Ten Days Late and Billions of Dollars Short: The Employment Effects of Delays in Paycheck Protection Program Financing

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Ten Days Late and Billions of Dollars Short: The Employment Effects of Delays in Paycheck Protection Program Financing Cynthia L. Doniger and Benjamin Kay 2021-003 Please cite this paper as: Doniger, Cynthia L., and Benjamin Kay (2021). “Ten Days Late and Billions of Dollars Short: The Employment Effects of Delays in Paycheck Protection Program Financing,” FinanceandEconomicsDiscussionSeries2021-003. Washington: BoardofGovernorsofthe Federal Reserve System, https://doi.org/10.17016/FEDS.2021.003. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Ten Days Late and Billions of Dollars Short: The Employment Effects of Delays in Paycheck Protection Program Financing Cynthia L. Doniger Benjamin Kay Federal Reserve Board Federal Reserve Board January 15, 2021—Latest draft on SSRN Abstract Delay in the provision of Paycheck Protection Program (PPP) loans due to insufficient initial funding under the CARES Act substantially and persistently reduced employment. DelayedloansincreasedjoblossesinMayandpersistentlyreducedrecalls throughout the summer. The magnitude and heterogeneity of effects suggest significant barriers to obtaining external financing, particularly among small firms. Effects are inequitably distributed: larger among the self-employed, less well paid, less well educated and—importantly for the design of future programs—in very small firms. Our estimates imply the PPP saved millions of jobs but larger initial funding could havesavedmillionsmore, particularlyifithadbeendirectedtowardthesmallestfirms. About half of the jobs lost to insufficient PPP funding are lost in firms with fewer than 10 employees, despite such firms accounting for less than 20 percent of employment. JEL Classification: E24: Employment Unemployment Wages Intergenerational Income Distribution Aggregat•e Human Capital• Aggreg•ate Labor Productivity H81: Go•vernmental Loans Loan Gu•arantees Credits Grants Bailouts J21: Labor Force and Empl•oyment, Size, and S•tructure • • G32: Financing Policy Financial Risk and Risk Management Capital and Ownership Structure V•alue of Firms Goodwill • • • Keywords: Paycheck Protection Program, CARES Act, Countercyclical Fiscal Policy, Covid-19, Kurzarbeit, Income Support, Small Business Lending, Small and Medium Enterprises (SMEs), Financial Frictions TheviewsexpressedheresolelyreflectthoseoftheauthorsandnotnecessarilythoseoftheFederalReserve Board, the Federal Reserve System as a whole, nor of anyone else associated with the Federal Reserve System. We thank Aditya Aladangady, Byron Lutz, Brendan Price, David Ratner, Michael Strain, and Missaka Warusawitharana for valuable comments, Jonathan Kay for his assistance with address mapping with ArcGIS, and Arazi Lubis, Mathew Jacob, John Kadlick, and Elizabeth Duncan for their excellent research assistance. We also thank participants in seminars at the Federal Reserve for helpful comments.

1 Introduction In the United States, the economic devastation of caused by COVID-19 has hit smaller firms particularly hard. Unlike larger firms with their better access to credit and substantial cash reserves(HankinsandPetersen(2020)), smallerfirmstypicallyhavelessthanamonthofcash reserves to insulate them (Farrell and Wheat (2016)). The shock of COVID-19, combined with these limited financial reserves, imperils the jobs of the 60 million workers or 48 percent of the workforce (SBA (2018)) employed at smaller firms. The CARES act1 provided funding for loans to small businesses under the Paycheck Protection Program (PPP) in response to the COVID-19 pandemic and the economic havoc wreaked by non-pharmaceutical interventions to fight the pandemic. Traditional de facto and de jure American income support programs like unemployment insurance, stimulus grants, disability insurance, and the Supplemental Nutrition Assistance Program do not promote or preserve the employee-employer relationship. In contrast, the PPP set aside hundreds of billions of dollars to keep employees attached to their existing small business employers. This paper asks three main questions. First, how effective was the PPP in preventing job losses? Second, were jobs saved by the PPP long lasting or did they disappear when the funding was exhausted? Third, were the effects equally distributed? These research questions fit within a broad literature evaluating the efficacy of counter-cyclical fiscal policy. In addition, the unique design, and design failures, of the PPP program allow our results to speak to a broader literature on the employment effects of firm liquidity, particularly during times of economic crisis and within very small businesses. We make three contributions to answering these questions. First, we find that expedient PPP loan issuance causes an economically large and a statistically robust improvement in labor market outcomes, with particularly large effects for employees of small businesses. These employment benefits extend through September, which is after the three months of payroll support provided by the PPP loan are exhausted, suggesting the PPP promoted lasting firm health. The effects of delayed PPP loans are inequitably distributed, with selfemployed workers and workers at the smallest firms particularly hard hit. Second, the natural experiment we study illustrates the importance of cash on hand for small businesses’ employment decisions. It is well documented that small businesses use limited external financing (Vos (1992), Bitler et al. (2001), Mach et al. (2006)). However, there is also evidence from surveys of small businesses (Vos et al. (2007)) that, under normal conditions, 90 percent or more of such firms have sufficient external financing. Of course, the period of the US COVID-19 epidemic is not normal conditions, and this paper contributes 1TheCARESactisalsoknownasH.R.748andtheCoronavirusAid,Relief,andEconomicSecurityAct 2

to a growing literature establishing that limited cash-on-hand and working capital lowers firm labor demand and increases the sensitivity of labor demand to shocks (Vall´ee et al. (mimeo), Barrot and Nanda (2020), Bacchetta et al. (2019), Mehrotra and Sergeyev (2020), and Ghaly et al. (2017)).2 Third, we use these estimates to model the aggregate jobs lost by the ten day delay in the PPP, the number of jobs saved by the PPP, and the cost per job saved by the PPP. Our calculations imply that the PPP saved millions of jobs, and did so in a cost effective way, particularly because of the temporarily elevated replacement rate of unemployment insurance during the crisis. Further, our results suggest that a program focused on getting funding quickly to the smallest firms would have saved jobs more cost effectively. The PPP met unexpectedly high demand and exhausted the initial congressionally allocated $ 349 billion in funding on April 16th, less than two weeks after opening. Congress $ subsequently allocated an additional 321 billion in PPP funds, and the program resumed making loans on April 27th. To measure the effects of the PPP on employment, we exploit the ten-day delay between April 16th and 27th in the approval of PPP loans. We identify the effect of delayed financing by examining the impact of loan timing in the window of time from April 14th to 28th. In particular, we examine the labor market consequences of delaying a PPP loan by ten days using a difference-in-difference with a dynamic treatment effects estimation strategy. Our key identifying assumption is that borrowers approved on the 16th and the 27th are identical (conditional on controls) and that delay does not alter loan demand. To validate our assumptions, we document that while the characteristics of PPP loan recipients vary dramatically over the course of the program these trends halt during the ten days of loan delay around which we construct our event window. Our data and identification strategy enable us to document heterogeneity along key dimensions. In particular, we investigate the differential effects of the PPP by employer size and self-employment as well as by education and earning ability. A ten-day delay in funding PPP loans had substantial negative labor market consequences. We find strong and persistent effects of loan delay from May through September. For a core-based statistical area (CBSA)3 with 1 percentage point fewer delayed loans within our event window, the unemployment rate is 10 basis points lower, the labor force participation rate is 5 basis points higher, and a measure of the nonemployment rate4 is 14 basis points higher in May with effects remaining above 10 basis points through the summer. 2Important additional papers in this literature include Chodorow-Reich (2014), Benmelech et al. (2011), and Greenstone et al. (2020). Key theory papers include Wasmer and Weil (2004), Eckstein et al. (2019), Gertler and Karadi (2015), and Petrosky-Nadeau and Wasmer (2013) 3One or more adjacent counties anchored by a city-like area. 4We construct a broader notion of labor force participation that includes nonparticipants who were recentlyparticipating. Inourview, thisbroadernotionofnonemploymentmoreaccuratelyreflectschanging 3

These point estimates imply that if the PPP funding under the CARES Act had been 10 percent larger, enabling the share of delayed loans in our event window to be 20 percent lower, the unemployment rate would have been about 2 percent lower, non-participation about 1 percent lower, and our measure of non-employment about 2.8 percent lower. Delayed financing drives labor market divergence through a combination of differential job-loss and job-finding rates. In the spring, job losses are relatively more important (like Bacchetta et al. (2019)); however, over the summer persistent differences in job-finding, primarily driven by the absence of recall from layoff, accumulate to become the dominant driver of differentials (like Mehrotra and Sergeyev (2020)). These results are consistent with conceiving of business owners as risk averse entrepreneurs with risky income deriving from their small business and facing a cash-in-advance constraint that must be internally financed. Delay in receiving PPP loans forced business owners to run down their buffer stock of liquidity, driving them closer to the internally financed cash-on-hand constraint and thus convexifying their value function (Carroll and Kimball (2001)). As a result, recipients of delayed loans recover employment less quickly as they seek to rebuild their buffer-stock. There are important heterogeneities in these effects. Employment effects for employees at smaller firms and among the self-employed are larger, potentially because they have smaller cash reserves (Farrell and Wheat (2016)) and poorer access to credit (Berger and Udell (2002). These differences, which our identification strategy and data uniquely allow us to observe, reconcile our estimated effects, which are large, with those of prominent competing studies, which are small (Autor et al., 2020; Chetty et al., 2020). Indeed, our point estimates suggestthatnearlyhalfofthejobslostwerelostduetoinsufficientfundingundertheCARES Act and were in firms with fewer than 10 employees while such firms typically account for less than 20 percent of employment. Our analysis also reveals greater job loss among the self-employed than the employees of private firms. We also find greater losses among those predicted to have low earnings and those with less education. While these individuals’ unemployment insurance replacement rates were most greatly increased by the CARES Act and subsequent federal interventions in the unemployment insurance system, our findings regarding differential job loss for these groups are robust to detailed controls for these variations.5 Our results are robust to alternative definitions of the event window. A narrower window produces job losses over time that are more persistent and more back-loaded, while a slightly labor market conditions during the COVID-19 shock because the nature of the shock induces larger than typical flows out of the labor force. 5Supplemental regressions, available upon request. show no causal evidence that PPP loan delay and variation in the unemployment insurance system had any reinforcing or offsetting effects on the baseline PPP effect. 4

wider event window produces nearly identical results as our preferred specification. We also estimate an alternative, counter-factual event windows away from the 10-day delay. Loan delay outside the window is likely correlated with unobserved attributes that predict employment differentials before treatment. In particular, for alternative event windows we observe non-trivial pre-trends, indicative of non-random selection. In addition, our results are robust to a broad set of controls for industry, occupation, demographics, non-pharmaceutical interventions (NPIs), other fiscal stimulus (unemployment insurance in particular), survey design, and the ongoing evolution of the COVID-19 pandemic itself. Section 2 overviews the features of the PPP essential for our analysis. Section 3 discusses the sources of our data and how we aggregate it. Section 4 details our identification strategy, econometric specifications, and describes our model. Section 5 presents the results of our empirical and modeling analysis. Section 6 shows the robustness of ours results to narrower and wider windows around the ten day delay and shows the results in other problematic windows. Section 7 explains how our paper fits in with the related PPP and Kurzarbeit literature and lays out our thought experiment which estimates the total jobs saved and costs per job saved of the PPP. We conclude with Section 8. 2 The Paycheck Protection Program The PPP, created by the CARES Act, provided small businesses, small nonprofit and religious organizations, and sole-proprietors and independent contractors with loans to help their employees during the extreme COVID-19 related economic difficulties they faced during the spring and summer of 2020. Here we provide a brief summary of the parameters of the program as they applied to the loans in our studied time-frame.6 As our study focuses on the consequence of exhaustion of the PPP funding supplied by the CARES Act, we then provide detail of the timing exhaustion and related events. 2.1 PPP Design The PPP was run by the SBA in consultation with the US Department of the Treasury. Borrowers had to meet a multi-part test to be eligible for a PPP loan. However, unlike for virtually all other forms of credit, they were not tested for collateral or ability to repay and the paperwork requirements were modest (estimated by the SBA at about two hours). Once a borrower had their PPP application approved, their lender was supposed to send the funds 6In Appendix A we provide significantly more detail about program as a whole. 5

to the borrower within 10 calendar days.7 The qualifying loan amount per employee was 2.5 times the average total monthly per-employee payments for payroll costs for the year $ before the loan date (or, at the option of the borrower, for 2019) including up to 100,000 in annualizedcashcompensation(wages,salaries,andcashtips)plusgroupinsurancepremiums (including health care benefits). The PPP loans studied in this paper had a term of two years at a cost of 1.0 percent per year interest, but were forgivable if they were used for qualified expenses during a specified period. For the loans we focus on, borrowers were told they had 8 weeks after fund distribution to spend the proceeds on qualified expenses and at least 75 percent of funds had to be spent on qualified payroll expenses. 2.2 Exhaustion of the CARES Act Funding Allocation Table1presentsatimelineofthemajoreventsinthefirstfewweeksofthePPP.TheCARES $ Act was signed into law on March 27, 2020, and appropriated 349 billion in PPP loans. The first PPP loans were approved on April 3rd, 2020, and it was almost immediately clear that these funds were insufficient to meet PPP loan demand, with major political figures discussing in the business press that additional appropriations were needed. Press accounts of the early days of the PPP highlight inequities and start up pains as banks introduce PPP offerings: some banks are slow participate, with many prioritizing their existing customers and larger firms. PPP loan volumes were high but could have been much higher. Capacity constraints seemed binding and many ready, qualified, and willing borrowers could not get a loan. These early approved loans were unusually large, to unusually large firms, from unusually small banks. Figure 4, Panel E shows that the average loan in the first week was for about three times the amount and saved three times the jobs of typical loans throughout the program. The ten largest US commercial banks serviced one-fourth of the PPP loans over the life of the program but only 8 percent in the first week. By the second week in April, it was clear that the program was overwhelmed. PPP loan applications poured into the banks, and banks responded by continuing to prioritize larger and, especially, existing customers. Banks were also directed to do so by guidance from the Treasury and the SBA. Funding was being depleted rapidly (see Figure 1). Congress was working on a second round of PPP appropriations, and it looked likely that it would happen but it was not clear when the additional appropriations would become available. On April 16th, the CARES Act PPP funding was depleted. Congressional discussions of 7This timing changed somewhat over the program. Initial program documentation said that lenders had ten days to make their distributions, but was unclear on what this timing directive actually required and when the clock started ticking. Eventually the Treasury and the SBA issued a rule that for loans approved onorbeforeApril28th,2020,lendershad10calendardaysfromApril28th,2020tofundtheloan. Forloans after that point, the loan had to fund in 10 calendar days. 6

a second PPP appropriation were ongoing, resulting in a Senate bill on April 21st, a House bill on April 23rd, and the President signing the PPP and Health Care Enhancement Act of 2020 (PPP Act) into law on April 24th. The PPP Act appropriated an additional $ 321 $ billion (for a total of 670 billion) for PPP loans, and banks began issuing additional PPP loans on April 27th. Loan demand was strong in the initial two weeks of the PPP Act funding before becoming subdued through the remainder of the summer, as can be seen in Figure 1. Importantly, strong trends in borrowers’ observables abated in the period around the ten days where there were no PPP loans. These trends are illustrated in Figure 4 and discussed in greater detail in Section 4. In addition, while there clearly is discernible geographic variation in areas that received relatively more loans early on from the CARES Act funding and that received funds later from the PPP Act funding (Figure 6), the variation in earlier andlaterloanswithinthenarrowwindowsurroundingtheexhaustionofCARESActfunding and resumption of lending under PPP Act funding has no clear geographic concentrations (Figure 3). 3 Data This section describes our main data sources: the SBA’s PPP Loan Level Data and the Bureau of Labor Statistics’ (BLS)Current PopulationSurvey (CPS). In Appendix B we provide information about auxiliary data sources utilized as controls and in robustness checks. 3.1 SBA PPP Loan Level Data The U.S. Department of the Treasury and the Small Business Administration (SBA) provide comprehensive loan-level data on PPP loans in all states, major territories (American Samoa, Virgin Islands, Guam, and Puerto Rico), and the District of Columbia (U.S. Treasury (2020)). In the August release of PPP data we use, for loans for amounts less than $ 150,000, the data include the loan amount, city, state, ZIP Code, NAICS code (industry), business type (incorporation type or non-profit status), number of jobs retained, approval data, and lender (bank).8 For the larger loans ( $ 150,000 to $ 10 million), the exact loan $ $ $ amounts are replaced with loan ranges (specifically 5-10 million, 2-5 million, 1-2 million, $ $ 350,000-1 million, 150,000-350,000), which we map to the middle of their ranges (e.g., we $ $ treat a loan coded as 1-2 million we treat as a loan for 1.5 million). We aggregate loan activity by date and the county or CBSA (in general, county and 8The loan data have additional fields devoted to borrower race, gender, and veteran status, but these fields are overwhelmingly unanswered, missing for more than 90 percent of loans. 7

