Chunyu Qu, Legacy Lending Relationships and Credit Rationing: Evidence from the Paycheck Protection Program (2025)
Summary
A research paper by Chunyu Qu of Fordham University and Dun & Bradstreet Inc., Legacy Lending Relationships and Credit Rationing: Evidence from the Paycheck Protection Program, dated Dec 24, 2025. Linking Small Business Administration loan records with Dun & Bradstreet microdata for over 26 million U.S. firms and applying entropy balancing, the study tests whether prior participation in the SBA 7(a) program served as a gateway to the $800 billion Paycheck Protection Program. It finds that during the initial April 2020 phase, firms with prior 7(a) relationships were approximately 29 percentage points more likely to receive PPP funding than comparable non-7(a) firms, whereas by June 2021 that advantage had largely disappeared.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“Chunyu Qu, Legacy Lending Relationships and Credit Rationing: Evidence from the Paycheck Protection Program (2025).”
Original at arxiv.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/chunyu-qu-legacy-lending-relationships-and-credit-rationing-evidence-f-54ac31aab3/
, SHA-256 ec5684e4abc794d6c02ed99c2306a70e0556226f0a86fd4c59ae80497c33bc55.Full text
Legacy Lending Relationships and Credit Rationing: Evidence
from the Paycheck Protection Program
Chunyu Qu
Dec 24, 2025
Abstract
This article examines how legacy lending relationships shape the allocation of emergency credit under severe
information frictions. Using a novel dataset linking Small Business Administration (SBA) loan records with Dun &
Bradstreet microdata for over 26 million U.S. firms, I investigate whether prior participation in the SBA 7(a)
program acted as a gateway to the Paycheck Protection Program (PPP). Employing entropy balancing to construct a
strictly comparable counterfactual group, I document a distinct dynamic evolution in credit rationing. In the
program’s initial "panic phase" (April 2020), banks relied heavily on legacy ties as a screening technology: firms
with prior 7(a) relationships were approximately 29 percentage points more likely to receive funding than
observationally identical non-7(a) firms. By June 2021, however, this "insider advantage" had largely vanished,
suggesting that policy adjustments and extended timelines eventually mitigated the initial intermediation frictions.
These findings highlight a fundamental trade-off between speed and equity in crisis response: while leveraging
existing credit rails accelerates deployment, it systematically excludes informationally opaque borrowers. I discuss
policy implications for designing future digital infrastructure to decouple verification from historical lending
relationships.
JEL Codes: H25; H81; G28; L26; C25
Key words: Paycheck Protection Program, Small Business Finance, Credit Access, SBA 7(a)
Loan, Financial Stress, COVID-19
Chunyu Qu holds a Ph.D. in Economics from Fordham University and is currently a Data Scientist with Dun &
Bradstreet Inc (D&B), 2900 Esperanza Crossing, Austin, TX 78758 (email; cqu9@fordham.edu).
The author thanks Sophie Mitra for her major feedback throughout the research. He also thanks Philip Shaw,
Subha Mani, Andrew Simons, Patricia Gomez-Gonzalez, and Johanna L. Francis from Fordham University for their
constructive feedback. Special thanks go to Yunbo Liu, Assistant Professor at CUHK Business School, for his
contributions to the modeling. The author further acknowledges Amber Jaycocks, Nalanda Matia, Julia Zhao,
Michelle Solomon, and Andrew Bynes at Dun & Bradstreet for their support, and Tuhin Batra, formerly of D&B and
now with Atlantic Technological University, Ireland, for their technical insights and helpful discussions. The author
acknowledges Qi He of Google LLC for her insightful input on the comparative assessment and selection of financial
metrics. Her expertise in designing augmented metric frameworks and in mechanisms optimization provided
valuable analytical choices in this study. All views expressed in this article are solely of the author and do not
represent the views of D&B or any affiliated institution.
1
Introduction
The Paycheck Protection Program (PPP) represented the largest fiscal intervention in U.S. history, channeling
over $800 billion in forgivable loans to small businesses. While the program’s scale was unprecedented, its
implementation faced a classic economic dilemma: how to rapidly allocate liquidity in a crisis when lenders face
severe information frictions regarding borrower viability. In the absence of perfect information, banks often rely on
"relationship lending" as a screening technology. This article examines how legacy lending relationships,
specifically prior participation in the Small Business Administration’s (SBA) 7(a) program, acted as a decisive, yet
unequal, gateway to emergency credit.
Existing literature on the PPP has largely focused on the supply side, documenting how bank branch networks
(Granja et al. 2022), lender technology (Howell et al. 2024), and banking relationships (Li and Strahan 2021) shaped
aggregate allocation. While these studies highlight geographic and institutional disparities, they often view the
problem through the lens of the lender. Much less is known about the firm-level selection mechanisms: specifically,
how a firm’s pre-crisis "institutional resume", its history of government-backed borrowing and verified
creditworthiness, determined its survival in the race for funds.
This article fills this gap by constructing a novel, near-universe dataset linking SBA records to microdata from
Dun & Bradstreet (D&B). I merge PPP loan records with historical SBA 7(a) data and D&B credit files for over 26
million U.S. firms. This granularity allows for a precise examination of how legacy government credit relationships
interact with firm fundamentals, such as commercial credit and financial stress, to determine credit access.
Distinguishing this analysis from prior work, I employ entropy balancing, a rigorous reweighting method that
improves upon standard propensity score matching, to estimate the causal effect of prior 7(a) participation. This
approach ensures that the estimated advantages of legacy borrowers are not merely artifacts of their size or
observable quality, but reflect the specific value of established institutional ties.
The empirical results reveal a striking dynamic evolution in credit allocation, highlighting a "learning curve" in
crisis response. First, in the chaotic early phase (April 2020), PPP access was heavily skewed toward "insiders."
Firms with prior 7(a) relationships were approximately 29 percentage points more likely to receive funding than
observationally identical non-7(a) firms. During this period, banks effectively used 7(a) history and high credit
scores as a proxy for verification speed, rationing credit away from opaque, albeit viable, borrowers.
Second, this screening mechanism relaxed significantly over time. By June 2021, as the program matured and
rules were adjusted, the predictive power of legacy ties and credit scores largely vanished. Late-stage recipients
resembled the general population of small businesses, suggesting that policy adjustments eventually mitigated the
initial information frictions.
This study makes two primary contributions to the literature on small business finance and crisis management.
First, it quantifies the "program-layering premium." I show that participation in peacetime guarantee programs (like
7(a)) creates an embedded advantage for accessing future emergency aid, a mechanism that race-neutral policies
may inadvertently reinforce. Second, it provides a firm-level dynamic assessment of the trade-off between speed and
equity. While leveraging legacy relationships expedited initial disbursements, it created significant access inequality
that was only corrected in later phases. These findings suggest that future emergency lending infrastructures must
move beyond reliance on legacy banking lists toward more inclusive, pre-qualified digital verification systems.
2
Section II. Literature Review
A rapidly growing literature examines how PPP funds were allocated and what effects they had on firms and
local economies. Autor et al. (2022b) provide an overview of the program’s design and aggregate distribution,
showing that a large share of subsidies accrued to business owners and high-income households rather than directly
preserving jobs. Hubbard and Strain (2021), Dalton (2023) and Autor et al. (2022a) use administrative and payroll
microdata to place PPP in the broader context of pandemic business support, finding modest employment effects
relative to the program’s fiscal cost.
One strand emphasizes bank supply, relationships, and information frictions. Humphries, Neilson, and Ulyssea
(2020) use survey data to document that many small firms, especially those without strong banking ties, faced
confusion about eligibility, difficulty finding participating lenders, and uncertainty around forgiveness, leading to
substantial take-up frictions. Li and Strahan (2021) show that PPP banks with more pre-existing small-business
relationships and greater engagement in commitment lending supplied more PPP loans. Granja et al. (2022) show
that funds flowed toward regions with denser branch networks and better-capitalized banks. Joaquim and Wang
(2022) show that recipiency of a loan substantially improved subsequent financial condition.
A second strand studies distributional disparities and the role of lender technology. Atkins, Cook, and Seamans
(2022) show that Black-owned firms received smaller PPP loans than otherwise similar firms. Chernenko and
Scharfstein (2024) and Chernenko et al. (2023) find that both lower application rates and differential approval
contribute to gaps, with administrative burdens and confusion playing important roles. Howell et al. (2024) show
that fintech and highly automated lenders were more likely to serve minority-owned firms and to extend smaller
loans.
These literatures provide a rich picture of where PPP dollars went and how lender behavior affected targeting,
however, three gaps remain. First, we know relatively little about how legacy program participation, in particular
prior SBA 7(a) borrowing, shaped access to PPP at the firm level, even though 7(a) relationships are a natural policy
lever and a core feature of the U.S. small-business credit infrastructure. Second, existing work rarely traces how the
mapping from firm characteristics to PPP access changed over the life of the program: most analyses focus on early
waves rather than the whole PPP window. Third, few studies combine near-universe firm-level data with modern
reweighting methods to approximate experimental comparisons between firms with and without prior program ties.
This article addresses these gaps by: (i) placing prior SBA 7(a) participation at the center of the analysis as a
predetermined, institutionally meaningful state variable; (ii) documenting how the joint roles of 7(a) ties, credit
quality, financial stress, and firm size in shaping PPP access evolved between April 2020 and June 2021; and (iii)
applying entropy-balancing ATT estimators to a merged D&B–SBA–PPP dataset covering virtually the entire U.S.
employer firm population. In doing so, it complements existing bank- and geography-focused studies with a firm-
level account of how legacy government programs and private credit hierarchies interacted to determine access to
crisis-era support.
Section III. Data and Variables
This study combines firm-level data from Dun & Bradstreet (D&B) with administrative records from the PPP
and SBA’s 7(a) lending program. Detailed information on matching procedures, variable construction, and
3
additional descriptive statistics is provided in Appendix A. To measure prior SBA borrowing experience, I use SBA
7(a) loan data from 2010–2019 and match these records to D&B firms. This yields a binary indicator of whether a
firm had obtained at least one 7(a) loan prior to the pandemic.
A. Outcome variables
I consider two main outcome variables:
PPP loan recipiency. A binary indicator equal to one if a firm is matched to at least one PPP loan by the
reference date and zero otherwise.
PPP loan amount. The total PPP loan amount (in U.S. dollars) received by each matched firm.
B. Treatment and key firm-level covariates
The central treatment variable in the analysis is firms’ prior experience with SBA’s main small-business lending
program:
SBA 7(a) experience. An indicator equal to one if a firm received at least one SBA 7(a) loan between 2010 and
2019 and zero otherwise. Because 7(a) loans predate the pandemic and the PPP rollout, this variable captures long-
standing relationships with SBA-backed lenders and familiarity with SBA procedures.
