The Paycheck Protection Program: Progressivity and Tax Effects (Splinter, Heiser, Love and Mortenson)

Summary

David Splinter and Jacob Mortenson of the Joint Committee on Taxation, with Eric Heiser and Michael Love, link Paycheck Protection Program loans to tax records to estimate who gained from the $800 billion program, in a paper forthcoming in the National Tax Journal. They estimate the PPP was progressive with respect to income: bottom-quintile incomes increased 18 percent, middle-quintile 4 percent and top-quintile 2 percent, though 44 percent of loan forgiveness accrued to the top quintile. About half of PPP relief benefited workers. The PPP also increased taxes and decreased unemployment compensation, reducing net program costs by one-quarter, and the authors estimate a gross cost per job-year retained of $133,000 to $146,000. The paper contrasts its progressivity finding with the view, cited to Autor et al., that the PPP was highly regressive.

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

Issuer: Source documents
Document type: The Ppp Progressivity And Tax Effects
Case: The Ppp Progressivity And Tax Effects

About this paper. Linking PPP loans to tax records, the authors find that bottom-quintile incomes rose 18 percent and top-quintile incomes 2 percent, and that about half of PPP relief benefited workers. They also find that 44 percent of PPP loan forgiveness accrued to the top quintile of earners. The paper qualifies the view that PPP mainly helped the well-off; it neither settles whether the program was cost-effective nor measures fraud.

Cite this document

“The Paycheck Protection Program: Progressivity and Tax Effects (Splinter, Heiser, Love and Mortenson).” Original at davidsplinter.com. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/the-ppp-progressivity-and-tax-effects/, SHA-256 a34b7f92a09f0c96d5789778721d1437c04a53c25034e738a0d0a9698166555b.

Full text

The Paycheck Protection Program:
Progressivity and Tax Effects
David Splinter Eric Heiser
Joint Committee on Taxation Columbia University

Michael Love Jacob Mortenson
Columbia University Law School Joint Committee on Taxation

Forthcoming in National Tax Journal

Abstract
The $800 billion Paycheck Protection Program (PPP) provided pandemic relief to businesses
retaining employees. Prior research has not directly estimated the PPP’s distributional or tax
effects. Linking PPP loans to tax records, we estimate progressive effects with respect to
income for both workers and business owners. Bottom-quintile incomes increased 18 percent
and top-quintile incomes increased 2 percent. About half of PPP relief benefitted workers. The
PPP also increased taxes and decreased unemployment compensation, reducing net program
costs by one-quarter. Net costs could have been even lower (and progressivity higher) without
the tax exclusion of PPP forgiveness.

Keywords: PPP, Covid-19, employment protection, countercyclical policy, stimulus checks,
unemployment insurance
JEL: E24, H22, H81

For helpful comments and feedback, we thank David Autor, Tom Barthold, Will Boning, David Cho, Jeff Clemens,
Michael Dalton, Connor Dowd, Chris Giosa, Lucas Goodman, Glenn Hubbard, Benjamin Kay, Adam Looney, Byron
Lutz, Rachel Moore, Brandon Pecoraro, William Peterman, David Ratner, John Sabelhaus, Andrew Samwick, Stan
Veuger, Eric Zwick, anonymous referees, and participants at the Harvard Kennedy School’s Mossavar-Rahmani
Center’s panel, National Tax Association annual conference, Grinnell College economics workshop, and Tax
Economists Forum. We thank William Gorman for help with the first figure. Splinter a nd Mortenson: This paper
embodies work undertaken for the staff of the Joint Committee on Taxation, but as members of both parties and both
houses of Congress comprise the Joint Committee on Taxation, this work should not be construed to represent the
position of any member of the Committee. Online data is available at https://davidsplinter.com/PPP.xlsx.

1
During the pandemic nearly $800 billion was spent to retain U.S. workers through the
Paycheck Protection Program (PPP). The PPP funded small business loans that were fully
forgiven if businesses spent a sufficient share on employee payroll costs. As with any
business-level subsidy or tax, the distributional impact is opaque. We provide estimates of
PPP incidence using administrative tax data, linking PPP loan data to business tax returns
and then linking businesses to the tax returns of employees and owners.
Our estimates indicate the PPP was progressive with respect to income: bottom-
quintile incomes increased 18 percent, middle-quintile incomes increased 4 percent, and top-
quintile incomes increased 2 percent. Based on differences in the timing of PPP loan
receipt, we estimate that about half of PPP loan forgiveness went to worker-retention costs.
Our estimated progressivity of the PPP, however, is relatively insensitive to the worker share
because owners and workers at PPP-receiving businesses occupied similar parts of the
income distribution. Despite the progressivity of PPP loan forgiveness, we find that 44
percent accrued to the top quintile of income earners. This estimate is a smaller than previous
estimates in the literature.1 Our progressive PPP result is robust to alternative assumptions—
even a counterfactual where we assign all the relief to business owners and none to affected
workers. That said, we do find that the PPP was less progressive than other major relief
programs during the pandemic, which is consistent with the PPP’s policy goal of retaining
employees regardless of distributional consequences.
Analysis of the PPP’s economic incidence has been complicated by it being a
subsidy targeted at workers but allocated through employers. Using links between
PPP-receiving businesses and the tax returns of their workers and owners, we present a
novel assessment of the PPP’s incidence. The current view is that the PPP was “highly
regressive” and had a “substantial” cost per job-year retained (Autor et al. 2022a, p. 56).
In contrast, we find the effects of the PPP on individual incomes were progressive.2 More
congruent with prior estimates, our estimated average gross cost per job-year retained by
the PPP ($133,000 to $146,000) was about three times median annual earnings. This does
seem substantial, but is similar to estimated costs for Great Recession job-retention
programs (Cho 2018; Chodorow-Reich 2019).
We also find that PPP costs were partially offset by positive fiscal externalities. This
results from retained workers paying higher income taxes and receiving less unemployment
compensation compared to a counterfactual where their employer did not receive PPP loans.
These fiscal externalities reduced the PPP’s net fiscal cost by one-quarter and had little
effect on progressivity. Our baseline estimates show fiscal externalities reduced the cost per
job-year retained from a gross cost of $133,000 to a net cost of $101,000.
1Autor et al. (2022a) estimated a top-quintile share of 72 percent.
2
This paper applies a consistent framework to evaluate progressivity—PPP relief as a share of income—allowing
comparisons of economically equivalent tax reductions and transfers.

2
Figure 1: Flow of first-draw PPP forgiveness to workers and owners ($ billions)

Notes: Owner share includes all PPP forgiveness from self-employed firms.
Source: Authors’ calculations using SBA and tax data.

Our paper contributes to the literature on PPP effects in several ways. Most
importantly, we leverage tax data to provide a richer and more accurate picture of the PPP’s
distributional impacts. Second, by observing both taxes paid and unemployment
compensation received by workers, we estimate the extent to which PPP’s loan forgiveness
costs were offset by higher taxes and lower unemployment payments. These positive fiscal
externalities saved or returned an estimated $133 billion of the program’s first-draw costs to
the government, as seen in Figure 1.3 Over $80 billion in additional taxes could have offset
program costs by making PPP forgiveness amounts taxable. Taxation of forgiven loans
would have also further increased the PPP’s net progressivity. Third, we account for non-
wage payroll costs ignored in prior estimates, suggesting 55 percent of PPP loan
forgiveness went to employee-retention costs. This estimated worker share is near the
loan forgiveness requirement that 60 percent go to payroll costs. 4 Fourth, we are the first
to apply differences-in-differences methodologies to tax microdata, producing estimates of
job-years saved that are consistent with prior work (Autor et al. 2022b; Dalton 2023).
Finally, we use administrative data to provide the first consistent estimates of the
distributional impacts of the three largest pandemic fiscal relief programs.
Thus, our estimates address two concerns of policymakers. First, they show the
different distributional impacts of major recession programs—those helping workers keep
their jobs (PPP), those helping workers who lost their jobs (unemployment compensation),
and those supporting economy-wide demand (stimulus payments). The PPP had
progressive impacts relative to income, but its progressivity was much less than that of

3
“First-draw” loans were the initial loans to eligible firms, mostly in 2020, accounting for the majority of PPP loans.
Firms exhausting the first draw and suffering a 25 percent revenue reduction were eligible for a “second draw” in 2021.
4 Despite this similarity, our estimates are based on economic incidence, not statutory incidence. The original 75-percent

forgiveness requirement was retroactively reduced to 60 percent on June 5th, 2020.

3
unemployment compensation or stimulus checks. Second, our estimates show the degree to
which employee-retention programs can have direct fiscal feedback effects that offset deficit
spending. As other pandemic programs were only partially taxable (unemployment
compensation) or not taxable (stimulus checks), accounting for fiscal feedback effects
provides a more apples-to-apples comparison of different policy options (Elmendorf,
Hubbard, and Williams forthcoming).
Our analysis has a number of limitations. Accounting for indirect effects, such as
consumption responses or employee scarring, would provide a more wholistic comparison
but is beyond the scope of this paper. 5 Considering longer-run and macroeconomic effects
would also change the PPP’s overall estimated impact. Benefits may be understated if the
PPP prevented firm-specific human capital losses or costs from restarting businesses that
would have closed. Further, if the PPP prevented a sustained recession and contributed to
the post-Covid recovery, this could imply additional benefits. Conversely, benefits may be
overstated if the PPP preserved less productive employment relationships and prevented
the formation of more productive employment matches. Finally, our estimates are for a
pandemic-driven recession and the applicability to standard downturns is unclear.
Section I discusses the policy background and reviews related studies. Section II
describes the PPP loan data, the tax data, and the linking method. Section III presents the
estimated effects of PPP loans on employment and wages. Section s IV and V show the
PPP’s distributional effects and fiscal externalities.

