A. Data Sources and Variable Construction

B. Summary Statistics

Summary statistics are reported in Table 1, with variable definitions listed in the Appendix. In the sample, there are 42,171 census tracts eligible for the Opportunity Zone designation; among these, 31,859 are LICs, while the rest are contiguous, non-LIC tracts. The sample includes 8,761 census tracts designated as Opportunity Zones, of which 8,531 are LICs.Footnote ¹⁸ For most analyses in the article, I include only the LIC tracts to make the treated and control groups more comparable. The main findings are all robust if non-LIC contiguous tracts are included. The sample period is from 2015 to 2019. All tract-level amount variables are winsorized at the 1st and 99th percentiles to avoid data errors involving extreme values that may drive the results.

TABLE 1Summary Statistics

TABLE 1 Long description

The table contains summary statistics for variables at the census-tract-year level.

Panel A. Private Investments and Entrepreneurship:

* Opportunity Zone dummy (O Z): N 154,563, Mean 0.247, Std. Dev. 0.431, Min 0.000, Median 0.000, Max 1.000.

* No. of investment deals (Num_Inv): N 154,563, Mean 0.118, Std. Dev. 2.257, Max 641.000.

* Dollar amount of investments (Amount_Inv, $millions): N 154,563, Mean 2.732, Std. Dev. 92.635, Max 15,422.180.

* No. of new firms (Num_NewFirm): N 147,638, Mean 10.158, Std. Dev. 46.537, Median 3.000, Max 4,628.000.

Panel B. Control Variables (N 154,492 for all):

* Population (thousands): Mean 4.046, Std. Dev. 1.892.

* Median income ($thousands): Mean 39.156, Std. Dev. 12.987.

* Median age: Mean 35.860, Std. Dev. 7.501.

* Poverty rate (%): Mean 22.181, Std. Dev. 9.976.

* White alone (%): Mean 61.206, Std. Dev. 28.727.

* Black alone (%): Mean 23.000, Std. Dev. 27.929.

* Unemployment rate (%): Mean 10.827, Std. Dev. 6.116.

* Without high school degree rate (%): Mean 20.867, Std. Dev. 11.883.

Panel C. Employment and Sales Related:

* Total employment (thousands): N 154,476, Mean 2.497, Std. Dev. 4.253.

* Total sales ($million): N 154,476, Mean 400.210, Std. Dev. 3,939.252.

* Employment HHI (Non-tradable): N 153,709, Mean 1,758.177, Std. Dev. 1,562.017.

* Employment HHI (Tradable): N 140,314, Mean 4,747.626, Std. Dev. 2,849.086.

* Sales HHI (Non-tradable): N 153,709, Mean 2,741.045, Std. Dev. 1,683.877.

* Sales HHI (Tradable): N 140,314, Mean 5,399.009, Std. Dev. 2,832.087.

Panel D. Other Economic Outcomes (N 147,638 for all):

* No. of firms dissolved: Mean 0.431, Std. Dev. 3.078, Max 436.000.

* No. of firms dissolved, age > 1 year: Mean 0.390, Std. Dev. 2.864, Max 407.000.

* No. of firms dissolved, age > 2 years: Mean 0.172, Std. Dev. 1.353, Max 193.000.

As shown in Table 1, 24.7% of the tracts are Opportunity Zones, while the remaining tracts are eligible but non-designated. On average, each year, a census tract has 0.12 private investments that average $2.732 million. In Supplementary Material Table A1, I report more details on the age and geographical distributions of private investment deals. An average census tract in the sample has about 11 new firms incorporated each year and has 4,046 people with a median income of $39,156, a poverty rate of 22%, a population that is 61% white and 23% black, an unemployment rate of 11%, and 21% without a high school diploma.

A. Impact on Local Private Investments

I start the empirical analysis by examining the impact of the Opportunity Zone policy on local private investments. Columns 1–4 in Table 2 report the ordinary least squares (OLS) regressions. The dependent variables are the natural logarithm of 1 plus the number of private investment deals in census tract i and year t, Ln(Num_Inv + 1) and the natural logarithm of 1 plus the dollar amount of private investments, Ln(Amount_Inv + 1). The coefficient estimates on $ OZ\ast Post $ are all positive and significant at the 1% level, either with or without control variables. The magnitude of the coefficient estimates suggests that the effects are also economically sizable: after the policy shock, the number and amount of private investments that flowed into treated tracts (the Opportunity Zones) increased by 10.5% and 16.1%, respectively, compared to tracts that were eligible but not designated.Footnote ²¹ Following Cohn, Liu, and Wardlaw (Reference Cohn, Liu and Wardlaw2022), I report the results of Poisson estimation in Supplementary Material Table A4 and find positive and statistically significant coefficient estimates on $ OZ\ast Post $ . When breaking down the impact on local private investments into three categories—finance, real estate, and business—based on the “industry” information in the Form D filings as shown in Supplementary Material Table A5, I find significant and positive impact on private investments across all three sectors, with the largest effect on the business sector.Footnote ²²