CBSA results are very similar but a CBSA better corresponds to a labor market and is our preferred level of aggregation). Neither county nor CBSA are fields in the PPP data. We use the HUD ZIP to FIPS and HUD ZIP to CBSA cross-walks respectively, to map each loan to a county and a CBSA (HUD (2020)). In some cases, the geographic areas of a ZIP Code is not contained entirely within a county (about 28 percent of ZIP Codes are in more than one county). More rarely, the geographic area of a ZIP Code is not contained entirely within a CBSA (about 17 percent of ZIP Codes are in more than one CBSA). In these cases, we map the ZIP Code to the county and CBSA that contains the plurality of the businesses of that ZIP Code. We want to minimize the measurement error introduced by these cross-walks. For the majority of the loans, we cannot check the resulting mapping. However, the larger loans ( $150,000) also report a street address in addition to State, City, and ZIP Code. Using ≥ ARC GIS (computerized mapping software), we map each one of these complete addresses (street information, state, city, and ZIP Code) to a county. In a comparison of the county mapped by our HUD cross-walk with the plurality of businesses approach to the ARC GIS approach, the results agree for more than 98 percent of these large loans, which gives us confidence that we are correctly attributing PPP loans to the county (and therefore CBSA) in which the borrowing firm operates. Table 2 reports basic summary statistics about the number of PPP loans, broken down by loan size. The PPP funded almost 4.9 million loans, of which 4.2 million were for less $ $ than 150,000. However, the roughly 650,000 (14 percent) PPP loans for 150,000 or more were awarded 74 percent of the funds. The distribution of PPP loans was very broad. Using the 2018 Census County Business Patterns to estimate the number if establishments, we find that in a typical county about 60 percent of small businesses received a PPP loan (Bureau (2020a)). To avoid selection issues induced by firm and managerial quality (much more on this in Section 4), we focus on the PPP loans issued from April 14th and 28th 2020 (the event window), about a third of the total. About 44 percent of the loans in this event window were late (approved on April 27th or 28th with PPP Act funding). Table 3 shows similar statistics of overall and late loan activity for the 20 most active CBSAs in the U.S. Treasury (2020) data. Even among large metropolitan areas there is considerable variation, with Miami having 48 percent of loans late in the window but Philadelphia having only 35 percent of loans late in the window. Across all CBSAs the variation is even larger, with a standard deviation of late loan share of about 10 percent. 8

3.2 Current Population Survey At the core of our analysis are the labor market outcomes for American workers, which we measure using microdata from the BLS’s CPS data from 2019 to 2020. As of the time of writing the most recent available microdata cover through September 2020. The CPS is a monthly rotating panel of approximately 65,000 U.S. households. Perhaps the most notable use of CPS micro-data is that the BLS use it calculate key official statistics, including the unemployment rate. The CPS is designed as short rotating panel with a core monthly survey administered to each household each wave. Each household is surveyed monthly for four consecutive months and then re-surveyed in the same four months one year later. Each monthly survey contains a core set of detailed questions about labor market activities during the week containing the 12th day of the month. These questions are used to classify each adult non-institutionalized, civilian into employed, unemployed, or not participating in the labor force. Linking the survey across waves enables us to observe whether an individual transits between these states. Inadditiontothevariablesneededtoconstructthenationalunemploymentrateandother aggregates, the core survey questions contain demographic and job characteristics such as industry, occupation, and class of worker (e.g. self-employed, private sector employee, public sector employee). In addition, each March, respondents report the size of their employer, inclusive of all of its establishments. Using the panel structure, we can link each response in December through June to the respondents employer size in the nearest March. Our empirical strategy, coupled with these data, allows us to identify the effects of the PPP on key sub-populations, namely employees of small firms and the self-employed that are either unmeasurable or unobserved or both in key complementary studies (Autor et al., 2020; Chetty et al., 2020). Additional details of the data, which we utilize in robustness checks, are discussed in Appendix B and include features related to unemployment insurance and the survey instrument’s sensitivity to the Covid-19 outbreak. 3.3 Geographic Merge Due to insufficient sample size, the CPS cannot be used to construct aggregate statistics at geographies smaller than a state. In addition, geographic information is suppressed in the micro-data when geographic identification would violate confidentiality requirements. In practice, this means that we can identify survey respondents’ county of residence for 281 counties. Just over 40 percent of the sampled population live in these counties. Grouping 9

counties into CBSAs improves geographic coverage since reidentification risk is reduced by pooling counties within a CBSA and thus a larger fraction of the sample resides in an identified CBSA. We can identify 258 CBSA in which 75 percent of the sampled population live. OurprimaryanalysismatchesloansintheSBA,NPI,andcase-countdatatoindividuals in the CPS according to CBSA.9 We match SBA loan data, to individuals residing outside of identified CBSAs according to their state excluding the identified areas, with a distinction between the rural and urban regions of the state.10 4 Identifying the Effects of Loan Delay We identify the effects of a delay in PPP funding on employment of current and recent employees of small firms. To measure the effects of PPP on employment, we exploit a ten day delay between April 16th and 27th in the approval of PPP loans caused by the exhaustion and replenishment of PPP funds. See Table 1 for an overview of the key events of the PPP as reported in the Wall Street Journal. The program met unexpectedly high demand and exhausted the initial appropriation of $ 349 billion on April 16th, less than two weeks after $ opening, but Congress subsequently appropriated an additional 320 billion in PPP funds, and the program resumed making loans on April 27th. In our view, due to bank processing constraints and other complications of participating in a novel program, firms did not have knowledge of the precise timing of the approval of their PPP loans. Whether a firm received an early or late PPP loan mostly reflected chance rather than selection (conditional on appropriate controls), allowing us to interpret the situation as a natural experiment on the effects of loan delay during an economic calamity. Identification requires two assumptions: Assumption 1. Delay in loan timing from the 16th to the 27th is independent of pre-existing firm attributes. 9Our general results are very similar when we repeat our analysis based on a county-level match. The county-level match affords greater geographic precision at the expense of sample size and a greater skew towardcountieswithgreaterpopulationdensity. Inaddition,inthecontextofourpaper,greatergeographic precision is not necessarily an advantage. Even if the county of residence of a respondent is disclosed in the CPS, the respondents relevant labor market is likely comprised of their county of residence and counties adjacent to or near where they reside. We view the CBSA as more appropriately capturing the relevant labor market. 10For example, Montana contains seven CBSAs but only a single identified CBSA in the CPS: Billings. First, we match outcomes for Billings CBSA residents in the CPS to PPP loans issued to firms in the Billings CBSA. Next, we match agglomerated outcomes for residents of the other six CBSAs in the CPS to agglomerated PPP loans across those six CBSAs. Finally, we match agglomerated outcomes for all other Montana residents (the rural ones) in the CPS to agglomerated loans to firms outside of the CBSAs. 10

More viable, productive, or better-run firms must not be more likely to get loans just before the April 16th exhaustion of initial PPP funds than other firms. If this assumption were badly violated, it would compromise our identification by conflating these selection effects with the effects of delay. Figure 4 plots the characteristics of the average CBSA receiving a PPP loan (CBSA characteristics weighted by share of loan dollars in that CBSA) by day from the advent of the PPP program until several weeks after our event window. These plots show clear trends in the characteristics of CBSA’ receiving loans over time. In particular, in the early days of the PPP, the places that received more loans had higher banking density and concentration. These places contained relatively more White people and were more likely to be rural. While there is a distinct time trend in these observables, the trend temporarily stabilizes across the 10-day gap of interest in our event window. This finding suggests that, while time trends exist over a longer horizon, our results are robust to spurious covariation with these observables. We conduct additional robustness checks of this assumption in Section 6. Meanwhile, as we discuss in Section 7, these trends cast doubt on the identification strategy of Granja et al. (2020). Assumption 2. Loan demand is not a function of loan timing. We investigate robustness to this second assumption in Section 6. Whether a firm gets early or late PPP funding (or none at all) is a firm-level characteristic. Ideally, we would have firm-level labor market outcomes for PPP-eligible firms matched to firm-level loan information (particularly if they received early or late PPP loans). Unfortunately, this is not possible in our data. The CPS does not capture whether individual firms utilized PPP loans, nor their timing if they did. To work around this data limitation, our analysis is at the CBSA level instead of firm level. We explore how the share of early loans across CBSAs drives CBSA labor market outcomes. To work at the aggregated level, we must further assume that loan delay is independently assigned after controlling for CBSA characteristics, In various specifications, we control for state-month fixed effects (capturing a variety of state level policies including most of the variation in NPIs), industrial mix, occupation mix, COVID-19 cases and deaths, county level NPIs, unemployment insurance replacement rates, and the CPS interview type. Our results are robust to all of these controls. The last day of loan-making under the initial appropriation was April 16th and the first day of loan issuance under the second appropriation was April 27th. Our preferred specification uses a measure of loan delay, share delayed, as the share of PPP loans issued to a CBSA in the window of April 14th to 28th on April 27th or 28th, which includes the last three days 11

of the original appropriation and the first two days of the second appropriation. Figure 1 depicts the cumulative loan issuance over the program’s life and the uptake during our event window. Our window is not symmetric because the data suggest that funding was exhausted mid-day on April 16th, and thus we use the last two full days under the first appropriation and the first two full days under the second appropriation. In a classic bias-variance tradeoff, the narrower the window, the more plausible that our results are uncontaminated by selection effects. However, the wider the window, the more precise our estimates are and the less measurement error we have in our proxy for financing delay at the county and CBSA level. Our preferred specification matches nicely with the information readily available to small business owners at the time, and largely reflects loans applied for before it became obvious that loan funds would be quickly exhausted. Table 1 shows an overview of the key PPP events and what potential PPP borrowers would have known about the program from reading the Wall Street Journal at the time. For those interested in who knew precisely what and when they knew it, Appendix D has a detailed timeline of news related to PPP and associated sources. In actuality, loan issuance spiked in the three days concluding with April 16th and in the two days beginning with April 27th. Figure 2 also shows that there was a large spike in loan demand around the April 17th-26th dates with no loans, and that loan issuance spike lasted longer than just one day on each side. This figure also shows that loan issuance on April 16th fell short of that on the 14thand 15th, indicating that funds were exhausted mid-day. In sensitivity analysis in Table 11, we find similar results for alternative event windows. Equation 1 shows our general linear probability model specification: 1 (cid:88) I [employed] = α+ 0[β I [month = m,year = 2020] sharedelayed + i,t m c { } m=1 I [month = m,year = 2020] controls Γ ]+ i,c,t m × × (1) 12 (cid:88) [ I [month = µ]controls Ψ ]+ i,c,t µ × µ=1 FE +ε i,t i,c,t where controls contains an indicator for census region; urban, suburban, rural; interview type; and a month fixed effect. Note that because the CPS is a location-based survey, c and i are co-linear and, as such, controlling for the individual fixed effect de facto controls for the CBSA fixed effect. We choose a linear probability model because our specification contains a 12

large number of fixed effects.11 Results are robust to adding additional fixed effects for twodigit NAICS industry and broad occupations, and state-by-time fixed effects; see Appendix C. We view state-by-time fixed effects as a flexible functional form that captures the effects of NPIs, public policy, and to some extent the effects of density, weather and climate, and use of public transportation. Our preferred specification controls for direct measures of these covariates, which, being more parsimonious, exhausts fewer degrees of freedom. The β (monthly sensitivity of the probability of employment to share delayed) are the m coefficients of interest. January 2020 is m = 0 and October 2020 is m = 9. Given the timing of the CPS reference week, April 2020 (m = 4) is a pre-event window observation. The coefficients β β trace out the evolving effect of loan delay through the summer and fall 5 9 − of 2020 while β β test for the presence of a pre-trend. We view β as comprising part of 1 3 4 − the pre-trend, but we refrain from interpreting this coefficient because of the coincidence of the CPS reference week and the exhaustion of the initial PPP funding allocation. This flexible specification allows us to test for heterogeneous effects by interacting share delayed or (sharedelayed) (month FE) with firm, worker, and neighborhood covari- × ates (e.g., race). In addition to using Equation 1 to measure employment sensitivities to loan delay, we can use the same specification to estimate the effect on labor force participation and flows into and out employment by replacing the left hand side variable with an indicator for participation, employment to non-employment transition, and so on. 5 Results 5.1 Employment We find that a 1 percentage point increase in loan delay increased unemployment by approximately 10 basis points in May (Table 5). The effect gradually fades through the summer. These effects are illustrated in Figure 5, Panel A.I, which plots coefficients β β and their 0 9 − 90, 95, and 99 percent confidence intervals from a model in which the only control is an indicator for the month interacted with the year 2020. Panel B.I plots the coefficients on the month 2020 fixed effects along with a 95 percent confidence interval, tracing out the × baseline COVID-19 shock. These coefficients are the model implied monthly unemployment 11Non-linear models such as probit and logit models may produce spurious results in a model with a large numberoffixedeffects,inparticular,whenthenumberofobservationsidentifyingaparticularconfiguration of fixed effects is small. 13

