Skip to main content


Log in

Evaluating the effects of Small Business Administration lending on growth

  • Published:
Small Business Economics Aims and scope Submit manuscript


Conventional wisdom suggests that small businesses are innovative engines of Schumpetarian growth. However, as small businesses, they are likely to face credit rationing in financial markets. If true, then policies that promote lending to small businesses may yield substantial economy-wide returns. We examine the relationship between Small Business Administration (SBA) lending and local economic growth using a spatial econometric framework and a sample of U.S. counties. We find evidence that a county’s SBA lending per capita is associated with direct negative effects on its income growth. We also find evidence of indirect negative effects on the growth rates of neighboring counties. Overall, a 10% increase in SBA loans per capita is associated with a cumulative decrease in income growth rates of about 0.02 to 0.03 percentage points.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2

Similar content being viewed by others



  2. de Rugy (2005) provides a concise overview of these government programs.

  3. This is from the SBA mission statement:

  4. It is important to note that while the SBA provides guarantees for these loans, the actual loan itself is underwritten by a private bank.

  5. See also: Blau (1987), Evans and Leighton (1989), Audretsch and Vivarelli (1995), Taylor (1996), Fairlie and Krashinsky (2012), Thurik et al. (2008), and Shane (2009).

  6. In developing economies, La Porta and Shleifer (2008) and Banerjee and Duflo (2011) both report evidence that most small businesses are started due to a lack of jobs at larger firms. These small businesses most often neither innovate nor grow.


  8. We also include various combinations of period (time), U.S. state, and county fixed effects.

  9. As described in detail in Section 2, we relate average SBA variable values over 5-year periods to average growth over subsequent 5-year periods.

  10. Though not statistically different than 0, the slope point estimate on an OLS best-fit line is negative.

  11. Craig et al. (2008) provide a similar study where local area employment rates are the dependent variable.

  12. Aside from SBA variables and time period dummies, Craig et al. only control for per capita income level and a measure of concentration in the deposit market.

  13. As we discuss below, our dependent variable is measured over a rolling 5-year window (t to t + 5), while our key independent variables are lagged over the previous 5-year window (t − 5 to t − 1). This latent structure helps us avoid issues of reverse causation. Moreover, if reverse causality were present, our priors are that the estimates would be biased upwards. As we report, however, our main effects are negative and significant.

  14. In order to address the possible presence of serial correlation, we also run models with the inclusion of a lagged dependent variable; our results remain robust.


  16. We always add 1 to the SBA loan amount before taking the log since there are observations of loans per capita that are equal to 0.

  17. Strictly speaking, we should take the rate differential times the amount of loans and then take the log of that product. However, the differential is not bounded above 0.

  18. While we only use SBA failure rates as a control, other work (e.g., Glennon and Nigro 2005) has focused on the hazard of SBA loan failure. Heretofore, one major limitation with studying failure rates directly is the limited firm-specific information available via the SBA. These constraints are slowly being removed with the availability of data (via special access) at Census and BEA. Future research could attempt to merge these proprietary firm-level data into this publicly available SBA loan data to explore more detailed correlates of failure.

  19. Following Samila and Sorenson (2011) for patent applications listing a number (n) of inventors, 1/n patents are assigned to the county of each individual inventor. This is described more fully in the Samila and Sorenson (2009) working paper.

  20. As we show in Table 1, there are 17 variables that come from decennial Census data. This creates a missing data issue for the midpoints of these variables in between each decennial census. As such, we follow Higgins et al. (2006) and Young et al. (2013) and use linear interpolation (Dezhbakhsh and Levy 1994) to generate the missing midpoint variables.