I. PPP Policy Background and Related Literature
Nearly $1 trillion of relief supported U.S. employee retention during the pandemic. Most
was from $800 billion of PPP loans.6 The PPP was created by the Coronavirus Aid, Relief,
and Economic Security Act in late March 2020 and was expanded and extended through
March 2021. Eligible firms generally had no more than 500 employees, met Small Business
Association size standards for their industry, or met an asset and net income test.7 For each
eligible firm, original first-draw loan amounts were for up to 2.5 months of average payroll
costs (after capping each employee’s annualized pay at $100,000) and a total loan of $10
million. There were 8.6 million first-draw loans totaling $589 billion, of which 95 percent was

5
Rose and Shem-Tov (2023) estimated cumulative earnings losses of $40,000 among full-time workers losing jobs in
firm-wide contractions. Davis and von Wachter (2011) estimated average male losses of 2.8 years of earnings in high-
unemployment mass-layoff events. Gertler, Huckfeldt, and Trigari (2022) found the PPP increased recalls from temporary
layoffs. Due to reduced scarring, Sledz (2025) found benefits from optimal PPP-like policies and Du (2024) found more
permanent layoffs (fewer recalls) during Covid would have caused large and sustained GDP effects.
6
Additionally, over $150 billion of employment retention tax credits targeted smaller businesses with gross receipts
declines (Internal Revenue Service 2023) and nearly $30 billion of grants went to businesses with the Restaurant
Revitalization Fund. Other employer-based programs included sick and family leave tax credits and payroll tax deferral
(Goodman 2021). Dilger, Lindsay, and Lowry (2021) provide PPP policy details.
7 See SBA’s “FAQs for Borrowers and Lenders” for the PPP and small business standards at

www.ecfr.gov/current/title-13/chapter-I/part-121.

4
forgiven. Second-draw loans totaling $209 billion were distributed in 2021 to firms with no
more than 300 employees and with revenue decreases.
In this paper, we focus on the $557 billion of first-draw PPP loan forgiveness.8
Firms were generally eligible for loan forgiveness if they maintained pre-Covid employment
and wages, had payroll and other eligible expenses—e.g., rent, mortgage, utilities—that
reached the loan amount, and had payroll costs that were at least 60 percent of the loan
amount. These requirements highlight the intent of retaining employees and preserving wage
levels, as well as preventing business closures by allowing some non-payroll expenses.

A. Prior PPP Analysis
PPP research has focused on estimating the share of loans spent on employee wages. Using
private payroll data, Autor et al. (2022a) estimated that up to 34 percent of the PPP went to
wages. Using population-level administrative data, Dalton (2023) estimated that 43 percent
of the PPP went to wages. Using similar methods, but with firm-level quarterly tax filings,
we estimate that 46 percent went to employee wages. Including missing tax and health
insurance payments for employees, suggests 55 percent of PPP forgiveness went to
employee-retention costs.9
Estimating the cost of retaining employees is another major research question. We
estimate a PPP cost per job-year retained of $133,000 to $146,000. This is similar to prior
estimates: $141,000 (Dalton 2023); $169,000 when accounting for smaller firm effects
(Autor et al. 2022b); and $175,000 based on regional differences in loan timing (Granja et
al. 2022). These estimates usually suggest the PPP retained about 3.5 million job-years,
similar to our estimate of 3.8 to 4.2 million job-years retained. Switching the focus to saved
jobs, which can be shorter duration than a year, suggests significantly lower costs of
$34,300 (Faulkender, Jackman, and Miran 2023) and $32,000 to $67,000 per job saved
(Bartik et al. 2021).10 These estimates imply the PPP saved over 10 million jobs, a relevant
estimate if the PPP’s primary goal was to maintain employer-employee relationships. This
large number of estimated jobs saved should be placed in context, as 45 million workers
received unemployment insurance in 2020 (Larrimore, Mortenson, and Splinter 2023a).
Prior estimates sometimes relied on an alternative estimation strategy, specifically,
comparing large firms just above and below the firm-size cutoffs for PPP eligibility,
usually at 500 employees (e.g., Chetty et al. 2023). Although this approach theoretically
provides a valid control group, it can have several sources of bias. First, because larger firms

8
Values are based on Small Business Association (2023). Some first-draw loans were in calendar year 2021. First-draw
forgiveness is therefore prorated to $500 billion in this paper’s estimates of 2020 income distributions.
9
Autor et al.’s results average across four ranges around employee firm -size discontinuities and the narrow range,
often preferred for regression discontinuity estimates, yields an estimate similar to our estimated employment effects
(see the appendix).
10
Doniger and Kay (2023) focus on spring 2020, when jobs were likely most sensitive to PPP funding. Their estimates
suggest a marginal cost-per-job-saved of only $17,000.

5
were less negatively affected by the pandemic than smaller firms, using larger firms as a
control group can systematically bias down the estimated number of jobs saved.11 Moreover,
the $10 million cap on PPP loans was binding for many large firms, attenuating job -retention
efficacy for larger firms. Additionally, tax data suggest there was no treatment discontinuity
at 500 employees and firm sizes were strategically misreported in SBA applications,
bunching just below the eligibility threshold (appendix Figure A9). Thus, firms with more
than 500 employees frequently received first-draw PPP loans, likely due to various
exemptions and inconsistent enforcement.12 Autor et al. (2022a, 2022b) used firm-size
cutoffs, but their job-year retained cost estimates are similar to our estimates. This is
because Autor et al. directly accounted for industry-based size exemptions, made
adjustments for other exemptions, and applied a correction to address the treatment
contamination issue around firm-size cutoffs. To address concerns about small firms, Autor
et al. (2022a) estimated an event-study analysis like that used here and found small-firm
employment effects of about 6%, as compared to our estimates of 7%.
Our analysis also enables a more accurate distributional analysis than prior work.
Autor et al. (2022a) estimated business owner distributional effects assuming a similar
distribution to that of capital income. But capital income largely results from dividends and
capital gains of large firms that were ineligible for PPP loans. In contrast, our direct links
between businesses and their owners reveal a significant number of PPP loans going to low-
income owners, including many self-employed individuals. Also, Autor et al. estimated
worker distributional effects by allocating counterfactual wage declines across the income
distribution to match observed pandemic-year employment losses. In contrast, we use
employer-employee linked wage data to produce more targeted estimates across firm size
and worker income distributions. Our results suggest a higher share of retained wages went
to low-income workers at smaller firms. This is consistent with the PPP’s targeting of smaller
firms. Finally, we use standard pre-tax (rather than after-tax) incomes to estimate
distributional effects.
Other research considered PPP’s impact on firm closures, additional uses of PPP
loans, and the need to access these loans through financial institutions. Hubbard and Strain
(2020), Bartik et al. (2021), Kurmann et al. (2022), Autor et al. (2022b), and Dalton (2023)
found evidence of reduced business closures from the PPP. This firm-preservation effect
contributes to preserved wages. Granja et al. (2022) showed that banks influenced PPP take
11
Faulkender, Jackman, and Miran (2023) found firms with more than one hundred employees had less than one-fifth the
effect on job preservation relative to smaller firms. Dalton (2023) found large firms had one-quarter the average effect on
wages. Cole (2024) estimated that medium to large firms had one-quarter to one-half the employment effect of small firms.
12
Beggs and Harvison (2023) found evidence of employee number misreporting, as the PPP application process essentially
used the “honor system.” Faulkender, Jackman, and Miran (2023) found many large firms without exceptions still
received PPP loans and discussed size-threshold approach issues. This paper includes two indirect indications of fraud.
First, the inability to match seven percent of PPP loan forgiveness by the name and location of the firms gives a
possible amount going to fabricated firms. Second, the appendix shows bunching of self -reported firm sizes at exactly
500 or just below in the PPP data, which suggests fraud along this margin.

6
up and that firms used PPP loans for buffer savings and non-wage expenses. Hubbard and
Strain (2020) showed the PPP helped firms pay bills on time. Gorbachev, Luengo-Prado,
and Wang (2024b) found the improved financial condition of PPP receiving firms aided
continued operation but not later expansion. The PPP also helped small businesses access
other credit (Karakaplan 2021), but this increased credit access was temporary (Mueller
and Spiegel 2023), was sensitive to prior banking relationships and firm size (Neilson,
Humphries, and Ulyssea 2020; Li and Strahan 2021; Duchin et al. 2022), had higher likely
misreporting among fintech lenders (Griffin, Kruger, and Mahajan 2023), and when
controlling for other characteristics, black-owned firms were less likely to apply and be
approved for PPP loans (Chernenko et al. 2023). Hamilton (2020) called for implementing
worker-retention programs through the IRS, which could disperse funds more directly and
assess loan forgiveness eligibility. Smart et al. (2023) studied the Canadian pandemic wage
subsidy. Strain and Veuger (2023) and Hong and Lucas (2023) compared Covid-era credit
programs across developed countries.

II. Data and Linking PPP Loans to Employers and Owners

PPP loan amounts and forgiveness amounts for each firm come from the Small Business
Administration (SBA 2023). We link these SBA data to Employer Identification Numbers
(EINs) in the tax data using firm names and addresses. Each firm’s EIN is linked to their
workers and owners to allocate the respective portions of the PPP. Population-wide income
data is used to estimate the PPP’s distributional effects on both workers and owners.

A. Linking PPP Loans to Firms’ Tax Filings
In the tax data, addresses, names, and firm EINs are reported on various forms depending
on the nature of the entity. These forms include quarterly payroll tax filings from employers
(Form 941) and annual entity-level tax returns (e.g., Forms 1120-S and 1065 for S
corporations and partnerships). For sole proprietorships, name and address information are
reported on Schedule C of individual tax returns. For SBA and tax data, the same procedure
cleans names and addresses. We link the SBA and tax data sequentially, beginning with exact
matches on address and name. Next, fuzzy matches use a similarity measure of the combination
of address and name. These matches begin at the zip-code level and sequentially expand to
the city, county, and state levels (appendix Table A1). To benchmark the validity of these
matches, we compare matched PPP loan forgiveness with certain PPP amounts captured in
tax data and find similar results (appendix Figure A2).
For first-draw loans, 93 percent of PPP forgiveness amounts are matched to a firm’s
EIN. The SBA data includes self-reported entity types and match rates are highest for C
and S corporations (97 percent), which represent about half of PPP forgiveness amounts.
Match rates are lowest for sole proprietorships, self-employed, and independent contractors
(67 percent) but these represent only about one-tenth of PPP forgiveness amounts. To adjust
7
for differential match rates, estimates are reweighted by the inverse match rate for each
entity type. For our regressions in the next section, firms with matched EINs are linked to
Forms 941, which report firms’ quarterly employee counts and wages. Overall, these firm-
level matches capture 74 percent of first-draw PPP forgiveness amounts in the SBA data,
as they exclude self-employed PPP recipients with no employees.

