The Nonimpact of Opportunity Zones on Home and Business Lending

Abstract

Opportunity Zones (OZs) promised to stimulate investment in over 8,500 “distressed” neighborhoods. Have OZs increased neighborhood investment and, if so, what types of neighborhoods have benefitted? This study uses a difference-in-differences (DID) design to compare small business and residential lending outcomes in designated OZs with areas that were eligible but not designated. Census tracts are stratified by pretreatment social and economic indicators of distress to search for heterogeneity in effects by neighborhood type. An event study framework is used to check the parallel trends assumption and census tracts are then matched to improve the counterfactual. Finally, an adjusted interrupted time series (AITS) analysis is introduced to further evaluate differences in outcome indicator levels and trends between target and control neighborhoods pre and post OZ. DID and AITS estimates suggest that OZs have had no statistically significant effects on business or residential loan growth. Heterogeneity modeling confirms a noneffect across neighborhood distress type. In conclusion, study limitations and future extensions for both policy and research are discussed.

Keywords:

• Opportunity Zones
• small business lending
• housing policy
• home mortgages
• development/revitalization
• tax policy
• neighborhood distress
• economic development

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Notes

1 This estimate was established using the Joint Committee on Taxation forgone revenue estimates from 2018 to 2021 to extrapolate the 10-year initial “term” of the incentive. This estimate does not include inflows of revenue expected in 2026 when the initial capital gains tax deferment may be due. However, it also does not include the likely much larger long-run cost to the Treasury in forgone tax revenues (thus making it a conservative estimate).

2 The OZ statute in the TCJA allowed census tracts to be designated OZs if they were LICs under the definitions established for the New Markets Tax Credit program; census tracts with at least 50% of households with incomes below 60% of the area median income or a poverty rate above 25%. Census tracts that were contiguous to LICs were also eligible for selection if their area median income did not exceed 125% of the adjacent qualifying LIC tract.

3 After passage of the TCJA, governors had 90 days—until March 21, 2018—to nominate zones or seek a 30-day extension. The first set of designations, covering 18 states, were made public by the Treasury on April 9, 2018. All qualified OZs were officially posted by the Treasury on July 9, 2018.

4 See Gelfond and Looney (2018) and Theodos et al. (2018).

5 See The Promise of Opportunity Zones, Senate Hearing 115-297 before the Joint Economic Committee (2018).

6 See Bernstein and Hassett (2015) for the original OZ white paper.

7 See Wessel (2021, “Choosing the Zones”) for the most in-depth review of OZ selection and its criticisms.

8 A few studies also find that neighborhoods receiving Empowerment Zone designation experienced improvements in labor market conditions. See Busso and Kline (2008) and Ham et al. (2011). For a comprehensive review of tax preferences, as well as direct expenditure initiatives, for place-based development see Foell and Pitzer (2020).

9 Note: Corinth and Feldman (2021) employ a fuzzy regression discontinuity design for evaluation robustness but find outcomes consistent with their DID modeling.

10 There is no mandatory federal requirement to report OZ equity investments to the US Treasury or a plan by the Treasury to release voluntary reporting data to the public (Wessel, 2021).

11 In small business development, debt finance typically quickly follows equity investments. While venture capital has received significant attention with the financialization of technology-oriented startups, debt finance remains fundamental to the overall capital structures of small businesses—including those for young firms. See Berger and Udell (2003), Cotei and Farhat (2017), and Ibrahim (2010).

12 Chen et al. (2019) support their analysis of home prices post OZ designation under a similar theoretical construction. A notable caveat to this theory is that small business and home investors may lack knowledge or be incorrect in their expectations of OZ.

13 Census tracts are frequently used as proxies for neighborhoods in policy evaluations or examinations of “neighborhood effects.” They are also frequently the geographic level by which place-based incentives—including OZ—are designated. For a history on census tracts as units of analysis for place-based evaluations, as well as a specific discussion of their validity in representing neighborhoods in regards to OZ policy assignment, see Fraker (2022).

14 CRA small business loan data has several noteworthy limitations. First, loans are reported by census tract but there is no way to confirm where loan capital is deployed (i.e., the funds might be used to support firm activities in other locations e.g., another office). There is evidence that the data contain a large number of P.O. boxes, compromising the geographical assignment. Second, commercial banks and savings associations typically do not fund startups, meaning the data more likely represent lending for existing businesses. Third, the data do not capture larger commercial real estate loans above $1 million. Finally, the CRA does not provide data on other factors that might be influencing supply and demand of credit (e.g., underwriting standards or small business age) and impacting loan extensions in any given community. See Canner (1999).

15 HMDA requires most depository lending institutions, including banks, credit unions, and savings associations, as well as non-depository institutions, like mortgage companies, to file mortgage loan applications. This includes investment/business loans for home purchases, improvements, and refinances. For the purposes of studying outcomes in LICs, however, it has two noteworthy limitations: the small number of loans issued in LICs and lack of data on the precise number of multifamily dwelling units associated with multifamily loans make it hard to interpret the effects on area rent levels and multifamily housing markets, specifically and respectively. For more on this and other HMDA limitations see Pettit and Droesch (2009). For the history of HMDA reporting requirements and criteria for lenders see FFIEC (2020).

16 Total and average number of residential loans and total and average loan amounts are then calculated for each census tract. Loans are separated by single family (1–4 units) and multifamily (over five units) homes and are also separated by race of borrower for sensitivity testing. This separation did not affect the findings.