ratesforaCBSAwiththeaveragepercentageofdelayedloansof42. ThebaselineCOVID-19 shock hits in April and the delayed PPP shock hits in May, as expected.12 Changes in the unemployment rate do not capture the full employment effects of the COVID-19 shock because that shock also triggered flows from employment to out of the labor force at a rate unprecedented in past recessions. To present a fuller picture of the true employment effects, we expand our sample to include those workers recently observed in employment or unemployment.13 Panels A.II and B.II trace out the effect of delayed PPP financing and the COVID-19 shock on the labor force participation rate of the expanded sample. Both shocks increase non-participation with the COVID-19 shock’s effect leading the PPP delay effect by about one month. Panels A.III and B.III trace out the effect of delayedPPPfinancingandtheCOVID-19shockonthenonemploymentrateoftheexpanded sample of potential workers. As anticipated given heavy flows from employment to out of the labor force, the effects on nonemployment in May 2020 are about a third larger than the effects solely on unemployment. A thought experiment translates these regression coefficients into macroeconomic aggregates. Increasing the funding allocation under the CARES Act by 10 percent—an increase in $ the funding allocation in the CARES Act of about 35 billion—would have enabled reducing the delay in our event window by roughly 20 percent in all locations. This change would have boosted employment by nearly 2.8 million jobs. Further, this money would then not have needed to be allocated in the PPP Act.14 Our results imply that these effects persisted long after the PPP funds were spent and through the end of our observation in the fall. As noted in Section 6, these figures are robust to the width of the event window. These effects are enormous. To understand these estimates, it is helpful to relate them to the recent and closely related contribution of Barrot and Nanda (2020). Barrot and Nanda (2020) find a 5.7 percent increase in employment for a payment of 100 percent of payroll 15 days earlier. This figure rises to roughly 7.2 percent in a “slack” labor market.15 We find approximately a 14 percentage point drop in nonemployment that, given the nonemployment 12Recall that the CPS survey is on or near the 12th of the month, so the workers at firms getting PPP loans in our window (both early and late) did not have their funds when their employees were surveyed in April. Similarly, most firms did not experience the serious March COVID-19 shock before March 12th. Thus, in our data, the COVID-19 shock hits in April 2020 and the PPP loan delay shock hits in May. 13Specifically, we include all respondents who reported employment or unemployment in any of the potential seven CPS interviews before the month in question. Hall and Kudlyak (2019) find that past labor market experience is a better predictor of job finding than any standard BLS notion of marginally attached, and therefore we use this more accurate measure. 14For additional context, the average CBSA has a 44 percent share delayed with a standard deviation of 8 percent. 15Barrot and Nanda (2020) provide no information about the degree of slackness that they call “slack” or the semi-elasticity of their estimate to slackness, but it is safe to assume that May through October of 2020 is “slack” by any metric. 14

rateof21percentinAprilof2020,impliesapproximatelya17percentincreaseinemployment for a payment of 250 percent of payroll ten days earlier. Meanwhile, assuming, valiantly, that everything is linear the Barrot and Nanda (2020) numbers suggest a 12.16 percentage point increase in employment for a payment of a payment of 250 percent ten days earlier in a “slack” labor market. This result places our estimated employment effect only slightly above theirs and, referring to Barrot and Nanda (2020) Table V, suggests an external financing cost of approximately 50 percent in spring 2020. This amount is not above the external financing costs implied by the estimates of Barrot and Nanda (2020)—which they argue pertain in non-crisis times—in an economically significant way. Our estimates reflect the effect of earlier spending but not greater spending, which complicates the comparison with other papers that focus on the aggregate effects of spending regardlessoftiming. Forexample, mostassessmentsoftheAmericanRecoveryandReinvestment Act (ARRA) also find large employment effects, but for much greater funding outlays. $ Wilson (2012) finds that the entire the 262 billion in ARRA government spending created 2.1 million jobs, or a reduction in the unemployment rate of 1.4 percentage points. Feyrer $ and Sacerdote (2011) find that the 85 billion in local ARRA spending reduced unemploy- $ ment by 0.4 percentage points. Chodorow-Reich et al. (2012) found that the 88 billion Medicaid grant as part of the ARRA lowered the national unemployment rate by about 2.1 percentage points. Tables5showsthattheunemploymentandnonemploymenteffectestimateswithcontrols for industry fixed effects; occupational exposure to the COVID-19 shock; new COVID-19 casesanddeathsineachCBSAeachmonth; controlsforvariationinsurveynon-responsedue to inability to conduct in-person interviews in spring and early summer; the Oxford index for the severity of COVID-19 state level non-pharmaceutical interventions; and controls for the state-level timing of federally mandated changes to unemployment insurance interacted with workers’ eligibility for the programs. The effects depicted in A.I and A.III are robust. Columns (1) and (3) present the coefficients plotted in Figure 5, while columns (2) and (4) present the coefficients in the model with controls. Appendix C shows that the results are robust to inclusion of various subsets of these controls; controlling for occupation fixed effects rather than occupational exposure to COVID-19; and controlling for state-by-month fixed effects in lieu of measures of state-level NPIs, local outbreak severity, and state-level variation in implementation of federal changes to unemployment insurance. These jobs gains, which stem solely from the timing of PPP loan allocation, are large relative to the largest alternative estimate of the efficacy of the entire PPP program (Autor etal.(2020)). Atfirstglancethisresultseemsshocking; however, itiseasilyreconciledbythe heterogeneity in effects. Estimated marginal effects for small firms and the self-employed far 15

exceed those for firms near the 500 employee cutoff: the variation that underlies the Autor et al. (2020) identification strategy. Indeed, our point estimates suggest that more than half of the jobs lost were lost due to insufficient funding under the CARES Act were in firms with fewer than 10 employees. In normal times, such firms account for less than 20 percent of employment. 5.2 Job Flows The nonemployment effects of the PPP loan delay are driven by excess layoffs and labor market exit in May, combined with depressed recalls through the summer and early fall. Table 6 quantifies these results. Column (1) and (5) record the effects of loan delay each month on outflows to unemployment and inflows to employment. Columns (2)-(4) dis-aggregate the outflows into outflows to job seeking (excluding layoff), labeled “Separation”; to nonparticipation, labeled “Exits”; and to layoff. Layoffs are the primary driver of outflows, with exit a close second. Columns (6)-(8) dis-aggregate the inflows into inflows from job seeking (excluding layoff), labeled ”Accessions”; from non-participation, labeled ”Reentry”; and from layoff, labeled ”Recalls.” The absence of recalls is the primary driver of depressed inflows. Note, due to the identification of flows based on information collected from nonemployed individuals, outflows are roughly the sum of constituent parts while inflows are the weighted average. On average, outflows from employment are 4 percent and inflows to employment are 40 percent (Shimer, 2012). Volatility around these figures is 0.4 and 5 percent, respectively. Outflows and inflows reached their pre-COVID-19 peak and trough of 5.8 and 22 percent during the Great Recession.16 Compared with these historical figures, our marginal effects, which imply a 0.95 percent increase in job loss and a 2.15 percent decrease in job finding for a 4 percent increase in the CARES Act funding allocation, are very large. The shock to outflows from employment, while dramatic, is temporary. In contrast, the shock to inflows to employment is sustained (the coefficients through September are jointly statistically significant at the 1 percent level). Further, the May outflow from unemployment coefficient—coupled with the fraction of individuals in employment, unemployment, and nonemployemnt at the onset of the pandemic—implies that the May non-employment effects documented in the preceding subsection are accounted for by excess job-losses. However, as documented in Shimer (2012) and elsewhere, job finding in the United States is rapid and, as a result, spikes in job loss do not result in persistently elevated unemployment. Historically, elevated unemployment stems in large part from depressed rates of job finding. 16Note, all numbers quoted here are based on the quarterly averaged series to avoid excessive volatility. 16

We find a similar pattern in the response to the PPP funding delay shock. By August the nonemployment effects we observe are nearly fully accounted for by the accumulated effects of depressed job finding rates over the summer months. 5.3 Firm Heterogeneity Effects of PPP loan delay are heterogeneous with respect to employer type and size. Table 7 reports effects by the self-employed and the employees of private companies.17 As a robustness check, it also reports the effects on employees of local, state, or federal governments, which are ineligible for PPP loans. The self-employed are highly sensitive to the timing of financing. These workers experience the most dramatic non-employment immediately following PPP loan delay, recover during the eight-week term of employment specified for loan forgiveness, and then fall into non-employment again. By comparison, employees of private firms experience smaller but more persistent increases in non-employment. Statistically and economically insignificant coefficients for government employees serve to validate our empirical strategy, as these employees are ineligible for the program. Table 8 reports effects by size for the months in which employer size can be identified using the CPS.18 This analysis reveals that the aggregate effect is driven almost exclusively by the effect of PPP loan delay on the smallest firms. Of the 2.8 million jobs we estimate could have been saved by increasing the funding allocation in the CARES Act by 10 percent, 1.6 million—more than half—would have been in firms with fewer than 10 employees. For context, firms with fewer than 10 employees account for roughly 20 percent of employment in normal times. 5.4 Worker Heterogeneity The rich worker characteristic data in the CPS enables us to explore the effect of PPP loan delay on different kinds of workers. Table 9, Columns I - III show that workers in the lowest two terciles of predicted wages suffered the most from PPP loan delay, even after controlling 17Note,weclassifyincorporatedself-employedasself-employedratherthanemployeesofprivatecompanies, due to their treatment as self-employed under the PPP forgiveness criteria. This is contrary to the usual BLS convention. 18AsnotedinSection3.2employersizeisreportedbyCPSrespondentswhoweresurveyedinMarch. This rotatingdesignoftheCPSpanelmeansthatweareabletoidentifyemployersizeasofthemostrecentMarch for nearly all respondents in March, at most 3/4ths of respondents in April and February; 1/2 of the sample inMayandJanuary;and1/4thofthesampleinDecemberandJune. Inpracticethemonth-to-monthrecord linkage is closer to 70 percent (Nekarda, 2009). Thus, the precision of our coefficients is not equal across months, and June sample sizes in particular were the smallest and therefore the estimates are weakest. 17

for occupational exposure to the COVID-19 shock.19 Controls for occupational exposure are important because essential workers are paid less than non-essential workers (McNicholas and Poydock (2020)) and because essential workers were less likely to work for firms that ceased operating because of the COVID-19 shock. Columns IV - VI show the employment effects by worker education. The effects of loan delay also hit less-educated workers harder, again controlling for occupational exposure to the COVID-19 shock. These controls are important since less-educated workers make up nearly 70 percent of the essential workforce (McNicholas and Poydock (2020)).20 This finding provides a counterpoint to our results on employment flows. While the employment flows point to longer-term scarring effects of the delay in PPP funding as entrepreneurs rebuild buffers stocks, these results show that the burden of displacement has fallen on workers for whom the long term costs of displacement are smallest.21 6 Robustness 6.1 Event Window Figure 3 shows that the window around April 14th to 28th helps minimize the confounding spatial variation in our data.22 Figure 3 shows there is a high degree of variation in the timing of PPP loans within this window between adjacent counties such that there are very few clusters or other obvious geographic patterns. Other periods of PPP loan issuance suffer from pronounced geographic clustering. Figure 6 shows that there is a strong spatial correlation to the share of loans made to a county from the second pool of PPP funding (Panel A) and the second half of an arbitrary window in late April and early May (Panel B). These spatial correlations are part of a general pattern (though not during our window) where obtaining earlier loans is correlated with urbanity and geographic region (along with firm size and sophistication) in likely spurious ways. Table 10 contrasts our baseline unemployment results with the results of estimating them on the problematic alternative windows shown in Figures 3. In contrast with our instrument, 19We predict wages based on an augmented Mincer (1974) regression augmented to include industry, occupation, race/ethnicity, and sex fixed effects. We prefer terciles of imputed wages because they are available for all respondents. 20Insupplementalregressions,notreportedbutavailableuponrequest,wefindeconomicallyimportantbut statisticallyinsignificantevidencethatabilitytoworkfromhomeandemploymentinanessentialoccupation insulated workers from the effects of PPP delay, at least in the short run. 21Doniger(2019)findsthatmorehighlyeducatedworkersexperiencemorepersistentnegativewageeffects from displacement during a recession. 22Figure 4 shows that it also helps control for a number of loan, borrower, public policy, and firm characteristics that vary over the life of the program. 18

these correlations present spurious pre-trends (particularly for the case of the late April and early May window) that likely reflect differences in the quality of loan applicants. However, ourresultsarerobusttomakingtheeventwindownarrowerorwider(subjectto the usual unknown bias-variance trade-off of varying the window width). Table 11 contrasts our baseline unemployment results with the results of estimating them on these wider and narrower windows around April 17th to 26th. The baseline specification coefficients are economically similar and statistically indistinguishable from those estimated on the two alternative windows. Unsurprisingly, estimates of the jobs that would have been saved via a 10 percent increase in the initial allocation of PPP funding under the CARES Act are very similar across the three windows. Our preferred specification suggests jobs savings on the order of 3.5 million persisting through September. Alternative windows suggest nearly identical values. While the more broadly defined window suggests mildly lesser employment effects through the summer than our baseline the more narrowly defined window suggests employment effects more than 1.5 times as large. 6.2 Adjusting for Firm Failure Our baseline results assume that loan demand is invariant to PPP loan delay. However, firms that missed the first tranche of PPP funding may have failed or found alternative financing on less favorable terms. If so, the delay from the exhaustion of PPP funds would reduce PPP loan demand in the second round and bias our estimate of the share of loans delayed. Specifically, we measure: L ShareDelayed = (2) E +L where L is the volume of late window PPP loans on April 27th-28th, E is the volume of early window PPP loans on April 14th-16th, and E +L is the volume of PPP loans in the April 14th-28th window. However, allowing for failure and alternative financing, the share of loan demand that existed during April 14th-16th and was not met is L+M UnmetDemand = (3) E +L+M where M is the missing loan demand from firms that either fail or find alternative financing by April 27th and consequently never seek a PPP loan. Next we make assumptions to adjust for this unobserved shifting PPP loan demand to recover the true effect of insufficient funding under the CARES Act on employment and isolate the true effect of a 10-day delay in financing on survivors. 19