In addition, I use several pre-PPP variables from D&B as key covariates to summarize firms’ financial
condition and organizational scale:
Commercial Credit Ranking. A percentile ranking from 1 to 100, where higher values indicate a lower predicted
probability of severe delinquency, bankruptcy, or closure over the next 12 months.
Financial Stress Ranking. A percentile ranking from 1 to 100, where higher values indicate a lower predicted
probability that a firm will seek legal relief or cease operations without repaying creditors within the next 12
months.
Firm size. The employment data reported by D&B, updated monthly. I use both the level and its natural
logarithm to allow for nonlinear size effects.
Business age. The number of years since the firm first entered the D&B database, capturing the maturity and
track record of the business.
These variables are predetermined or slow-moving relative to PPP approval decisions and provide a compact
representation of firms’ creditworthiness, failure risk, and scale. Construction details for the D&B risk rankings are
given in Appendix A.2.
C. Secondary demographic covariates and controls
The analysis also incorporates ownership demographics and additional controls. All regressions include
additional controls for organizational form and local economic conditions, such as ownership/lease status, prior
business failure flags, foreign or public status, industry codes, and state fixed effects. These variables help absorb
broad differences in sectoral composition and geographic exposure to the pandemic. Appendix A.3–A.5 provide
formal definitions and additional descriptive statistics for these controls.
Section IV. Empirical Strategy and Identification
The analysis also incorporates ownership demographics and additional controls. All regressions include
additional controls such as ownership/lease status, prior business failure flags, foreign or public status, industry
4
codes, and state fixed effects. These variables help absorb broad differences in sectoral composition and geographic
exposure to the pandemic. Appendix A.3–A.5 provide formal definitions and additional descriptive statistics for
these controls.
A. Baseline models
(1) PPP recipiency
I describe the econometric model used to investigate the association between PPP recipiency and selected
characteristics. The baseline linear probability model is
𝑌𝑖𝑗𝑠 = 𝛽 ′ 𝑍𝑖 + 𝛾 ′ 𝑋𝑖 + 𝛼𝑗 + 𝜌𝑠 + 𝜀𝑖𝑗𝑠 . (1a)
where 𝑍𝑖 can be 7𝑎𝑖 (indicator for having at least one SBA 7(a) loan between 2010–2019), commercial credit and
financial-stress rankings (1–100, higher values = lower risk), ln (𝑓𝑖𝑟𝑚𝑆𝑖𝑧𝑒) (log number of employees), and 𝐴𝑔𝑒𝑖
(firm age). Let 𝑋𝑖 collect additional controls: ownership status, business-failure history, foreign and public status,
county population, log sales, subsidiary indicator, central-city and rural/suburban region dummies, and other
structural characteristics. All specifications include state fixed effects 𝜌𝑠 and broad industry fixed effects 𝛼𝑗 .
Equation (1a) is estimated by OLS with standard errors clustered at the county level. Coefficients on 𝑍𝑖 are
interpreted as percentage-point changes in PPP recipiency associated with a unit change in the corresponding
characteristic, conditional on the full set of controls and fixed effects. For expositional clarity, I also report a set of
specifications where equation (1a) is re-written as
𝑌𝑖𝑗𝑠 = 𝛽𝐷𝑖 + ∑ 𝛾𝑘 𝑋𝑖𝑘 + 𝛼𝑗 + 𝜌𝑠 + 𝜀𝑖𝑗𝑠 , (1b)
𝑘
and the “highlighted” regressor 𝐷𝑖 is rotated across the key interpretive variables, while the remaining firm
characteristics stay in 𝑋𝑖𝑘 . Columns based on (1b) make it easier to compare the marginal association of each
characteristic with PPP recipiency, but the joint specification (1a) is treated as the preferred baseline.
(2) PPP loan amounts
To study the intensive margin among PPP recipients, let LoanAmt𝑖𝑗𝑠 denote the total PPP loan amount received
by firm 𝑖. The baseline specification for loan amounts is the following OLS model in levels:
LoanAmt𝑖𝑗𝑠 = 𝛿 ′ 𝑍𝑖 + 𝜙 ′ 𝑋𝑖 + 𝛼𝑗 + 𝜌𝑠 + 𝑢𝑖𝑗𝑠 . (2a)
δ is a vector of coefficients on the core characteristics, ϕ collects coefficients on the additional controls, the other
terms keep still. Similarly, (1b) turns to be
LoanAmt 𝑖𝑗𝑠 = 𝜃𝐷𝑖 + ∑ 𝜅𝑘 𝑋𝑖𝑘 + 𝛼𝑗 + 𝜌𝑠 + 𝜈𝑖𝑗𝑠 , (2b)
𝑘
with all the terms the same as (1b). (2b) facilitate direct comparison of the marginal association between each firm
characteristic and the dollar level of PPP support, while (2a) remains the preferred model.
B. Identification Strategy
To move closer to a causal effect of 7(a) experience on receiving PPP loans, I adopt a treatment–effects
framework that compares firms with and without 7(a) experience that are otherwise similar in observed pre-PPP
5
characteristics. The goal is to estimate the average effect of 7(a) relationships on PPP access for firms that actually
had such relationships.
Let 𝑇𝑖 = 𝟏{loan7a𝑖 = 1} indicate whether firm 𝑖had at least one SBA 7(a) loan approved between 2010 and
2019, and let 𝑌𝑖 = 1{PPP𝑖 = 1} be an indicator for receiving at least one PPP loan by the reference date. The causal
estimand of interest is the average treatment effect on the treated (ATT),
𝐴𝑇𝑇 = 𝐸[𝑌𝑖 (1) − 𝑌𝑖 (0) ∣ 𝑇𝑖 = 1], (3a)
where 𝑌𝑖 (1) and 𝑌𝑖 (0) denote firm 𝑖’s potential PPP recipiency with and without prior 7(a) borrowing. Identification
relies on two standard assumptions: (i) selection on observables, {𝑌𝑖 (0), 𝑌𝑖 (1)} ⊥ 𝑇𝑖 ∣ 𝑋𝑖 , where 𝑋𝑖 collects f
covariates, and (ii) overlap in the propensity to obtain 7(a) loans, 0 < Pr (𝑇𝑖 = 1 ∣ 𝑋𝑖 ) < 1 for firms in the support
of 𝑋𝑖 .
The covariate vector 𝑋𝑖 includes the core firm characteristics used in the baseline regressions, commercial credit
ranking, financial-stress ranking, log firm size, firm age, and log sales, together with indicators for premise
ownership, prior business-failure history, foreign and public status, metropolitan and rural/suburban location, and
industry and state fixed effects. All covariates are measured prior to the start of the PPP program.
For the main causal estimates, I use entropy balancing to construct a synthetic control group of non-7(a) firms
whose covariate distribution matches that of 7(a) firms. Intuitively, rather than discarding most controls and keeping
only a few nearest neighbors, entropy balancing retains the full sample of non-7(a) firms but assigns them
continuous weights so that their weighted covariate means coincide with those of the treated group.
Formally, let 𝑤𝑖 denote the weight assigned to firm 𝑖. Treated firms (𝑇𝑖 = 1) receive unit weight, 𝑤𝑖 = 1. For
control firms (𝑇𝑖 = 0), entropy balancing chooses non-negative weights 𝑤𝑖 that minimize the Kullback–Leibler
divergence from uniform weights subject to linear balancing constraints on 𝑋𝑖 :
𝑤
min ∑𝑖:𝑇𝑖=0 𝑤𝑖 log ( 𝑖 ) (3b)
𝑤𝑖 : 𝑇𝑖 =0 𝑤
ˉ
subject to
∑𝑖:𝑇𝑖=0 𝑤𝑖 𝑋𝑖 = ∑𝑖:𝑇𝑖=1 𝑋𝑖 , (3c)
and ∑𝑖:𝑇𝑖=0 𝑤𝑖 = 𝑁𝑇 , where 𝑤ˉ is a normalizing constant and 𝑁𝑇 is the number of treated firms. In practice, I impose
balance on the first moments of all covariates listed in Table 4 and verify robustness to also balancing selected
second moments. Panel A of Table 4 shows substantial raw imbalance between 7(a) and non-7(a) firms in credit
rankings, financial-stress rankings, size, age, and ownership structure. Panel B demonstrates that the entropy-
balancing weights reduce all standardized differences to below 5 in absolute value, indicating excellent covariate
balance. Given the entropy-balancing weights, the ATT on PPP recipiency is estimated as a weighted difference in
mean outcomes:
1 1
̂
𝐴𝑇𝑇 𝐸𝐵 = ∑𝑖:𝑇𝑖=1 𝑌𝑖 − ∑𝑖:𝑇𝑖=0 𝑤𝑖 𝑌𝑖 . (3d)
𝑁𝑇 𝑁𝑇
using the entropy-balancing weights for control firms and unit weights for treated firms, and interpret 𝜏̂ as a
regression-adjusted ATT. Standard errors are computed using robust variance estimators with clustering at the
county level. Table 5 reports the resulting ATT estimates for PPP recipiency; they remain large and highly
6
significant and are only moderately smaller than the corresponding baseline LPM coefficients, suggesting that a
substantial 7(a) advantage persists even after aligning the observable characteristics of 7(a) and non-7(a) firms.
C. Robustness checks
As a robustness check for model (1) and (2), I estimate probit and logit models for PPP recipiency and report
average marginal effects. The signs and magnitudes closely mirror the LPM results; see Appendix B.
As a robustness check for causal inference model (3), I also implement nearest-neighbor propensity-score
matching. I first estimate the propensity to obtain a 7(a) loan, Pr (𝑇𝑖 = 1 ∣ 𝑋𝑖 ), using a logit model with the same
covariates 𝑋𝑖 as in the entropy-balancing design. Each treated firm is then matched to the nearest untreated firm in
terms of the estimated propensity score, within a common-support region. The ATT is computed as the average
difference in PPP recipiency between treated firms and their matched controls. The matching-based ATT estimates,
reported in Appendix D, are very close in magnitude to the entropy-balancing estimates, reinforcing the conclusion
that prior 7(a) relationships had a sizeable causal impact on PPP access for otherwise similar firms.
Section V. Results
A. Descriptive Statistics
Tables 1 and B1 describe firm characteristics in April 2020 and June 2021, separately for PPP recipients and
non-recipients. In April 2020, 26.7 million firms are observed, of which about 22.9 million received a first-draw
PPP loan. Recipients are much more likely to have prior 7(a) borrowing, display higher commercial-credit rankings
and lower predicted financial stress, and are larger and older on average than non-recipients. By June 2021, the
sample expands to 29.8 million firms, but only 1.6 million obtain a first-draw loan in that later window, and the gaps
in credit quality, stress rankings, firm size, and age between recipients and non-recipients shrink sharply. These
patterns suggest that early PPP lending was concentrated among better-established, lower-risk firms, while late-stage
recipients more closely resemble the broader small-business population. Moreover, PPP recipiency exhibits an
inverted-U relationship with firm size: mid-sized firms (101–250 employees) have the highest probability of
receiving a loan, whereas very small and very large firms are much less likely to obtain PPP funds (Figure 1).