B. Linking Firms to Workers and Owners
We match each firm to its employees and owners. Employees are matched to all their
employers based on the EINs of Forms W-2 reporting wages of at least $250. Among firms
for which we match to an EIN, firms linked to Form W-2 wages represent 72 percent of
PPP forgiveness amounts. For employees of a given firm, we aggregate unemployment
compensation recipient counts and amounts observed on Form 1099-G to firm-level.
Owner portions of PPP loan forgiveness are allocated proportional to each owner’s
share of the firm. Sole-proprietorship owners, including the self-employed and contractors,
are identified with Schedule C of individual tax returns. C-corporation owners are identified
with Schedule G of Form 1120, which reports the individuals or entities owning 20 percent
or more of the corporation’s voting stock, along with their ownership shares. S-corporation
and partnership owners and ownership interests are identified with Schedules K. Tiered
ownership of partnerships are based on Love (2021). Among firms for which we match to
an EIN, these links account for 68 percent of PPP forgiveness. Non-profits cannot be linked
to owners and explain nearly one-third of the incomplete matches. Some incomplete matches
are also due to owners without individual tax returns, such as trusts and foreigners, or C-
corporation owners with small shares of the firm. To account for incomplete links, we gross
up the matched distributions for each SBA-based entity type to fully account for the PPP’s
worker and owner portions. Finally, we use taxpayer identification numbers of workers and
owners to match them to their respective individual tax returns and information returns.
This allows us to assign each worker and owner to an income group.

C. Income Groups
Workers and business owners are placed into fiscal income groups. Fiscal income is
essentially market income observed in tax data plus Social Security benefits. This income
definition parallels other studies using tax data (Larrimore, Mortenson, and Splinter 2021;
Congressional Budget Office 2022). For tax return filers, fiscal income is adjusted gross
income with the following changes: add adjustments, nontaxable interest, and nontaxable
Social Security benefits; remove taxable unemployment compensation (most was already
excluded from taxable income in 2020) and negative other income; and replace taxable
retirement income with total distributions less rollovers. For non-filers, fiscal income
includes wages, dividends, interest, some miscellaneous income, private retirement income,
Social Security benefits, and partnership income. These amounts come from various third-
party reported information returns. Following a common approach in income distribution

8
studies, we estimate income-group thresholds and totals after bottom-coding incomes at
zero, size-adjusting incomes for ranking, and creating groups based on the number of
individuals. This resembles the U.S. Census’ equivalized-income distributions and estimates
using tax data (see appendix Table A3).

III. Effects on Employment and Wages: Dynamic Difference-in-Differences Analysis

We first estimate the effects of the PPP on employment and wages using the dynamic
difference-in-differences estimator developed by Callaway and Sant’Anna (2021). This
estimator is used to estimate job and wage changes in similar settings (Miller 2023). Our
approach and sample restrictions, broadly follow Dalton (2023), and are described in detail
in the appendix.13 We estimate an average treatment effect on the firms receiving PPP loans
by comparing firms receiving the treatment against firms yet to receive the treatment. More
precisely, we compare proportional changes in employment or wages:
𝑦𝑖,𝑞,𝑡,𝑠,𝑗,𝑓
𝑌𝑖,𝑞,𝑡,𝑠,𝑗,𝑓 = (1)
(𝑦𝑖,𝑞,𝑡,𝑠,𝑗,𝑓 | 𝑡=2018)

In Equation 1, yi,q,t,s,j,f is the employment or wages in firm i, quarter q, year t, state s,
two-digit industry code j, and firm-size f. To estimate proportional changes, each firm’s
employment or wages is divided by its value in the same quarter in 2018. We further
“demean” the dependent variable by the average employment or wage growth for state,
two-digit industry code, and firm-size groups (s,j,f) to control for geographic variation in
Covid-related employment shocks and state-level policy. Firm sizes are based on the
minimum employment in any quarter in 2018.
To measure employment and wages for outcome variables, we use Form 941
quarterly payroll filings from 2018 to 2021. Form 941 data presents an advantage over self-
reported PPP data, which is subject to misreporting (appendix Figure A9). The regression
sample is limited to firms with Form 941 quarterly observations that were matched to first-
draw PPP loans in the SBA data. This limits the identifying variation to differences in the
timing of first-draw PPP receipt, with treatment mostly occurring in the second and third
quarters of 2020 but sometimes in the first and second quarters of 2021. While this
approach’s identifying variation could be problematic to the degree timing of PPP receipt
is endogenous to wage and employment growth, this approach avoids the limitations of the
size-threshold approach discussed in section I.A. Moreover, this approach shows clear
effects in the quarter of treatment, small pre-trend effects, and similar results to prior
methods using various methods.
13
Firms are dropped if they: (1) had zero or one employee in any quarter of 2018 or 2019, (2) had no valid industry
classification code, (3) had no observed state, (4) had an industry classification of public administration or utilities
due to ineligibility, (5) are restaurants because of simultaneous industry-specific relief, or (6) never received first-
draw PPP loans.

9
Figure 2: PPP effects on firm-level employment and wages
Panel A: Employment changes Panel B: Wage changes

Notes: Only first-draw PPP loans are considered. Dots are average treatment on treated coefficient estimates and ranges
are 95 percent confidence intervals associated with quarterly employment growth pre- and post-treatment. Wages are
from Form 941 indexed to 2020 dollars. Source: Authors’ calculations using SBA and tax data.

Table 1: Total PPP effects on employment and wages
Panel A: Retained Employees from PPP
Avg. Base Retained PPP $ of PPP per
Firm-Size
treatment Employees Job Years Forgiven Job-Year
Group
#employees (millions) (millions) ($billions) Retained
Self-empl. --- --- --- 53 ---
1-9 0.08 14.4 1.2 135 116,000
10-49 0.07 18.2 1.3 164 130,000
50-99 0.07 7.4 0.6 70 127,000
100-249 0.07 8.2 0.5 78 143,000
250+ 0.06 10.6 0.7 57 86,000
Total 0.07 58.9 4.2 557 133,000
Panel B: PPP worker share (costs to retain employees)
Avg. PPP employee PPP PPP employee
Firm-Size Base employee
treatment retention costs Forgiven retentions costs
Group costs ($billions)
for wages ($billions) ($billions) (% of PPP)
Self-empl. --- --- --- 53 ---
1-9 0.10 697 71 135 53
10-49 0.10 1,162 111 164 68
50-99 0.09 408 38 70 55
100-249 0.10 453 46 78 59
250+ 0.07 529 39 57 69
Total 0.09 3,249 306 557 55
Notes: Average treatment effects are averaged across four quarters and totals are base-weighted averages. PPP are forgiven
amounts of first-draw loans scaled to reflect a 72 percent match rate. Firm sizes and base wages are from the 2018 quarter
corresponding to quarter of treatment in 2020. Base employee costs are from Form 941 wages, indexed to 2020 dollars, scaled
by the match rate, and increased 20 percent to account for non-wage employee costs. Firms with one employee are excluded
from the regression and the size 2–9 coefficients are applied above. Source: Authors’ calculations using SBA and tax data.

10
Base-year wages are adjusted to account for non-wage costs of employee retention.
This is because many direct payroll costs from retaining employees are missing from our
measure of quarterly wages (the maximum of Form 941 total compensation or Medicare
wages capped at $100,000 average per employee). Base-year wages are scaled up 20 percent
to account for employer-paid federal payroll taxes, health insurance premiums, and
retirement contributions (appendix Table A4). This ignores other employer costs from
retaining employees, suggesting we may modestly underestimate the PPP’s worker share.
Figure 2A displays coefficient estimates and confidence intervals associated with
quarterly employment growth pre- and post-treatment. The coefficient in the period of
initial treatment is around 0.11, which can be interpreted as firms receiving first-draw PPP
in that quarter resulted in 11 percentage points higher employment growth relative to firms
that received first-draw PPP in a future quarter. Figure 2B displays analogous estimates for
total wage growth, and the coefficient during the quarter of treatment is around 14 percent.
Attenuated effects persist into the following three quarters. The small anticipatory effects are
likely due to widespread knowledge of the program a few weeks before enactment (Autor et
al. 2022b).
The overall impact on employment and wages is estimated by multiplying firm-size
regression coefficients by real base-year levels. Table 1A suggests the first-draw PPP
forgiveness retained 4.2 million job-years at a cost of $133,000 per job-year retained. When
adjusting for quarter-before-treatment levels across firm sizes, first-draw PPP forgiveness
retained 3.8 million job-years at a cost of $146,000 per job-year retained. Table 1B shows
that first-draw PPP loans that were forgiven induced $306 billion of additional employee-
retention costs. This represents 55 percent of PPP loan forgiveness. 14 Excluding PPP
amounts going to self-employed firms suggests 61 percent of forgiveness supported
employee payroll costs. Further robustness checks are discussed in the appendix .

IV. Distribution of PPP Effects
To allocate PPP loan forgiveness over the income distribution, we divide forgiveness into
worker and owner portions. The worker portion accounts for employee-retention costs. The
owner portion is the remaining amount (less non-profits) and allocated by firm ownership.
Allocations are across eight annual 2020 income groups: the bottom four quintiles, P80–
90, P90–95, P95–99, and the top one percent.
The worker portion’s distribution varies by allocation method, as seen in Figure 3A.
Allocating lump sum, where each worker within a firm is allocated the same amount (up
to their observed wage), gives a more progressive result. Allocating proportional to wages
gives a less progressive result. The lump-sum allocation would be more appropriate when
the PPP helped firms retain all employees—but for additional time among low-wage
14
When adjusting for quarter-before-treatment levels across firm sizes, these are $258 billion and 46 percent of PPP
loan forgiveness.