17 For additional details on how this index was established, see Table 1 and Theodos et al. (2018).

18 Non LICs represent roughly 20% of OZ eligible census tracts. However, due to limitations on how many LICs could be selected, less than 3% (230 census tracts) were designated as OZs. Maintaining non LICs for the general comparison would result in a disproportionate number of less distressed neighborhoods being used as controls. Instead, non LIC census tracts are compared as a separate and additional evaluation.

19 Outliers were identified using box plots. Specifically, people per housing unit was converted to a logarithmic scale and census tracts 1.5 times below the interquartile range or 1.5 times above the third quartile range were removed. This resulted in a sample of census tracts with 1.26 to 4.28 people per housing unit.

20 Models including census tracts with very low population and housing statistics are also checked. This did not substantially alter the findings.

21 This model is replicated from Arefeva et al. (2020), Chen et al. (2019) and Freedman et al. (2021): 10,000 simulations that randomly select ⅓ of the census tracts for the DID analysis are run to test and confirm that our model results are not due to chance. We also test the dummy for just 2019, as 2018 is not a full treatment year, but this does not alter the findings.

22 Three additional robustness and sensitivity steps are taken. First, based on these indicators, control and treatment groups are split into just two intervals (i.e., less impoverished, and more impoverished), and four intervals (e.g., highest unemployment, high unemployment, less unemployment, least unemployment) and in the case of the UI scores by individual (1–10) scores. Second, Jenks Natural Breaks are considered to partition the data, but this results in poor model fit and is dropped from the analysis. Third, as HMDA data include residential loan originations by race, the dependent variable of home loan originations is stratified to test whether outcomes differ by the race of the borrower. None of these models substantially alter the findings.

23 This event study design framework is replicated from Freedman et al. (2021).

24 Plot graphs are constructed using the 5-year ACS estimate described in the data section. The United States Census Bureau (2018) recommends conceptualizing these as estimates, not counts. Count data are more accurately recorded in the Census Bureau’s Population and Housing Units Estimates (PEP) pages. We use ACS data, however, because we do not seek to predict specific counts but to visualize trends of these covariates and because the PEP pages do not provide yearly data.

25 PSM designs are roughly modeled after Chen et al. (2019) and Freedman et al. (2021).

26 As homeownership and therefore rate of loan originations may be depressed in lower income census tracts, we also add home ownership percentage as a matching criterion, but this does not affect the findings.

27 To maintain model stability, we also only keep census tracts with an R squared of the regression of at least .2.

28 We focus on these methods because just 230 of a possible 10,285 non-LIC contiguous census tracts were designated OZs. Full-sample comparisons and matching by pretreatment lending outcomes were not suitable for establishing a counterfactual.

29 Simulation sampling supports our findings of non-effect. Of the 10,000 simulations, only 2.1% of models demonstrate a statistically significant (p < .05) effect regarding the growth of small business loans. Similarly, we find only .02%, 1.1%, and 3.3% of models show any statistically significant effect when considering small business loan amount, home loans, and home loan amount, respectively.

30 This modeling can be provided upon request from the first author.

31 There is one noteworthy exception: the average number of business loans and loan amounts per capita is higher in OZs than in the control group. A likely explanation is that localities favored census tracts where existing business districts and commercial land uses were in place as opposed to selecting neighborhoods with heavier residential land use—as the ostensible goal of the legislation was to encourage business development.

32 These results are similar to those of Chen et al. (2019) who rule out increases on housing prices above 1.3%, and Freedman et al. (2021) who rule out effect sizes on employment for zone residents above .2% and reductions in the poverty rate larger than 1.6% with 95% confidence. To some contrast, Pierzak (2021) finds statistically significant positive effects on apartment prices by 7.8% and Arefeva et al. (2021) find statistically significant positive effects on employment between 3 and 4.5%.

33 Evaluations of place-based policies suggest that original residents are unlikely to be the main beneficiaries (Busso et al., 2013; Freedman, 2015) and OZ has been structured so most benefits accrue to the capital gain holding class—who are unlikely to be the residents of distressed communities (Eastman & Kaeding, 2019).

34 For example, Sage et al. (2021) find that OZ increased the value of vacant and deteriorating but not other properties; Pierzak finds statistically significant and considerable OZ effects on existing market-rate apartment properties; and Atkins et al. (2020) find OZ may have greater, albeit still nominal, job growth effects on urban over rural census tracts.

35 Likewise, we do not account for other subsidies or incentives that OZ census tracts may receive.

36 As discussed in the literature review, limited research has documented statistically significant outcomes for neighborhoods receiving NMTC program designation (Freedman, 2015; Harger & Ross, 2016) and investment (Theodos et al., 2022).

37 See Tankersley (2021) and Wessel (2021) regarding current support from Congress and the Biden Administration for OZ.

38 Qualitative reporting indicates that OZ might lead to increased investment in LICs if the incentive were increased only in the most distressed census tracts (see Snidal & Newman, 2022) and expanded to benefit community actors, like community development financial institutions, that have a history of investing in LICs (see Kim 2022; Snidal & Newman, 2022; Theodos et al. 2020).

Additional information

Notes on contributors

Michael Snidal

Michael Snidal is the principal of Snidal Real Estate, a Baltimore based real estate and property management company and a doctoral candidate at Columbia University's Graduate School of Architecture, Planning and Preservation.

Guanglai Li

Guanglai Li is a data scientist.