The descriptive statistics show that early PPP funds flowed disproportionately to larger, older, lower-risk firms
with prior SBA relationships, whereas by mid-2021 the marginal PPP recipient was closer to the typical small
business in the population, both in financial condition and in observables. The regression and matching analyses that
follow quantify these patterns and examine how much of the early advantage associated with 7(a) relationships and
stronger credit profiles persists after conditioning on observables.
Finally, the monthly timing patterns reinforce the layering interpretation and highlight how quickly 7(a) firms
reacted. Table C2 shows that about 50% of all PPP recipients with prior 7(a) borrowing obtained their first PPP loan
in April 2020, compared with roughly 32% of all PPP recipients. Over the full first year, around 69% of 7(a)–PPP
firms entered the program in 2020 versus 55% of all PPP recipients. By contrast, in the 2021 rounds the monthly
share of new PPP loans going to 7(a) firms is consistently below their share in the overall PPP pool. This front-
7
loaded pattern indicates that legacy SBA participants not only had higher approval chances, but also moved into the
program much earlier, when capacity constraints and processing bottlenecks were most severe.
B. Baseline Regressions
Table 2 presents the estimates for the program’s initial phase (April 2020), a period characterized by extreme
liquidity demand and administrative bottlenecks. The results indicate that credit allocation was heavily governed by
pre-existing institutional ties and verifiable creditworthiness. As shown in column (6), having a prior SBA 7(a)
relationship raised the probability of receiving a PPP loan by approximately 31.6 percentage points, conditional on
all controls. This magnitude suggests that during the "panic phase" of the crisis, banks effectively rationed credit by
prioritizing "known" borrowers to minimize processing time and information costs. Similarly, commercial credit
ranking and firm size served as strong positive predictors, reinforcing the view that early allocation followed a
"flight to quality" pattern, where informationally opaque or riskier firms were systematically crowded out.
By June 2021, the allocation landscape had shifted fundamentally. Table 3 reveals that the screening
mechanisms dominant in the early rounds had largely dissipated. The coefficient on 7(a) experience drops to near
zero (and turns slightly negative), implying that the initial advantage of legacy borrowers did not persist into the
program's maturity. Furthermore, the marginal effects of credit rankings and financial stress scores shrank by an
order of magnitude compared to April 2020. This convergence indicates that as the liquidity crisis eased and policy
rules were relaxed to encourage broader participation, the "information rent" associated with legacy banking
relationships was competed away. The distinct inverted-U relationship between firm size and recipiency, visualized
in Figure 2(a), further confirms that while mid-sized firms were served early, the program eventually saturated the
eligible population, reaching the smallest micro-firms and those with weaker credit profiles in later waves.
Comparing the coefficients across the two periods highlights a critical time-varying trade-off in emergency
lending. The sharp decline in the predictive power of 7(a) ties and credit scores (from Table 2 to Table 3) suggests a
rapid relaxation of information frictions. In April 2020, legacy relationships acted as a necessary screening
technology, a "fast track" that facilitated speed but generated inequality. By June 2021, the disappearance of these
effects signals that the combination of fintech entry, policy recalibration, and extended deadlines successfully
mitigated the initial rationing. This temporal heterogeneity challenges static evaluations of the PPP; it suggests that
"misallocation" was largely a feature of the program's initial design for speed, which was subsequently corrected as
the mechanism of delivery evolved from relationship-based triage to broad-based distribution.
Credit quality and failure risk also matter, but on a smaller scale. A one-point increase in the commercial-credit
ranking (on its 1–100 scale) is associated with roughly a 0.10 percentage-point increase in the probability of
receiving a loan, so moving a firm one decile up the national credit distribution raises PPP access by about 1
percentage point. The financial-stress ranking has a similar positive association. Firm size is another key predictor:
in column (6) the coefficient on log firm size implies that a one-log-point increase in firm size (approximately a
170% firm size increase) raises the probability of PPP recipiency by about 4–5 percentage points. Firm age plays a
minor role once size and risk are controlled for.
The “rotated” specifications in columns (2)–(5) confirm that each characteristic is individually powerful: when
entered one at a time, higher credit ranking, lower financial stress, and larger firms are all significantly more likely
8
to obtain PPP loans. Importantly, the magnitudes in column (6) are close to those in columns (1)–(5), suggesting that
multicollinearity is not driving the results and that the large 7(a) effect reflects an incremental advantage even after
conditioning on observable risk and size. Overall, these regressions reinforce the descriptive patterns: in the
program’s first month, PPP loans flowed disproportionately to larger, older, lower-risk firms with existing SBA
relationships.
By June 2021, these patterns change dramatically. In Table 3, column (6), the coefficient on 7(a) experience is
now small and slightly negative (around –0.8 percentage points), indicating that among firms still applying for first-
draw loans in mid-2021, having prior 7(a) borrowing no longer confers an advantage and may even be mildly
associated with lower recipiency. The coefficients on credit and financial-stress rankings remain positive but are an
order of magnitude smaller than in April 2020: a 10-point improvement in either ranking now corresponds to less
than a one-percentage-point increase in PPP access. The effect of firm size nearly vanishes and turns slightly
negative, consistent with the program having largely saturated larger firms earlier.
Taken together, the April 2020 and June 2021 regressions show a clear evolution. Initially, PPP allocation was
heavily shaped by pre-existing SBA relationships, credit quality, and firm size, with 7(a) borrowers enjoying a very
large edge. Over time, as program rules evolved and remaining applicants became more marginal, these advantages
eroded: the predictive power of 7(a) status and risk rankings weakened sharply, and PPP recipiency among new
entrants in 2021 looked much closer to that of the average firm in the economy. The next subsection uses
propensity-score and entropy-balancing methods to test whether the early 7(a) advantage survives when comparing
treated and untreated firms with very similar observable characteristics.
C. Intensive margin - Loan amounts
Tables B4 and B5 report OLS estimates of equation (2a) for the dollar amount of PPP loans among recipient
firms. In April 2020, prior 7(a) participation is associated not only with a higher probability of receiving a PPP loan
but also with substantially larger loans conditional on recipiency. In the full April specification, firms with 7(a)
experience receive on average about $25,000 more in PPP funding than comparable non-7(a) firms, roughly a 20
percent increase relative to the mean first-draw loan. Higher commercial-credit rankings, lower financial stress, and
larger firm size are also associated with larger loan amounts, in line with the program’s payroll-based formula.
By June 2021, this intensive-margin advantage has essentially disappeared. In the June specification, the 7(a)
coefficient is close to zero and statistically insignificant, indicating no meaningful difference in loan size between
7(a) and non-7(a) firms among late-stage recipients. The coefficients on credit rankings, financial stress, and firm
size remain positive but are smaller in magnitude than in 2020, mirroring the convergence in observable
characteristics documented in the descriptive statistics.
D. Causal impact of 7(a) relationships on PPP access
Table 4 documents how firms with prior 7(a) borrowing differ from those without such relationships before any
reweighting. Panel A shows substantial raw imbalance in pre-PPP characteristics. Relative to non-7(a) firms, 7(a)
borrowers have much stronger credit rankings and lower predicted financial stress, are larger and older on average,
and are more likely to operate in metropolitan locations. The standardized mean differences for credit ranking,
9
financial-stress ranking, log firm size, and firm age all exceed 50 in absolute value, with values around 70 for the
two risk scores. These magnitudes indicate strong positive selection of financially healthier, more established firms
into the 7(a) program.
Panel B of Table 4 demonstrates that the entropy-balancing weights are highly successful in aligning the
observable characteristics of non-7(a) firms with those of 7(a) borrowers. After reweighting, the weighted means for
the control group closely track those of the treated group, and all standardized differences fall below 7, with most
below 5 in absolute value. In other words, conditional on the rich set of firm-level controls, the weighted non-7(a)
sample becomes observationally similar to the 7(a) sample along dimensions of credit quality, financial stress, size,
age, ownership, prior failure history, public and foreign status, and location. This provides a transparent check that
the entropy-balancing design delivers excellent covariate balance.
Table 5 reports the resulting average treatment effects on the treated (ATT) of prior 7(a) borrowing on the
probability of receiving a PPP loan, using the full firm population observed in June 2021. The raw difference in
means indicates that 7(a) firms were far more likely to obtain PPP funding: about 60.7 percent of firms with 7(a)
experience received a PPP loan, compared with only about 19.6 percent of firms without such experience, yielding
an unadjusted gap of roughly 41 percentage points. Once industry and state fixed effects and the full set of firm-
level controls are included in the baseline linear probability model, the estimated 7(a) coefficient declines but
remains very large at 0.316 (column “OLS”). The causal estimators based on matching and weighting produce
slightly smaller but tightly clustered effects: nearest-neighbor propensity-score matching yields an ATT of 0.305,
while the entropy-balancing estimator, the preferred specification, implies that prior 7(a) relationships raise the PPP
recipiency probability by about 0.292. Relative to the 19.6 percent PPP take-up rate among comparable non-7(a)
firms, this implies an increase of roughly 150 percent. The stability of the estimated treatment effect across OLS,
matching, and entropy-balancing specifications suggests that, even after accounting for substantial selection on
observable credit quality and firm characteristics, pre-existing 7(a) relationships had a large and economically
meaningful causal impact on firms’ access to PPP funds.
E. Robustness and Sensitivity Results
As a robustness check, I re-estimate the PPP-recipiency models using probit specifications (Tables B2–B3). The
probit coefficients closely mirror the linear-probability results: in April 2020, prior 7(a) experience remains strongly
and positively associated with PPP access, while higher commercial-credit and financial-stress rankings and larger
firm size also significantly raise the likelihood of receiving a loan. By June 2021, the marginal effect of 7(a)
becomes small and negative and the effects of credit, stress, and size all shrink substantially, reproducing the pattern
of weakening selection documented in Tables 2–3. Overall, the probit estimates confirm that the main findings are
not driven by the choice of a linear-probability model.
Descriptive patterns also reveal pronounced heterogeneity in the underlying 7(a) coverage rate (Appendix Table
C1). Only about 0.06 percent of micro firms with 1–5 employees have any 7(a) history in the comprehensive
population of US small firms, compared with roughly 1.2 percent of firms with 21–50 workers, before the share
declines again for 200+ employee firms. A similar gradient appears in business age, rising from about 0.2 percent
among firms aged 1–5 years to 0.8 percent for those older than 15 years, and along pre-PPP credit and sales
10
distributions. Legacy 7(a) relationships were thus already concentrated among larger, older, financially stronger and
higher-sales businesses before the pandemic, which helps explain why these firms enjoyed disproportionate PPP
access once the program opened.