TABLE 2Impact of OZs on Private Investment and Entrepreneurship

TABLE 2 Long description

The table presents regression results for three dependent variables: natural log of number of investments plus 1, natural log of investment amount plus 1, and natural log of new firms plus 1. Each variable is tested in two models, one without and one with control variables.

Key findings for the OZ times Post interaction term:

* For Ln(Num_Inv + 1): Models 1 and 2 show a positive coefficient of 0.011, significant at the 1% level.

* For Ln(Amount_Inv + 1): Models 3 and 4 show positive coefficients of 0.151 and 0.149 respectively, significant at the 1% level.

* For Ln(Num_NewFirms + 1): Models 5 and 6 show negative coefficients of -0.018 and -0.019, significant at the 10% and 5% levels respectively.

Control variables included in even-numbered models:

* Population: Positive and significant for investment models, negative and non-significant for new firms.

* Median_Income: Positive and significant only for new firms (0.085).

* Median_Age: Positive and significant only for new firms (0.151).

* %White and Poverty_Rate: Small but statistically significant positive coefficients for investment models.

* %Black and %NoHighSchool: Negative and significant for new firms.

Model Statistics:

* Number of observations range from 147,565 to 154,563.

* R-squared values are 0.733 for investment count, 0.565 for investment amount, and 0.849 for new firms.

* All models include Tract and Year Fixed Effects.

To further analyze the impact on local investments, I examine how firm age influences investment patterns. As outlined in the model (Section VI.C), Opportunity Zone investors are likely to favor existing firms relative to new firms, likely due to their lower risk and more information available. To test this hypothesis, I categorize firms based on whether they have been operational for at least 1 year and compare the local private investments received by each group. The 1-year threshold captures the “up or out” dynamic among new firms (Wiens and Jackson (Reference Wiens and Jackson2015)), as evidenced by Bureau of Labor Statistics data showing that roughly 25% of new firms fail within their first year, with exit rates declining thereafter.

The results are shown in Table 3, where the dependent variables are the number of investment deals (Ln(Num + 1)) and the dollar amount of deals (Ln(Amount + 1)) with other empirical specification similar to equation (1). The first two columns in Table 3 suggest that the Opportunity Zone policy has invited more private investment deals for older firms than newly formed firms. The difference between these two types of firms is statistically significant at the 1% level (z-stat being 3.578). Similarly, the dollar amount of investment increased for both older and newly formed firms, but the increase was significantly larger for the older ones in the Opportunity Zones than for the newly formed ones. It is worth noting that the increase in investment amount is greater than that in deal count, suggesting investors were deploying large sums into fewer firms. This pattern could be driven by the tight timeline to meet the qualification requirements and is consistent with the interpretation that investment flows favored existing firms.

TABLE 3Impact on Private Investments by Firm Age

TABLE 3 Long description

The table is organized into four columns under two main dependent variables. Columns 1 and 2 measure the natural logarithm of one plus the number of investments L n ( N u m _ I n v plus 1). Columns 3 and 4 measure the natural logarithm of one plus the investment amount L n ( A m o u n t _ I n v plus 1).

Within each pair, the first column represents firms One Year and Above, and the second column represents firms Less Than 1 Year.

Row 1: O Z asterisk P o s t interaction term.

- Column 1: 0.010 with significance at 1 percent level. Standard error 0.002.

- Column 2: 0.002 with significance at 5 percent level. Standard error 0.001.

- Column 3: 0.140 with significance at 1 percent level. Standard error 0.031.

- Column 4: 0.035 with significance at 5 percent level. Standard error 0.016.

Row 2: Number of observations is 154,492 for all columns.

Row 3: R-squared values.

- Column 1: 0.727.

- Column 2: 0.579.

- Column 3: 0.576.

- Column 4: 0.450.

Row 4 to 6: Controls, Tract F E, and Year F E are marked Yes for all columns.

Row 7: Difference.