Assumption 3. Only a fraction S of loan demand not met by April 16th eventually seeks a PPP loan. With this assumption we can write L = (1 E)S and M = (1 E)(1 S). (4) − − − So we have: (1 E)S ShareDelayed = − E +(1 E)S − and solving for (1 E) gives − ShareDelayed UnmetDemand = 1 E = . − ShareDelayed(1 S)+S − Now, the linear probability model I [notemployed] =α+ i,c,t 8 (cid:88) [β I [month = m,year = 2020] UnmetDemand + m c { } m=0 I [month = m,year = 2020] controls Γ ]+ (5) i,c,t m × × 12 (cid:88) [ I [month = µ]controls XΨ ]+ i,c,t µ µ=0 FE + ε , i,c,t recovers the effect of loan delay on the nonemployment rate, accounting for firm failure. In simulations (not shown), we found that if enough firms fail as a result of the delay of PPP loans (and never get PPP loans or get them outside of our event window), this outcome can meaningfully bias our estimates even if all other identification assumptions are met. However, empirically, not enough US firms have failed to make this bias a problem. Bartik et al. (2020a) estimate that 100,000 US firms failed during this period.23 Bialik and Gole (2020) estimate essentially the same figure, with 97,966 permanently closed during this period. Thisnumbermightnotevenbeunusuallyhigh: theSBAestimatesthatabout45,000 smallfirmsclosemonthlyundernormalconditions(Marks(2020)). But, evenassumingthese 23They find that 100 of their surveyed 4,969 firms is permanently closed, which is equivalent to about 100,000 of the 4.6 million small businesses they sample from. 20

failuresareentirelyattributabletotheCOVID-19shock, theeffectsonourestimatesarevery small. Circa2017, theCensusestimatesthattherearesixmillionUSfirms(Bureau(2020b)). If all firm failures and permanent closures were caused by the PPP running out of funding, these figures imply that 1.7 percent of firms permanently closed as a result of the delay and never received PPP funding.24 A 2 percent bias is well below the threshold at which our estimates are biased in any meaningful way. Further, using county bankruptcy data obtained from RAND State Statistics (2020), we test if county-level loan delay is associated with changes in the fraction of businesses filings for bankruptcy in federal courts. We find no statistically or economically significant relationship between share delayed and increased incidence of any chapter of business bankruptcy. 7 Discussion Here we provide additional context for the results in Section 5. First, we compare our employment effects of PPP loan delay to other papers studying the efficacy of the PPP using other metrics and identification strategies. Second, we compare our findings with estimates of the benefits of the long established Kurzarbeit job programs in Germany. From all counter-cyclical jobs support programs in developed countries, the Kurzarbeit is the most similar in design to the PPP. Third, we use our estimates of the effect of a ten day delay to roughly estimate of the aggregate job effects of the PPP program. Fourth, we use our aggregate PPP job savings to estimate the cost per job saved of the PPP program and compare it with the stimulus literature. 7.1 Comparison with Other Studies WhenstudyingthelabormarketconsequencesofthePPP,otherauthorshavefoundavariety of effects and effect sizes, which vary considerably in their magnitudes and timing. On the lower end, Chetty et al. (2020) find almost no employment effect for the program (albeit with large standard errors). Granja et al. (2020) find no employment effects in April and very small effects in May and June, on the order of 0.8 percent to 1.3 percent of hours worked, 24We estimate the bias in our coefficients resulting from unobserved firm failure. We find that bias becomes a serious problem (coefficients off by more than 10 percent) only when the probability of firm failure conditional on loan delay is about 50 percent. Even if 100 percent of the firm failures were caused by loandelay(thatisallfailedfirmswerePPPeligibleandwouldhavebeeninoureventwindowwithoutfailure), and these failures were entirely in the smallest firms where we find the largest effects, these circumstances would only allow for a 13 percent failure rate. In our $150,000 or smaller loan category, there are 1.4 million in the event window, with 46 percent late or 644,000 late loans. If 100,000 firms fail and thus have no loan demand, the conditional failure probability is 100 / (100 + 644) or 13 percent. In our simulations, this calculation biases coefficients by less than 2 percent. 21

which is roughly equivalent to one to two million full time jobs. On the higher end, Autor et al. (2020) find the PPP saved 1.4 million to 3.2 million jobs—using data through June—by which point 98 percent of PPP loans had been issued. To put these figures into context, PPP loan participants said that the loans would support about 51 million jobs in aggregate and SBA (2018) estimates that about 59 million workers work at at firms with 500 or fewer employees. These papers use different data and vary in their econometric specifications and controls, and this contributes to differences between their estimates and ours. Granja et al. (2020), similarly to our paper, uses geographic variation in loan level timing to measure the effects of the PPP. They use the Homebase and Earning databases for their primary labor force analysis, which are opportunity samples of predominantly lower income workers who sign up using a cell phone app. The extent to which the geographic, education, race, and industry mixture of the workers in these databases matches the broader population is unknown. Selection into the use of these apps may have further unknown effects on the external validity of their results. In contrast, we use the CPS, is a large stratified random sample run by the Bureau of Labor Statistics and its methodology has been vetted over decades. In addition, as we see in Figure 6, there is a strong geographic component to the timing of loan funding, with Southwest, upper Midwest, and urban East Coast all substantially more delayed than the country as a whole. Larger firms were substantially less $ delayed. Forexample,55percentofPPPloansover 5millionwereinthefirstfundinground, $ but only 17 percent of loans under 150,000 were. Furthermore, as we showed in Section 3, the demographic, economic, educational, and geographic composition of the areas where PPP loans occur shifts considerably over the PPP sample. By comparing the counties by their exposure to the first and second PPP waves, Granja et al. (2020) assume that the ZIP Codes they study are otherwise comparable, conditional on their fixed effects specification. However, we show in Section 3 that the areas receiving early and late loans differ in several important areas that influence the changes (and not just the levels, which would be absorbed by fixed effects), and these differences may instead be driving their effects. Table 10, Column (2) shows estimates using our preferred CPS derived outcome variable with a first wave share instrument, similar to what is done in Granja et al. (2020). We obtain results similar to our baseline. However, in Table 10, Column (3) we show that loans outside our narrowly defined event window co-vary with unobserved covariates and deliver spurious pre-trends. This outcome illustrates why our event-window design is more robust. Instead of looking at cross county differences, Chetty et al. (2020) and Autor et al. (2020) use a difference-in-difference estimation strategy to compare the abnormal employment effects in earnings data of firms employing 100-500 employees (most of whom were PPP eligible) with firms employing 501-800 employees (generally PPP ineligible). Similar to 22

Granja et al. (2020), Chetty et al. (2020) use an opportunity sample of lower-income wage earners that makes it difficult to extrapolate to broader effects. Because larger firms secured PPP loans relatively early in the sample and the PPP likely saturated eligible demand (because it ultimately expired in August with excess funds), their estimates are contaminated by time-varying treatment in the two firm groups they study. Figure 4, Panel E shows that $ early in the program the average loan size was for about 500,000 and saved 35 jobs. In $ our study window, thse numbers are stable at about 200,000 and 10 jobs per loan, and in $ the later part of the sample they fall to 30,000 and three jobs per loan. At no point were the sorts of larger loans (to firms with 100 employees or more) considered by Chetty et al. (2020) a “typical” loan, and, indeed, there is evidence (along with the results in Figure 4) of big shifts in loan composition through the PPP. Autor et al. (2020) measure the effects of the PPP using payroll data from ADP, a payroll processing firm, and an intent-to-treat approach, comparing PPP-eligible and PPPineligible firms using industry-level size thresholds. This is also an opportunity sample and not designed to be statistically representative. For example, Cajner et al. (2018) note that theADPdataarerelativelyskewedtowardtheNortheastandtowardlargerfirms. According to Grigsby et al. (2019), only about half of firms use payroll processing services, and the ADP sample skews toward large (but not very large) firms, and pays hourly workers about 6 percent more than comparable workers in the CPS. We are particularly concerned about the use of ADP by very small firms. In a survey of firms with 20 or fewer employees, Surepayroll (2020) found only a third use payroll software (like ADP). The most significant drawback the approach of Chetty et al. (2020) and Autor et al. (2020), however, is that the exogenous variation—ineligibility of firms with more than 500 employees—that each exploits does not allow them to measure how PPP loans might vary in job-creating effectiveness by firm size, particularly for the smaller firms that received most of the loan dollars and indicated they would save most of the jobs. Our identification strategy and econometric specification allows us to test whether the employment effects on larger PPP-eligible firms are representative of overall PPP loan effects on smaller firms.25 Table 8 shows the effects on private sector employment by number of firm employees in May, where the large sample size of CPS workers observable pre- and post-PPP and identify firm size and thus allows the most precisely estimated comparison. The substantially larger coefficients for smaller firms show that the PPP had the biggest employment effects on the smallest firms, the very same ones for which Chetty et al. (2020) cannot estimate a causal effect. In addition, we report no effect–as expected–for firms larger than 500 employees.26 25Because we do not observe loan receipt, our estimates constitute intent-to-treat effects. 26TheCPSandthePPPdefinefirmsizedifferently. CPSrecordstotalemployeesacrossallestablishments 23

Hubbard et al. (2020) also compare firms above and below the 500 employee PPP loan cutoff and find large employment effects, about 1.8 percentage points, on employment. Their $ results, using Dun & Bradstreet data on firms with PPP loans of 150,000 or greater, do not use a narrow size window around the 500 cutoff, and therefore may include some of the smaller (and more employment sensitive) firms that drive our results. We view our paper as providing strong and precisely identified evidence that the employment effects they find are indeed large and robust. Like in our paper, Humphries et al. (2020) finds that early loans went to disproportionately larger firms and that early loan sizes were disproportionately large. Their paper uses an opportunity sample internet survey to study the effects of the PPP. Like in our administrative data, they also find an employment effect of PPP funding. Their results inform our thinking regarding the aggregate effects implied by our estimates. Our estimates show that small firms benefit the most from expediently delivered loans. Unfortunately, the roll out of the PPP favored expedient delivery of funds to larger firms. In our opinion, this substantively blunted the efficacy of the program. Similarly, Li and Strahan (2020) finds that proxies for relationship lending at banks (using a smaller banks, being in a bank’s core market, proximity to older branches, using a bank with higher levels of small business loans) predict which areas get relatively more early PPP loans. 7.2 Comparison with Kurzarbeit Programs By sharing the burden of keeping temporarily less-productive workers employed between employers and the state, the PPP has more in common with German Kurzarbeit programs than traditional US income support programs. Under the Kurzarbeit program, (literally German for “short work”), the German government compensates employers for most of the wage, pension, and insurance costs of keeping them employed during recessions. After the relativeout-performanceoftheGermanlabormarketrelativetopeersintheGlobalFinancial Crisis, Contessi et al. (2013) and Felter (2012) argued that a Kurzarbeit-style program could be cost-effective tool for fighting unemployment in recessions. In a calibrated model of the labor market effects of fiscal policy in Europe, Faia et al. (2013) find that hiring subsidies (like Kurzarbeit and the PPP) of 0.5 percent of GDP, about the same size as the PPP loans in our windows, could reduce unemployment by 1.8 percentage points. The United States had similar programs with limited availability in the great recession (17 states and about1percentofhoursofunemploymentclaimed), called“short-timecompensation”(STC) programs (Abraham and Houseman (2014)). Unlike the PPP, the Kurzarbeit program is whilePPPsizerestrictionsareappliedattheestablishmentlevelinkeyindustries. However,thisdiscrepancy would bias us against finding stronger effects at smaller establishments and, thus, reinforces our findings. 24

open to firms of any size, provides funding for much longer than three months, and does not pay the full wage and benefits costs of workers (Felter (2012)). Nevertheless, the PPP does represent the United States’ first national and widely available attempt to implement a Kurzarbeit-style program, and the results appear to be very successful and in line with theoretical predictions. 7.3 Estimates of Aggregate Jobs Saved by the PPP Where possible, the natural inclination in program evaluation is to attempt to find whether the benefits exceed the costs. Cost-benefit analysis of programs intended to save jobs, as in the case of the PPP, is difficult since jobs fit neatly neither in the costs nor the benefits column. Indeed, while the more salient features of jobs (e.g. salaries and benefits) are generally classified as costs, measures of job creation and preservation also proxy for less salient features that represent benefits (e.g. match quality). Thus, while PPP supported jobs are associated with benefits like increased productivity all we see clearly are the costs. 27 Still, creating and preserving jobs was the purpose of the PPP, so instead we consider a cost-effectiveness analysis: how much the PPP cost per job it saved.28 To estimate the cost per job saved necessitates estimating the jobs saved by the program as a whole. This section lays out the (perhaps valiant) assumptions necessary to translate our results in Section 5 into an estimate of jobs saved by the PPP and then calculates that estimate. Section 7.4 translates this estimate of the number of jobs saved into an estimate of costs per job saved (our measure of cost effectiveness) and compares that result with the cost-effectiveness of job-saving programs in the US financial crisis. First, the calculation of jobs saved requires assuming that our estimate of the cost in jobs of loan delay is invariant across the entire horizon of the PPP program. For several reasons, this external validity assumption is questionable. For one, the observed and unobserved attributes of PPP loan recipients are not time invariant. For another, the economic environment, and in particular the availability and interest rates of alternative loans or grants varied over time. We partially address this problem by allowing for different parameters for small and large firms in our aggregation. While imperfect, this solution partly addresses both concerns, as the average size of PPP recipients fell dramatically over time and access 27Labor not used in one endeavor can be used in another or consumed as leisure. Therefore, in a typical cost-benefit analysis, jobs and associated labor costs are in the cost part of the analysis. As we understand it, the jobs saved by the PPP are proxies for benefits we actually care about but cannot measure—stable productive matches between employees that produce significant output that can then be consumed. 28Cost effectiveness analyses are a common alternative to cost-benefit analysis (CBA). It is common to use cost effectiveness analysis when a full CBA is impossible, such as when there are measures of policy effectiveness (here the number of jobs saved) but the measures capture only part of the social benefits of a policy (Boardman et al. (2001)). 25

to alternative funding sources likely covaries with size. To implement this in our jobs saved figures, we construct our job-saving estimates by binned firm size, taking seriously the coefficients on delay obtained by firm size reported in Table 8, and then aggregate over all the firm size bins. Second, converting our point estimates, which measure jobs lost due to the delay in funding, into estimates of jobs saved due to the existence of PPP loans requires us to make an assumption about the amount of time required for a business to obtain funding from its next-best funding source relative to the time required to receive funding from the PPP. We know of no estimates of this time to alternative funding (alone or by firm size). Instead, we posit a difference in the timing of the receipt of funding from a PPP and the next best alternative of 15 days and assume that firms could have started either application on the opening date of the PPP program. Using these assumptions, we weight PPP loan amounts by hypothesized difference in the number of days between PPP loan receipt and the next best alternative. Using these weights, we cumulate delay weighted loans within firm size categories.29 This calculation suggests that the PPP program was equivalent to accelerating the delivery of funding to businesses with less than 10, 10 to 99, and 100-499 employees respectively by an equivalent of 8, 15 and 20 days respectively. Applying these factors to the coefficients on effect on nonemployment of a 10-day delay in funding implies that the PPP program preserved nearly 13 million jobs.3031 Of these 13 million nearly 9 million, or two thirds, were in firms with fewer than 10 employees. For context, firms of this size account for roughly 20 percent of employment in normal times. Meanwhile, these assumptions suggest approximately 1.3 million were in firms with 100-499 employees, which is comparable to the estimates of Autor et al. (2020) for PPP job savings for firms of this size. 7.4 Costs per Job Saved by the PPP $ $ Although the PPP was appropriated a total of 670 billion that ( 349 billion through the $ CARES Act and 321 billion under the PPP Act), not all funds were ultimately dispersed. 29Larger firms may need less time than smaller firms to line up alternative funding. Though larger firms havebetteraveragecreditmarketaccess,theyalsoneedmuchlargerloans,whereassmallerfirmsmaybeable to rely on faster retail credit products like home equity lines of credit and credit card loans. For simplicity, our primary estimate holds this delay to alternative funding sources fixed by firm size. 30Whenweinsteadassumethatittakes100-499employeefirms5days,10-99employeefirms10days,and lessthan10employeefirms15daystosecurealternativefunding,theresultsaresimilar(becausethelargest effects are in the smallest firms), with the program estimated to save 10 million instead of 12.7 million jobs. 31This calculation assumes that individuals who were nonemployed at the onset of the pandemic and subsequently found jobs as a result of the PPP program found them in proportion to historical levels of employment in firms of these sizes. 26