11
employees more likely to become unemployed or have longer unemployment spells without
PPP loans. The proportional allocation would apply when the PPP prevented firm closure
or reduced unpaid furloughs of equal time for all workers at a firm. Estimates suggest that
low-wage workers had disproportionately large wage declines in 2020 (Larrimore,
Mortenson, and Splinter 2023b), which supports a lump-sum allocation. Given the small
differences between the allocations, we simplify the main exposition by using the average of
lump-sum and proportional allocations.
The owner portion represents the 43 percent of PPP forgiveness left after deducting
the estimated worker portion and the implied amount going to non-profit owners ($18
billion). Overall, the owner portion of PPP tends to decline with income (see appendix
Figure A3) in a similar pattern to that seen for the worker portion in Figure 3A.
A progressive effect results from the combined worker and owner portions. Figure
3B shows the shares of each PPP amount by income quintile: the bottom quintile receives 10
percent while the top quintile receives 44 percent.15 Due to the lower incomes of the bottom
quintile, their smaller share of PPP still represents a higher share of income than for other
income groups. Figure 3C shows that PPP amounts represent 18 percent of bottom-quintile
income, 4 percent of middle-quintile income, and 2 percent of top-quintile incomes. While we
estimate that the PPP had progressive impacts on income, we also find it was less
progressive than other major pandemic relief with unemployment compensation and
stimulus checks. Figure 3D shows this differential progressivity with redistribution rates.
PPP progressivity may be underestimated in our analysis for several reasons. First,
we only consider wage effects for one year after PPP receipt. However, the share of the
PPP going to wages increases as one considers a longer post-period (Dalton 2023; Autor
et al. 2022b). Second, non-payroll costs supported by the PPP, such as from rents and
utilities, are not captured in our analysis and imply too much is allocated to owners. Third,
our income definition used to rank individuals includes PPP-retained wages. While this
allows for conventional distribution measures using observed wages, it understates the PPP’s
progressivity because increased wages moved some individuals up the income distribution.
Our progressive PPP result is robust to alternative assumptions. To simplify, we
present distribution-wide progressivity summary measures. 16 A positive value means the
PPP was progressive and our baseline progressivity is 0.38. Worker portions allocated
proportionally or lump-sum produce progressivities of 0.31 and 0.46. Worker shares of zero
and 100 percent lead to progressivities of 0.22 and 0.63. Allocating the seven percent of PPP
forgiveness among unmatched firms to the top or bottom quintiles gives progressivities of
0.31 to 0.46.

15 These estimates resemble those using linked employer-employee tax data for the Employee Retention Tax Credit.
Using an equal-owner/employee assumption and a lump-sum approach, about 12 percent of these credits went to
incomes below $10,000 and 39 percent to incomes above $100,0000 (Goodman 2023).
16
Elasticity-based progressivity is one minus the slope from regressing the natural log of relief on the natural log of
income. Progressivity is zero when relief is proportional to income across the distribution. See the appendix.

12
Figure 3: PPP Distributions, 2020
Panel A: Worker portion by allocation method Panel B: Shares by income quintile

Panel C: Redistribution rates Panel D: Redistribution rates by relief program
100

Unemployment
80

Relief (%income)
60

40
Stimulus
Payments

PPP
20

0
0 20 40 60 80 100
Income percentile
Notes: Income is essentially market income plus Social-20 Security benefits reported in tax data (fiscal income)
for both filers and non-filers, as described in the text. Quintiles have the same number of individuals ranked by
size-adjusted income. Panel D: Total 2020 relief amounts were $577 billion for unemployment compensation,
$445 billion for stimulus payments (including true-ups claimed on tax returns), and $500 billion of PPP first-
draw forgiveness. Source: Authors’ calculations using SBA and tax data.

While we estimate that the PPP had progressive impacts on income, we also find it
was less progressive than other major relief. In 2020, stimulus payment progressivity was
1.6, unemployment compensation progressivity was 1.4, and PPP progressivity was 0.4 (see
appendix). Stimulus payments were more progressive than the PPP because stimulus was
essentially a universal lump-sum transfer (Splinter 2023). Unemployment compensation
targets individuals with job losses and was especially progressive in the pandemic (Cortes
and Forsythe 2023; Larrimore, Mortenson, and Splinter 2023a). As each relief program
was about half a trillion dollars in 2020, these progressivity differences translate into

13
redistribution differences. 17 Unemployment compensation lowered the Gini coefficient by
4 points, stimulus payments by 2 points, and the PPP by only 1 point.

V. Net Fiscal Cost of PPP: Accounting for Fiscal Externalities
The gross amount spent on PPP loan forgiveness overstates the net fiscal cost. Leveraging
our links to individual tax returns, we estimate taxes on PPP-retained wages. In addition,
we estimate the amount of reduced unemployment compensation resulting from PPP loans.
When considering these positive fiscal externalities, the estimated net fiscal cost of the
2020 first-draw PPP relative to the gross amount falls by 24 percent.
PPP-induced taxes are estimated by multiplying PPP-retained wages by applicable
average marginal tax rates (AMTRs). These are the Table 1B employee costs reduced by
the non-wage costs excluded from taxation. For individual income taxes, a simple federal
payroll and income tax calculator based on taxable income from workers’ tax returns provides
AMTRs of wage decreases of $10,000. Refundable credit phase-ins result in bottom-quintile
negative AMTRs. Payroll tax rates include both the employee and employer shares, for a
total of 12.4 percent up to the taxable maximum of $137,700 and 2.9 percent on all wages
for Medicare taxes. We assume only the latter applies to the top five percent. Estimated taxes
are aggregated to the firm-level for the eight income groups and weighted by wages. For
2020, we estimate positive fiscal externalities from income taxes of $46 billion and from
payroll taxes of $42 billion, for a total of 18 percent of gross PPP forgiveness.
PPP loans that successfully preserved employment possibly lowered the amount of
unemployment compensation paid. Using annual unemployment compensation data, for
each income group we compare unemployment compensation as a share of wages of
similar-sized firms receiving first-draw PPP in the second quarter of 2020 to those receiving
it later (results are similar when comparing firms receiving and never receiving the PPP).18
See the appendix for details. For 2020, we estimate positive fiscal externalities from
reduced unemployment compensation of $32 billion, or 6 percent of the gross PPP costs.
This suggests $8,000 in avoided unemployment compensation per retained job-year,
although this estimated reduction has significant uncertainty.
These estimated externalities from taxes and unemployment compensation reduce
the net cost to about 76 percent of the gross cost, suggesting the cost per job-year retained
falls from $133,000 to $101,000 (from $146,000 to $111,000 with a pre-treatment
adjustment). The latter estimates are about twice median annual worker earnings. These
externalities have some distributional effects (Figures 4A and 4B) but little overall impact
on progressivity (Figure 4C).
The PPP had other effects beyond the two externalities considered here. Our
analysis excludes additional program costs from overhead and bank fees, although most
loans only had a one-percent fee. It also ignores small offsetting effects from the taxation on
unemployment compensation because in 2020 up to $10,200 per spouse was excluded from
income taxation for those with modified adjusted gross incomes under $150,000. It also

17
Progressivity measures are independent of program size, but redistribution measures are size dependent (Splinter 2020).
18
These unemployment compensation amounts in tax data are comprehensive, include expansions, and are over twice the
amount in the CPS (Larrimore, Mortenson, and Splinter 2023a). Many firms receiving PPP laid off some workers for limited
spells, allowing them to claim unemployment compensation.

14
excludes some sources of lower net costs: (1) additional taxes or reduced unemployment
compensation in later years, (2) additional state taxes from increased wages, (3) additional
taxes from induced taxable business profits, and (4) reductions in spending on other
government programs such as Medicaid, disability, and SNAP.
An embedded net fiscal cost is that forgiven PPP loans were excluded from firms’
taxable income, even though the expenses they funded were deductible for tax purposes.
This “double” benefit implies a tax expenditure. Multiplying 2020 loan forgiveness by
owner AMTRs, we estimate a tax-exclusion tax expenditure of $84 billion, or nearly one-
fifth of the PPP’s gross cost. About 90 percent of this tax expenditure accrued to the top
quintile (appendix Figure A10). This suggests an alternative PPP policy in which forgiven
loans were taxed would eliminate much of top-quintile net relief and further reduce net costs.

Figure 4: Net Fiscal Impact of the PPP, 2020 redistribution rates
Panel A: Worker taxes Panel B: Reduced unemployment compensation

Panel C: Total PPP net of externalities

Notes: See Figure 3 notes. Source: Authors’ calculations using SBA and tax data.

15
VI. Conclusion
It is unclear who bears the burden of business-level taxes. Business-level subsidies present
a similar puzzle—the PPP was a subsidy targeted at workers but allocated through
employers. Using links between PPP-receiving businesses and the tax returns of their
workers and owners, we estimate the distributional impacts of the PPP.
We present a novel assessment of the PPP’s incidence. Autor et al. (2022a, p. 56)
summarize the current view—the “incidence of the program across the household income
distribution was highly regressive.” Using administrative tax data, we estimate that the
overall distribution of the PPP was less concentrated than previous assumptions suggested,
and that the PPP’s effects on individual incomes was progressive. The PPP had a less
progressive impact than targeted relief like unemployment compensation, but this is
consistent with the PPP’s objective of retaining employees across the distribution.
The cost of retaining jobs with the PPP appears similar to other programs. Using an
event-study approach, we estimate an average gross cost per job-year retained of $133,000
to $146,000. For comparison, certain Great Recession programs in the Recovery Act had an
estimated cost per job-year of about $196,000 when using firm-level matched data (Cho
2018), which is similar to our approach, and usually between $50,000 and $200,000 when
using local and state-level employment changes that capture macroeconomic effects (see
review in Chodorow-Reich 2019).19 Meanwhile, Covid aid to state and local governments
had an estimated $850,000 cost per job-year (Clemens, Hoxie, and Veuger 2022). Finally,
we account for positive fiscal externalities ignored in prior analysis. When considering
PPP-induced decreases in unemployment compensation and increases in taxes paid,
progressivity is unchanged and the net cost of the PPP declines by one quarter. An
alternative policy making forgiven PPP loans taxable would increase net progressivity and
further reduce net costs.
During the Covid pandemic, a broad group of workers and firms faced shocks, with
45 million workers receiving unemployment insurance. In mild recessions, however,
shocks are smaller and likely more concentrated among marginal workers and businesses.
Therefore, rather than business-level subsidies like the PPP, countercyclical fiscal policy
typically relies on individual-level transfers, like unemployment compensation. These
transfers can target those with job or income losses but do not impede job changes or firm
exits like the PPP. The unique pandemic circumstances motivated the less-targeted
approach of the PPP, but means it is unclear to what degree this paper’s distributional
estimates would apply to another PPP-like program. Still, the $800 billion PPP can provide
some lessons about the effects of job-retention programs during deep economic shocks.

19
When considering macroeconomic effects, unemployment compensation and stimulus payments were associated
more increase in county-level employment than the PPP (Gorbachev, Luengo-Prado, and Wang 2024a).