- Columns 1 and 2: 0.008.

- Columns 3 and 4: 0.105.

Row 8: z-statistics.

- Columns 1 and 2: 3.578 with significance at 1 percent level.

- Columns 3 and 4: 3.010 with significance at 1 percent level.

Table 3 presents the results where the dependent variables are the number of investment deals (Ln(Num + 1)) and the total dollar amount (Ln(Amount + 1)), using specifications consistent with equation (1). Columns 1 and 2 indicate that the Opportunity Zone policy attracted significantly more private investment deals to older firms compared to newly formed ones and the difference is statistically significant at the 1% level (z-stat = 3.578). A similar divergence appears in investment amount, while both groups experienced an increase, the capital injection into older incumbents was substantially larger than that into new firms. Notably, the response in investment amount is much stronger than that in the investment count, suggesting that investors are investing more money into fewer firms. This pattern is consistent with the program’s tight qualification timeline, which likely incentivized investors to prioritize “shovel-ready” projects in established firms over new ventures that require longer lead times.

The previous findings show that the policy had a distributional effect on private investments between incumbents and newly formed firms. Despite bringing more private investments into economically distressed communities, incumbents benefited significantly more than newly formed firms.

B. Impact on Local Entrepreneurship

After verifying that the Opportunity Zone policy has indeed increased private investments in the treated census tracts, I next examine its impact on local entrepreneurship, measured by new business formation. Table 2, columns 5 and 6, report the results where the dependent variable is the total number of new businesses registered in census tract i and year t (Ln(Num_NewFirm + 1)), without and with controls, respectively. The coefficient estimate on $ OZ\ast Post $ is negative and statistically significant at the 1% significance level. This suggests that, after the introduction of the policy, new business formation declined an average of 2.1% more in census tracts designated as Opportunity Zones, compared to other eligible but non-designated tracts.Footnote ²³ Poisson regression also reports a significantly negative estimate, as shown in Supplementary Material Table A4, column 3.

Next, I test whether there is a decline in local new business formation in sectors with greater local competition. The intuition is that as investors allocate more capital to existing, older firms in the treated areas, the potentially widened gap in financial resources would discourage a potential entrepreneur from entering the market in sectors that compete mostly locally.

Using the industry classification from Mian and Sufi (Reference Mian and Sufi2014), I divide sectors into non-tradable (e.g., restaurants, grocery stores, and retail services), tradable (e.g., manufacturing and wholesale), construction, and other. Intuitively, firms in the non-tradable sector mostly compete locally. Table 4 presents the effect of the policy on the formation of new businesses in these sectors. I observe that the coefficient estimate on $ OZ\ast Post $ is negative and statistically significant for the non-tradable and other sectors, while it is insignificant for the tradable and construction sectors. This finding suggests that the Opportunity Zone policy could help incumbent firms gain financial resources and maintain their competitive advantages, such as store renovations, as shown by Sage, Langen, and Van de Minne (Reference Sage, Langen and Van de Minne2023), who find that commercial property prices rose significantly in treated areas where upgrades were needed. Potential entrepreneurs, particularly in non-tradable sectors, may avoid starting businesses if they anticipate being unable to match the financial capacity of incumbent firms.Footnote ²⁴ Consistent with this, I later show that local employment decreased in Opportunity Zones, suggesting that private investments were directed toward non-labor expenditures like renovations rather than job creation.

TABLE 4Impact on Local Entrepreneurship By Sector

TABLE 4 Long description

The table presents regression results for the dependent variable L n (Num underscore NewFirms plus 1) across four sectors: (1) Non-Tradable, (2) Tradable, (3) Construction, and (4) Other.

For the primary interaction term O Z times Post:

- Non-Tradable: -0.016 with three asterisks (standard error 0.005).

- Tradable: -0.003 (standard error 0.002).

- Construction: -0.005 (standard error 0.006).

- Other: -0.021 with two asterisks (standard error 0.009).

Model statistics for all four columns:

- Number of observations: 147,565.

- R-squared values: 0.568 for Non-Tradable, 0.388 for Tradable, 0.722 for Construction, and 0.847 for Other.

- Controls, Tract F E, and Year F E are included (Yes) for all models.

Also, one should expect the decline in new business formation to be greater in Opportunity Zones that have received private investments than those that have not, compared to non-treated census tracts. Supplementary Material Table A7 confirms the previous conjecture that the magnitude of the coefficient on $ OZ\ast Post $ in a sample restricted to only positive-investments Opportunity Zones is about twice that in the baseline regressions.

