addition, LIHTC projects, like all investments in multi- family and commercial real estate, receive favorable federal tax benefits. Through the Rural Housing Service's Section 515 program, the federal government also has been a major provider of first mort- gages at deeply subsidized interest rates. In addition, the federal government provides tenant- and project-based assistance through Section 8.
State and local governments, as we saw, are significant providers of first mortgages for LIHTC projects, usually at subsidized rates.
State and local governments also provide significant levels of sub- sidy through below-market first mortgages and gap financing that may take the form of subsidized debt, soft debt, and grants. State and local subsidies often may be funded by federal subsidies such as Community Development Block Grant or HOME funds.Finally, pri- vate banks, nonprofit organizations, and retirement and insurance companies together provide 29 percent of first mortgages and 17 percent of gap financing across all of our projects. More than 60 percent of these loans from nonprofit organizations are below mar- ket, and about 20 percent of the loans from private banks are below market.
Federal tax subsidies The 2,397 projects in our sample for which we have information on federal tax credits generate a total of $5.32 billion in tax credits to investors (in nominal dollars). Because the tax credits are distrib- uted over 10 years, this represents $532 million in tax credits is- sued each year.
Table 7 builds an estimate of total subsidies per LIHTC unit in our sample. This table includes only the 2,145 projects for which we have accurate information on tax credits and interest rates. The first column shows the net present value (NPV) of the tax credits, which averages $28,910 per unit for this subsample. 48 Across metro- politan areas, there is wide variation in tax credits per unit. For ex- ample, Philadelphia projects generate $77,405 in tax credits, on av- erage, which is 3.5 times more than the tax credits per unit in Cleveland projects. 49 48We calculate the NPV of the 10-year stream of the annual tax credit payments using the GAO's annual discount rate of 6.7 percent, and then convert to 1996 dol- lars using the CPI.
49U.S. GAO (1997) showed a wide range in state allocation of credits: For the years 1992 to 1994, state tax-credit allocations per unit ranged from a high of $67,220 in California to $10,120 in Mississippi (in undiscounted, nominal dollars).
The Low-Income Housing Tax Credit: The First Ten Years299 Table 7.Estimated Total Subsidies Per LIHTC Unit Subsidies Tax Benefits Financing NPV of Subsidies NPV of from Grants and Subsidy/ Tax Credits Concessionary Loans Total TDC TDC(%) Units All Projects $28,910 $14,635 $43,545 $64,059 68.0 124,061 With Section 515 13,043 31,579 44,622 48,680 91.7 31,148 Without Section 515 34,229 9,104 43,333 69,215 62.6 92,913 Without Section 515 By Location Central City 36,609 15,375 51,984 76,221 68.2 42,676 Suburban 31,923 3,104 35,027 63,522 55.1 25,044 Nonmetropolitan 33,092 6,271 39,363 59,581 66.1 5,035 By Metropolitan Area Atlanta, GA 28,886 1,859 30,745 53,126 57.9 2,670 Baltimore, MD 33,220 11,704 44,924 72,715 61.8 2,111 Boston, MA 43,514 7,603 51,117 117,498 43.5 1,347 Chicago, IL 35,628 16,586 52,214 71,069 73.5 2,321 Cincinnati, OH 33,385 7,449 40,834 66,993 61.0 1,163 Cleveland, OH 22,113 10,647 32,760 50,874 64.4 1,392 Detroit, MI 30,636 4,824 35,460 68,880 51.5 1,431 Kansas City, KS/MO 37,037 8,334 45,372 65,134 69.7 1,630 Los Angeles, CA 52,003 21,336 73,339 104,410 70.2 3,231 Miami, FL 34,454 6,224 40,678 73,585 55.3 1,383 Minneapolis±St. Paul, MN 35,940 17,324 53,263 75,054 71.0 1,809 New York, NY 47,516 39,388 86,904 104,591 83.1 8,115 Philadelphia, PA 77,405 19,551 96,956 110,025 88.1 798 Washington, DC 31,215 6,314 37,529 69,740 53.8 2,075 Notes:This table is based on a subset of our data for which we have accurate information on tax credits and interest rates. We consider metropolitan areas only where the sample has at least 15 LIHTC projects. For these calculations we assume that all projects are held for 20 years. Section 515 loans are assumed to have 50-year terms; all other loans are assumed to have 20-year terms. Grants are assumed to be paid in full in the initial period. Loans with 0 percent interest are considered loans, not grants. Annual subsidies from concessionary loans are calculated based on the difference between the loan rate and the 30-year CMT. NPVs are calculated using a discount rate of 6.7 percent. All dollar figures are in 1996 dollars.