16
References
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B.
Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz. 2022a. “ 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.
Autor, David, David Cho, Leland D Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B
Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz. 2022b. “ An Evaluation of the
Paycheck Protection Program using Administrative Payroll Microdata.” Journal of Public
Economics 211, 104664.
Bartik, Alexander W., Zoe B. Cullen, Edward L. Glaeser, Michael Luca, Christopher T. Stanton,
and Adi Sunderam. 2021. “The Targeting and Impact of Paycheck Protection Program Loans
to Small Businesses.” NBER Working Paper 27623.
Beggs, William, and Thuong Harvison. 2023. "Fraud and Abuse in the Paycheck Protection Program?
Evidence from Investment Advisory Firms.” Journal of Banking & Finance 147: 106444.
Callaway, Brantly, and Pedro H.C. Sant’Anna. 2021. “Difference-in-Differences with Multiple
Time Periods.” Journal of Econometrics 225 (2): 200–230.
Chernenko, Sergey, Nathan Kaplan, Asani Sarkar, and David S. Scharfstein. 2023. “Applications or
Approvals: What Drives Racial Disparities in the Paycheck Protection Program?” NBER
Working Paper 31172.
Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and the Opportunity
Insights Team. 2023. “How Did COVID-19 and Stabilization Policies Affect Spending and
Employment? A New Real-Time Economic Tracker Based on Private Sector Data.”
https://opportunityinsights.org/wp-content/uploads/2020/05/tracker_paper.pdf
Cho, David. 2018. “The Labor Market Effects of Demand Shocks: Firm-Level Evidence from
the Recovery Act.” Working paper available at www.david-cho.com.
Chodorow-Reich, Gabriel. 2019. “Geographic Cross-sectional Fiscal Spending Multipliers: What
Have We Learned?” American Economic Journal: Economic Policy 11 (2): 1–34.
Clemens, Jeffrey, Philip G. Hoxie, and Stan Veuger. 2022. “ Was Pandemic Relief Effective
Fiscal Stimulus? Evidence from Aid to State and Local Governments.” NBER Working
Paper 30168.
Cole, Allison. 2024. “The Impact of the Paycheck Protection Program on (Really) Small
Businesses.” SSRN working paper. https://ssrn.com/abstract=3730268.
Congressional Budget Office. 2022. “The Distribution of Household Income and Federal Taxes,
2019.” Congressional Budget Office.
Cortes, Guido Matias, and Eliza C. Forsythe. 2023. “Distributional Impacts of the Covid-19
Pandemic and the CARES Act.” Journal of Economic Inequality 225 (2): 200–230.
Dalton, Michael. 2023. “Putting the Paycheck Protection Program into Perspective: An Analysis
Using Administrative and Survey Data.” National Tax Journal 76 (2): 393–437.
Davis, Steven J., and Till von Wachter. 2011. “Recessions and the Costs of Job Loss.” Brookings
Papers on Economic Activity 42 (2): 1–73.
Dilger, Robert Jay, Bruce R. Lindsay, and Sean Lowry. 2021. “ COVID-19 Relief Assistance to
Small Businesses: Issues and Policy Options.” Congressional Research Service Report R46284.

17
Doniger, Cynthia and Benjamin S. Kay. 2023. “Long Lived Employment Effects of Delays in
Emergency Financing for Small Businesses.” Journal of Monetary Economics 140: 78–91.
Duchin, Ran, Xiumin Martin, Roni Michaely, and Ivy Wang. 2022. “Concierge treatment from
banks: Evidence from the Paycheck Protection Program.” Journal of Corporate Finance 72:
102124.
Elmendorf, Douglas, Glenn Hubbard, and Heidi Williams. Forthcoming. “Dynamic Scoring: A
Progress Report on Why, When, and How.” Brookings Papers on Economic Activity.
Faulkender, Michael W., Robert Jackman, and Stephen Miran. 2023. “The Job Preservation
Effects of Paycheck Protection.” Available at SSRN: https://ssrn.com/abstract=3767509.
Gertler, Mark, Christopher K. Huckfeldt, and Antonella Trigari . 2022. “Temporary Layoffs,
Loss-of-Recall and Cyclical Unemployment Dynamics.” NBER Working Paper No. 30134.
Goodman, Lucas. 2023. “Delivering Aid to Business through the Payroll Tax System: The Case
of the Employee Retention Credit.” National Tax Journal. Early access
https://doi.org/10.1086/724133.
Gorbachev, Olga, Maria Luengo-Prado, and J. Christina Wang. 2024a. “Cities Disrupted: The
Diverse Impact of the PPP and Other Pandemic Support Programs.” Working Paper.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4724191
Gorbachev, Olga, Maria Luengo-Prado, and J. Christina Wang. 2024b. “Effects of the Paycheck
Protection Program on Small Businesses’ Financial Health and Real Activity.” Working
Paper. http://luengoprado.net/pdfs/PPP_SG.pdf
Griffin, John M., Samuel Kruger, and Prateek Mahajan. 2023. “Did FinTech Lenders Facilitate
PPP Fraud?” Journal of Finance 78 (3): 1777–1827.
Granja, João, Constantine Yannelis, Christos Makridis, and Eric Zwick. 2022. “Did the Paycheck
Protection Program Hit the Target?” Journal of Financial Economics 145 (3): 725–761.
Doniger, Cynthia and Benjamin S. Kay. 2023. “Long Lived Employment Effects of Delays in
Emergency Financing for Small Businesses.” SSRN Working Paper.
http://dx.doi.org/10.2139/ssrn.3747223.
Du, William. 2024. “The Macroeconomic Consequences of Unemployment Scarring.” Working paper.
Hamilton, Steven. 2020. “From Survival to Revival: How to Help Small Businesses through the
COVID-19 Crisis.” Brookings Institution. Policy Proposal 2020-14.
Hubbard, Glenn, and Michael R. Strain. 2020. “Has the Paycheck Protection Program
Succeeded?” Brookings Papers on Economic Activity 51 (2): 335–378.
Internal Revenue Service. 2023. Data Book, 2022. Available at www.irs.gov/pub/irs-
pdf/p55b.pdf
Karakaplan, Mustafa U. 2021. “This time is really different: The multiplier effect of the
Paycheck Protection Program (PPP) on small business bank loans.” Journal of Banking &
Finance 106223.
Kurmann, André, Etienne Lalé, and Lien Ta. 2022. “Measuring Small Business Dynamics and
Employment with Private-Sector Real-Time Data.” IZA Discussion Papers No. 15515.

18
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2021. “Household Incomes in Tax Data: Using
Addresses to Move from Tax Unit to Household Income Distributions.” Journal of Human
Resources 56 (2): 600–631.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2022. “ Unemployment Insurance in
Survey and Administrative Data.” Journal of Public Economics 206: 104597.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2023a. “Unemployment Insurance in Survey
and Administrative Data.” Journal of Policy Analysis and Management 42 (2): 571–579.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2023b. “Earnings Business Cycles: The
Covid Recession, Recovery, and Policy Response.” Journal of Public Economics 225: 104983.
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: 2411–2438.
Love, Michael. 2021. “Where in the World Does Partnership Income Go? Evidence of a Growing Use of Tax
Havens.” SSRN Working Paper. https://ssrn.com/abstract=3985535
Miller, Douglas L. 2023. “An Introductory Guide to Event Study Models.” Journal of Economic
Perspectives 37 (2): 203–230.
Neilson, Christopher A., John Eric Humphries, and Gabriel Ulyssea. 2020. “Information
Frictions and Access to the Paycheck Protection Program.” Journal of Public
Economics 190: 104244.
Rose, Evan K., and Yotam Shem-Tov. 2023. “How Replaceable Is a Low-Wage Job?” NBER
Working Paper 31447.
Sledz, Shannon. 2025. “Whom to Insure — Firms of Workers?” Working paper. Available at
https://users.ssc.wisc.edu/~ssledz/JMP_Sledz_Shannon.pdf .
Small Business Association. 2023. “All PPP Loan Data” File last updated on January 1, 2023.
Accessed on February 24, 2023, from https://data.sba.gov/dataset/ppp-foia
Smart, Michael, Matthew Kronberg, Josip Lesica, Danny Leung, and Huju Liu. 2023. “The
Employment Effects of a Pandemic Wage Subsidy.” CESifo Working Paper No. 10218.
Splinter, David. 2020. “U.S. Tax Progressivity and Redistribution.” National Tax Journal 73(4):
1005–1024.
Splinter, David. 2023. “Stimulus Checks: True-Up and Safe-Harbor Costs.” National Tax Journal
76(2): 349–366.
Strain, Michael R., and Stan Veuger. 2023. Preserving Links in the Pandemic: Policies to
Maintain Worker-Firm Attachment in the OECD. Washington, DC: AEI Press.

19
Online Appendix
1. Linking Data
The SBA data are linked to tax data by name and address. Addresses, names, and Employer
Identification Numbers (EINs) are reported on a variety of tax filings, depending on the nature of
the entity. These include quarterly payroll tax filings for employers (Form 941) and annual entity-
level tax returns (Forms 1120, 1120-S, 1065, and 990). For sole proprietorships, name and address
information are reported on individual tax returns (Form 1040) filing a Schedule C and account
for either business or individual names, including both spouses’ names on joint returns. For both
SBA and tax data, we clean all names and addresses using the same cleaning procedure, removing
special characters and spaces and standardizing address and business name endings.
We link the SBA and tax data sequentially, with exact matches on address and name. Next,
fuzzy matches use a similarity measure of the combination of address and name. These matches start
at a small geographic level (zip codes) and sequentially expand to consider matches at the city,
county, and state levels. As the geographic level expands, the match criteria become stricter. Table
A1 summarizes the proportion of loans matched at each step. Table A2 shows match rates by firm
size, and Figure A1 shows first-draw PPP take-up rates by firm size.
For first-draw loans, 93 percent of PPP forgiveness amounts, 93 percent of employees, and
79 percent of loans are matched to a firm’s EIN. The SBA data includes self-reported entity types
(e.g., partnership, C corporation, etc.) that help show which entities are matched well, even though
these self-reported entity types do not completely align with tax entity types. Match rates are
highest for corporations (97 percent), and these represent over half of PPP forgiveness amounts.
About half of loans counts are to certain single-person firms (sole proprietorships, self-employed,
and independent contractors). This group’s low match rate (67 percent) pulls down the percent of
loans matched from 93 percent without these small firms to 79 percent with them. However, as
these firms receive only about one-tenth of total PPP forgiveness amounts, they only push the
forgiveness match rate down from 96 percent to 93 percent.
For our regressions, firms with matched EINs are then linked to Forms 941, which have
quarterly employee counts and wage amounts. Overall, these matches capture 74 percent of first-
draw PPP forgiveness amounts. For comparison, Dalton (2023) matched 87 percent of certain 2020
PPP loans amounts to quarterly wages, but this excluded entities that tend to have lower match
rates (i.e., sole proprietorships, self-employed, independent contractors, and non-profit firms).