Section VI. Discussion and Implications
A. Summary of Main Findings and Mechanisms
The evidence suggests three central messages. First, pre-existing SBA 7(a) relationships created a very large
and robust advantage in accessing PPP funds. In the raw data for June 2021, roughly 60.7 percent of firms with at
least one 7(a) loan had received a PPP loan, compared with only 19.6 percent of firms without such borrowing,
implying an unadjusted gap of about 41 percentage points. After conditioning on rich firm-level controls and state
and industry fixed effects, the baseline linear probability model still attributes a 31.6-percentage-point higher
probability of PPP recipiency to prior 7(a) borrowing. Entropy-balancing weights that align non-7(a) firms with 7(a)
borrowers on observables reduce the estimate only modestly, to an ATT of roughly 29 percentage points. This
pattern indicates that legacy SBA relationships, rather than observable differences in risk, size, or location alone,
had an independent and economically large effect on whether firms accessed PPP funds.
A simple back-of-the-envelope calculation illustrates the policy magnitude of this relationship channel. By the
end of PPP (June 2021), there were 187,762 businesses with at least one historical 7(a) loan. Combining this count
with the preferred entropy-balancing ATT estimate of 0.292 implies that roughly 0.292 × 187,762 ≈ 55,000
additional firms received a first-draw PPP loan because they had a legacy 7(a) relationship rather than
observationally similar characteristics alone. Using the average first-draw PPP loan size among June recipients of
about USD 20,000, this translates into roughly USD 1.1 billion in incremental, largely forgivable credit effectively
earmarked for firms already inside the federal guarantee system. It highlights how legacy program participation can
quietly steer tens of thousands of marginal awards and a non-trivial amount of fiscal support toward established
borrowers, with obvious implications for the design of future emergency lending schemes.
Second, the strength of selection on 7(a) ties, credit quality, and firm size changed markedly over time. In April
2020, PPP recipiency was heavily tilted toward firms that had prior 7(a) loans, scored high on commercial credit and
low on financial stress, and employed more workers. By June 2021, these coefficients shrank dramatically: the 7(a)
effect in the baseline regressions becomes small and slightly negative, and the marginal effects of credit, stress, and
size are an order of magnitude smaller. Combined with the descriptive statistics, this suggests that the first tranche of
PPP funds primarily reached financially stronger, more established firms that were already inside the SBA and
banking networks, whereas later rounds extended support to a pool of applicants that more closely resembled the
broader small-business population. The causal analysis shows, however, that even in 2021, conditional on
observables, firms with 7(a) ties still enjoyed a sizable residual advantage.
Third, the results highlight a speed–targeting–equity trade-off that is central to crisis-era business lending. Early
in the pandemic, when avoiding mass layoffs was paramount, channeling funds through entities with existing
documentation and relationships may have been the fastest way to move money. Yet doing so entrenched pre-
existing hierarchies in credit access and program familiarity. Subsequent reforms and outreach broadened eligibility
11
and diffused information, but they could not fully erase the institutional advantage of firms already integrated into
SBA lending programs.
Finally, linking these findings on access to broader program outcomes suggests that the reliance on legacy rails
may have generated a "double distortion." In companion work, Qu (2025) finds that the marginal impact of PPP
loans on employment retention, financial stress alleviation, and commercial credit improvement was significantly
smaller for firms with 7(a) experience compared to their non-7(a) counterparts. This implies that the SBA’s
infrastructure implicitly prioritized "insiders", who were already financially stronger, over "outsiders" for whom the
marginal utility of liquidity was arguably higher. Therefore, the structural inequality documented in this article
likely came at the cost of aggregate allocative efficiency, diverting emergency funds away from the firms that could
have derived the greatest stabilizing benefit. This reinforces the urgency of decoupling future crisis verification from
historical lending relationships, not merely for the sake of equity, but to maximize the economic multiplier of fiscal
intervention.
B. Policy Implications for Crisis-Response Lending
The empirical findings of this article highlight a fundamental structural tension in crisis response: the mismatch
between universal liquidity needs and niche transmission channels. While relying on the SBA 7(a) infrastructure
allowed for rapid initial deployment, it essentially privatized the allocation decision to commercial banks. As
established in the macro-finance literature, financial intermediaries naturally prioritize relationships that minimize
their own information costs and credit risks, leading to a "flight to quality" that may not align with the social
objective of broad stabilization. To resolve the distortions observed in the PPP’s "first-come, first-served" (FCFS)
design, future policy must move beyond ad-hoc adjustments toward structural reform in three specific areas.
1. The "Pre-Qualification" Model The reliance on commercial banks to process applications created a
bottleneck where firms with pre-existing 7(a) ties were prioritized, while "outsiders" faced administrative queues.
To mitigate this intermediation friction, future emergency programs should establish a direct "public option" rail that
bypasses bank underwriting for standardized relief.
2. Redesigning Allocation Mechanisms: Alternatives to "First-Come, First-Served" This article shows that the
FCFS mechanism acted as a force multiplier for inequality. Because 7(a) borrowers could navigate the application
process faster (often with lender assistance), the FCFS rule effectively reserved the initial funding tranche for them,
leaving the fund depleted for others. Future programs should abandon strict FCFS for oversubscribed facilities in
favor of Algorithmic Tranching or Lottery-Based Processing within the first window (e.g., the first 72 hours). Rather
than processing applications linearly, the administrator could enforce predetermined quotas based on firm size,
geography, or lack of prior credit history from Day 1. For instance, reserving 20% of the initial tranche specifically
for "non-relationship" borrowers would structurally prevent the crowding-out effect observed in April 2020.
3. Investing in Digital Public Infrastructure (DPI) to Reduce Information Asymmetry The core reason banks
relied on 7(a) history was the lack of verifiable information on other small businesses. The "water" of liquidity failed
to reach the "cracks" of the economy because the plumbing, the information infrastructure, was incomplete.
Policymakers should accelerate the development of Open Banking APIs and a standardized Digital Business ID
system. Similar to successful implementations in other economies, a unified digital identity that links a firm's tax
12
filing, utility payments, and bank account would allow lenders (including non-bank FinTechs) to verify a firm’s
existence and scale instantaneously, without relying on manual document review or prior relationship history. By
reducing the marginal cost of verifying "unknown" firms, such infrastructure would reduce the lender's incentive to
ration credit solely to legacy clients.
4. The reliance on private intermediaries created a principal-agent conflict: while the government's objective
was broad coverage, the lenders' objective was to minimize processing costs. Since verifying "outsider" firms (those
without prior 7(a) history) entails significantly higher marginal costs than processing legacy borrowers, a flat fee
structure implicitly penalizes lenders for serving the most vulnerable firms. Future emergency facilities should adopt
a Tiered Origination Fee Structure that compensates lenders based on the complexity of verification rather than just
loan size. Lenders should receive a "discovery premium" (e.g., a higher percentage fee) for originating loans to first-
time borrowers or underserved demographics. By explicitly pricing the "information friction," the government can
align the banks' profit motive with the social goal of equitable distribution, ensuring that the extra effort required to
reach non-networked firms is economically viable for intermediaries.
C. Contribution to the PPP and Small-Business Finance Literature
This study adds to the rapidly growing literature on the design and consequences of PPP by shifting the focus
from “where the money went” to “through which institutional channels it flowed.”
First, I place prior SBA 7(a) borrowing at the center of the analysis as a key state variable in small-business
finance. Whereas most work emphasizes firmographics, geography, banks, or race and community characteristics, I
show that participation in a long-standing SBA program strongly conditions access to a new crisis instrument
launched years later. Using entropy-balancing weights and propensity-score matching as a robustness check, I find
that prior 7(a) borrowing raises the probability of receiving a PPP loan by about 30 percentage points, far larger than
the effect of moving a firm several deciles up the credit-risk distribution. This “program-layering premium” is race-
neutral in definition but access-unequal in effect, adding an institutional channel to debates on fairness and targeting
in PPP allocation.
Second, I recast PPP as a dynamic policy that shifted from relationship-driven triage to more inclusive
allocation. Much empirical work evaluates PPP using a single cross-section or focuses on the initial lending waves.
By comparing April 2020 with June 2021 in a consistent firm-level framework, I show that the large 7(a) advantage
and strong selection on credit quality essentially disappear by mid-2021, on both the extensive margin (who receives
a loan) and the intensive margin (loan amounts). This time-variation provides micro-level evidence that reforms,
such as priority windows for very small borrowers and relaxed documentation, did more than expand volume; they
changed who the marginal dollar reached. The article thus complements macro-level assessments of PPP’s aggregate
employment and firm-size effects by documenting how its targeting mechanics evolved over the life of the program.
Third, the data and methods offer a portable template for evaluating emergency business credit using
observational microdata. I combine near-universe coverage of U.S. firms from Dun & Bradstreet with SBA 7(a) and
PPP administrative records, constructing firm-level measures of risk, size, ownership, organizational form, and
legacy program participation at an unusually fine scale. Methodologically, I apply entropy-balancing weights to
estimate average treatment effects on the treated of prior 7(a) participation and benchmark these against linear-
13
probability and propensity-score-matching estimators. While other PPP studies exploit geographic or bank-level
variation and difference-in-differences designs, this article shows how modern reweighting tools can approximate
experimental comparisons between firms with and without prior program ties in a purely observational setting. The
framework can be adapted to study layering and spillovers in other SBA initiatives, state credit guarantees, or future
crisis-lending facilities.
By foregrounding legacy SBA relationships, tracing their time-varying influence, and demonstrating a scalable
causal-weighting design, the article advances the PPP literature from documenting ex post distributions toward
understanding the institutional channels that grant some firms privileged access to public credit in times of crisis.
Section VII. Conclusions
This article leverages a near-universe panel of 26 million U.S. firms to analyze how legacy credit relationships
conditioned the allocation of emergency liquidity. By linking SBA administrative records with commercial
microdata, I document that the "first-come, first-served" design of the Paycheck Protection Program effectively
devolved into a relationship-based rationing mechanism during the crisis onset.
Three key insights emerge. First, information frictions dictated the initial allocation. In April 2020, banks
utilized prior SBA 7(a) participation as a primary screening technology: firms with legacy ties were approximately
29 percentage points more likely to receive funding than observationally identical non-7(a) firms, a causal estimate
derived via entropy balancing. Second, this "insider advantage" was dynamic. As the program matured and policy
rules evolved, the predictive power of legacy ties and credit scores dissipated, indicating that the initial rationing
was a function of administrative bottlenecks rather than a permanent structural barrier. Third, while the program
eventually achieved broad coverage, the reliance on private intermediaries created a "program-layering premium,"
where participation in peacetime guarantees inadvertently privileged firms for future emergency aid.