Finally, I validate the entrepreneurship finding using data from the ZBPs provided by the Census Bureau. Instead of counting the number of firms registered, the ZBP database provides statistics on the total number of establishments in a zip code. I calculate the changes in the number of establishments as a proxy for local net creation of businesses. As shown in column 1 in Supplementary Material Table A6, the Opportunity Zone policy had a negative net effect on the creation of local businesses, confirming the previous finding using OpenCorporates data. The ZBP data also provide the number of employees of established businesses. The decrease in the net creation of local establishments was mainly among smaller firms (with fewer than 10 employees or with 10–50 employees) but not in large ones (with more than 50 employees).

C. Identification Assumptions and Robustness Tests

The DiD approach compares outcome variables before and after the policy between designated census tracts and eligible but non-designated tracts. This identification strategy relies on two main assumptions. First, it assumes that changes in private investments and entrepreneurship would have been the same across the treated and control areas, absent the Opportunity Zone policy change (i.e., the parallel trend assumption). Second, it assumes that the Opportunity Zone policy was not determined based on the level and growth of local private investments and new business formation before the policy.

I take several steps to verify these assumptions and address potential concerns. One concern is that the selection of Opportunity Zones was not random and that outcome variables such as private investments may have evolved differently between designated and non-designated tracts in the absence of the policy. To address this concern, I plot the coefficient estimates around the introduction of the Opportunity Zone policy at the end of 2017. As shown in Figure 2, local private investments and entrepreneurship did not diverge before the policy’s introduction, as the 95% confidence intervals all cover 0.Footnote ²⁵ The difference between the treated and control tracts started to enlarge significantly only after the policy was introduced. The figure provides support for the parallel trend assumption required by the DiD approach.

FIGURE 2Semi-annual Coefficient Estimates Plotting

Figure 2 shows the coefficient plot around the Opportunity Zone policy by estimating the following model:

$$ {Y}_{i,t}=\alpha +\sum \limits_{t=2015,t\ne 2017H2}^{2019}{\beta}_t{OZ}_i\ast {Semester}_t+{OZ}_i+{Controls}_i+{\delta}_t+{\eta}_c+{\unicode{x025B}}_{i,t}, $$

where $ {Semester}_t $ is a set of indicator variables that equals 1 in semester $ t $ . The benchmark group comprises observations from the second semester of 2017 (2017H2) when the Opportunity Zone policy was signed into law. $ {OZ}_i $ is a dummy that equals 1 if the census tract $ i $ was designated as an Opportunity Zone and equals 0 if the tract was eligible but not selected. Graph A shows the plot of coefficient estimates of $ {\beta}_t $ when the outcome variable is the natural logarithm of 1 plus the number of private investments. The outcome variable in Graph B is the natural logarithm of 1 plus the amount of private investments. The outcome variable in Graph C is the natural logarithm of 1 plus the number of new for-profit firms. The center points show the point estimates of $ {\beta}_t $ and the vertical lines denote the 95% confidence intervals of $ {\beta}_t $ estimates. Control variables are specified as in the baseline regressions in Table 2. Census tract fixed effects and semester fixed effects are included. Standard errors are clustered by census tracts.

FIGURE 2 Long description

A multi-panel figure containing three coefficient plots, each with a Y-axis labeled beta sub t and an X-axis showing semi-annual periods from 2015 to 2019 H 2. A vertical dashed line marks the policy implementation at 2017 H 2, where the coefficient is normalized to zero.

* Graph A, Number of Private Investments: Before 2017 H 2, coefficients are near zero or slightly negative. After the policy, there is a clear upward trend, with point estimates rising from approximately 0.004 in 2018 H 1 to 0.008 in 2019 H 1, before a slight dip in 2019 H 2. Confidence intervals for post-policy periods mostly stay above the zero line.

* Graph B, Amount of Private Investments: Similar to Graph A, the pre-policy period shows coefficients near zero. Post-policy, the estimates increase significantly, peaking at approximately 0.14 in 2019 H 1. The 95 percent confidence intervals are wider than in Graph A but show a positive shift after 2017 H 2.

* Graph C, Number of New Firms: This plot shows a different trend. Pre-policy estimates fluctuate around zero. Post-policy, the coefficients drop into negative territory, with point estimates between negative 0.02 and negative 0.03 for 2018 and 2019, suggesting a decrease in new firm creation relative to the benchmark.