$ We calculate (because of binning) that the PPP funded 555 billion in loans. Because we estimate the PPP saved 13 million jobs (Section 7.3), we calculate a cost per job saved $ of about 43,000. Our PPP job-saving estimates is substantially larger than that of other papers, and so our estimate of the cost per job saved is also much lower. Chetty et al. (2020) $ estimates a cost per job saved of 289,000, and Autor et al. (2020) imply a cost per job $ $ saved of 173,000 to 396,000. However, our cost per job saved figure is very much in line with estimates of the job creation costs of the US stimulus programs of the financial crisis. Chodorow-Reich (2019) reviews seven studies of the American Recovery and Reinvestment $ Act (ARRA) and finds the ARRA programs save jobs at a cost of about 50,000 each, very much in line with our estimates. If anything, PPP loans might be expected to be even cheaper than these ARRA programs. The PPP is explicitly a job saving program, and it seems reasonable that such a program should save jobs at a lower cost (of course, other programs have other benefits). Given the estimates of the cost per job saved under the ARRA, the smaller employment results of other PPP papers are surprising. Estimates from $ the ARRA imply a typical government stimulus program of 555 billion should save about 11 million jobs. We estimate that the PPP saved 13 million jobs or about 15 percent more than expected from ARRA estimates, which seems reasonable, as, again, the PPP had a primary objective to save jobs. However, while asking “what is the gross-cost per job saved of the PPP?” is a well posed question, it not the right question, even when we are unable to measure the benefits of jobs saved by the PPP. Rather, the right question is, “what is the opportunity cost to the government of saving these jobs?” The CARES Act, like the Emergency Unemployment Compensation Act of 2008 (EUC 2008) passed by the United States during the US during the financial crisis, extended unemployment insurance by 13 weeks for a total in most states of 39 weeks. The federal government further enhanced the unemployment insurance program during the COVID-19 recession (also in the CARES Act) in two important ways above and beyondthechangesintheEUC2008. First,itexpandedthesetofworkerswhoareeligiblefor unemployment insurance to include self-employed and contract workers. Second, it increased $ $ all weekly benefits by 600, which alone is worth as much as 10,000 to a worker collecting it for the maximum of 3/27/20 to 7/25/20. Combined, this payout was a substantial increase in the generosity of the unemployment insurance system that gave three-fourths of workers eligible for unemployment insurance an income replacement rate above 100 percent (Ganong et al. (2020)). To the extent that the alternative to PPP funding is workers at small firms collecting unemployment insurance, these costs should be subtracted from the cost of the PPP to estimate a more accurate net-cost per job saved that better captures the opportunity cost 27

of the federal government of implementing the PPP. Within this institutional setting of a greaterthan100percentunemploymentinsurancereplacementrate, payingworkerstheirfull salaries may, even if they are completely unproductive for the spell of employment funded by the PPP loan, save the government money if the alternative is unemployment. In actuality, such workers are not completely unproductive, and a PPP-style program preserves quality employee-employer matches, both of which add to the benefits of the PPP.32 Thequantitativeeffectsaresubstantial. AtypicaljobfundedbyaPPPloan, asmeasured $ by the PPP data, pays 43,000 per year. The 3 month wage bill (the PPP loan size) for this $ worker is 11,000. Such a worker is likely eligible for a 145 percent unemployment replacement rate for about three months and regular unemployment benefits after that through the 39th week (Ganong et al. (2020)). If we assume that workers retained due to PPP loans are 50 percent as productive as during normal times (and their normal output is at least their compensation) and it otherwise takes the workers 6 months to find a new job with the same productivity as the job they would have kept with a PPP loan, we can estimate the net- $ $ opportunitycostofsavingthatjob. The 43,000costperjobsaved, lessthe 5,000costofsix months of unemployment insurance (3 months at a replacement rate of 145 percent (Ganong et al. (2020)) and 3 months under the normal average rate of 45 percent (Evermore (2019))) $ and 11,000 in additional output for the retained workers gives an opportunity cost per job saved of about $ 27,000.33 Given that, like in Kurzarbeit programs, these funds preserve workers’ idiosyncratic human capital and other valuable aspects of the employer-employee match, this solution is an attractive alternative to paying workers unemployment insurance, which effectively pays them to look for work, instead of the PPP, which pays them to do that work or at minimum remain productively matched in the face of transient shocks. 8 Conclusion Our estimates indicate a very substantial effect of PPP loan delay with the majority of the effect being concentrated in the smallest firms and in self-employed, lower-income, and lesseducated workers. In particular, we conclude that a more targeted program that got more 32This is true for any period where the unemployment insurance system is unusually comprehensive or generous. The gross-costs per job saved in ARRA estimates are probably also an overestimate of the true opportunitycostperjobsaved,becauseofthesesupplementstounemploymentinsurance,buttheCOVID-19 era changes were larger than those in the EUC 2008. 33If job losses would be for longer or the PPP retained workers were more productive, the opportunity cost perjob saved would be even lower. If workers would otherwiseremain unemployedfor nine months and the workers are fully productive when funded under the PPP, the social cost of the program per job saved is about$4,000,evenwithoutincludinganymultipliereffects. Byayeartofindanewjobandfullproductivity for PPP funded workers, the opportunity cost per job saved is negative (meaning the the program pays for itself). 28

money to the smallest firms faster could have delivered most of the job saving of the PPP as a whole for a fraction of the price. References Abraham, Katharine G, and Susan N Houseman. 2014. “Short-Time Compensation as a Tool to Mitigate Job Loss? Evidence on the US Experience During the Recent Recession.” Industrial Relations: A Journal of Economy and Society 53 (4): 543–567. Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B. Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz. 2020. “An Evaluation of the Paycheck Protection Program Using Administrative Payroll Microdata.” working paper, MIT. Bacchetta, Philippe, Kenza Benhima, and C´eline Poilly. 2019. “Corporate cash and employment.” American Economic Journal: Macroeconomics 11 (3): 30–66. Barrot, Jean-No¨el, and Ramna Nanda. 2020. “The Employment Effects of Faster Payment: Evidence from the Federal Quickpay Reform.” The Journal of Finance 75 (6): 3139–3173. https://doi.org/10.1111/jofi.12955. Bartik, Alexander W, Marianne Bertrand, Zo¨e B Cullen, Edward L Glaeser, Michael Luca, and Christopher T Stanton. 2020a. “How are small businesses adjusting to covid-19? early evidence from a survey.”Technical report, National Bureau of Economic Research, https://www.nber.org/papers/w26989.pdf. Bartik, Alexander W, Zoe B Cullen, Edward L Glaeser, Michael Luca, Christopher T Stanton, and Adi Sunderam. 2020b. “The targeting and impact of Paycheck Protection Program loans to small businesses.”Technical report, National Bureau of Economic Research. Benmelech, Efraim, Nittai K Bergman, and Amit Seru. 2011. “Financing labor.”Technical report, National Bureau of Economic Research. Berger, Allen N, and Gregory F Udell. 2002. “Small business credit availability and relationship lending: The importance of bank organisational structure.” The economic journal 112 (477): F32–F53. Bialik, Carl, and Daniel Gole. 2020. “Yelp: Local Economic Impact Report: September 2020.”Technical report, Yelp, https://www.yelpeconomicaverage.com/ business-closures-update-sep-2020.html. Bitler, Marianne P, Alicia M Robb, and John D Wolken. 2001. “Financial services used by small businesses: Evidence from the 1998 Survey of Small Business Finances.” Fed. Res. Bull. 87 183. Boardman, Anthony, David Greenberg, Aidan Vining, and David Weimer. 2001. Cost-Benefit Analysis Concepts and Practice Second Edition. Prentice Hall. 29

Bureau, US Census. 2020a. “County Business Patterns (CBP).” Electronic, June, https: //www.census.gov/programs-surveys/cbp.html. Bureau, US Census. 2020b. “SUSB Annual Data Tables by Establishment Industry.” Electronic, March, https://www.census.gov/data/tables/2017/econ/susb/ 2017-susb-annual.html. Cajner, Tomaz, Leland Dod Crane, Ryan Decker, Adrian Hamins-Puertolas, Christopher Johann Kurz, and Tyler Radler. 2018. “Using Payroll Processor Microdata to Measure Aggregate Labor Market Activity.”Technical report, Federal Reserve, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3106216. Carroll, Christopher D, and Miles S Kimball. 2001. “Liquidity Constraints and Precautionary Saving.” Working Paper 8496, National Bureau of Economic Research. 10.3386/w8496. Census Bureau. 2020. “2018 County Business Patterns.” Electric, June, https://www. census.gov/programs-surveys/cbp.html. Chetty, Raj, John N Friedman, Nathaniel Hendren, Michael Stepner, and The Opportunity Insights Team. 2020. “How Did COVID-19 and Stabilization Policies Affect Spending and Employment? A New Real-Time Economic Tracker Based on Private Sector Data.” Working Paper 27431, National Bureau of Economic Research. 10.3386/ w27431. Chodorow-Reich, Gabriel. 2014. “The employment effects of credit market disruptions: Firm-level evidence from the 2008–9 financial crisis.” The Quarterly Journal of Economics 129 (1): 1–59. Chodorow-Reich, Gabriel. 2019. “Geographic cross-sectional fiscal spending multipliers: What have we learned?” American Economic Journal: Economic Policy 11 (2): 1–34. Chodorow-Reich, Gabriel, Laura Feiveson, Zachary Liscow, and William Gui Woolston. 2012. “Does state fiscal relief during recessions increase employment? Evidence from the American Recovery and Reinvestment Act.” American Economic Journal: Economic Policy 4 (3): 118–45. Contessi, Silvio, Li Li et al. 2013. “Translating Kurzarbeit.” Economic Synopses 17. Doniger, Cynthia L.2019.“DoGreasyWheelsCurbInequality?”Technicalreport, Federal Reserve, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3362342. Eckstein, Zvi, Ofer Setty, and David Weiss. 2019. “Financial risk and unemployment.” International Economic Review 60 (2): 475–516. Evermore, Michele. 2019. “Are State Unemployment Systems Still Able to Counter Recessions?”Technical report, National Employment Law Project, https://www.nelp.org/ publication/state-unemployment-systems-still-able-counter-recessions/. 30

Faia, Ester, Wolfgang Lechthaler, and Christian Merkl. 2013. “Fiscal stimulus and labor market policies in Europe.” Journal of Economic Dynamics and Control 37 (3): 483–499. Farrell, Diana, and Chris Wheat. 2016. “Cash is King: Flows, Balances, and Buffer Days Evidence from 600,000 Small Businesses.”Technical report, JPMorgan Chase & Co Institute, https://institute.jpmorganchase.com/institute/research/ small-business/report-cash-flows-balances-and-buffer-days. Felter, Megan. 2012. “Short-Time Compensation: Is Germany’s Success with Kurzarbeit an Answer to US Unemployment.” BC Int’l & Comp. L. Rev. 35 481. Feyrer, James, and Bruce Sacerdote. 2011. “Did the stimulus stimulate? Real time estimates of the effects of the American Recovery and Reinvestment Act.”Technical report, National Bureau of Economic Research, https://www.nber.org/system/files/ working_papers/w16759/w16759.pdf. Ganong, Peter, Pascal J Noel, and Joseph S Vavra. 2020. “US Unemployment Insurance Replacement Rates During the Pandemic.”Technical report, National Bureau of Economic Research. Gertler, Mark, and Peter Karadi. 2015. “Monetary policy surprises, credit costs, and economic activity.” American Economic Journal: Macroeconomics 7 (1): 44–76. Ghaly, Mohamed, Viet Anh Dang, and Konstantinos Stathopoulos. 2017. “Cash holdings and labor heterogeneity: the role of skilled labor.” The Review of Financial Studies 30 (10): 3636–3668. Granja, Joao, Christos Makridis, Constantine Yannelis, and Eric Zwick. 2020. “Did the Paycheck Protection Program Hit the Target?” Working Paper 27095, National Bureau of Economic Research. 10.3386/w27095. Greenstone, Michael, Alexandre Mas, and Hoai-Luu Nguyen. 2020. “Do Credit Market Shocks Affect the Real Economy? Quasi-experimental Evidence from the Great Recession and” Normal” Economic Times.” American Economic Journal: Economic Policy 12 (1): 200–225. Grigsby, John, Erik Hurst, and Ahu Yildirmaz. 2019. “Aggregate nominal wage adjustments: New evidence from administrative payroll data.”Technical report, National Bureau of Economic Research, https://www.nber.org/papers/w25628. Hale, Thomas, Sam Webster, Anna Petherick, Toby Phillips, and Beatriz Kira. 2020.“OxfordCOVID-19GovernmentResponseTracker.”https://github.com/OxCGRT/ covid-policy-tracker/. Hall, Robert E, and Marianna Kudlyak. 2019. “Job-Finding and Job-Losing: A Comprehensive Model of Heterogeneous Individual Labor-Market Dynamics.” Working Paper 25625, National Bureau of Economic Research. 10.3386/w25625. 31

Hankins, Kristine W., and Mitchell Petersen. 2020. “Why Are Companies Sitting on So Much Cash?” Harvard Business Review 1 (17): . Hubbard, Glenn, Michael R Strain et al. 2020. “Has the Paycheck Protection Program Succeeded?” In Brookings Papers on Economic Activity Fall Conference ‘COVID-19 and the Economy’, Online, Volume 24. HUD. 2020. “HUD USPS ZIP Code Crosswalk Files.” Electronic, June, https://www. huduser.gov/portal/datasets/usps_crosswalk.html. Humphries, John Eric, Christopher A Neilson, and Gabriel Ulyssea. 2020. “Information frictions and access to the Paycheck Protection Program.” Journal of public economics 190 104244. Jackson, Osborne. 2020. ““An Approach to Predicting Regional Labor Market Effects of Economic Shocks: The COVID-19 Pandemic in New England,” Current Policy Perspectives..”Technical report, Federal Reserve Bank of Boston, https://www.bostonfed.org/publications/current-policy-perspectives/ 2020/the-covid-19-pandemic-in-new-england.aspx. JHU CSSE. 2020. “COVID-19 Data Repository by the Center for Systems Science and Engineering(CSSE)atJohnsHopkinsUniversity.”Electronic, October, https://github. com/CSSEGISandData/COVID-19. Keystone Strategy.2020.“CoronavirusCityandCountyNon-PharmaceuticalIntervention Rollout Date Dataset.” Electronic, June, https://github.com/Keystone-Strategy/ covid19-intervention-data. of Labor, Department. 2020. “State Law Information.” Electronic, October, https:// oui.doleta.gov/unemploy/statelaws.asp. Leibovici, Fernando, Santacreu, Ana Maria, and Matthew Famiglietti. 2020. “Social Distancing and Contact-Intensive Occupations.”Technical report, Federal Reserve Bank of St. Louis, https://www.stlouisfed.org/on-the-economy/2020/march/ social-distancing-contact-intensive-occupations. Li, Lei, and Philip E Strahan. 2020. “Who Supplies PPP Loans (And Does it Matter)? Banks, Relationships and the COVID Crisis.” NBER Working Paper (w28286): . Mach, Traci L, John D Wolken, Courtney M Carter, John A Holmes, and Lieu N Hazelwood. 2006. “Financial services used by small businesses: Evidence from the 2003 Survey of Small Business Finances.” Federal Reserve Bulletin A167. Marks, Gene. 2020. “Coronavirus is forcing 100,000 small firms out of business? That needs a closer look.” The Guardian 5 (21): , https://www.theguardian.com/business/ 2020/may/21/coronavirus-forcing-small-firms-out-of-business-analysis. 32