These findings offer a clear roadmap for future crisis-response architecture. The trade-off between speed and
equity is not inevitable; it is a consequence of relying on commercial underwriting rails for public transfers. To
mitigate this, future policy should decouple verification from historical lending relationships. This requires investing
in Digital Public Infrastructure for ex-ante verification and establishing pre-qualification mechanisms based on
administrative tax data. By reducing the marginal cost of verifying "informationally opaque" firms, policymakers
can ensure that the next injection of emergency liquidity reaches the most vulnerable businesses as rapidly as the
most connected ones.
14
References
Atkins, Rachel, Lisa D. Cook, and Robert Seamans. 2022. “Discrimination in Lending? Evidence from the Paycheck
Protection Program.” Small Business Economics 58 (2): 843–865. https://doi.org/10.1007/s11187-021-00523-1
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B. Peterman, David
Ratner, Daniel Villar, and Ahu Yildirmaz. 2022a. “An Evaluation of the Paycheck Protection Program Using
Administrative Payroll Microdata.” Journal of Public Economics 211: 104664.
https://doi.org/10.1016/j.jpubeco.2022.104664
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B. Peterman, David
Ratner, Daniel Villar, and Ahu Yildirmaz. 2022b. “The $800 Billion Paycheck Protection Program: Where Did
the Money Go and Why Did It Go There?” Journal of Economic Perspectives 36 (2): 55–80.
https://doi.org/10.1257/jep.36.2.55
Chetty, Raj, John N. Friedman, Michael Stepner, and the Opportunity Insights Team. 2024. “The Economic Impacts
of COVID-19: Evidence from a New Public Database Built Using Private Sector Data.” Quarterly Journal of
Economics 139 (2): 829–889.
https://doi.org/10.1093/qje/qjad036
Chernenko, Sergey, and David S. Scharfstein. 2024. “Racial Disparities in the Paycheck Protection Program.”
Journal of Financial Economics 160: 103911. https://doi.org/10.1016/j.jfineco.2023.103911
Chernenko, Sergey, Nathan Kaplan, Asani Sarkar, and David S. Scharfstein. 2023. “Applications or Approvals:
What Drives Racial Disparities in the Paycheck Protection Program?” Federal Reserve Bank of New York Staff
Report No. 1060. https://doi.org/10.59576/sr.1060
Dalton, Michael R. 2023. “Putting the Paycheck Protection Program into Perspective: An Analysis Using
Administrative and Survey Data.” National Tax Journal 76 (3): 521–550. https://doi.org/10.1086/726972
Granja, João, Christos Makridis, Constantine Yannelis, and Eric Zwick. 2022. “Did the Paycheck Protection
Program Hit the Target?” Journal of Financial Economics 145 (3): 785–813.
https://doi.org/10.1016/j.jfineco.2022.05.011
Hubbard, R. Glenn, and Michael R. Strain. 2020. “Has the Paycheck Protection Program Succeeded?” Brookings
Articles on Economic Activity (Fall 2020): 335–390. https://www.brookings.edu/wp-
content/uploads/2020/09/Hubbard-article.pdf
Howell, Sabrina T., Theresa Kuchler, David Snitkof, Johannes Stroebel, and Jun Wong. 2024. “Lender Automation
and Racial Disparities in Credit Access.” Journal of Finance 79 (2): 809–853.
https://doi.org/10.1111/jofi.13274
Humphries, John Eric, Christopher A. Neilson, and Gabriel Ulyssea. 2020. “Information Frictions and Access to the
Paycheck Protection Program.” Journal of Public Economics 190: 104244.
https://doi.org/10.1016/j.jpubeco.2020.104244
Joaquim, Gustavo, and J. Christina Wang. 2022. “What Do 25 Million Records of Small Businesses Say about the
Effects of the PPP?” Federal Reserve Bank of Boston Research Department Working Article No. 22-23.
https://doi.org/10.29412/res.wp.2022.23
Li, Lei, and Philip E. Strahan. 2021. “Who Supplies PPP Loans (and Does It Matter)? Banks, Relationships, and the
COVID Crisis.” Journal of Financial and Quantitative Analysis 56 (7): 2413–2449.
https://doi.org/10.1017/S0022109021000440
Qu, Chunyu. 2025. “Structure, Risk, and Access to Credit: Reassessment of the Paycheck Protection Program
Effectiveness.” SSRN Working Paper No. 5928880.
15
Tables
Table 1. Summary of Firm Characteristics in April 2020
PPP Recipients Non-Recipients
Variable Median Mean Std err Median Mean Std err
Independent Variables
7a Experience 0.000 0.032 0.177 0.000 0.004 0.060
Commercial Credit Ranking 70.000 60.844 32.198 37.000 42.277 27.847
Financial Stress Ranking 51.000 50.948 27.273 33.000 36.634 23.315
Firm size 4.000 10.431 78.178 2.000 5.207 68.073
log(firm size) 1.609 1.736 0.892 1.099 1.405 0.660
Firm age 13.000 19.341 19.120 7.000 12.210 12.782
Control Variables
Ownership 0.000 0.092 0.288 0.000 0.022 0.146
Business Failure History 0.000 0.422 0.494 0.000 0.143 0.350
Foreign Ownership 0.000 0.002 0.049 0.000 0.001 0.030
Public Status 0.000 0.001 0.022 0.000 0.002 0.013
County Population 9.000 7.820 1.584 9.000 7.937 1.491
Annual Sales ($) 165,249 5,628,610 386,381,894 94,459 1,636,740 241,172,630
log(sales) 12.015 12.373 1.749 11.456 11.384 1.389
Subsidiary 0.000 0.009 0.092 0.000 0.008 0.090
Metropolitan Location 0.000 0.055 0.229 0.000 0.011 0.104
Rural/Suburb Location 0.000 0.040 0.195 0.000 0.009 0.092
Industry Location 0.000 0.025 0.155 0.000 0.004 0.061
Sensitivity Test Dependent
Approved Loan Amount ($) 28,539 116,282 376,609
Notes: This table summarizes firm characteristics in April 2020. The dataset includes 26,704,411 firms, of which 22,907,491
received a PPP loan and 3,796,920 did not. These figures provide a descriptive snapshot of PPP participation during the
program’s first month. Significance levels are denoted by *, **, and ***, corresponding to the 10%, 5%, and 1% levels,
respectively.
16
Table 2. Linear probability model of PPP Recipiency (April 2020)
Variables (1) (2) (3) (4) (5) (6)
7a Experience 0.388*** 0.316***
(0.004) (0.006)
Commercial Credit 0.002*** 0.001 *
Ranking (0.000) (0.000)
Financial Stress 0.002*** 0.001 *
Ranking (0.000) (0.000)
log(firm size) 0.05 *** 0.004**
(0.000) (0.002)
Firm age -0.001*** 0.019
(0.000) (0.016)
Ownership 0.048*** 0.035*** 0.036*** 0.033*** 0.055*** 0.038***
(0.002) (0.000) (0.002) (0.000) (0.002) (0.002)
Business Failure 0.126*** 0.103*** 0.098*** 0.123*** 0.137*** 0.080*
History (0.001) (0.000) (0.001) (0.000) (0.001) (0.017)
Foreign Ownership 0.098*** 0.093*** 0.091*** 0.101*** 0.094*** 0.077***
(0.011) (0.002) (0.011) (0.002) (0.011) (0.029)
Public Status -0.169*** -0.141*** -0.146*** -0.202*** -0.176*** -0.102*
(0.023) (0.004) (0.023) (0.004) (0.023) (0.062)
County Population 0.001*** 0.003*** 0.002*** 0.001*** 0.001*** 0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
log(sales) 0.034*** 0.030*** 0.032*** 0.021*** 0.036*** 0.029
(0.000) (0.000) (0.000) (0.000) (0.000) (0.020)
Subsidiary -0.210*** -0.195*** -0.192*** -0.247*** -0.219*** -0.121***
(0.004) (0.001) (0.004) (0.001) (0.004) (0.003)
City Central Region 0.113*** 0.117*** 0.118*** 0.100*** 0.115*** 0.072***
(0.003) (0.001) (0.003) (0.001) (0.003) (0.006)
Rural/Suburb Region 0.085*** 0.081*** 0.083*** 0.066*** 0.088*** 0.043
(0.003) (0.001) (0.003) (0.001) (0.003) (0.038)
Industry Region 0.159*** 0.168*** 0.173*** 0.139*** 0.164*** 0.045
(0.004) (0.001) (0.004) (0.001) (0.004) (0.048)
R2 0.107 0.112 0.108 0.103 0.098 0.216
Notes: This table reports estimates from model (1a) and (1b) for first-draw PPP loans in April 2020. All columns use 26,704,411
observations. Column (6) includes all firm characteristics (7(a) experience, credit ranking, financial stress ranking, log firm size,
and firm age) simultaneously. Columns (1) – (5) report specifications that include only one of these characteristics at a time,
always with the same set of controls. All regressions include state and industry fixed effects. Significance levels are denoted by *,
**, and ***, corresponding to the 10%, 5%, and 1% levels, respectively.
17
Table 3. Linear probability model of PPP Recipiency (June 2021)
Variables (1) (2) (3) (4) (5) (6)
7a Experience -0.009*** -0.008***
(0.003) (0.003)
Commercial Credit 0.000*** 0.000**
Ranking (0.000) (0.000)
Financial Stress 0.000*** 0.000*
Ranking (0.000) (0.000)
log(firm size) -0.001*** -0.000
(0.000) (0.000)
Firm age 0.000 0.000
(0.000) (0.000)
Ownership -0.005*** -0.007*** -0.007*** -0.005*** -0.005*** -0.002
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Business Failure 0.011*** 0.008*** 0.008*** 0.011*** 0.011*** 0.002*
History (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Foreign Ownership 0.035*** 0.034*** 0.034*** 0.035*** 0.035*** 0.028***
(0.008) (0.008) (0.008) (0.008) (0.008) (0.004)
Public Status 0.203*** 0.204*** 0.204*** 0.204*** 0.203*** 0.274***
(0.016) (0.016) (0.016) (0.016) (0.016) (0.038)
County Population -0.003*** -0.002*** -0.003*** -0.003*** -0.003*** 0.002
(0.000) (0.000) (0.000) (0.000) (0.000) (0.004)
log(sales) 0.000* 0.000 0.000 0.000*** 0.000** 0.001*
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Subsidiary 0.039*** 0.041*** 0.041*** 0.040*** 0.039*** 0.026*
(0.003) (0.003) (0.003) (0.003) (0.003) (0.016)
City Central Region 0.004** 0.004** 0.004** 0.004** 0.004** 0.002**
(0.002) (0.002) (0.002) (0.002) (0.002) (0.001)
Rural/Suburb -0.006** -0.006** -0.006** -0.005** -0.005** 0.005
Region (0.002) (0.002) (0.002) (0.002) (0.002) (0.004)
Industry Region -0.008** -0.009*** -0.008** -0.008*** -0.009*** -0.006
(0.003) (0.003) (0.003) (0.003) (0.003) (0.005)
R2 0.014 0.015 0.014 0.014 0.014 0.063
Notes: This table reports estimates from model (1a) and (1b) for first-draw PPP loans in June 2021. All columns use 28,172,912
observations. Column (6) includes all firm characteristics (7(a) experience, credit ranking, financial stress ranking, log firm size,
and firm age) simultaneously. Columns (1)–(5) report specifications that include only one of these characteristics at a time,
always with the same set of controls. All regressions include state and industry fixed effects. Significance levels are denoted by *,