Another concern is related to the differences in characteristics between the designated and non-designated tracts.

Even though the DiD approach requires only pre-trends to be similar instead of the levels, some may still worry that non-balanced covariates may threaten the parallel trend assumption. To address this concern, I include tract-level lagged-one-year control variables, including population, median income, median age, percentage of white or black population, poverty rate, and unemployment rate in the baseline regressions.

I also perform a PSM on pre-treatment characteristics and keep the control tract with the highest propensity score within the same county as the treated tract (an Opportunity Zone) as its matched control. I add state dummies to control for the political affiliation of state governors and their potential impact on the selection of tracts. To further address the concern that the variation in political ideology drove the difference in private investments between treated and control, Supplementary Material Table A8 shows the sub-sample tests of splitting the sample census tracts by whether the Democratic or Republican party won their located states in the 2016 presidential election, and no significantly different impact of the policy was found between the two groups of tracts.

Column 1 of Supplementary Material Table A9 presents the results of running the Logit regression in the full sample to produce the propensity scores. The results show that the policy was successful in general at targeting areas in need: census tracts with lower income, higher poverty, and a greater percentage of black people were more likely to become Opportunity Zones, alleviating the concern that the increase in private investments was driven by governors’ favoritism toward areas that had already been gentrified.Footnote ²⁶ Column 2 shows the Logit regression results using the matched sample, and one sees that the observation numbers decreased to 15,210.Footnote ²⁷ The independent variables lost significance in the matched sample, and the pseudo $ {R}^2 $ decreased from 0.0558 to 0.0011, suggesting that the pre-shock characteristics are comparable in the matched sample and not likely to explain the selection of the Opportunity Zone in the matched sample. Supplementary Material Table A10 shows the DiD results with a sub-sample of matched pairs from PSM, and the main findings are robust in the matched sample.

Some critics of the policy argue that certain census tracts designated as Opportunity Zones are close to colleges and universities where students are considered low-income residents. Thus, I conduct a robustness test by excluding census tracts with colleges and universities. The campus address and population information are collected from Homeland Infrastructure Foundation-Level Data. The results in Supplementary Material Table A11 show that the previous results are not driven purely by treated census tracts that have or are close to a campus of colleges and universities, thereby alleviating the concern that the economic characteristics of college towns and students might affect the level of investment in Opportunity Zones.

Concerns may arise regarding whether private investments were influenced by other tax reform measures enacted simultaneously with the TCJA, particularly the tax deductions for multinational corporations (MNCs). However, it is unlikely that other provisions of the TCJA would confound the results presented in this article. The primary benefit of Opportunity Zones stems from capital gains tax relief for investors, whereas other TCJA provisions target corporate expenses and profits, which are less likely to directly affect distressed areas like Opportunity Zones. Moreover, as demonstrated by Albertus, Glover, and Levine (Reference Albertus, Glover and Levine2025), there was no significant response in U.S. firms’ investment and employment due to increased access to inexpensive capital, further mitigating concerns about confounding effects in this context. To alleviate the specific concern about tax deductions for MNCs, I conduct a robustness test by excluding census tracts containing headquarters or at least 20 subsidiaries of any MNC. A firm is classified as multinational in a given year if its pretax foreign income (PIFO) or foreign tax expenses (TXFO) are non-missing and greater than 0, following the approach in Lampenius, Shevlin, and Stenzel (Reference Lampenius, Shevlin and Stenzel2021). The coefficient estimates on OZ*Post in Supplementary Material Table A12 remain statistically significant, suggesting that the original findings were not substantially impacted by tax reform provisions of the TCJA specifically benefiting MNCs.

Finally, I examine whether the distance from a control tract to a treated tract affects the results. The analysis aims to determine whether the policy had any spillover effects on local investments and entrepreneurship in nearby control tracts, which could potentially bias the estimation. I split the control tracts, eligible but non-designated as OZs, based on whether there is an Opportunity Zone within 3 km and rerun the baseline regressions. Supplementary Material Table A13 shows that the baseline results are robust regardless of whether benchmarking against control groups closer to a treated census tract or not. This suggests little spillover effect on local private investments or entrepreneurship.