300Jean L. Cummings and Denise DiPasquale Concessionary financing The federal government provides substantial subsidies for Section 515 projects by offering below-market interest rates on first mort- gages. For Section 515 projects in our data, the average interest rate on a first mortgage is 1.2 percent. To estimate the size of this subsidy in our sample, we compare the interest rate on a Section 515 mortgage with the rate on 30-year CMT for the year the project was purchased by the syndicator. The 30-year CMT rate is a very conservative estimate of market interest rates. 50 By using this rate we are likely to understate the level of subsidy. On average, the Section 515 rate is 677 basis points below the CMT. Taking the spread between the Section 515 rate and the CMT times the out- standing mortgage amount, we calculate a subsidy implied by the rate discount. 51 We then calculate the NPV of the subsidy stream using a 6.7 percent discount rate and convert to 1996 dollars. As- suming a mortgage term of 50 years for the Section 515 projects in our data (the typical Section 515 mortgage term), and that the mortgages are not prepaid, the total subsidy from the Section 515 mortgage averages $38,944 per unit, a substantial subsidy given that TDC per unit averages $48,680 for these projects. In table 7, we assume that the mortgage is prepaid in year 20 when the prop- erty is sold. The average subsidy from the Section 515 mortgage in this case falls to $31,000 (the remaining $590 per unit comes from additional concessionary financing).
We perform similar calculations for the concessionary financing from state and local governments, private banks, nonprofits, and other lenders. We assume a mortgage term of 20 years (again, a conservative estimate since many LIHTC mortgages are for 30 or more years), and treat loans with 0 percent interest rates as loans rather than grants, resulting in a smaller subsidy. Using this defini- tion, 41 percent of the non±Section 515 first mortgages have conces- sionary rates. In addition, 86 percent of non±first mortgages have concessionary rates. We then add grant values to the NPV of the concessionary loans to get total financing subsidies. The value of these subsidies ranges from $40 per unit to nearly $100,000 per unit. 50Comparing the 30-year CMT with the average contract rate on apartment mort- gages provided by the American Council of Life Insurers (ACLI), we find, on aver- age, the ACLI rate is 113 bps above the CMT between 1988 and 1996. The spread ranges from a low of 70 bps in 1988 to a high of 140 bps in 1991 and 1993.
51We assume a fixed payment, self-amortizing mortgage with annual payments.
We calculate the interest in each period assuming an interest rate equal to the CMT and the actual interest rate. The difference is the annual subsidy.
The Low-Income Housing Tax Credit: The First Ten Years301 Total costs to society In table 7, we summarize the various subsidies provided to LIHTC projects by federal, state, and local governments and by private or- ganizations. The costs to the federal government for these tax credit units are the NPVs of the tax credits. The cost to government to ac- tually bring these units on line includes the cost of subsidized fi- nancing. Again, we are conservative in our assumptions in calculat- ing the subsidies: We assume that all projects are held for 20 years, treat loans with 0 percent interest rates as loans rather than as grants, and use the CMT as a measure of market-rate interest levels.
Even with these conservative assumptions, the data provided in ta- ble 7 show that the total costs to society of providing affordable rental housing through the LIHTC program are quite high. The evi- dence presented in table 7 indicates that, including the tax credits and subsidies from concessionary financing and grants, the projects in our sample receive a total of $43,545 per unit in subsidies, on av- erage, which represents 68 percent of average TDC per unit. The federal government provides 66 percent of these subsidies through tax credits. Subsidies for Section 515 projects represent a larger portion of TDC than for non±Section 515 projects. Section 515 proj- ects receive significantly less in tax credits (29 percent of total sub- sidies) but more in loan subsidies because of the deep discount on Section 515 mortgages. Across central-city, suburban, and rural lo- cations, non±Section 515 projects receive similar tax credits per unit. Central-city projects, however, receive significantly more in subsidized loans and grants, bringing the total subsidy to $51,984, on average. The average central-city subsidy is more than 30 per- cent higher than the average subsidy on rural units and 48 percent higher than the average suburban unit subsidy.