Table A1: Linking Method: SBA and tax data, first-draw loans
Match Share of loan Share of loan Share of
Type counts (%) dollars (%) employees (%)
Exact 22 39 38
Zip code 15 26 26
City 2 3 3
County 2 2 2
State 4 2 3
Schedule K 0 0 0
Schedule C 35 21 21
Total 79 93 93
Sources: Authors’ calculations using SBA and tax data .

20
Table A2: SBA to tax EIN match rates: First-draw PPP loans by firm size
Forgiveness Amount Number of Loans
Firm
Size Match Match
$Billions Thousands
Rate (%) Rate (%)
1-9 138 84 7,281 76
10-49 180 96 1,103 95
50-99 78 96 132 95
100-249 89 96 71 95
250+ 72 97 25 95
Total 557 93 8,613 79
Sources: Authors’ calculations using SBA and tax data .

Figure A1: PPP take-up rate of firms by firm size (%)

Notes: Within each firm-size bin of 10 employees, take-up rate is the number of SBA-matched firms with first-draw
PPP loans divided by the total number of firms with Form 941. Firm sizes are based on the average number of
employees reported on quarterly filed 2019 Forms 941. Source: Authors’ calculations using SBA and tax data.

21
2. Comparison with PPP amounts in tax data
Comparisons of the SBA loan data with forgiven amounts reported on schedule K for S
corporations and partnerships (Forms 1120-S and 1065) suggest that our procedure results in a
high rate of correct matches. Despite being excluded from taxation, S corporations and
partnerships were instructed to report this forgiveness as other tax-exempt income (schedule K lines
18b and 16b, respectively). While other types of income could be reported here, total other tax-
exempt income reported by S corporations and partnerships increased from $14 billion (0.1 million
records) in 2019, to $148 billion (1.2 million records) in 2020, and in 2021 to $298 billion (1.5
million records) in 2021. This suggests the vast majority of this income reported on these lines
was forgiven PPP loans. Of the 2.1 million firms with relevant schedule K’s and reporting other
tax-exempt income, we match 75 percent to a PPP loan. Aggregating across loans and across tax
years 2020 and 2021, we find that 69 percent of the matched had PPP loan amounts within 10
percent (most others were just below this range, as seen in Figure A2).

Figure A2: Ratio of all PPP forgiveness in SBA and tax data, matched firms

Notes: Ratio of first- and second-draw PPP loan forgiveness in SBA data to other tax-exempt income reported on
Schedule K in tax data for 2020 and 2021, only among matched firms. Bins are for 20-percent intervals. Firms with
twice as much SBA amounts than tax amounts suggests that some EINs were linked to two different PPP-receiving
firms. Source: Authors’ calculations using SBA and tax data.

3. Differences-in-Differences Estimation
We estimate the effects of the PPP on employment and wages using a dynamic difference-
in-differences estimator developed by Callaway and Sant’Anna (2021). We use Form 941
quarterly payroll filings from 2018 to 2021, which include counts of employees and wa ges paid in
that quarter.
We make several sample restrictions. Firms are dropped if they have: (1) zero or one
employee in any quarter of 2018 or 2019, (2) no valid industry classification code in any year, (3)
no observed state in any year, (4) industry classifications of public administration or utilities due
22
to ineligibility, (5) are restaurants because of simultaneous industry -specific relief paralleling the
PPP, or (6) never received first-draw PPP loans. 20 Finally, the panel is balanced by setting the
number of employees and wages to zero in quarters for which a firm was inactive. Wages are
adjusted to 2020 price levels.

ATTp,q = E [ Yq (PPP) – Yq (No PPP) ] (A.1)

The coefficient estimated using the dynamic difference-in-differences estimator is an
average treatment effect on the treated (ATT). In Equation A.1, for firms receiving PPP in quarter
p, ATTp,q is the average difference in quarter q of the dependent variable Y (employment or wages)
between firms receiving and not receiving first-draw PPP loans. This approach, after demeaning
to control for time-invariant covariates, accounts for different treatment timing and heterogeneous
treatment effects to estimate an uncontaminated ATT (Dalton 2023).
𝑦𝑖,𝑞,𝑡,𝑠,𝑗,𝑓
𝑌𝑖,𝑞,𝑡,𝑠,𝑗,𝑓 = (A.2)
(𝑦𝑖,𝑞,𝑡,𝑠,𝑗,𝑓 | 𝑡=2018)

𝑌̃𝑖,𝑞,𝑡,𝑠,𝑗,𝑓 = 𝑌𝑖,𝑞,𝑡,𝑠,𝑗,𝑓 − 𝑌̅𝑠,𝑗,𝑓 (A.3)

In Equation A.2, yi,q,t,s,j,f is the employment or wages in firm i, quarter q, year t, state s, two-
digit industry code j, and firm-size f. To estimate proportional changes, each firm’s employment
or wages is divided by its value in the same quarter in 2018. Following Dalton (2023), in Equation
A.3, the dependent variable is “demeaned” by the average employment or wage growth for state,
two-digit industry code, and firm-size groups (s,j,f). This is similar to using fixed effects, although
it allows for faster processing with the full population. Demeaning by state helps control for
geographic variation in Covid-related employment shocks and state-level policy. The firm-size
bins are defined based on the minimum employment observed in any quarter in 2018. The
regression sample is limited to firms with Form 941 quarterly observations that were matched to
first-draw PPP loans in the SBA data. This limits the identifying variation to differences in the
timing of first-draw PPP receipt, with treatment mostly occurring in the second and third quarters
of 2020 but sometimes in the first and second quarters of 2021. While this identifying variation
could be problematic to the degree timing of PPP receipt is endogenous to wage and employment
growth, this approach avoids the limitations of the size-threshold approach discussed in section
I.A of the main paper and section 8 of this appendix.
Base-year wages are adjusted to account for non-wage costs of employee retention. This is
because many direct payroll costs from retaining employees are missing from our measure of
quarterly wages using Form 941 (the maximum of total compensation or Medicare wages and then
capped at $100,000 average per employee). Base-year wages are scaled up to account for
employer-paid federal payroll taxes, health insurance premiums, and retirement contributions. 21

20
The restrictions result in the following changes: (1) dropping zero- or one-employee firms reduces PPP amounts by
about one-fifth (including from many Sch. C sole proprietors), (2) removing firms without industry codes reduces the
firm count and PPP amounts by 3% (codes come from firm tax returns), (3) removing firms with no reported state has
negligible effects, (4 and 5) removing public admin., utilities, or restaurants reduces the firm count and PPP amounts
by 8%, and (6) removing firms never receiving first-draw PPP loans reduces the firm count by about one-third.
21
Average costs with respect to wages are based on national accounts: 8 percent for employer-paid payroll taxes, 8
percent for health insurance, and 4 percent for retirement contributions. See appendix Table A4.

23
This approach ignores other employer costs from retaining employees, suggesting we may
modestly underestimate the PPP’s worker share.
Figure 3A of the main paper displays coefficient estimates and confidence intervals
associated with quarterly employment growth pre- and post-treatment. The coefficient in the
period of initial treatment is around 0.11, which can be interpreted as firms receiving first-draw
PPP in that quarter resulted in 11 percentage point higher employment growth relative to firms
that received first-draw PPP in a future quarter. Figure 3B of the main paper displays analogous
estimates for total wage growth, and the coefficient during the quarter of treatment is around 14
percent. Attenuated effects persist into the following three quarters. Although we observe a slight
pre-trend in the two quarters prior, the magnitude is tiny in comparison to the size of the estimate
effect, and a portion is attributable to anticipatory effects (Autor et al. 2022).22

4. Expanded Fiscal Income
Workers and business owners are placed into 2020 fiscal income groups, which is essentially
market income observed in tax data and Social Security benefits. This income definition parallels
other studies using tax data: Piketty and Saez (2003); Larrimore, Mortenson, and Splinter (2021,
2022); and Congressional Budget Office 2022). For tax return filers, fiscal income is adjusted gross
income reported on individual tax returns adjusted as follows: add nontaxable interest, adjustments,
and non-taxable taxable Social Security; remove taxable unemployment compensation (most was
excluded from taxation in 2020) and negative other income to account for net operating loss
carryovers from prior-year losses; and replace taxable private retirement income with gross private
retirement income (total distributions less rollovers). For non-filers, fiscal income includes wages,
dividends, interest, miscellaneous income (half to account for missing deductions), gross private
retirement income, Social Security benefits, and partnership income from Forms W-2, 1099-DIV,
1099-INT, 1099-NEC, 1099-MISC, 1099-R, 5498, SSA-1099, and 1065 and 1120-S Sch. K-1s. In
2020, our measure of fiscal income totals $14.0 trillion, or 79 percent of national income. Fiscal
income is usually about 60 percent of national income (Auten and Splinter 2024), but our measure
includes capital gains and certain Social Security benefits.23
In the 2020 tax data, we observe 332.7 million domestic individuals, which matches the
U.S. Census resident population. 24 This includes filers and dependents on domestic tax returns and
non-filers with at least one domestic information return. This fits with prior analysis, where 2010
tax data included 99.8 percent of the U.S. Census resident population (Larrimore, Mortenson, and
Splinter 2021).
Following the approach in prominent income distribution studies, we estimate income-
group thresholds and totals after bottom coding incomes at zero, size -adjusting incomes, and
creating groups based on the number of individuals. This resembles the equivalized-income
distributional estimates presented by the U.S. Census and estimates using tax data by Auten and
Splinter (2024) and the Congressional Budget Office. Our income shares resemble prior estimates
(see Table A3). When ranking tax units across the distribution, income is size adjusted to account
for economies of scale and sharing. This adjustment divides income by the square-root of the
22
Along with anticipation before overall enactment, some of the treatment firms in later periods likely anticipated
receipt or adapted behavior in months prior to receipt anticipating funds would be available.
23
Our 2020 income is 95 percent of the 2019 Congressional Budget Office (2022) measure of expanded fiscal income,
which includes additional income sources (e.g., corporate taxes and Medicare benefits).
24
U.S. Census July 1st population estimates for 2020 and 2021 are 332.0 and 333.3 million and average to 332.7 million,
the more comparable end-of-year level for IRS data . See www.census.gov/quickfacts/fact/table/US/PST045222.