The previous analyses confirm that the Opportunity Zone policy is a plausible quasi-natural shock to local private investments: The policy was effective in introducing private investments to treated areas that otherwise would not have happened. Some may question why the effect showed up quickly after the policy implementation and the importance of the effect if it was only short term. Two points worth discussing when interpreting this result. First, the immediate decrease in new firms in 2018 should not be surprising because the law requires investors to reinvest their capital gains within 180 days after realization. Second, the negative effect, which was greater in 2018 compared to 2019, may be due to the timeline of the tax benefits offered: the latest time to max out all the tax benefits provided by the policy is 2018. Without the pandemic in 2020, the negative impact may have faded over the years, given that investors would have had more time to do due diligence on their investment targets. In this case, their preference for older businesses, as discussed in the following section, might diminish. Although the policy’s negative effect on local new business formation might be relatively short term, decreases in local business formation could significantly impact local market competition and employment, as I will show in the following sections of the article.

A. Reduced Employment and Sales

Table 5 presents the result of examining how the Opportunity Zone policy affects local employment. Panel A of column 1 shows that local employment decreased by 1.7% in the treated areas after the policy and the effect is statistically significant at the 1% level. When breaking down by firm age (i.e., newly formed or more than 1 year) and ownership type (i.e., whether it is a standalone firm or a subsidiary of the parent company), the total employment by Age ≤ 1&Standalone firms decreased significantly by 4.1% and the other type of firms also decreased significantly by 1.6%. However, the total employment for Age >1&Subsidiary firms that have the most information available to investors and the lowest risk did not change significantly. When replacing the dependent variable with sales generated by local firms, I observe similar findings that the total sales also declined significantly. The only types of firms that survived this decrease were those older than 1 year and subsidiaries.Footnote ²⁸

TABLE 5Impact on Local Employment and Sales

TABLE 5 Long description

The table is divided into two vertical sections.

Panel A: L n (Employment plus 1).

This panel measures the impact on jobs across four columns:

1. Total: O Z asterisk Post coefficient is minus 0.017 with 1 percent significance level.

2. Age greater than 1 and Subsidiary: coefficient is minus 0.004, not significant.

3. Age less than or equal to 1 and Standalone: coefficient is minus 0.041 with 1 percent significance level.

4. Others: coefficient is minus 0.016 with 1 percent significance level.

All columns have 154,476 observations and R-squared values ranging from 0.827 to 0.995.

Panel B: L n (Sales plus 1).

This panel measures the impact on sales across the same four columns:

1. Total: O Z asterisk Post coefficient is minus 0.028 with 1 percent significance level.

2. Age greater than 1 and Subsidiary: coefficient is minus 0.008, not significant.

3. Age less than or equal to 1 and Standalone: coefficient is minus 0.066 with 1 percent significance level.

4. Others: coefficient is minus 0.031 with 1 percent significance level.

All columns have 154,476 observations and R-squared values ranging from 0.670 to 0.983.

Both panels include Controls, Tract F E, and Year F E. Standard errors are provided in parentheses below each coefficient.

To validate that the decline in new business formation is the main driver of the observed decrease in aggregate employment, I first examine whether there is indeed a large gap between new firms and existing firms in job creation in Opportunity Zones. New entrants create, on average, 11,827 jobs per firm, whereas incumbents create only 0.048 jobs per firm. This confirms that new firms are the primary engine of local job growth in these distressed neighborhoods. Using these statistics, I then perform a back-of-the-envelope calculation to test the mechanism. Given that new firms constitute approximately 8% of total local establishments, and the policy reduced new business formation by 1.8% (based on the coefficient of −0.018 in Table 2), the predicted impact on aggregate employment driven by this decline in new firms is approximately 0.08 $ \times $ (−0.018) $ \times $ 11.827 $ \approx $ −1.703%. This calculated magnitude matches the actual aggregate employment decline of 1.7% observed in Panel A of Table 5 (the coefficient on $ OZ\ast Post $ being −0.017). This result supports the hypothesis that the aggregate employment loss was driven by the “missing generation” of high-growth new firms, which incumbents failed to replace.

To explain why incumbents failed to offset this job loss, I draw on studies showing that these established firms directed tax-advantaged capital toward non-employment-related assets and property renovations rather than expanding their workforce. Specifically, Freedman et al. (Reference Freedman, Khanna and Neumark2021) find that the policy’s impact on the average earnings of local residents is statistically indistinguishable from 0, while Sage et al. (Reference Sage, Langen and Van de Minne2023) document an increase in local commercial property prices for assets requiring renovation in the treated areas. Combining these findings with the negative employment results from my analysis suggests that the tax-induced capital inflows were likely spent toward physical assets rather than labor.Footnote ²⁹

To further validate that the decline in entrepreneurship and employment is driven by incumbent crowding-out in locally competitive markets, I examine whether the policy’s impact varies based on the region’s industrial composition. If the negative real effects indeed stem from incumbents edging out new firms through local competition, one would expect these effects to be mitigated in regions dominated by tradable sectors, where demand is not constrained by the local market and firms do not compete locally.