**, and ***, corresponding to the 10%, 5%, and 1% levels, respectively.
18
Table 4. Covariate Balance for Firms With and Without Prior 7(a) Loans (June 2021)
Panel A. Raw differences (unweighted)
Variable Mean, 7(a)=1 Mean, 7(a)=0 std, 7(a)=1 std, 7(a)=0 Std. diff ×100
Commercial Credit Ranking (1–
66.223 45.245 30.91 27.569 71.629
100)
Financial Stress Ranking (1–100) 53.245 35.812 27.818 21.916 69.616
log(firm size) 1.792 1.327 0.847 0.627 62.383
Firm age (years) 21.891 12.676 18.929 12.143 57.949
Ownership (=1) 0.104 0.02 0.282 0.142 37.619
Business failure history (=1) 0.212 0.198 0.504 0.354 3.217
Foreign ownership (=1) 0.002 0.001 0.051 0.032 2.344
Public status (=1) 0.001 0.001 0.024 0.013 0
log(annual sales) 12.532 10.412 1.697 1.361 137.838
Metropolitan location (=1) 0.462 0.011 0.22 0.103 262.737
Rural/suburban location (=1) 0.041 0.083 0.193 0.09 27.877
Panel B. After propensity-score weighting / entropy balancing
Weighted mean, Weighted std,
Variable Mean, 7(a)=1 std, 7(a)=1 Std. diff ×100
7(a)=0 7(a)=0
Commercial Credit Ranking (1–
66.223 64.291 30.91 30.632 6.279
100)
Financial Stress Ranking (1–100) 53.245 52.121 27.818 25.647 4.201
log(firm size) 1.792 1.758 0.847 0.726 4.309
Firm age (years) 21.891 20.35 18.929 17.895 8.366
Ownership (=1) 0.104 0.101 0.282 0.263 1.1
Business failure history (=1) 0.212 0.211 0.504 0.49 0.201
Foreign ownership (=1) 0.002 0.002 0.051 0.042 0
Public status (=1) 0.001 0.001 0.024 0.018 0
log(annual sales) 12.532 12.398 1.697 1.667 7.968
Metropolitan location (=1) 0.462 0.454 0.22 0.187 3.918
Rural/suburban location (=1) 0.041 0.04 0.193 0.129 0.609
Notes: This table reports covariate balance between firms with and without prior SBA 7(a) borrowing in the matched sample
(𝑁𝑇𝑟𝑒𝑎𝑡𝑒𝑑 = 187,762; 𝑁𝐶𝑜𝑛𝑡𝑟𝑜𝑙 = 27,985,150). Treatment (𝑇𝑖 = 1) is defined as having at least one 7(a) loan approved between
2010 and 2019. Panel A reports unweighted raw differences. Panel B reports differences after applying entropy-balancing
weights to the control group to match the first moments of the covariate vector. Standardized differences are calculated following
Austin (2009); absolute values below 10 indicate negligible imbalance. See Section IV.B for methodological details.
19
Table 5. Estimated Effect of Prior 7(a) Borrowing on PPP Recipiency (June 2021)
Estimator ATT on PPP recipient (=1)
Raw difference in means 0.410*** (0.001)
OLS (baseline LPM with controls) 0.316*** (0.006)
Propensity-score matching (NN, 1:1) 0.305*** (0.007)
Entropy-balancing weights (EB-ATT) 0.292*** (0.007)
Notes: The table reports average treatment effects (including model (4)) on the treated (ATT) of prior SBA 7(a) borrowing on
PPP recipiency, using the full population by June 2021. 𝑇𝑖 = 1 is defined as having at least one 7(a) loan approved (2010-2019);
𝑌𝑖 = 1 indicates firm received at least one PPP loan. The raw difference in means equals
Pr (𝑌 = 1 ∣ 𝑇 = 1) − Pr (𝑌 = 1 ∣ 𝑇 = 0),
where Pr (𝑌 = 1 ∣ 𝑇 = 1) ≈ 0.607 and Pr (𝑌 = 1 ∣ 𝑇 = 0) ≈ 0.196 based on the 2×2 distribution of PPP and 7(a) status. LPM
(1a) estimated on the full sample with industry and state fixed effects and the full set of controls. PSM uses nearest-neighbor 1:1
matching without replacement on the estimated propensity score 𝑝̂ (𝑋𝑖 ) = Pr (𝑇𝑖 = 1 ∣ 𝑋𝑖 ), where 𝑋𝑖 denotes the same set of
pretreatment covariates. Entropy-balancing weights (EB-ATT) reweight non-7(a) firms so that their weighted covariate means
match those of 7(a) firms, as documented in Table 4, Panel B. Coefficients are reported as ATT effects on the PPP recipiency
probability; robust standard errors are in parentheses. The symbols ***, **, and * denote statistical significance at the 1%, 5%,
and 10% levels, respectively.
20
Figures
Figure 1. Daily Approved Loans over different stages of PPP.
Notes: This figure depicts the daily count of approved Paycheck Protection Program (PPP) loans across both major disbursement
periods: Round 1 (April–August 2020) and Round 2 (January–June 2021). The blue area represents the total number of daily loan
approvals (including both first and second draws), while the orange area identifies second-draw loans only, available exclusively
in Round 2. The pattern reveals a highly front-loaded distribution of approvals in each round. During the first round, loan
approvals surged immediately after program launch, peaking on May 1, 2020, with 800,530 loans approved in a single day,
followed by smaller peaks on April 15 (315,720) and April 28 (441,184). Activity declined sharply after May 2020 as initial
funds were exhausted. In the second round, approvals resumed on January 12, 2021, with notable spikes on January 23 (89,400)
and March 20 (138,762), driven largely by second-draw loans. The figure underscores the intense initial demand for relief
funding and the more gradual, sustained lending pattern during the program’s later phase.
Sources: Authors' calculation based on the described data sources.
21
Figure 2. Binscatters of PPP Loan Recipiency Likelihood vs Business Determinants
(a) (b)
(c) (d)
Notes: Figure B5 displays four binscatter plots illustrating the relationship between firm or owner characteristics and the
probability of receiving a PPP loan. In all panels, the y-axis represents the ratio of firms that successfully obtained a PPP loan
within each bin, while the x-axis varies across panels to capture different explanatory dimensions. Panel (a) plots the share of
approved firms against the logarithm of firm firm size. The relationship follows an inverted-U pattern, with loan approval rates
increasing among small-to-medium firms and gradually declining among the largest firms. Panel (b) shows a strong positive
association between firms’ credit percentage rank and loan approval probability, indicating that higher-credit firms were
substantially more likely to obtain PPP funding. Panel (c) plots approval rates by financial stress ranking. The relationship is
relatively flat across most of the distribution but rises in the upper percentiles, suggesting that highly stressed firms exhibited
slightly greater loan access, possibly reflecting targeted relief efforts. Panel (d) displays the relationship between firm age and
PPP loan approval. The probability of approval rises with firm age up to roughly mid-life firms and then levels off, indicating
that older and more established firms had systematically higher access to PPP loans. All panels are constructed from firm-level
data, with observations grouped into evenly spaced bins along the x-axis. The plots illustrate broad, unconditional patterns prior
to multivariate adjustment and serve as descriptive evidence of heterogeneity in PPP loan accessibility across firm characteristics.
Sources: Authors' calculation based on the described data sources.
22
Appendix A. Data Construction and Variable Definitions
This appendix documents the construction of the merged PPP–SBA–D&B dataset, the definition and
transformation of key variables, and additional descriptive evidence on sample coverage and firm characteristics. It
is intended to complement the brief overview in Section III by providing technical details that are not essential for
following the main argument.
D&B maintains a large commercial database on U.S. businesses, including firm identifiers, industry codes,
employment, location, and a rich set of risk and ownership indicators derived from financial statements, trade-credit
histories, and public records. I use D&B snapshots as of April 2020 and June 2021 to represent firms’ pre- and mid-
program conditions.
PPP loan records, released by the U.S. Small Business Administration, report the name and location of each
borrower, loan amount, self-reported jobs retained, and lender identity for all loans approved between April 2020
and mid-August 2021. I merge these records to D&B using standardized firm names, ZIP codes, and industry
information to identify which D&B firms received PPP loans and in what amounts.
To measure prior SBA borrowing experience, I use SBA 7(a) loan data from 2010–2019 and match these
records to D&B firms. This yields a binary indicator of whether a firm had obtained at least one 7(a) loan prior to
the pandemic. The analysis focuses on firms that are active in D&B, have valid identifiers, and possess non-missing
values for core risk and employment variables at each reference date. The exact matching rules, exclusion criteria,
and sample counts are reported in Appendix A.1–A.4.
A.1 Construction of SBA 7(a) experience
SBA 7(a) data provide loan-level information on borrower names, locations, loan amounts, and approval dates.
I restrict attention to loans approved between 2010 and 2019 to ensure that 7(a) experience predates the COVID-19
shock and PPP rollout. The 7(a) file is matched to D&B firms using the same standardized name and address
procedures as for PPP. For each D&B firm, I define:
7(a) experience = 1 if the firm is matched to at least one 7(a) loan during 2010–2019, and 0 otherwise.
In robustness checks (not reported in the main text), I experiment with alternative definitions based on the
number of 7(a) loans or cumulative 7(a) volume. These yield similar patterns, underscoring that having any 7(a)
history is the most salient dimension for PPP access. Summary statistics on 7(a) participation by industry and state
are provided in Appendix Table A2.