McNicholas, Celine, and Margaret Poydock. 2020. “A comprehensive look at their wages, demographics, and unionization rates.”Technical report, Economic Policy Institute, https://www.epi.org/blog/ who-are-essential-workers-a-comprehensive-look-at-their-wages-demographics-and-unionization-rates/. Mehrotra, Neil, and Dmitriy Sergeyev. 2020. “Financial shocks, firm credit and the Great Recession.” Journal of Monetary Economics In Press. Mincer, Jacob. 1974. Schooling, Experience, and Earnings. Human Behavior & Social Institutions No. 2.. ERIC. Nekarda, Christopher J. 2009. “A longitudinal analysis of the current population survey: Assessing the cyclical bias of geographic mobility.”Technical report, Federal Reserve. Nunn, Ryan, Jana Parsons, and Jay Shambaugh. 2020. “Incomes have crashed. How much has unemployment insurance helped?”Technical report, Brookings, https://www.brookings.edu/blog/up-front/2020/05/13/ incomes-have-crashed-how-much-has-unemployment-insurance-helped/. Petrosky-Nadeau, Nicolas, and Etienne Wasmer. 2013. “The cyclical volatility of labor markets under frictional financial markets.” American Economic Journal: Macroeconomics 5 (1): 193–221. RAND State Statistics. 2020. “Bankruptcies and Bankruptcy Filing Rates.” https:// randstatestats.org/us/stats/bankruptcies-and-bankruptcy-filing-rates.html. SBA. 2018. “2018 Small Business Profile.”Technical report, Small Business Administration, https://www.sba.gov/sites/default/ files/advocacy/2018-Small-Business-Profiles-US.pdf?fbclid= IwAR2mlIFIyWGX0RwJtDR6CDT21cpreCyXtS15N2XJgHNlJ7n8SVYKBcxfd7U. SBA. 2020. “Table of size standards: The SBA’s table of small business size standards helps small businesses assess their business size.”Technical report, Small Business Administration, https://www.sba.gov/sites/default/files/2019-08/SBA%20Table% 20of%20Size%20Standards_Effective%20Aug%2019%2C%202019.xlsx. Shimer, Robert. 2012. “Reassessing the ins and outs of unemployment.” Review of Economic Dynamics 15 (2): 127–148. $ Singh, Andy. 2020. “When Will 300 LWA Unemployment Start Being Paid in My State? News and Updates.”Technical report, $ aving to Invest, https://www.savingtoinvest. com/when-will-300-lwa-unemployment-start-being-paid-in-my-state/. Surepayroll. 2020. “The Hiden Impact of Payroll.”Technical report, Surepayroll, https://www.surepayroll.com/globalassets/documents/nocrawl/surepayroll_ taxesandregulations_datareport_v3.0.pdf. 33

U.S. Treasury. 2020. “SBA Paycheck Protection Program Loan Level Data.” Electronic, August, https://home.treasury.gov/ policy-issues/cares-act/assistance-for-small-businesses/ sba-paycheck-protection-program-loan-level-data. Vall´ee, Boris, Julien Sauvagnat, Thorsten Martin, and Jean-Noel Barrot. mimeo. “Employment Effects of Alleviating Financing Frictions: Worker-level Evidence from a Loan Guarantee Program.” ”working paper, february 2020.”. Vos, Ed. 1992. “Unlisted businesses are not financial clones of listed businesses.” Entrepreneurship Theory and Practice 16 (4): 57–68. Vos, Ed, Andy Jia-Yuh Yeh, Sara Carter, and Stephen Tagg. 2007. “The happy story of small business financing.” Journal of Banking & finance 31 (9): 2648–2672. Ward, Jason, and Kathryn Anne Edwards. 2020. “Statistics in the Time of Coronavirus: COVID-19-related Nonresponse in the CPS Household Survey.”Technical report, Rand, https://www.rand.org/pubs/working_papers/WRA842-1.html. Wasmer, Etienne, and Philippe Weil. 2004. “The macroeconomics of labor and credit market imperfections.” American Economic Review 94 (4): 944–963. Wilson, Daniel J. 2012. “Fiscal spending jobs multipliers: Evidence from the 2009 American Recovery and Reinvestment Act.” American Economic Journal: Economic Policy 4 (3): 251–82. 34

Table 1: Timeline of Major PPP Events Period MajorEvents PPPPassedIntoLawaspartoftheCARESACT(3/27). March27thtoApril6th SenatorRubioannouncesPPPexpectedtorunoutinlateMayandwill requiremoreappropriations. Congressexpressesintenttoappropriate$250billioninadditionalPPP fundingbyendofweek. April7thto13th SBAloaninfrastructurecannotprocessvolume,experiencingnotablecrashes. Bankslimitloansprimarilytofirmswithexistingcreditrelationshipsdueto availabilityoffundingandprograminterest. BusinessesstillreportsignificantdelaysinreceivingPPPfunds. Somebanksstoptakingnewapplicationsinanticipationoffundingexhaustion. April14thto20th HeavyPPPissuanceuntilprogramexhaustsinitialappropriation(4/16). CongressionalnegotiationsoversecondPPPappropriationsresume. PoliticaldiscussionsoversecondPPPappropriationscontinue. Billinthehouseandsenate,thensignedintolaw(4/24). April21thto30th PPPresumesmakingloansandathighvolumes(4/27). SBAencouragessmallbusinessestoapplyforPPPfundsassoonaspossible. MajorslowdowninPPPloans. CongressexploreschangingPPPloanrequirementstogivemoretimeand May flexibilitytospendfunds. SenateadjournsforrecesswithoutextendingtermforPPPloans. June InitialendofPPP(6/30). PPPtermsadjustedinlawtoextendloanforgivenesstimelinefromeightto twentyfourweeks. July Reducespayrollrequirementstoallow60%oftheloantogotowardspayroll, comparedtothepreviousrequirementof75%. ExtensionofPPPloandeadline(8/8). August PPPloanprogramexpires(8/8). Sources: VariousWSJarticlesin2020. 35

Table 2: Summary Statistics of PPP Loans by Headcount and Loan Size. Loan Loan Loan Jobs Window† Window† Late‡ PPPAct§ Size (#) ($) Retained Loans($) Loans(#) Loans(%) Loans(%) A)[$0,$150k] 4,224,172 141,804,112,463 19,669,493 42,537,113,452 1,124,214 56 27 B)($150k-$350k] 379,062 75,812,379,056 8,727,023 24,420,472,101 122,102 44 32 C)($350k-$1M] 199,456 134,632,999,456 10,015,580 42,439,337,873 62,873 40 32 D)($1M-$2M] 53,030 79,545,053,030 5,906,794 24,027,016,018 16,018 39 30 E)($2M-$5M] 24,838 86,933,024,838 5,167,844 25,935,007,410 7,410 38 30 F)($5M-$10M] 4,840 36,300,004,840 1,639,326 11,797,501,573 1,573 41 32 AllLoanSizes 4,885,398 555,027,573,683 51,126,060 171,156,448,427 1,334,190 54 27 Loan Loan Jobs Window† Window† Late‡ PPPAct§ Headcount* (#) ($) Retained* Loans($) Loans(#) Loans(%) Loans(%) 1)[0,9] 3,630,053 96,520,713,219 11,768,615 28,125,628,870 939,940 57 26 2)[10,99] 1,172,183 286,449,242,433 29,412,071 90,865,561,688 370,358 46 32 3)[100,499] 79,380 158,655,830,222 15,001,340 47,529,721,213 22,897 40 29 4)[500] 3,782 13,401,787,809 1,891,000 4,635,536,656 995 47 26 AllHeadcount 4,885,398 555,027,573,683 58,073,026 171,156,448,427 1,334,190 54 27 Source: SBA/Treasury(2020) * - Actual headcount (“jobs retained” in the SBA / Treasury data) if headcount>0 or loansize/headcount< 40,000, which is 81 percent of the observations (reasonable jobs saved sample). Where headcount is missing, reportedaszero,ortheaverageloansizeperemployeeiswelloutsidethePPPloanparameters,weimputethe headcount with a linear model calibrated on the reasonable jobs saved sample using an OLS regression of the of logjobsretained on logloansize and dummy variables for major industries. Within this sample, the model predictsthereportedemployeesizegroupsinTable2withabout90percentaccuracy. †-Loansissuedonorbetween4/14/20and4/28/20. ‡-Theshareofloansonorbetween4/14/20and4/28/20issuedon4/27/20or4/28/20. §-Theshareofloansissuedonorafter4/17/20. 36

.scitsitatS naoL ASBC 02 poT :3 elbaT ‡snaoLetaL% †snaoLwodniW ASBC deniateRsboJ eziSnaoL .gvA sralloDnaoL tnuoCnaoL 79.74 374,92 LF,hcaeBmlaPtseW-eladreduaLtroF-imaiM 888,761,1 745,48 182,412,740,31 913,451 1 36.74 87633 XT,notgnilrA-htroWtroF-sallaD 755,442,1 217,121 731,219,335,31 691,111 2 63.74 658,09 AP-JN-YN,ytiCyesreJ-kraweN-kroYweN 441,492,3 512,621 561,859,025,44 937,253 3 30.74 392,52 AG,llewsoR-sgnirpSydnaS-atnaltA 686,909 950,201 312,758,964,01 685,201 4 53.64 73941 ZA,eladsttocS-aseM-xineohP 240,696 323,811 173,524,540,7 445,95 5 07.54 012,35 AC,miehanA-hcaeBgnoL-selegnAsoL 080,065,2 607,711 392,049,723,82 666,042 6 40.54 77691 IM,nrobraeD-nerraW-tiorteD 875,507 633,441 655,737,481,8 607,65 7 57.44 268,82 XT,dnaLraguS-sdnaldooWehT-notsuoH 792,903,1 900,911 581,486,499,11 887,001 8 43.44 00141 OC,doowekaL-aroruA-revneD 832,515 687,911 056,706,552,6 322,25 9 91.44 09561 AW,euvelleB-amocaT-elttaeS 047,115 368,141 898,107,930,8 276,65 01 47.34 81911 AC,dabslraC-ogeiDnaS 695,046 537,121 085,959,562,6 274,15 11 36.34 785,41 LI-OM,siuoL .tS 126,974 586,821 119,393,064,5 234,24 21 44.34 146,81 IW-NM,notgnimoolB-luaP .tS-silopaenniM 479,517 997,941 791,349,345,8 630,75 31 93.34 975,12 AC,drawyaH-dnalkaO-ocsicnarFnaS 326,078 645,841 966,869,202,21 941,28 41 98.14 947,11 AW-RO,orobslliH-revuocnaV-dnaltroP 069,373 003,921 391,743,210,5 567,83 51 54.14 102,82 VW-DM-AV-CD,airdnaxelA-notgnilrA-notgnihsaW 784,699 049,541 495,328,691,31 624,09 61 47.83 063,34 IW-NI-LI,niglE-ellivrepaN-ogacihC 888,907,1 538,031 000,777,769,91 716,251 71 65.73 181,33 HN-AM,notweN-egdirbmaC-notsoB 280,528 538,041 927,743,919,11 336,48 81 01.53 516,23 DM-ED-JN-AP,notgnimliW-nedmaC-aihpledalihP 933,268 758,731 579,029,469,11 297,68 91 47.43 768,31 DM,noswoT-aibmuloC-eromitlaB 184,064 328,631 424,743,59,35 334,93 02 ))0202(yrusaerT.S.U :ecruoS( .02/82/4dna02/41/4neewtebronodeussisnaoL-† .02/82/4ro02/72/4nodeussi02/82/4dna02/41/4neewtebronosnaolfoerahsehT-‡ 37

Table 4: Comparison of Counties by Share of Delayed PPP Loans. CharacteristicsofCountiesbyShareofPPPLoansDelayed. BelowMedian AboveMedian Above-Below Mean Std. Dev. Mean Std. Dev. Difference t-stat Demographics RuralCounty 0.48 0.50 0.69 0.46 0.22 10.70 FractionBlackResidents 0.10 0.14 0.08 0.15 -0.01 -2.01 FractionWhiteResidents 0.75 0.19 0.75 0.21 0.01 0.75 FractionHispanicResidents 0.09 0.13 0.10 0.14 0.01 1.97 Republican2016Pres. VoteShare 0.65 0.16 0.68 0.17 0.03 4.81 PerCapitaIncome 26,888 6,923 25,851 6,248 -1,037 -3.75 Population(Non-institutional) 140,982 335,878 98,344 417,790 -42,638 -2.68 LogPerCapitaIncome 10.17 0.24 10.13 0.24 -0.04 -3.63 LogPopulation(Non-institutional) 10.53 1.63 9.78 1.54 -0.75 -11.30 FractionResidents<H.S.Edu. 0.14 0.06 0.14 0.07 0.00 1.37 FractionResidentsH.S.orSomeCollege 0.55 0.08 0.57 0.07 0.02 5.56 FractionResidentsCollegeDeg. orMore 0.31 0.11 0.29 0.09 -0.02 -4.89 Frac. WorkerswithEmployerInsurance 0.31 0.07 0.30 0.06 -0.02 -5.89 FractionSelf-employedWorkers 0.11 0.04 0.13 0.06 0.02 7.88 FractionPrivateWorkers 0.65 0.08 0.62 0.09 -0.03 -8.00 FractionPublicSectorWorkers 0.16 0.06 0.18 0.07 0.01 5.21 Banking Branchesper100kPopulation 7.65 15.60 12.57 19.55 4.92 6.60 Depositsper100kPopulation 304,949 696,544 464,737 634,684 159,788 5.69 CountyHHI(BranchShare) 2,667 2,171 3,188 2,307 521 5.52 CountyHHI(DepositShare) 3,234 2,191 3,746 2,296 512 5.42 COVID-19 Casesper100kPopulation(4/11/20) 62 118 63 176 1 0.18 Deathsper100kPopulation(4/11/20) 2 5 2 10 0 1.32 NPI:SchoolClosure(4/11/20) 1.00 0.00 1.00 0.00 0.00 . NPI:AnySerious(4/11/20) 1.00 0.00 1.00 0.00 0.00 . Casesper100kPopulation(9/11/20) 1,593 1,330 1,511 1,369 -82 -1.45 Deathsper100kPopulation(9/11/20) 38 44 37 54 -1 -0.34 Business Bankruptcies Chapter7per100kPopulation(12/31/19) 4.01 8.02 3.65 8.21 -0.36 -1.05 Chapter11per100kPopulation(12/31/19) 1.19 8.59 0.92 4.19 -0.27 -0.94 Chapter13per100kPopulation(12/31/19) 0.63 2.36 0.64 3.08 0.01 0.07 Chapter7per100kPopulation(6/30/20) 3.40 6.29 3.55 8.35 0.15 0.48 Chapter11per100kPopulation(6/30/20) 0.95 3.66 0.73 2.53 -0.22 -1.65 Chapter13per100kPopulation(6/30/20) 0.62 3.63 0.60 3.03 -0.02 -0.14 Observations 1139 1138 2277 Sources: CensusBureau(2020),KeystoneStrategy(2020),JHUCSSE(2020) RANDStateStatistics(2020),MITElectionDataandScienceLab(2018)andSBA(2020). 38