Table 6 reports the results of estimating the baseline DiD model in two subsamples: census tracts where the tradable sector is the largest employer sector and accounts for more than 20% of the workforce (“Tradable-Dominant”), and those where the tradable sector is not the dominant sector. As shown in columns 1 and 3, in tradable-dominant tracts, the coefficients for both new business formation and employment are statistically indistinguishable from 0. On the contrary, the negative and statistically significant impact on both new business formation and total employment is concentrated mainly in tracts where the tradable sector is not dominant. This analysis confirms that the crowding-out of entrepreneurship and employment is specific to markets where firms compete locally.

TABLE 6Heterogeneous Effects by Tradable Sector Dominance

TABLE 6 Long description

The table consists of four columns of regression results. Columns 1 and 2 use the natural logarithm of the number of new firms plus 1 as the dependent variable. Columns 3 and 4 use the natural logarithm of employment plus 1. The columns are further divided by whether the census tract is Tradable-Dominant (Yes for columns 1 and 3; No for columns 2 and 4).

* Row 1: OZ asterisk Post interaction term.

- Column 1 (Tradable-Dominant: Yes): 0.012 with a standard error of 0.032.

- Column 2 (Tradable-Dominant: No): negative 0.020 with a standard error of 0.010, significant at the 5 percent level.

- Column 3 (Tradable-Dominant: Yes): negative 0.006 with a standard error of 0.004.

- Column 4 (Tradable-Dominant: No): negative 0.017 with a standard error of 0.001, significant at the 1 percent level.

* Row 2: Number of observations.

- Column 1: 10,818.

- Column 2: 136,749.

- Column 3: 11,493.

- Column 4: 142,972.

* Row 3: R-squared values.

- Column 1: 0.847.

- Column 2: 0.849.

- Column 3: 0.995.

- Column 4: 0.995.

* Row 4 and 5: Tract F E and Year F E are marked as Yes for all four columns.

B. Decreased Market Competition and Exits for Incumbents

Next, I examine whether and how market competition and firm exits changed after the policy’s implementation.

First, I study the Opportunity Zone policy’s impact on local market competition. In Table 7, Panel A, the dependent variables are the concentration ratios of the top 3 and top 5 employers in a census tract in a year (CR3 and CR5) as well as the HHI based on employment and use them as dependent variables in different sectors.Footnote ³⁰ One would expect to see that market competition would decrease significantly in the non-tradable sector, where most of the decline in business formation is observed, and see little impact in the tradable sector. This is the case shown in Table 7: The coefficient estimates on $ OZ\ast Post $ are all positive and significant at a 1% significance level for the non-tradable sector, and I do not observe significance in the tradable sector. In Panel B, I use sales instead of employment to construct the previous three measures. The results are similar to those in Panel A.

TABLE 7Impact on Local Competition

TABLE 7 Long description

Panel A. Local Market Competition in Employment.

For the Non-Tradable sector, the O Z asterisk Post coefficient is 0.378 for C R 3, 0.289 for C R 5, and 30.957 for H H I, all significant at the 1 percent level. Standard errors are 0.096, 0.084, and 8.423 respectively.

For the Tradable sector, coefficients are 0.053 for C R 3, 0.008 for C R 5, and 17.060 for H H I, none of which are statistically significant.

Number of observations ranges from 140,231 to 153,709 with R-squared values between 0.890 and 0.954.

Panel B. Local Market Competition in Sales.

For the Non-Tradable sector, the O Z asterisk Post coefficient is 0.262 for C R 3 significant at 5 percent, 0.250 for C R 5 significant at 5 percent, and 23.415 for H H I significant at 10 percent. Standard errors are 0.128, 0.098, and 12.585 respectively.

For the Tradable sector, coefficients are negative 0.055 for C R 3, negative 0.016 for C R 5, and 11.782 for H H I, none of which are statistically significant.

Number of observations ranges from 140,231 to 153,709 with R-squared values between 0.835 and 0.945.

Both panels include Controls, Tract F E, and Year F E.