24
number of individuals on a tax return (filers and dependents).25 Income groups are created based
on the number of individuals such that each quintile has the same number of individuals (as
compared to the same number of tax filing units).
Our income rankings include PPP-induced wages, although redistribution measures should
use pre-policy income. However, assuming the estimated 4.6 million affected workers fall two
quintiles without PPP would only increase elasticity-based progressivity from 0.40 to 0.41.

Table A3: Fiscal Income Share Comparisons
Bottom 2nd 3rd 4th Top
Quintile Quintile Quintile Quintile Quintile
AS: no cap gains, 2019 2% 7% 13% 21% 57%
CBO: cap gains, 2019 4% 9% 14% 20% 55%
Fiscal Income: cap gains, 2020 2% 7% 12% 19% 60%
Notes: Our measure is fiscal income, adjusted gross income plus nontaxable interest, Social Security benefits, and
non-rollover retirement income minus taxable unemployment compensation and net operating loss carryovers.
Congressional Budget Office (CBO) income is “income before taxes and transfers” (includes realized capital
gains, Social Security, unemployment compensation, and Medicare benefits) and has a larger sharing unit for size
adjusting than other measures (household vs. tax units), which increases bottom-quintile incomes. Auten and
Splinter (AS) fiscal income is from the step just after grouping by size-adjusted income and number of individuals
(excludes realized capital gains and transfers). All measures define groups using the number of individuals and
size-adjusted income. Sources: CBO (2022), Auten and Splinter (2024), and authors’ calculations.

Table A4: Non-wage employee cost adjustments using 2020 values in NIPA tables 7.8 and 2.2B
Type Amount Base Wages Percent
Payroll taxes 717 9,457 8%
Health insurance 797 9,457 8%
Retirement contributions 295 7,963 4%
Total 20%
Notes: Payroll taxes include $454 billion of OASDI Social Security taxes, $131 billion of Medicare taxes, $87 billion of
workers’ compensation, and $45 billion of unemployment taxes. Wages and salaries are $9,457 billion and for retirement,
private wages and salaries are $7,963 billion. Source: Bureau of Economic Analysis (NIPA Tables 7.8 and 2.2B).

5. Income Variability: Redistribution or Stabilization
PPP distributional estimates appear robust to one-year income changes. Redistribution rates are
based on annual income, but income variability can push individuals into different income groups in
surrounding years. This means fiscal relief from PPP and other programs can represent
“redistribution” for individuals with persistently low incomes, but “stabilization” for those with
short-term losses (Larrimore, Mortenson, and Splinter 2016). To assess low-income dynamics, we
consider 2020 tax returns with less than $10,000 of size-adjusted income. In both the preceding and
subsequent years, about 30 percent of these filers have higher incomes but only 0.5 percent have
surrounding-year incomes above $100,000. Splinter (2022) estimated similar low-income dynamics.
When limiting to returns with business income or losses on Schedule E, which accounts for S

25
As non-filing tax units are not defined in the tax data and only about one-tenth of non-filing tax units are married
(Auten and Splinter 2024), we do not combine non-filers into synthetic tax units. Linking non-filers, however, would
re-rank few non-filers across our broad income groups.

25
corporations and partnerships, about 45 percent have higher incomes and 9 percent have
surrounding-year incomes above $100,000. The bottom quintile also appears insensitive to removing
certain business losses. Among the 2020 tax returns with zero or negative income, removing
Schedule E losses, leaves 96 percent with incomes below $10,000 and moves less than one percent
above $100,000.

6. Estimating Avoided Unemployment Compensation from the PPP
If workers are retained as a result of the PPP, they do not need to apply for unemployment
compensation. Because we estimate a sizeable number of workers were retained due to PPP, we
also estimate the amount of unemployment compensation avoided because of the PPP. It is not
possible, however, to run differences-in-differences regressions to estimate avoided unemployment
compensation in the same manner as estimating retained employees or wages, which are reported
quarterly in tax data. Unemployment compensation reported in tax data is only available annually,
and thus we cannot exploit quarterly variation. Instead, we use a simple procedure that relies on
the following assumption: had the PPP not happened, workers in the same portion of the income
distribution at similar-sized firms would have had the same aggregate unemployment-compensation-
to-wage ratios whether the firm received the PPP in the second quarter of 2020 (treated) or after
this quarter (untreated).
The following equations more precisely state our procedure. Take the aggregate wages of
the untreated workers (𝑤0) and the aggregate unemployment compensation of the untreated
workers (𝑢0) in a single “bin,” where bins are by firm size and by the placement of the worker
across our eight income groups. For each bin, these can be represented as an unemployment-
compensation-to-wage ratio 𝑟0 :
𝑢0
𝑟𝑜 = (A.4)
𝑤0

We assume that 𝑟0 = 𝑟𝑡𝑛 , where 𝑟𝑡𝑛 is the same ratio for the treated firms had the PPP not happened.
That is,
𝑢0 𝑢𝑛
𝑟𝑜 = = 𝑡
= 𝑟𝑡𝑛 (A.5)
𝑤0 𝑤𝑛
𝑡

where 𝑢𝑛𝑡 and 𝑤𝑡𝑛 represent the unemployment compensation and wages of the treated workers
had the PPP not occurred (thus the superscript n).
The challenge is that we cannot observe 𝑢𝑛𝑡 or 𝑤𝑡𝑛 , so we must infer them. However, we
can observe 𝑢𝑎𝑡 and 𝑤𝑡𝑎 , the actual unemployment compensation and wages of the treated workers.
We find in our empirical estimates that the PPP retained workers on net, thus increasing wages
and reducing unemployment compensation. We write the actual observed ratio for treated workers
𝑟𝑡𝑎 as:
𝑢𝑛 𝑠
𝑡 −𝑢𝑡
𝑟𝑡𝑎 = (A.6)
𝑤𝑛 𝑠
𝑡 +𝑤𝑡

where 𝑢𝑠𝑡 and 𝑤𝑡𝑠 represent the unemployment compensation and wages of the treated workers that
were saved due to the PPP (i.e., the unemployment compensation was avoided and the wages were
saved).

26
Using the three equations above, we can solve for the desired term, unemployment
compensation avoided (𝑢𝑠𝑡 ), as a function of observable aggregate parameters within each bin:

𝑢𝑠𝑡 = 𝑟𝑜 ⋅ 𝑤𝑡𝑛 − 𝑢𝑎𝑡 (A.7)

In other words, the avoided unemployment compensation (𝑢𝑠𝑡 ) equals the amount of unemployment
compensation that would have been paid for the treated workers without PPP (because 𝑢𝑛𝑡 = 𝑟𝑡𝑛 ⋅
𝑤𝑡𝑛 by definition and 𝑟𝑡𝑛 = 𝑟𝑜 by assumption, thus 𝑟𝑜 ⋅ 𝑤𝑡𝑛 = 𝑢𝑛𝑡 ) minus the actual observed
unemployment compensation paid to the treated workers (𝑢𝑎𝑡 ). We can estimate 𝑤𝑡𝑛 using the
results of our regression analysis on saved wages, because 𝑤𝑡𝑛 = 𝑤𝑡𝑎 − 𝑤𝑡𝑠 by definition).
Practically speaking, we take the results of our regression analysis of estimated saved wages by
firm size, apply those saved wages proportionally across the income distribution categories, and
then estimate the unemployment compensation avoided by each firm size and worker income
distribution bin. Note that this excludes self -employed workers and may therefore present an
underestimate of the PPP’s effect on unemployment compensation.

7. Elasticity-Based Progressivity
The elasticity-based progressivity is one minus the slope from regressing the natural log of relief
on the natural log of income (average per capita amounts for eight income groups). Splinter (2020)
discusses this measure as applied to taxes. A progressivity of zero would result from relief that
was proportional to income. Positive progressivity results from relief that decreases with income
(negative slope) or increases more slowly than income (positive slope of less than one). Note that
progressivity measures control for the size of the relief, making different relief programs more
comparable (in contrast, redistribution measures are sensitive to the total amount of relief ; see
Lambert 1993 and Kakwani 1977).
Figure A3 shows redistribution rates (PPP forgiveness as a share of income) of the owner
portion by entity type. The PPP targeted smaller firms and therefore less went to C corporations
and partnerships than to S corporations and sole proprietorships. Over the income distribution,
the C corporation portion is relatively proportional. The partnership portion and S corporation
portions have inverse J-shapes with larger benefits for the bottom of the distribution. The sole
proprietorship share is strongly progressive. Overall, we observe an inverse J-shape pattern of the
owner portion of PPP over the income distribution: 10 percent of income for the bottom quintile,
about 1 percent for middle quintiles, and 2 percent for the top percentile of the income distribution.
Figure A4 shows the average log relief and log incomes for the eight income groups in
2020. The figure shows a negative slope, meaning absolute amounts of unemployment insurance
decrease as one moves up the income distribution, hence the large progressivity of 1.4. The top -
right figure shows the same for stimulus payments and a large progressivity of 1.6 (although the
fit is poor because this was a flat benefit amount with an income phase out). The bottom -left figure
shows a positive slope of less than one, meaning absolute amounts of unemployment insurance
increase but increase more slowly than income as one moves up the income distribution, hence the
small progressivity of 0.4 for the overall PPP. The bottom-right figure shows a flatter positive
slope for a moderate progressivity of 0.7 for the wage portion of PPP. Re ynolds–Smolensky
redistribution indexes show similar results: unemployment compensation lowers the Gini
coefficient by 4 points, stimulus payments by 2 points, and the PPP by 1 point.

27
Figure A3: PPP owner portion by entity type, 2020 (redistribution rates)

Notes: Income is essentially market income plus Social Security benefits reported in tax data (fiscal income)
for both filers and non-filers, as described in the text. Quintiles have the same number of individuals ranked by
size-adjusted income. Businesses are grouped by tax-data based entity types where unmatched types are not
shown. Source: Authors’ calculations using SBA and tax data.