Table 5: Employment Effects of Delay PanelA: PanelB: (p.p,Unemployment (p.p,Nonemployment perp.pDelay) perp.pDelay) Sharedelayed×Monthin2020× January -0.00912 -0.0108 -0.0138 -0.0150 (0.0154) (0.0154) (0.0217) (0.0215) February -0.0193 -0.0224 -0.0155 -0.0179 (0.0191) (0.0189) (0.0251) (0.0245) March 0.00459 -0.00432 0.0227 0.0121 (0.0231) (0.0226) (0.0286) (0.0276) April -0.00208 -0.0399 0.0112 -0.0453 (0.0541) (0.0429) (0.0604) (0.0461) May 0.116** 0.105*** 0.157*** 0.141*** (0.0503) (0.0387) (0.0553) (0.0421) June 0.115** 0.0932** 0.123** 0.108*** (0.0499) (0.0372) (0.0545) (0.0405) July 0.113*** 0.0827** 0.115** 0.0827** (0.0439) (0.0347) (0.0481) (0.0389) August 0.109*** 0.0842*** 0.140*** 0.106*** (0.0379) (0.0307) (0.0428) (0.0359) September 0.126*** 0.103*** 0.134*** 0.103*** (0.0363) (0.0305) (0.0435) (0.0361) October 0.0799** 0.0662** 0.0850** 0.0541 (0.0316) (0.0273) (0.0411) (0.0348) r2 0.665 0.675 0.639 0.649 N 792,197 792,197 831,737 831,737 FixedEffects Individual X X X X Month X X X X Month-in-2020 X X X X Controls IndustryandOccupationa X X CasesandDeathsb X X Non-pharmaceuticalInterventionsc X X UnemploymentInsuranced X X CovidInducedMeasurementChangese X X Note: Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. Source: CurrentPopulationSurvey,SBAdata,andauthors’calculations. Additionaldatanotedbelow. a Fixedeffectsfor2-DigitNAICSandcontrolsforoccupationalexposuretotheCovidshockfully interactedwithmonth-in-2020fixedeffects. Occupationalexposureismeasuredbyabilitytowork fromhome,requiredproximitytoothers,andessentialness. CodingfollowsLeibovicietal.(2020) andJackson(2020). b: New cases and new deaths in the CBSA. New York City is an outlier, as a result we allow for thecoefficientoncasesanddeathstodifferforthisCBSA.Data:JHUCSSE(2020). c: StringencyIndex: OxfordCOVID-19GovernmentResponseTracker. Haleetal.(2020). d: Indicator for whether PUC, PUA, and/or LWA were made in the respondent’s state in the referenceweekinteractedwithclassofworker. Data: Nunnetal.(2020)andSingh(2020). e: Fixed effects for month in which the respondent was first interviewed by the CPS and for interviewtypefullyinteractedwithmonth-in-2020fixedeffects. 39

Table 6: Employment Flows (p.p, Flow per p.p Delay).. PanelA: PanelB: OutflowfromEmployment InflowtoEmployment All Separation Exit Layoff All Accession Reentry Recall Sharedelayed× January -0.0154 -0.00307 -0.00910 -0.00447 0.108 0.201 0.0204 -0.781 (0.0200) (0.00957) (0.0173) (0.00603) (0.165) (0.258) (0.235) (0.625) February -0.0241 -0.00122 -0.0168 -0.00720 -0.262* -0.265 -0.257 -0.235 (0.0183) (0.00782) (0.0157) (0.00623) (0.157) (0.236) (0.251) (0.498) March 0.00719 -0.00217 -0.00177 0.0115 -0.0498 0.0114 -0.0937 -0.422 (0.0219) (0.00779) (0.0183) (0.00984) (0.160) (0.237) (0.225) (0.599) April -0.0290 -0.00751 0.0317 -0.0581* 0.0749 0.137 0.135 -0.618 (0.0392) (0.0129) (0.0281) (0.0325) (0.134) (0.217) (0.188) (0.451) May 0.116*** 0.00598 0.0479** 0.0753*** -0.263** -0.220 -0.175 -0.500** (0.0273) (0.00814) (0.0201) (0.0186) (0.117) (0.254) (0.205) (0.238) June -0.0229 -0.00644 0.00554 -0.0232 -0.194 -0.502* 0.179 -0.548** (0.0294) (0.00994) (0.0209) (0.0195) (0.134) (0.276) (0.206) (0.249) July -0.0182 -0.000635 -0.0172 -0.00134 0.0549 0.175 0.219 -0.330 (0.0284) (0.00818) (0.0211) (0.0182) (0.141) (0.220) (0.223) (0.277) August -0.00546 0.00361 -0.00854 -0.0000866 -0.0934 0.0628 -0.604** -0.351 (0.0270) (0.00915) (0.0227) (0.0122) (0.133) (0.219) (0.266) (0.273) September 0.0176 -0.000408 0.000595 0.0194** -0.0969 -0.125 0.497** -0.740*** (0.0244) (0.0104) (0.0208) (0.00867) (0.135) (0.199) (0.232) (0.282) October -0.0166 -0.00819 -0.0169 0.00846 0.0386 -0.0742 -0.00285 -0.246 (0.0194) (0.00928) (0.0164) (0.00653) (0.134) (0.205) (0.214) (0.347) R2 0.043 0.005 0.019 0.079 0.042 0.079 0.074 0.150 N 600,468 575,129 591,162 577,377 50,541 17,955 20,788 11,798 Note: Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. Source: CurrentPopulationSurvey,SBAdata,andauthors’calculations. Additionaldatanotedbelow. Controls: Individual,month,andmonth-in-2020fixedeffects. Fixedeffectsfor2-DigitNAICSandcontrolsforoccupationalexposuretothe Covid shock fully interacted with month-in-2020 fixed effects. Occupational exposure is measured by ability to work from home, required proximitytoothers,andessentialness. CodingfollowsLeibovicietal.(2020)andJackson(2020). NewcasesandnewdeathsintheCBSA. NewYorkCityisanoutlier,asaresultweallowforthecoefficientoncasesanddeathstodifferforthisCBSA.Source: JHUCSSE(2020). StringencyIndexofnon-pharmesuticalinterventions. Source: OxfordCOVID-19GovernmentResponseTracker. Haleetal.(2020). Indicator forwhetherPUC,PUA,and/orLWAweremadeintherespondent’sstateinthereferenceweekinteractedwithclassofworker. Data: Nunn etal.(2020)andSingh(2020). FixedeffectsformonthinwhichtherespondentwasfirstinterviewedbytheCPSandforinterviewtypefully interactedwithmonth-in-2020fixedeffects. 40

Table 7: Unemployment by Class of Worker (p.p, Nonemployment per p.p Delay). Sharedelayed×... Self-employed×... PrivateEmployee×... PublicEmployee×... January 0.00787 -0.0161 0.0107 (0.0692) (0.0222) (0.0448) February 0.0243 -0.0221 0.00788 (0.0734) (0.0259) (0.0451) March 0.0402 0.0133 -0.0322 (0.0774) (0.0291) (0.0547) April 0.0164 -0.0412 -0.0820 (0.115) (0.0474) (0.0768) May 0.233** 0.133*** 0.00123 (0.0968) (0.0440) (0.0766) June 0.159 0.105** -0.0320 (0.104) (0.0436) (0.0704) July 0.130 0.0738* -0.0417 (0.0934) (0.0423) (0.0796) August 0.122 0.104*** -0.0378 (0.0899) (0.0385) (0.0781) September 0.210** 0.0925** 0.0998 (0.0889) (0.0390) (0.0655) October 0.0109 0.0606* 0.0694 (0.0855) (0.0367) (0.0593) Observations 915488 R-squared 0.643 Note: Standarderrors,clusteredattheindividualXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. Source: CurrentPopulationSurvey,SBADATA,andauthors’calculations. Additionaldatasetsnotedbelow. Controls: Individual,month,andmonth-in-2020fixedeffects. Fixedeffectsfor2-DigitNAICSandcontrolsfor occupationalexposuretotheCovidshockfullyinteractedwithmonth-in-2020fixedeffects. Occupationalexposure ismeasuredbyabilitytoworkfromhome,requiredproximitytoothers,andessentialness. CodingfollowsLeibovici etal.(2020)andJackson(2020). NewcasesandnewdeathsintheCBSA.NewYorkCityisanoutlier,asaresult weallowforthecoefficientoncasesanddeathstodifferforthisCBSA.Source: JHUCSSE(2020). Stringency Indexofnon-pharmesuticalinterventions. Source: OxfordCOVID-19GovernmentResponseTracker. Haleetal. (2020). IndicatorforwhetherPUC,PUA,and/orLWAweremadeintherespondent’sstateinthereferenceweek interactedwithclassofworker. Data: Nunnetal.(2020)andSingh(2020). Fixedeffectsformonthinwhich therespondentwasfirstinterviewedbytheCPSandforinterviewtypefullyinteractedwithmonth-in-2020fixed effects. 41

Table 8: Nonemployment by Firm Size (p.p, Nonemployment per p.p Delay). Eligiblea Ineligiblea Unemployed NumberofEmployees Sharedelayed× lessthan10× 10to99× 100to499× 500to999× 1,000ormore January -0.611 0.0106 -0.0722 -0.0632 0.151 -0.0147 (0.426) (0.0997) (0.0849) (0.118) (0.207) (0.0656) February -0.383 0.0582 -0.0973 0.000185 0.145 -0.0375 (0.367) (0.0937) (0.0784) (0.112) (0.204) (0.0604) March -0.0989 0.0666 -0.0761 -0.0217 0.0100 0.0195 (0.361) (0.0896) (0.0785) (0.120) (0.207) (0.0577) April -0.261 0.0302 -0.139 -0.0584 0.173 0.0333 (0.380) (0.145) (0.133) (0.162) (0.264) (0.0931) May 0.144 0.454*** 0.0282 0.0468 0.113 0.0788 (0.376) (0.168) (0.144) (0.167) (0.287) (0.0975) June 0.0552 0.218 -0.111 0.217 -0.0729 0.0705 (0.476) (0.225) (0.190) (0.224) (0.377) (0.119) Observations 0.624 R-squared 263,330 Note: Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. Source: CurrentPopulationSurvey,SBAData,andauthors’calculations. Additionaldatasetsnotedbelow. Controls:Individual,month,andmonth-in-2020fixedeffects. Fixedeffectsfor2-DigitNAICSandcontrolsforoccupationalexposure totheCovidshockfullyinteractedwithmonth-in-2020fixedeffects. Occupationalexposureismeasuredbyabilitytoworkfrom home,requiredproximitytoothers,andessentialness. CodingfollowsLeibovicietal.(2020)andJackson(2020). Newcasesand newdeathsintheCBSA.NewYorkCityisanoutlier,asaresultweallowforthecoefficientoncasesanddeathstodifferforthis CBSA.Source: JHUCSSE(2020). StringencyIndexofnon-pharmesuticalinterventions. Source: OxfordCOVID-19Government ResponseTracker. Haleetal.(2020). IndicatorforwhetherPUC,PUA,and/orLWAweremadeintherespondent’sstateinthe referenceweekinteractedwithclassofworker. Data: Nunnetal.(2020)andSingh(2020). Fixedeffectsformonthinwhichthe respondentwasfirstinterviewedbytheCPSandforinterviewtypefullyinteractedwithmonth-in-2020fixedeffects. 42

Table 9: Nonemployment by Worker Characteristics (p.p, Nonenemployment per p.p Delay). TercileofPredictedWagesa EducationalAttainment Sharedelayed× I II III Lessthan HighSchool SomeCollege January 0.00980 -0.0168 -0.0402 0.135 -0.0210 -0.0344 (0.0450) (0.0340) (0.0271) (0.102) (0.0296) (0.0276) February -0.0275 -0.0103 -0.0162 0.0221 -0.0420 0.0122 (0.0528) (0.0403) (0.0299) (0.121) (0.0340) (0.0328) March 0.0276 0.00817 -0.00203 0.127 0.0158 -0.0154 (0.0611) (0.0465) (0.0317) (0.124) (0.0406) (0.0330) April -0.154 0.0338 -0.0123 0.0709 -0.0983 0.00509 (0.0939) (0.0760) (0.0508) (0.175) (0.0661) (0.0511) May 0.177** 0.204*** 0.0407 0.318* 0.149** 0.0960* (0.0872) (0.0646) (0.0452) (0.166) (0.0598) (0.0511) June 0.123 0.156** 0.0474 0.313* 0.0938 0.0894* (0.0867) (0.0623) (0.0477) (0.162) (0.0598) (0.0474) July 0.126 0.122** 0.00379 0.430** 0.0702 0.0457 (0.0872) (0.0623) (0.0400) (0.173) (0.0577) (0.0446) August 0.211*** 0.0482 0.0517 0.255 0.102* 0.0943** (0.0777) (0.0614) (0.0393) (0.159) (0.0544) (0.0455) September 0.139* 0.0915* 0.0709* 0.0889 0.0995* 0.119*** (0.0771) (0.0553) (0.0408) (0.134) (0.0518) (0.0431) October 0.0435 0.0506 0.0661 -0.00548 0.0527 0.0719 (0.0734) (0.0577) (0.0411) (0.142) (0.0485) (0.0449) Observations 830,401 830,401 R-squared 0.649 0.649 Note: Standarderrors,clusteredattheCBSAXIndustry,inparenthesise. *p<0.10,**p<0.05,***p<0.01. Source: CurrentPopulationSurvey,SBADATA,andauthors’calculations. Additionaldatasetsnotedbelow. a Predicted wages conditional on education, potential experience and it’s square, sex, race, ethnicity, broad industriesoccupations,andCBSA. Controls: Individual,month,andmonth-in-2020fixedeffects. Fixedeffectsfor2-DigitNAICSandcontrolsfor occupationalexposuretotheCovidshockfullyinteractedwithmonth-in-2020fixedeffects. Occupationalexposure ismeasuredbyabilitytoworkfromhome,requiredproximitytoothers,andessentialness. CodingfollowsLeibovici etal.(2020)andJackson(2020). NewcasesandnewdeathsintheCBSA.NewYorkCityisanoutlier,asaresult weallowforthecoefficientoncasesanddeathstodifferforthisCBSA.Source: JHUCSSE(2020). Stringency Indexofnon-pharmesuticalinterventions. Source: OxfordCOVID-19GovernmentResponseTracker. Haleetal. (2020). IndicatorforwhetherPUC,PUA,and/orLWAweremadeintherespondent’sstateinthereferenceweek interactedwithclassofworker. Data: Nunnetal.(2020)andSingh(2020). Fixedeffectsformonthinwhich therespondentwasfirstinterviewedbytheCPSandforinterviewtypefullyinteractedwithmonth-in-2020fixed effects. 43

Table10: FalsificationTests(p.p, Unemploymentperp.pDelay). BaselineSpecification SecondRound May2-3 TotalLoans April30-May3 ShareDelayedXblankspace January -0.0150 -0.0101 -0.00754 (0.0215) (0.00996) (0.00968) February -0.0179 -0.00926 -0.0183 (0.0245) (0.0111) (0.0112) March 0.0121 0.00127 -0.00233 (0.0276) (0.0128) (0.0120) April -0.0453 0.0285 0.0644*** (0.0461) (0.0206) (0.0211) May 0.141*** 0.118*** 0.0606*** (0.0421) (0.0208) (0.0203) June 0.108*** 0.0936*** 0.0561*** (0.0405) (0.0176) (0.0190) July 0.0827** 0.0733*** 0.0274 (0.0389) (0.0184) (0.0169) August 0.106*** 0.0711*** 0.0297* (0.0359) (0.0167) (0.0169) September 0.103*** 0.0972*** 0.0532*** (0.0361) (0.0166) (0.0168) October 0.0541 0.0609*** 0.0305* (0.0348) (0.0164) (0.0164) Observations 0.649 0.649 0.649 R-squared 831,737 831,737 831,737 Note: Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,*** p<0.01. Source: Current Population Survey, SBA DATA, and authors’ calculations. Additional data sets notedbelow. Controls: Individual,month,andmonth-in-2020fixedeffects. Fixedeffectsfor2-DigitNAICSand controls for occupational exposure to the Covid shock fully interacted with month-in-2020 fixed effects. Occupational exposure is measured by ability to work from home, required proximity to others, and essentialness. Coding follows Leibovici et al. (2020) and Jackson (2020). New cases andnewdeathsintheCBSA.NewYorkCityisanoutlier,asaresultweallowforthecoefficient oncasesanddeathstodifferforthisCBSA.Source: JHUCSSE(2020). StringencyIndexofnonpharmesuticalinterventions. Source: OxfordCOVID-19GovernmentResponseTracker. Haleetal. (2020). IndicatorforwhetherPUC,PUA,and/orLWAweremadeintherespondent’sstateinthe referenceweekinteractedwithclassofworker. Data: Nunnetal.(2020)andSingh(2020). Fixed effectsformonthinwhichtherespondentwasfirstinterviewedbytheCPSandforinterviewtype fullyinteractedwithmonth-in-2020fixedeffects. 44