Second, I examine whether the policy affected incumbents’ dissolution (i.e., firm exit). The logic is that increased private investments in incumbents may help their survival. In Table 8, the dependent variable is the natural logarithm of the number of firms that dissolve in a census tract each year. In column 1, the coefficient on OZ*Post is negative and insignificant, suggesting that the total number of companies going out of business did not change significantly. However, as shown in columns 2 and 3, there was a significant decrease in the number of firms that were at least 1 or 2 years old when they dissolved. The results suggest that older incumbents in the treated areas were more likely to survive (less likely to go out of business) after implementing the policy. The finding is consistent with the hypothesis that the additional financing resources, brought in by the tax incentives, helped incumbents build up their competitive advantage over potential newcomers and reduced local competition.

TABLE 8Impact on Existing Firms’ Dissolution by Firm Age

TABLE 8 Long description

The table consists of three columns under the primary dependent variable L n (Num_Dissolved plus 1).

Column 1: All Dissolved. The O Z times Post coefficient is minus 0.006 with a standard error of 0.004. R-squared is 0.510.

Column 2: Firm Age greater than 1 Year. The O Z times Post coefficient is minus 0.007 with a significance level of 10 percent and a standard error of 0.004. R-squared is 0.483.

Column 3: Firm Age greater than 2 Years. The O Z times Post coefficient is minus 0.014 with a significance level of 1 percent and a standard error of 0.003. R-squared is 0.437.

All three columns have 147,565 observations and include Controls, Tract F E, and Year F E. Standard errors are clustered at the census tract level.

Overall, the results in this section show that the Opportunity Zone policy has negatively impacted local employment and market competition. The negative economic effects highlight the importance of the previous findings in the article that the Opportunity Zone policy negatively impacted the formation of local businesses.Footnote ³¹

C. Stylized Model: Place-Based Tax Incentives, Business Formation, and Employment

To understand why tax-induced investments negatively affect local business formation and employment, I present a stylized framework. Rather than attempting to capture every institutional detail of the Opportunity Zone program, the model isolates a specific economic channel, the tension between funding established incumbents versus riskier new ventures. While this framework provides useful intuition for the observed displacement effects, I acknowledge it may not be the only mechanism at work. Other institutional factors, such as the program’s short qualification window, likely operated alongside this channel to discourage firm entry. The full model and proofs are in Supplementary Material Section B. A brief summary of the model predictions is discussed later.Footnote ³²

The model explores how an investor (i.e., a QOF) allocates capital between existing firms and new firms within an Opportunity Zone. Most of the firms located in these low-income and high-poverty neighborhoods are private companies with severe information asymmetry between owners and investors (Leland and Pyle (Reference Leland and Pyle1977)) due to limited availability of information and uncertainty (Ivković and Weisbenner (Reference Ivković and Weisbenner2005), Seasholes and Zhu (Reference Seasholes and Zhu2010)). One important observable factor related to the risk of failure that investors would consider is the company’s age (Dunne, Roberts, and Samuelson (Reference Dunne, Roberts and Samuelson1989)). The investor faces a trade-off between the safety of investing in established companies, which guarantees a fixed return, and the higher but riskier potential returns from investing in new firms conditional on success. The model incorporates two key features to mirror the conditions in Opportunity Zones. First, the expected profitability of new ventures being successful is constrained by a scarcity of high-quality entrepreneurial projects. As more capital is invested in the distressed tracts, the marginal probability of success diminishes, reflecting a limited supply of productive local opportunities. Second, local competition from better-financed existing firms further reduces the likelihood of success for new firms, with the effect being particularly pronounced in non-tradable sectors, where local competition is more intense.

The model shows that the investor’s optimal investment in new firms depends on the size of the available fund (K). When the fund size is small, the investor directs almost all capital toward new firms because the marginal success probability remains high. However, as the fund size grows, competition between firms intensifies, and the optimal investment in new firms decreases. If the fund becomes too large, the investor eventually allocates all the capital to existing firms, as the competitive pressure makes investments in new firms unprofitable. The model predicts that in sectors with higher local competition, such as restaurants and grocery stores in the non-tradable sector, the Opportunity Zone policy can have a negative impact on new business formation.

The model also suggests that employment outcomes are tied to the allocation of capital between existing and new firms. New firms are generally more effective in creating jobs than established firms and if the gap in job creation potential between new and old firms is large enough, an increase in the fund size could lead to a decrease in total employment. This occurs because the investor increasingly favors established firms, which contribute less to employment, as the fund grows. Thus, the model predicts that the Opportunity Zone policy could reduce employment if new firms’ job creation advantage is sufficiently large.