Figure A4: Elasticity-based progressivity by program, 2020

28
8. Robustness Checks
The main regression compares firms that received a first-draw PPP loan in a given quarter with firms
that receive first-draw PPP loans in subsequent quarters. In this section, we present robustness checks
after describing our main analysis. Our main analysis is based on variation in treatment over time.
This has some strengths and weaknesses due to likely reasons firms delay applying for or receiving
PPP loans. Some firms may have delayed loan applications due to incomplete information, the
belief that they did not initially qualify because they were not affected by the pandemic (and later
they thought they were affected), or they delayed their application because they initially had no
banking relationship. As owners learned of the widespread take up and billions spent on PPP loans,
all three of these frictions should decrease over time. The latter two reasons would be problematic
if application timing was correlated with real economic shocks or if those with no banking
relationships are different. Our summary statistics in Table A4 suggest the firms receiving the PPP
in different calendar quarter have similar average wages, although sizes vary over time. Note that
low average wages in Table A4 are because tax data do not allow us to separate part-time and full-
time workers or remove workers joining or leaving a firm during that quarter.
The first robustness check we consider includes separate estimates for restaurants, which
are excluded from the main analyses due to simultaneous relief from a separate program. Relative
to our main results, Figure A5 shows nearly identical first and second quarter effects and lower
third and fourth quarter effects. Second, we show the results by calendar year time, as the main
figures pool results of treatments starting in various quarters. Figure A6 shows that initial-quarter
treatment effects were positive for all three timing groups but dissipated for later groups, consistent
with the waning of the pandemic’s economic shock over time. Third, we include firms that have
600 employees or less in the base year and never received PPP loans, which is closer to the Dalton
(2023) approach. Compared to our baseline estimates, Figure A7 shows that including firms never
receiving PPP results in lower one-quarter-after-treatment effects and higher four-quarter-after-
treatment effects. These results seem less reasonable given the expectation of declining effects
over time and the pre-trends also look problematic.
Some papers exploit the nearly two-week period during which PPP loans were not approved
during mid-April 2020 due to allocated funds running out. Appropriations were added soon
afterwards, but this provides an alternative identification of the effect of PPP loans on employment.
The SBA data show that approximately no applications were approved between April 16 and April
26, 2020. We limit our sample to two groups of firms: those approved for PPP loans the week before
the approval pause (April 9 to April 16) and those approved for PPP loans the week after the pause
(April 27 to May 3). Using a standard difference-in-difference regression, we compare the
employment trends in subsequent quarters for those treated the week before the pause with those
treated the week after the pause. This suggests firms approved for PPP loans in the week just before
the approval pause had average employment growth 1.5 percentage points higher than firms
approved in the week just after the pause (p-value less than 0.1%). This is about one-fifth the
estimated 7 percentage points higher employment effect with our baseline approach, which captures
much more heterogeneity across time than this one-week approach. Therefore, we view the findings
from the approval-pause discontinuity approach as being broadly consistent with our baseline findings.
Covid intensity early in the pandemic may be associated with the effectiveness of PPP
loans at mitigating employment loss. To highlight this mechanism, we divide our sample between
firms located in New York and New Jersey, states that were impacted early in the pandemic, and
all other states. Figure A8 shows that first-quarter employment effects of PPP loans are larger in

29
New York and New Jersey, the states affected earlier and more intensely by Covid. Later-quarter
effects are similar when comparing these states and other states.
Finally, we replicate the threshold method from Autor et al. (2022, hereafter Autor et al.)
and find no statistically significant effect on wages with this method when using the matched tax
data. The threshold method compares firms just above and below the original firm-size cutoffs for
PPP eligibility. This approach theoretically provides a valid control group. However, many firms
with more than 500 employees received first-draw PPP loans and there was no take-up discontinuity
at this size threshold, as seen above in Figure A1. This resulted from various exemptions and likely
inconsistent enforcement of the threshold, meaning some firms in the control group received
treatment and this may have attenuated results. We also find evidence of misreported firm size in the
self-reported SBA data. Figure A9 (right side) shows clear evidence of firm size bunching just below
the 500 employee eligibility threshold that is not observed in the matched tax data (left side).
Specifically, for first-draw PPP applications in the SBA data, over 2,000 firms self-reported having
491 to 500 employees, while fewer than 500 firms reported having 481 to 490 employees.
Autor et al. estimated intent-to-treat effects for four employee ranges, comparing firms
within 50, 100, 150, and 250 employees of the eligibility threshold. They then averaged results
across the four employee ranges, resulting in their mid-May average intent-to-treat effect on
employment of 3 percent. Next, to account for take-up, they multiplied this intent-to-treat estimate
by two to yield average treatment on the treated (ATT). Their peak 6 percent ATT falls to 2.4
percent by the end of 2020. 26 The 6 percent estimate is essentially an average of ATTs across the
four employee ranges (50, 100, 150, and 250 employees) of 10, 7, 4, and 2 percent. Hence, the
effects differ by a factor of five. Note that the 50-employee range peak ATT of 10 percent is similar
to our peak ATT of 11 percent in Figure 3A.
We run the same regression except for three changes: we use quarters rather than weeks,
we use a base period of 2019 rather than Feb 2020, and we directly estimate ATTs rather than
intent-to-treat that require adjustments to get ATTs.27 Using the threshold method and linked tax
data, our average estimated peak effect is essentially zero. The lack of discontinuity at the 500-
employee threshold may contribute to the sensitivity of estimates to different employee ranges, as
well as attenuate results. To address this concern, we replicate the 50-employee range estimate but
add a 50-employee donut around the 500-employee threshold. This had no effect on our zero-
estimate result using the threshold method.

Table A4: Firm characteristics by quarter of treatment
Mean
Quarter of Mean
Employee
Treatment Wage
Count
Q2 2020 19 13,779
Q3 2020 12 11,390
Q1 2021 15 12,070
Q2 2021 32 11,572

26
Note that Autor et al. extend these ATTs to account for small-firm and post-2020 effects.
27
Our regression is: employment growth relative to 2019 average with respect to an indicator for receiving first-draw
PPP (at all), state-by-quarter fixed effect, industry-by-quarter fixed effect, and quarter-by-PPP fixed effect. Autor et
al. account for higher threshold in select industries. Our results may be a bit attenuated because we do not account for
higher thresholds in select industries.

30
Figure A5: PPP effects on firm-level employment and wages, only restaurants
Panel A: Employment changes Panel B: Wage changes

Figure A6: PPP effects on firm-level employment and wages, by calendar quarter
Panel A: Employment changes Panel B: Wage changes

Notes: See above. Source: Authors’ calculations using SBA and tax data.

31
Figure A7: PPP effects, including firms never receiving first-draw loans as controls
Panel A: Employment changes Panel B: Wage changes

Notes: See above. Source: Authors’ calculations using SBA and tax data.

Figure A8: PPP effects on firm-level employment, by NY-NJ and other states
Panel A: New York & New Jersey Panel B: All other states

Notes: See above. Source: Authors’ calculations using SBA and tax data.

32
Figure A9: Firm counts by reported firm size in PPP loan application
Panel A: Tax Data Panel B: PPP applications (SBA data)

Notes: Firm counts are for ten-employee wide bins. In first-draw PPP applications (SBA data) on the right, over 2,000
firms report being in the 10-employee range just below 500 employees (i.e., 491 to 500 employees), while fewer than
500 firms report being in the 10-employee range just below that (i.e., 481 to 490 employees). Source: SBA data and
authors’ calculations merging both SBA and tax data.

Figure A10: PPP forgiveness exclusion from taxation, 2020
Panel A: PPP forgiveness tax expenditure Panel B: PPP net of externalities and taxable forgiveness

Notes: An embedded net fiscal cost of the PPP is that forgiven PPP loans were excluded from firms’ taxable income,
even though the expenses that the loans funded were still deductible for tax purposes. This implies a tax expenditure.
We multiply 2020 first-draw loan forgiveness by average marginal tax rates of business owners to estimate a tax-
exclusion tax expenditure of $84 billion that is distributed as in Panel A. Average marginal tax rates are estimated for
owner profit increases of $25,000 using the simple tax calculator and weighting by ownership. A 21 percent tax rate is
applied to C corporations. Tax expenditures ignore responses that would decrease the estimate, such as any expected loan-
forgiveness underreporting if forgiven amounts were taxable. For income definitions, see Figure 4 notes.
Source: Authors’ calculations using SBA and tax data.

33
References
Auten, Gerald, and David Splinter. 2024. “Income Inequality in the United State: Using Tax
Data to Measure Long-term Trends.” Journal of Political Economy 132(7): 2179–2227.
Congressional Budget Office. 2022. “The Distribution of Household Income and Federal Taxes,
2019.” (supplemental tables) Congressional Budget Office. www.cbo.gov/system/files/2022 -
11/58353-supplemental-data.xlsx accessed on March 20, 2023.
Dalton, Michael. 2023. “Putting the Paycheck Protection Program into Perspective: An Analysis
Using Administrative and Survey Data.” National Tax Journal. Early access
https://doi.org/10.1086/724591.
Doniger, Cynthia L., and Benjamin Kay. 2023. “Long Lived Employment Effects Kakwani,
Nanak C. 1977. “Measurement of Tax Progressivity: An International Comparison.”
Economic Journal 87 (345): 71–80.
Lambert, Peter J. 1993. The Distribution and Redistribution of Income. (2nd ed.) Manchester:
Manchester University Press.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2016. “Income and Earnings Mobility in
U.S. Tax Data.” in Federal Reserve Bank of St. Louis and the Board of Governors of the
Federal Reserve System (Eds.) Economic Mobility: Research & Ideas on Strengthening
Families, Communities & the Economy, 481–516.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2021. “Household Incomes in Tax Data:
Using Addresses to Move from Tax Unit to Household Income Distributions .” with Jeff
Larrimore and Jacob Mortenson. Journal of Human Resources 56(2): 600–631.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2022. “Income Declines During
Covid-19.” AEA Papers and Proceedings 112: 340–344.
Larrimore, Jeff, Jacob Mortenson, and David Splinter. 202 3. “Earnings Business Cycles: The
Covid Recession, Recovery, and Policy Response.” Working paper.
Piketty, Thomas, and Emmanuel Saez. 2003. “Income Inequality in the United States, 1913 –
1998.” Quarterly Journal of Economics 118(1): 1–39. Updated estimates accessed from
https://eml.berkeley.edu/~saez/ on February 8, 2023.
Splinter, David. 2020. “U.S. Tax Progressivity and Redistribution.” National Tax Journal 73(4):
1005–1024.
Splinter, David. 2022. “Income Mobility and Inequality: Adult-Level Measures from the U.S.
Tax Data since 1979.” Review of Income and Wealth 68(4): 906–921.

File and source

File: the-ppp-progressivity-and-tax-effects.pdf
Size: 1,282,548 bytes
SHA-256: a34b7f92a09f0c96d5789778721d1437c04a53c25034e738a0d0a9698166555b
Our copy: the-ppp-progressivity-and-tax-effects.pdf
Original: davidsplinter.com