  21. Intuitively, consider the example of three counties (“A,” “B,” and “C”) that lay in succession along a line: A, then B, and then C; and consider that growth rates are spatially dependent on neighboring growth rates. We want to use a SAR model to estimate that spatial dependence. We could consider a 1-closest neighbor weighting scheme. In that case, A would affect B, which would affect C; and then there would be feedback effects. Alternatively, we could consider a two-closest neighbor scheme where A affects both B and C; and then there are feedback effects. In either case, a particular spatial autoregressive root (ρ) that is less than 1 in absolute value will imply a finite cumulative effect. If there is a true cumulative effect, then there will be a root (ρ1) such that a one-closest neighbor SAR specification is consistent with it; and there will be another root (ρ2) such that a two-closest neighbor SAR specification is consistent with it. From either specification, then, we can in principle estimate that true cumulative effect.

  22. In addition, Bramoullé et al. (2009) show that the endogenous and exogenous peer effects are identified in spatial econometric models under easily verifiable conditions.

  23. Full OLS results are provided in Appendix A and full SDM results are provided in Appendix B. Appendices are available upon request.


  25. These regressions are unreported in the text for space consideration, but the full regression results are available upon request.

  26. The interpretation of the rate differential coefficient can be confusing given that the average value of the prime rate net of the SBA rate in our sample is negative (− 0.009). A negative coefficient on the rate differential may suggest a positive growth effect at that average value. However, the marginal growth effect is negative. For example, starting from the average (− 0.009), a decrease in the SBA rate lowers the rate differential in absolute value, implying that the marginal effect on growth is negative when the coefficient is negative.

  27. Firm-level evidence also links venture capital to positive economic outcomes. For examples, Jain and Kini (1995) and Engel and Keilbach (2007) find that venture capital-funded firms experience higher sales growth and employment; Kortum and Lerner (2000) find that manufacturing firms that receive venture capital have higher patenting rates. However, Gompers and Lerner (2003) also find that during boom periods, venture capital tends to overfund particular sectors and it becomes less effective.

  28. With the inclusion of county fixed effects, we drop the 27 additional control variables because their inclusion results in a near singular design matrix.

  29. Results also remain robust to the inclusion of clustered standard errors. We argue that models 4, 5 and 6, with the inclusion of county-level fixed effects, alleviate any potential concerns over a county-level omitted variable bias.

  30. Initial income is, of course, dropped from the control variable set when we employ the income level as the dependent variable.

  31. This issue can be more clearly understand with the following example. Vietnam has a lower level of patenting and a higher growth rate than the USA. Putting both in a regression in this example would lead to precisely the same result. Our data show a similar pattern. For example, counties in the top 5% in terms of per capita income growth have patenting levels in the lowest 10%. Likewise, counties in the top 5% in terms of patenting have below mean (and median) levels of per capita income growth. As such, we maintain that our level specifications are consistent with Samila and Sorenson (2011).

  32. As we shall see, this is also true for the analogous SDM estimation (see Table 5C). We will return to its discussion and interpretation below.

  33. In addition to the panel of 3-year periods, we also considered a panel of 2-year periods, which we report in Appendix Table C. Across all three models, the coefficients on SBA lending (i.e., direct, indirect, or total) are not significant.

  34. In a panel of 2-year averages, the coefficient on (log) SBA loans per capita was not statistically significant in any specification.


Download references


We thank Jerry Thursby, Bart Hamilton, Daniel Levy, Jeremy Watson, Marco Ceccagnoli, Alex Oettl, and seminar participants at Mississippi State University for valuable comments and discussions. We are grateful to Alexander Kritikos (editor) and two anonymous referees for their helpful comments and direction. We acknowledge the Small Business Administration for their help in our FOIA request. Authors are listed alphabetically and contributed equally.


Higgins acknowledges the funding from the Sorenson Center for Discovery and Innovation.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Briana S. Stenard.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material


(DOCX 38 kb)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Higgins, M.J., Lacombe, D.J., Stenard, B.S. et al. Evaluating the effects of Small Business Administration lending on growth. Small Bus Econ 57, 23–45 (2021).

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


JEL codes