Subsidies vary widely across local markets. Our estimated subsidies per unit are more than 80 percent of TDC per unit in New York and Philadelphia. In Atlanta, Chicago, Kansas City, and Philadelphia, the NPV of the tax credits per unit represents more than 50 percent of the TDC per unit. The importance of subsidized loans and grants varies widely from $1,859 per unit or 3 percent of TDC in Atlanta to $39,388 per unit or 38 percent of TDC in New York City. The first mortgages alone provided by the city of New York carry substantial subsidies averaging nearly $36,000 per unit. In contrast, first mort- gages provided by state and local government lenders in Boston bring much more modest subsidies of $5,000 per unit.
The total subsidies provided range from $30,745 in Atlanta to $96,956 in Philadelphia. These subsidies seem quite large, espe- cially given that these estimates do not include many tenant- or 302Jean L. Cummings and Denise DiPasquale project-based Section 8 subsidies. The size of the subsidy may well reflect local housing policy. The large subsidies in Los Angeles, for example, may reflect the high costs of providing SROs for special- needs populations.
The figures in table 7 do not include all subsidies used to bring LIHTC units into the marketplace. These excluded subsidies can be large. In many cases, the rents required to make LIHTC projects vi- able are higher than the rents affordable to the target population.
As a result, many projects rely heavily on tenant-based and addi- tional project-based subsidies. Our data do not permit us to identify tenant-based and certain project-based subsidies such as Section 8.
However, U.S. GAO (1997) indicates that the 152,658 households in its study received $299 million in rental subsidies annually, which averages to roughly $1,500 per household. 52 The vast majority of tenant-based rental subsidies are short-term (e.g., one year) com- mitments but are renewable. To get an idea of the potential size of this subsidy, assume that residents of LIHTC projects receive $1,500 per year in rental subsidy for the life of the project (assumed to be 20 years in this calculation). The present discounted value of these annual rental subsidies is substantial, roughly $18,200 per unit or 28 percent of average TDC per unit. If we add this subsidy to the amounts in table 7, the average subsidy per unit reaches 96 percent of TDC.
In addition, investors in LIHTC projects, like other real estate in- vestors, receive benefits in the form of favorable treatment for fed- eral income tax purposes. As discussed earlier in this article, an im- portant component of the return to equity investors in real estate is the deduction of depreciation from taxable income. Our data permit the calculation of the value of this deduction to investors in the LIHTC projects in our sample. On average, the value of the depreci- ation deduction in the reduction of income taxes paid is $8,900 per unit, or 14 percent of TDC, after taking into account the capital gains tax paid on depreciation recaptured when the property is sold in year 20. 53 We did not include this calculation of the depreciation benefit in table 7 because the rationale for the depreciation deduc- tion is to permit investors to realize the real costs associated with a 52According to U.S. GAO (1997), roughly 40 percent of households in its study ac- tually received rent-based subsidies. Therefore, the average annual subsidy to each of the households that received rental subsidies is $3,750.
53We assume a corporate tax rate of 40 percent and that the investor holds the property for 20 years. We assume that all depreciation taken is recaptured at sale.
The value of the depreciation deduction is the present discounted value of the re- duction in tax due to depreciation over the 20 years the property is held minus the present value of the capital gains tax paid on the depreciation recaptured at sale.
The Low-Income Housing Tax Credit: The First Ten Years303 structure wearing out over time. 54 However, because a building's tax life (e.g., 27.5 years) is generally considerably shorter than its economic life, some portion of the depreciation benefit may be viewed as a subsidy.
These subsidy calculations raise an important issue concerning the efficiency of the LIHTC program. Our price calculations in the pre- vious section indicate that, over time, more of every tax-credit dol- lar provided by the federal government ends up in the project, which we interpret as evidence that the tax-credit program is be- coming more efficient. The subsidy calculations in table 7 and above are estimates of thetotal cost to societyof LIHTC projects, including all sources of subsidies. While we find increased efficiency in get- ting more of every tax-credit dollar into the project, the overall structure of the LIHTC program, with its dependence on many sources and types of subsidy, can make it expensive. The fact that the program requires deep subsidies to provide housing that serves low-income householdsÐbut certainly not the poorest of the poorÐ is not a failure of the programper se. But because it is the major federal housing production vehicle, it is important for policy makers to recognize the full costs of delivering affordable units through the LIHTC program. Conclusion The LIHTC program has been thede factofederal production pro- gram for affordable housing since its inception in 1987. Has it been successful? Measuring success requires comparing outcomes with goals. Because it brings so many players to the table, the program has many goals, some of which are in conflict.
The LIHTC program certainly has generated unitsÐroughly 550,000 to 600,000 units were put in place in its first 10 years. The program has produced a wide variety of housing types and served a broad range of populations, suggesting that the flexibility in the program's design to permit state and local governments to pursue their own policy goals is working.