A.2 Risk score ranking
D&B produces proprietary risk measures that estimate the likelihood of severe delinquency, financial distress,
or business failure over the coming 12 months. The raw scores are used internally and by lenders but are not directly
comparable across time or industries. To make them more interpretable and to facilitate regression analysis, I use
them with their percentile rankings:
Commercial Credit Ranking (CCS). Firms are ranked by their Commercial Credit Score within the national
cross-section and rescale the ranks to lie between 1 and 100, where 100 corresponds to the lowest predicted
probability of severe delinquency and 1 to the highest. This ranking is monotone in the underlying score and can be
interpreted as a credit-quality percentile.
23
Financial Stress Ranking (FS). Similar as CCS, so 100 denotes the lowest predicted probability of failure or
severe distress and 1 the highest.
A.3 Additional descriptive statistics
Ownership. The ownership binary reflects the ownership status of the firm and is equal to one if the
management owns the business, and zero if it rents or leases the business.
Foreign Ownership. The foreign ownership variable is equal to one if the business is owned by a party abroad: it
differentiates businesses that are domestically owned and those owned by parent companies based abroad.
Business Failure History. The business failure history variable is a binary equal to one if the firm or its
management has had any past bankruptcies or business failures and zero otherwise.
Public and Private. D&B labels each business as public or private and updates this information monthly.
Although the PPP is designed for small businesses, some large and publicly listed companies were also eligible for
PPP loans.
County Population. The County Population category reflects the residential population of the county where the
business is located, providing demographic information on the business's operational environment. This is based on
the most recent U.S. Census data, with codes ranging from 0 (denoting populations under 1,000) to 9 (indicating
populations of 500,000 and over). A one unit increase in the index indicates a double of population scale.
Subsidiary. The Subsidiary section indicates whether the business is a corporation that is majority-owned by
another company.
Location. Location classification provides information about the type of area where the business operates. I
categorize areas into four groups: metropolitan, suburban/rural, industrial and other (residential and other types of
areas).
Sales. The model utilizes the logarithm of annual sales in USD to control the revenue magnitude.
Industry and State. The model includes 1-digit NAICS industry codes for industry-fixed effects and state fixed
effects.
24
Appendix B. Robustness Check – Probit Models
In addition to the baseline models (1) and (2), a sensitivity analysis was conducted using a probit model. Probit
models, unlike LPMs, do not produce predicted values outside the 0-1 range, making them more suitable for
modeling probabilities. I re-estimate equation (1a) using probit (and logit) models for PPP recipiency. For each
specification, I report the average marginal effects of 𝑍𝑖 evaluated at the sample means. The signs, magnitudes, and
significance levels of the marginal effects are nearly identical to the LPM coefficients, confirming that the results
are not driven by the linear probability functional form. The probit model can be expressed as follows
𝑃(𝑌𝑖𝑗𝑠 = 1 | 𝐷𝑖 , 𝑋𝑖𝑘 , 𝛼𝑗 , 𝜌𝑠 ) = 𝛷(𝑎 + 𝛽𝐷𝑖 + ∑𝑘 𝛾𝑘 𝑋𝑖𝑘 + 𝛼𝑗 + 𝜌𝑠 ) (4)
is the population Probit model, where Φ denotes the cumulative distribution function of the standard normal
distribution. It's important to note that the probit model assumes that the error term 𝜖𝑖 follows a standard normal
distribution. The model estimates the probability that 𝑌𝑖𝑗𝑠 equals 1 given the independent and control variables.
25
Table B1. Summary of Firm Characteristics in June 2021
PPP Recipients Non-Recipients
Variable Median Mean Std err Median Mean Std err
Independent Variables
7a Experience 0.000 0.006 0.079 0.000 0.007 0.084
Commercial Credit Ranking 49.000 50.479 31.840 41.000 46.425 29.431
Financial Stress Ranking 37.000 42.065 25.582 33.000 39.323 23.795
Firm size 3.000 5.442 22.095 3.000 4.081 10.726
log(firm size) 1.386 1.207 0.917 1.386 1.219 0.838
Firm age 8.000 13.463 15.369 8.000 12.582 13.514
Control Variables
Ownership 0.000 0.031 0.174 0.000 0.027 0.163
Business Failure History 0.000 0.199 0.399 0.000 0.163 0.369
Foreign Ownership 0.000 0.002 0.049 0.000 0.001 0.030
Public Status 0.000 0.001 0.031 0.000 0.000 0.012
County Population 9.000 7.681 1.824 9.000 7.960 1.470
Annual Sales ($) 142,073 10,410,737 509,045,293 150,000 1,364,023 145,627,829
log(sales) 11.864 12.145 1.701 11.918 12.083 1.829
Subsidiary 0.000 0.014 0.118 0.000 0.007 0.084
Metropolitan Location 0.000 0.018 0.133 0.000 0.014 0.119
Rural/Suburb Location 0.000 0.012 0.111 0.000 0.011 0.104
Industry Location 0.000 0.006 0.077 0.000 0.006 0.075
Sensitivity Test Dependent
Approved Loan Amount ($) 16,665 20,703 115,137
Notes: This table summarizes firm characteristics for first-draw PPP loans in June 2021. The full sample includes 29,765,361
firms, of which 1,592,449 received a PPP loan and 28,172,912 did not. These figures provide a descriptive overview of PPP
participation in mid-2021. Significance levels are denoted by *, **, and ***, corresponding to the 10%, 5%, and 1% levels,
respectively.
Sources: Authors’ calculations based on D&B data and SBA loan records.
26
Table B2. Probit Model of PPP Recipiency (April 2020)
Variables (1) (2) (3) (4) (5) (6)
7a Experience 1.187*** 0.921***
(0.006) (0.008)
Commercial Credit 0.007*** 0.002***
Ranking (0.000) (0.000)
Financial Stress 0.007*** 0.006***
Ranking (0.000) (0.000)
log (firm size) 0.118*** 0.082**
(0.001) (0.05)
Firm age -0.003*** -0.001
(0.000) (0.000)
Ownership 0.062*** 0.025*** 0.023*** 0.040*** 0.100*** 0.052***
(0.003) (0.003) (0.003) (0.003) (0.003) (0.009)
Business Failure 0.385*** 0.308*** 0.279*** 0.394*** 0.433*** 0.124***
History (0.002) (0.002) (0.002) (0.002) (0.002) (0.022)
Foreign Ownership 0.486*** 0.457*** 0.452*** 0.511*** 0.452*** 0.169***
(0.019) (0.019) (0.019) (0.019) (0.016) (0.046)
Public Status -1.058*** -0.889*** -0.914*** -1.057*** -1.072*** -0.323***
(0.039) (0.039) (0.039) (0.039) (0.033) (0.100)
County Population 0.001** 0.011*** 0.004*** 0.001 -0.000 0.000
(0.001) (0.001) (0.001) (0.001) (0.000) (0.000)
log(sales) 0.203*** 0.172*** 0.184*** 0.159*** 0.213*** 0.042
(0.001) (0.001) (0.001) (0.001) (0.001) (0.031)
Subsidiary -1.066*** -0.96*** -0.958*** -1.136*** -1.124*** -0.620*
(0.009) (0.009) (0.009) (0.009) (0.008) (0.351)
City Central Region 0.224*** 0.250*** 0.249*** 0.213*** 0.225*** 0.971*
(0.004) (0.004) (0.004) (0.004) (0.004) (0.579)
Rural/Suburb Region 0.144*** 0.142*** 0.137*** 0.121*** 0.155*** 0.042
(0.005) (0.005) (0.005) (0.005) (0.004) (0.061)
Industry Region 0.320*** 0.367*** 0.376*** 0.304*** 0.322*** 0.111
(0.007) (0.007) (0.007) (0.007) (0.006) (0.189)
Notes: This table reports estimates from model (4) for first-draw PPP loans in April 2020. All columns use 26,704,411
observations. Column (6) includes all firm characteristics (7(a) experience, credit ranking, financial stress ranking, log firm size,
and firm age) simultaneously. Columns (1) – (5) report specifications that include only one of these characteristics at a time,
always with the same set of controls. All regressions include state and industry fixed effects. Significance levels are denoted by *,
**, and ***, corresponding to the 10%, 5%, and 1% levels, respectively.
Sources: Authors’ calculations based on D&B data and SBA loan records.
27
Table B3. Probit Model of PPP Recipiency (June 2021)
Variables (1) (2) (3) (4) (5) (6)
7a Experience -0.078*** -0.072***
(0.026) (0.026)
Commercial 0.002*** 0.003*
Credit Ranking
(0.000) (0.002)
Financial Stress 0.001*** 0.001*
Ranking
(0.000) (0.000)
log (firm size) -0.010*** 0.006
(0.003) (0.004)
Firm age -0.000 -0.000
(0.000) (0.001)
Ownership -0.042*** -0.056*** -0.052*** -0.04*** -0.035** -0.034**
(0.014) (0.014) (0.014) (0.014) (0.014) (0.020)
Business Failure 0.097*** 0.071*** 0.073*** 0.094*** 0.105*** 0.125
History
(0.006) (0.006) (0.007) (0.006) (0.007) (0.097)
Foreign 0.192*** 0.187*** 0.186*** 0.194*** 0.19*** 0.231
Ownership (0.056) (0.056) (0.056) (0.056) (0.056) (0.180)
Public Status 0.954*** 0.968*** 0.966*** 0.968*** 0.959*** 0.422
(0.102) (0.102) (0.102) (0.102) (0.101) (0.308)
County -0.019*** -0.017*** -0.018*** -0.019*** -0.019*** -0.02*
Population (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
log(sales) 0.002* -0.000 0.001 0.004*** 0.002* 0.002*
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Subsidiary 0.285*** 0.304*** 0.302*** 0.296*** 0.283*** 0.282***
(0.022) (0.022) (0.022) (0.022) (0.022) (0.053)
City Central 0.038** 0.036** 0.037** 0.04** 0.038** 0.029*
Region
(0.017) (0.017) (0.017) (0.017) (0.017) (0.010)
Rural/Suburb -0.046** -0.052** -0.051** -0.044** -0.043** -0.059
Region (0.021) (0.021) (0.021) (0.021) (0.021) (0.051)
Industry Region -0.072** -0.075*** -0.072** -0.07** -0.074*** -0.055
(0.029) (0.029) (0.029) (0.029) (0.029) (0.049)
Notes: This table presents the results for probit model (4), pertaining to the first draw of PPP loans in June 2021. The number of
observations for the first 6 regressions is 28,172,912. All regressions include state and industry fixed effects. Significance levels
are denoted by *, **, and ***, corresponding to the 10%, 5%, and 1% levels, respectively.
Sources: Authors’ calculations based on D&B data and SBA loan records.
28
Table B4. OLS of the Loan Amount (April 2020)
Variables (1) (2) (3) (4) (5) (6)
7a Experience 34812*** 25023***
(1634) (1520)
Commercial 98*** 392***
Credit Ranking (6) (77)
Financial Stress 17** 98
Ranking (7) (82)
log (firm size) 46272*** 2722***
(304) (388)
Firm age -72*** -212
(14) (184)
Ownership 26219*** 25465*** 26073*** 14229*** 27200*** 3503*
(913) (914) (914) (906) (933) (2207)
Business 14169*** 12873*** 14047*** 9766*** 15551*** 1404
Failure History (454) (462) (470) (450) (507) (920)
Foreign 55289*** 54892*** 55104*** 57364*** 54836*** 7252***
Ownership (4620) (4620) (4621) (4568) (4621) (2926)
Public Status 32 1310 -49 -26506*** -178 898
(9877) (9878) (9879) (9768) (9879) (6089)
County 2178*** 2302*** 2199*** 1772*** 2144*** 1038
Population (114) (114) (114) (113) (114) (795)
log(sales) 16790*** 16519*** 16803*** 5215*** 16927*** 4546*
(112) (113) (113) (134) (113) (2712)
Subsidiary -35361*** -34607*** -35566*** -62076*** -36311*** -5291***
(1719) (1720) (1722) (1708) (1722) (9719)
City Central 52327*** 52667*** 52614*** 40180*** 52586*** 2815
Region (1168) (1168) (1168) (1158) (1168) (2169)
Rural/Suburb 49016*** 48873*** 49029*** 33895*** 49440*** 9001
Region (1352) (1352) (1352) (1340) (1354) (7349)
Industry 125208*** 125973*** 125840*** 105273*** 125575 10501*
Region (1822) (1822) (1822) (1806) ***(1822) (6203)
R2 0.064 0.064 0.063 0.085 0.064 0.183
Notes: This table presents the results for linear model (2a) and (2b), pertaining to the first draw of PPP loans in April 2020. The
number of observations for the first 6 regressions is 26,704,411. All regressions include state and industry fixed effects.
Significance levels are denoted by *, **, and ***, corresponding to the 10%, 5%, and 1% levels, respectively.
Sources: Authors’ calculations based on D&B data and SBA loan records.
29
Table B5. OLS of the Loan Amount (June 2021)
Variables (1) (2) (3) (4) (5) (6)
7a Experience -139.44
-50.28 (192.30)
(317.43)
Commercial 5.19*** 8.22***
Credit Ranking
(0.99) (1.13)
Financial Stress 3.98*** 12.98***
Ranking
(1.30) (4.30)
log(firm size) 471.05*** 121.76
(36.29) (98.39)
Firm age 20.53*** 39.98
(2.57) (32.06)
Ownership 1482.25*** 1437.52*** 1450.57*** 1418.94*** 1174.24*** 808.83***
(181.42) (181.62) (181.71) (181.47) (185.47) (218.45)
Business 664.42*** 581.89*** 591.66*** 794.74*** 300.35*** 214.53
Failure History (83.84) (85.25) (87.01) (84.41) (95.35) (184.20)
Foreign -1013.22 -1027.29 -1029.41 -1112.26 -934.82 -596.25
Ownership (903.06) (903.06) (903.07) (903.02) (903.09) (601.91)
Public Status 1429.85 1465.27 1464.76 759.62 1280.82 1067.52
(1944.46) (1944.44) (1944.47) (1944.98) (1944.49) (1287.87)
County 25.94 30.87 26.21 23.811 39.95 10.11
Population (21.46) (21.48) (21.46) (21.46) (21.53) (16.45)
log(sales) 194.12*** 188.230*** 192.52*** 113.21*** 190.38*** 164.12*
(15.20) (15.24) (15.21) (16.42) (15.21) (9.91)
Subsidiary 1948.64*** 2002.87*** 1994.44*** 1486.79*** 2060.52*** 1743.28**
(332.29) (332.42) (332.58) (334.16) (332.54) (871.00)
City Central 1547.00*** 1540.74*** 1545.00*** 1446.16*** 1516.90*** 1504.03***
Region (235.18) (235.16) (235.16) (235.27) (235.18) (235.31)
Rural/Suburb 1044.94*** 1025.31*** 1029.84*** 944.73*** 911.25*** 741.76***
Region (270.67) (270.69) (270.71) (270.75) (271.18) (730.86)
Industry Region 617.66* 609.70* 619.28* 475.23 630.89* 441.09
(368.05) (368.02) (368.02) (368.15) (368.01) (465.75)
R2 0.001 0.001 0.001 0.001 0.001 0.0069
Notes: This table presents the results for OLS model (2a) and (2b) for PPP loan amounts in June 2021. The number of
observations for the first 6 regressions is 28,172,912. All regressions include state and industry fixed effects. Significance levels
are denoted by *, **, and ***, corresponding to the 10%, 5%, and 1% levels, respectively.
30
Appendix C. SBA 7(a) loans recipiency by firm characteristics and timeline
Table C1. Pre-PPP SBA 7(a) Participation by Firm Characteristics
Panel A. By employment size
Employment Total firms 7a Recipients 7(a) loan (%)
1–5 21,174,425 128,293 0.61
6–20 6,248,267 52,896 0.84
21–50 447,567 5,324 1.19
51–100 182,338 978 0.54
101–200 82,596 278 0.34
201–500 37,719 106 0.28
Panel B. By firm age
Firm age group (years) Total firms 7a Recipients 7(a) loan (%)
1–5 4,425,317 8,588 0.19
6–10 11,991,351 83,731 0.70
11–15 4,683,242 36,149 0.78
16+ 7,073,002 59,407 0.84
Panel C. By pre-PPP financial-stress percentile
Financial-stress percentile Total firms 7a Recipients 7(a) loan (%)
<10 1,953,991 24,333 1.25
11–30 8,989,954 30,077 0.03
31–50 8,887,936 45,619 0.53
51–70 4,382,218 48,602 1.11
71–90 3,104,797 32,754 1.05
91–99 854,016 6,490 0.76
Panel D. By pre-PPP commercial-credit percentile
Commercial credit percentile Total firms 7a Recipients 7(a) loan (%)
<10 2,522,190 21,336 0.85
11–30 8,806,435 21,068 0.24
31–50 5,889,887 19,428 0.83
51–70 4,054,070 23,431 0.58
71–90 4,270,697 56,611 1.33
91–99 2,629,633 46,001 1.75
Panel E. By log annual sales
Log(sales) Total firms 7a Recipients 7(a) loan (%)
0–5 2,178,441 5,389 0.25
5–10 471,307 1,210 0.26
10–15 24,959,967 170,715 0.68
15–20 556,259 10,520 1.89
>20 6,938 41 0.59
Notes: This table reports pre-PPP participation in SBA 7(a) loans by firm characteristics for the complete PPP period. Results are
derived from the D&B–PPP–SBA matched dataset. 7a Recipients count firms with at least one 7(a) approval between 2010 and
2019. 7(a) loan (%) equals 𝑛7(𝑎) /𝑛obs , expressed in percent. Financial-stress and commercial-credit groups are based on decile
bins of the respective national rankings for the firm at the time when it received the 7(a) loan. The change of the stratified
variables over year is ignorable during the period of time, thus making ignorable difference for group totals.
Sources: Authors' calculation based on the described data sources.
31
Table C2. Monthly distribution of PPP recipients and regarding prior 7(a) experience
Monthly weight of 7(a) Monthly weight
Date
and PPP recipients of PPP recipients
Apr-20 50.23% 32.45%
May-20 14.54% 16.23%
Jun-20 2.44% 3.92%
Jul-20 0.90% 1.88%
Aug-20 0.50% 0.98%
2020 cumulative 68.60% 55.47%
Jan-21 9.14% 6.10%
Feb-21 10.14% 8.98%
Mar-21 6.75% 10.94%
Apr-21 3.60% 11.29%
May-21 1.75% 7.19%
Jun-21 0.02% 0.03%
2021 cumulative 31.40% 44.53%
Notes: This table reports the monthly distribution of PPP loan recipients and of PPP recipients with prior SBA 7(a) borrowing in
the matched D&B–PPP–7(a) sample. “Monthly weight of 7(a) and PPP recipients” is the share of all firms that both hold a pre-
2020 7(a) loan and receive at least one PPP loan whose first PPP approval falls in the given month, summing to 100 percent over
all listed months. “Monthly weight of PPP recipients” is defined analogously as the share of all PPP-recipient firms (regardless of
7(a) status) whose first PPP approval occurs in that month, again summing to 100 percent. “2020 cumulative” and “2021
cumulative” rows report the total shares within each calendar year; minor discrepancies from 100 percent reflect rounding.
32
Appendix D. Clarification on 7(a) and PPP
The PPP and the 7(a)-loan program are both instrumental in supporting small businesses, but they serve
different purposes and have distinct terms and conditions.
One key difference between the two loans is regarding the credit check requirement. While the SBA's
traditional 7(a) loan program usually mandates a credit examination for approval, this requirement is currently
suspended for PPP loans.
Specifically, the funds from PPP loans should be directed towards payroll costs, employee benefits, mortgage
interest, rent, and utilities for accounts that were in effect prior to February 15, 2020. A unique feature of PPP loans
is the provision for loan forgiveness. Borrowers can qualify for loan forgiveness if they allocate the loan towards
certain expenses, with a stipulation that at least 60% of the funds must be directed towards payroll costs. PPP loans
can reach up to $10 million, with the SBA backing 100% of the loan amount. The maturity of PPP loans is 2 years
for loans issued before June 5, and 5 years for loans issued on or after that date. Furthermore, PPP loans have a flat
interest rate of 1%.
In contrast, SBA 7(a) loans are intended to facilitate business growth during economically stable times. The
funds from these loans can be used for a wide array of purposes, including buying real estate, renovation, new
construction, purchasing inventory, investing in machinery or equipment, working capital, refinancing debt, and
financing the buyout of a partner. Unlike PPP loans, SBA 7(a) loans do not offer loan forgiveness and require
collateral for loans of more than $25,000. The maximum amount for SBA 7(a) loans is $5 million, with the SBA
guaranteeing 85% for loans up to $150,000 and 75% for loans greater than $150,000. The loan maturity for SBA
7(a) loans can be up to 5 or 10 years, or even 25 years for real estate financing. The interest rates for SBA 7(a) loans
are variable, with a ceiling placed by the SBA on the amount that lenders can charge.
33
File and source
- File
- qu-2025-legacy-lending-relationships-and-credit-rationing-evidence-from-the-paycheck-prote.pdf
- Size
- 1,010,131 bytes
- SHA-256
- ec5684e4abc794d6c02ed99c2306a70e0556226f0a86fd4c59ae80497c33bc55
- Our copy
- qu-2025-legacy-lending-relationships-and-credit-rationing-evidence-from-the-paycheck-prote.pdf
- Original
- arxiv.org