Our data suggest that the program has been used most often to pro- vide better housing in poor neighborhoods rather than housing op- portunities for poor households in higher-income neighborhoods. A significant portion of projects in our sample are in neighborhoods 54The unique feature of depreciation is that the deduction does not match an ac- tual cash expenditure. However, the physical decline in the structure over time represents a real cost to the investor.
304Jean L. Cummings and Denise DiPasquale that are quite racially concentratedÐmore than 30 percent of the projects are in neighborhoods with a population that is at least 90 percent white, and nearly 18 percent are in neighborhoods with a population that is at least 90 percent nonwhite. Some policy makers use the LIHTC program to encourage new investment in deteriorat- ing or struggling neighborhoods. Our data show that in many of the neighborhoods where LIHTC projects are built, LIHTC units repre- sent the only new residential construction in recent years.
LIHTC projects can be expensive. For our sample, TDC per unit av- erages $65,307, but 11 percent of the units in the sample have per unit TDC that exceeds $100,000. State and local officials often favor nonprofit developers because they bring community support and commitment to projects. These are clearly valuable attributes, but nonprofits generally bring additional costs. Controlling for project size, unit type, and location, units developed by nonprofit sponsors, on average, cost 20.3 percent more than for-profit units in our sample.
While our data suggest that LIHTC projects serve low- and moderate-income households, these projects generally do not serve the poorest households. Pegging income limits for the LIHTC pro- gram to HUD-adjusted Section 8 income limits pushes the LIHTC program toward higher-income renters who generate higher rents, which enhances financial viability but excludes very poor tenants.
All participants in the LIHTC program are concerned with the fi- nancial viability of projects. While the majority of projects in our sample were financially sound in 1995, 16 percent of the projects had operating expenses greater than 105 percent of operating reve- nues. In some markets, LIHTC vacancy rates exceed market rates, suggesting that developers or state agencies may have misjudged the market. Many of the projects in our sample are dependent on additional subsidies to be financially viable.
A major goal of the LIHTC program was to attract the private sec- tor to the building of affordable housing in order to leverage scarce subsidy dollars, which the program certainly has done. Private banks have been a significant source of debt financing, and the number of investors participating in the program has increased steadily. We find strong evidence that the growth of the program and the increased competition from developers and investors has led to increased efficiency in raising LIHTC equity. Over time, more of each tax credit dollar has gone directly into the projects. In addi- tion, returns to investors have decreased over time, especially for central-city projects, perhaps indicating decreased perception of risk for these projects.
The Low-Income Housing Tax Credit: The First Ten Years305 The LIHTC program's total costs to society, however, are quite high.
We estimate an average per-unit subsidy of $43,545, which we know is an underestimate because it does not include tenant-based and some project-based subsidies. Tax credits represent two-thirds of the total subsidies per unit. Subsidies are considerably higher in central cities and in some of the larger metropolitan areas.
Going forward, the LIHTC program may face some difficult chal- lenges. The program is only 11 years old, which means that most of its housing has not yet required major capital for renovation or sys- tem replacement. The ability to meet these needs over time will de- termine the long-term financial viability of these projects and de- pends critically on funding reserves. From a policy perspective, the issue of maintaining affordability over time also must be consid- ered. The oldest units built under the LIHTC program are required to remain affordable for 15 years and many projects now require 30 years of affordability. In 5 to 10 years, many units could convert to market-rate units, creating a problem similar to the expiring-use problem of the late 1980s. While many LIHTC projects have provi- sions granting nonprofit organizations the right of first refusal to purchase the buildings at expiration, these organizations still will have to find the funds to purchase and maintain the buildings. The substantial subsidies beyond the tax credit required to make many of these projects work may become more scarce over time. In addi- tion, because the LIHTC program is virtually the only housing pro- duction program left, pressure on state and local officials to reach lower-income households may increase. Without substantial addi- tional subsidy, this goal is difficult to achieve while still meeting the requirements of the private sector participants.
Given the important role of the LIHTC program in the provision of affordable rental housing, an increased effort to collect data that would permit ongoing analysis of the program is needed. At this point, 13 years after it began, we lack some basic facts on the pro- gram, such as the total number of units produced. The analysis in this article suggests that more information is required in several ar- eas: on tenant profiles, to shed more light on who is served by the program; on project characteristics, such as unit size in square feet, project amenities, and tenant support services, to provide a better understanding of development costs; and on financial variables, such as operating reserves and the use of project- and tenant-based subsidies, to provide a more in-depth examination of financial via- bility. Public information on project characteristics, costs, financial viability, and returns to the investors are essential to the develop- ment of an efficient program and an assessment of the extent to which the program is achieving its goals.
306Jean L. Cummings and Denise DiPasquale Authors Jean L. Cummings cofounded and is currently a Principal at City Research, a re- search and consulting firm that focuses on urban economics and policy. Denise Di- Pasquale is President and cofounder of City Research.
The authors would like to thank Dennis Fricke, Jeff Goldstein, Peter Levavi, Mat- thew Kahn, Stephen Malpezzi, Mark Willis, Benson Roberts, and anonymous refer- ees for their comments on a previous draft. Research on which this article is based was funded in part by the National Community Development Initiative, a consor- tium of private foundations, corporations, and public sector funders; however, the analyses and conclusions presented are solely those of the authors.
References Abt Associates. 1993.Nonprofit Housing: Costs and Funding.Report prepared for the U.S. Department of Housing and Urban Development, Office of Policy Development and Research.
Abt Associates. 1996. Development and Analysis of the National Low-Income Housing Tax Credit Database. U.S. Department of Housing and Urban Development, Office of Policy Development and Research, Washington, DC. World Wide Web page,http://www.huduser.org..
Case, Karl E. 1991. Investors, Developers, and Supply-Side Subsidies: How Much is Enough?Housing Policy Debate2(2):341±56.
Congressional Record.1986. S8146±8158, Senate Floor Amendment 132 (June 23).
Cummings, Jean L., and Denise DiPasquale. 1998a.Building Affordable Rental Housing: An Analysis of the Low-Income Housing Tax Credit. City Research. World Wide Web page,http://www.cityresearch.com..
Cummings, Jean L., and Denise DiPasquale. 1998b. Developing a Secondary Market for Affordable Rental Housing: Lessons From the LIMAC/Freddie Mac and EMI/Fannie Mae Programs.Cityscape: A Journal of Policy Development and Research4(1):19±41.
DiPasquale, Denise, and Jean L. Cummings. 1992. Financing Multifamily Rental Housing: The Changing Role of Lenders and Investors.Housing Policy Debate 3(1):77±116.
Ellen, Ingrid. 1996. Sharing America's Neighborhoods. Ph.D. diss. Harvard University.
Ernst & Young. 1997.The Low Income Housing Tax Credit: The First Decade.E&Y Kenneth Leventhal.
Guggenheim, Joseph. 1994.Tax Credits for Low Income Housing. 8th ed., Glen Echo, MD: Simon.
ICF, Inc. 1991.Evaluation of the Low-Income Housing Tax Credit, Final Report.
ICF.
The Low-Income Housing Tax Credit: The First Ten Years307 Keyes, Langley C., Alex Schwartz, Avis C. Vidal, and Rachel G. Bratt. 1996.
Networks and Nonprofits: Opportunities and Challenges in an Era of Federal Devolution.Housing Policy Debate7(2):201±29.
NAREIT World Wide Web page,http://www.nareit.com..
National Council of State Housing Agencies (NCSHA). 1996.State HFA Factbook:
1995 NCSHA Annual Survey Results.Washington, DC: NCSHA.
Newman, Sandra J., and Ann B. Schnare. 1997. Housing Programs Fail to Deliver on Neighborhood Quality.Housing Policy Debate8(4):703±41.
Stegman, Michael A. 1991. The Excessive Costs of Creative Finance: Growing Inefficiencies in the Production of Low-Income Housing.Housing Policy Debate 2(2):357±73.
U.S. General Accounting Office. 1997.Tax Credits: Opportunities to Improve Oversight of the Low-Income Housing Program.
U.S. General Accounting Office. 1999.Tax Credits: Reasons for Cost Differences in Housing Built by For-Profit and Nonprofit Developers.
Walker, Christopher. 1993. Nonprofit Housing Development: Status, Trends, and Prospects.Housing Policy Debate4(3):369±414.
Wallace, James E. 1991. Who Benefits from Preservation: The People Behind the Numbers.Housing Policy Debate2(2):219±44.
Wallace, James E. 1998. Evaluating the Low-Income Housing Tax Credit. In Evaluating Tax Expenditures: Tools and Techniques for Assessing Outcomes.
chap. 3, 43±62. San Francisco: Jossey-Bass.