Table 11: Narrower and Wider Windows. (p.p, Nonemployment per p.p Delay). BaselineSpecification 27 27-30 15-27 13-30 January -0.0150 -0.0125 -0.0193 (0.0215) (0.0190) (0.0214) February -0.0179 -0.00442 -0.0207 (0.0245) (0.0219) (0.0245) March 0.0121 0.0205 0.0167 (0.0276) (0.0247) (0.0273) April -0.0453 -0.0549 -0.0148 (0.0461) (0.0394) (0.0465) May 0.141*** 0.0764** 0.189*** (0.0421) (0.0358) (0.0435) June 0.108*** 0.0696** 0.140*** (0.0405) (0.0345) (0.0404) July 0.0827** 0.0770** 0.101** (0.0389) (0.0338) (0.0395) August 0.106*** 0.0895*** 0.121*** (0.0359) (0.0331) (0.0356) September 0.103*** 0.0735** 0.119*** (0.0361) (0.0317) (0.0366) October 0.0541 0.0417 0.0689* (0.0348) (0.0308) (0.0355) Observations 0.649 0.649 0.649 R-squared 831,737 831,737 831,737 Note: Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. Source: CurrentPopulationSurvey,SBADATA,andauthors’calculations. Additionaldatasetsnotedbelow. Controls: Individual,month,andmonth-in-2020fixedeffects. Fixedeffectsfor2-DigitNAICSand controls for occupational exposure to the Covid shock fully interacted with month-in-2020 fixed effects. Occupationalexposureismeasuredbyabilitytoworkfromhome, requiredproximityto others, and essentialness. Coding follows Leibovici et al. (2020) and Jackson (2020). New cases andnewdeathsintheCBSA.NewYorkCityisanoutlier,asaresultweallowforthecoefficient oncasesanddeathstodifferforthisCBSA.Source: JHUCSSE(2020). StringencyIndexofnonpharmesuticalinterventions. Source: OxfordCOVID-19GovernmentResponseTracker. Haleetal. (2020). IndicatorforwhetherPUC,PUA,and/orLWAweremadeintherespondent’sstateinthe referenceweekinteractedwithclassofworker. Data: Nunnetal.(2020)andSingh(2020). Fixed effectsformonthinwhichtherespondentwasfirstinterviewedbytheCPSandforinterviewtype fullyinteractedwithmonth-in-2020fixedeffects. 45

Figure 1: Evolution of PPP Loan Issuance PPP Funds PPP PPP Application Opens Depleted Reopens Deadline $700B PPP funding under the CARES and PPP Acts $600B $500B $400B PPP funding under the CARES Act $300B $200B $100B Event Window $0B Mar−30 Apr−13 Apr−27 May−11 May−25 Jun−08 Jun−22 Jul−06 Jul−20 Aug−03 Source: U.S. Treasury (2020) and authors’ Calculations Figure 2: PPP Loan Timing. Panel A: Loan Count Panel B: Loan Dollars snoilliM 5. 4. 3. 2. 1. 0 01apr2020 01may2020 01jun2020 01jul2020 snoilliM 00005 00004 00003 00002 00001 0 01apr2020 01may2020 01jun2020 01jul2020 Source: U.S. Treasury (2020) and authors’ Calculations 46

Figure 3: Geographic Distribution of Loan Timing. Panel A: Window Share Late (All Counties) (0.59,1.00] (0.45,0.59] (0.35,0.45] [0.00,0.35] Panel B: Window Share Late (Observable in CPS) (0.60,1.00] (0.46,0.60] (0.35,0.46] [0.00,0.35] Source: U.S. Treasury (2020) and authors’ Calculations 47

Figure 4: Time Variation in Characteristics of CBSAs Receiving Loans. Panel A: Banking Panel B: Demographics secidnI namhcsriH-lhadnifreH 0002 0081 0061 0041 0021 0001 5.1 1 5. 0 noitalupoP k001 rep ytisneD Left Axis Right Axis Branches Branches Deposits 100k Deposits April 3 April 16 April 27 May 10 noitalupoP dna emocnI 5 4 3 2 1 0 6. 5. 4. 3. 2. 1. 0 noitalupoP s'ASBC fo erahS Left Axis Right Axis Per Capita Income (10K) Rural Population (100K) Republican April 3 April 16 April 27 May 10 Panel C: Race/Ethnicity Panel D: NPI, School, UI ASBC fo puekcaM cinhtE/laicaR 8. 6. 4. 2. 0 White Black Hispanic April 3 April 16 April 27 May 10 3 yaM 03 lirpA ... 31 lirpA 11 lirpA 61 hcraM 31 hcraM Left Axis Right Axis PUC Schools PUA NPI April 3 April 16 April 27 May 10 Panel E: Loan Size and Jobs Retained Panel F: Cases and Deaths 000006 000004 000002 0 04 03 02 01 0 Left Axis Right Axis Loan Size Jobs Per Loan April 3 April 16 April 27 May 10 noitalupop K001 rep shtaeD / sesaC 004 003 002 001 02 51 01 5 0 Cases Deaths April 3 April 16 April 27 May 10 Source: U.S. Treasury (2020), FDIC Summary of Deposits (2019), Keystone Strategy (2020), Nunn et al. (2020), Singh (2020). Census DemographicData(2018),JHUCSSE(2020)andauthors’Calculations. Note: hashedlinesplottheaverageofeachseriesintheintervalApril 14-28. 48

Figure 5: Employment Effects (Coefficient Plots in p.p, Unemployment per p.p Delay). Panel A: marginal effect of delayed PPP financing I: unemployment II: labor force participation III: nonemployment stnioP egatnecreP 52. 51. 50. 50.- 51.- Jan. Feb.MarchAprilMayJuneJulyAug.Sept.Oct. stnioP egatnecreP 52. 51. 50. 50.- 51.- Jan. Feb.MarchAprilMayJuneJulyAug.Sept.Oct. stnioP egatnecreP 52. 51. 50. 50.- 51.- Jan. Feb.MarchAprilMayJuneJulyAug.Sept.Oct. Panel B: baseline COVID-19 shocka I: unemployment II: labor force participation III: nonemployment stnioP egatnecreP 02 51 01 5 0 5- Jan. Feb.MarchAprilMayJuneJulyAug.Sept.Oct. stnioP egatnecreP 02 51 01 5 0 5- Jan. Feb.MarchAprilMayJuneJulyAug.Sept.Oct. stnioP egatnecreP 02 51 01 5 0 5- Jan. Feb.MarchAprilMayJuneJulyAug.Sept.Oct. Note: The average potential worker lives in a CBSA where the share of delayed loans is 42%. The standard deviation is 7.5%. a controls contains month fixed effects. The baseline COVID-19 shock traces out the interaction between month fixed effects and year 2020. 49

Figure 6: Geographic Distribution of Loan Timing within other Windows. Panel A: Relative Share of Second Round (0.47,1.00] (0.34,0.47] (0.24,0.34] [0.05,0.24] Panel B: Relative Lateness within Second Round (0.57,1.00] (0.38,0.57] (0.20,0.38] [0.00,0.20] Source: U.S. Treasury (2020) and authors’ Calculations 50

A Additional Details about the Design and Implementation of the PPP This section explains on more detail how the PPP operated. These details are not directly pertinent to the design of our study, but may help the reader understand more precisely the terms of the PPP and more generally be valuable to other researchers. The PPP program was run by the SBA in consultation with the US Department of the Treasury. Borrowers had to meet a multi-part test to be eligible for a PPP loan. First, borrowers had to employ 500 or fewer employees in most industries.34 Second, firms are subject to net worth and income limitations. Firms had to have tangible net worth of less $ $ than 15 million and an average net income after federal taxes of less than 5 million. Third, borrowershadtobeindependentlyownedandoperatedandlocatedwithintheUnitedStates. Fourth, firms have to be engaged in a legal activity and not be a household employer nor a hedge or private equity fund. Fifth, they could not be delinquent or in default on an existing SBA loan, nor could they be bankrupt. However, unlike virtually all other forms of credit, these were not tested for collateral or ability to repay and the paper work requirements were modest (estimated by the SBA at about two hours).35 Once a borrower had their PPP application approved, their lender was supposed to send the funds to the borrower within 10 calendar days. This changed somewhat over the program, but not in our event window. Initial program documentation said that lenders had 10 days to make their distributions, but was unclear on what this actually required and when the clock started ticking. Eventually the Treasury and the SBA issued a rule that for loans approved on or before April 28, 2020, lenders had 10 calendar days from April 28, 2020 to fund the loan. For loans after this, the loan had to fund in 10 calendar days. The qualifying loan amount per employee was 2.5 times the average total monthly peremployee payments for payroll costs for the year prior to the loan date (or, at the option of $ the borrower, for 2019) including up to 100,000 in annualized cash compensation (wages, 34The SBA publishes North American Industry Classification System Codes (NAICS codes) based definitionsforsmallbusinessesbasedontheirheadcountsorannualrevenues(SBA(2020)). Forabout310NAICS code these limits are larger 500 employees and in these cases, the PPP is available for firms with more than 500 employees. However, there are no instances in the PPP data of any firms getting loans to support more than 500 employees, suggesting that perhaps firms meeting the less restrictive headcount requirements in these industries were unable to satisfy the other requirements or got multi-establishment loans like those available to hospitality firms. 35Despite the lack of traditional loan underwriting, Bartik et al. (2020b) find that 12-25 percent of small businesses had their PPP denied or wanted to apply for PPP loans but either did not qualify or were told that they did not. 51

salaries, and cash tips) plus group insurance premiums (including health care benefits).36 The qualifying loan amount is the total over all employees for covered firms, up to a limit of $ 10 million.37 PPP loans were forgivable if they were used for qualified expenses during a specified period. For the loans we focus on, borrowers were told they had 8 weeks after fund distribution to spend the proceeds on qualified expenses and at least 75% of funds had to be spent on qualified payroll expenses. The vast majority (92%) of PPP loans are made before June 5, 2020. Before that, the US Treasury and the SBA guidance was firms had 8 weeks to spend their funds to qualify for forgiveness and 75% of expenses had to be for payroll. On June 5, 2020, the Paycheck Protection Program Flexibility Act of 2020 passed and the US Treasury and the SBA created rules under that act that relaxed the requirement to allow 24 weeks to spend the PPP funds and 60% of expenses had to be for payroll to qualify for loan forgiveness. The proceeds of any PPP loans that did not qualify for forgiveness are due in two years from distribution at the cost of one percent per year interest. A.1 Detailed Timeline of the PPP in the Popular Press 36Therefore,aPPPloanforanemployeewithnobenefitsandanannualsalaryof$48,000wouldbe$10,000 ($48,000 2.5), but for an no-benefits employee making $100,000 or more, the loan would be for $20,834. 37The m × ax 12 imum loan amount from compensation expenses for employees was $46,154 per employee plus health and retirement benefits. For sole proprietors, independent contractor, or the self-employed, the maximum of compensation component of the loan was $20,833 and health care and retirement benefits are not covered. That said, most PPP loans were much smaller: the average loan was for $9,600 per employee and the average loan size was for $114,000, while the median loan was only $25,000. 52

Tabular Summary of PPP Timeline: Week of March 29th to April 4th Date Topic Reporting Summary 3/27/2020 PPP Opening PPP Passed Into Law. Coronavirus relief bill appropriates $350 Billion for PPP. Loan applications open Friday, April 3rd. Davidson [2020] 3/28/2020 Loan Private Directives from Treasury. Treasury Department Allocation privately encourages lenders to prioritize their existing customer base over new applicants in call with American Bankers Association. Omeokwe [2020a] 4/2/2020 Program Ad- Sec. Mnuchin Press Conference Error. Treasury Secretary ministration Mnuchin gives wrong address for SBA application portal in a Coronavirus Task Force briefing, instructing applicants to go to SBA.com instead of SBA.gov, the application portal. Larson [2020] Loan Applications handled through SBA approved lenders. Lenders Applications were not ready to take loan applications. Treasury was late to issue program rules. Rudegeair et al. [2020a] Loan Loans made for 8 weeks at 1% interest to help businesses make Requirements payroll and rent. The expectation of loan forgiveness was widely reported. Nilsen [2020] Loan Banks often prioritize larger businesses with established banking Allocation relationships. Rudegeair and Simon [2020] 4/4/2020 Funding Congress Announces Expectations about Funding. Senator Expectations Marco Rubio tweets funding for PPP expected to run out in late May and will require more appropriations. Rubio [2020a] Week of April 5th to 11th Date Topic Reporting Summary 4/6/2020 PPP Fed and Treasury Announce Program to Back PPP Financing Loans. The Federal Reserve Board, in cooperation with Treasury, announces a new 13(3) facility to offer term financing backed by PPP loans. Davidson and Timiraos [2020] Loan Large restaurants, franchises, and hotel chains, notably the New Recipients York based Shake Shack, gain exemption from employment limits and announce their intent to seek funding. There is a growing concern over these businesses competing with smaller firms for funds. Davis and Haddon [2020]

Week of April 5th to 11th (continued) Date Topic Reporting Summary 4/7/2020 Funding Congress Announces Intentions about Funding. Congress Expectations expresses intent to appropriate $250 Billion in additional PPP funding by end of week. Congress also expresses the intent to approve new lenders. Rudegeair et al. [2020b] Loan Processing applications costly for banks and borrowers. Banks Application take steps to limit applications to firms with existing accounts. Banks frequently change application rules responding to Treasury guidance, public criticism, and program capacity. Borrowers uncertain about application rules and loan availability. SBA loan infrastruture cannot process volume, experiencing notable crashes. Zhou [2020] Loan Banks limit loans primarily to firms with existing credit Allocation relationships due to availability of funding and program interest. Unbanked and smaller firms at a significant disadvantage when seeking loans, as banks have a large incentive to only make loans to firms with existing credit lines whose ability to repay is better known. Zhou [2020] Loan Receipt Small businesses report significant delays receiving funds. Firms attempt to substitute towards other lending out of desperation for funds. Hansen [2020], Hayashi [2020a] Week of April 12th to 18th Date Topic Reporting Summary Loan Wells Fargo reports reaching its capacity and stops making PPP Availability loans. However, the bank still accepts applications expecting fund replenishment. Niquette and Jacobs [2020], Warmbrodt [2020] Loan Receipt Businesses still report significant delays in receiving PPP funds. Senate Democratic Caucus [2020], The Editorial Board [2020] 4/14/2020 Program Ad- Cease and Desist Issued to SBA.com. Cease and Desist ministration issued to SBA.com by New York Attorney General Letitia James. Larson [2020] 4/15/2020 Funding Congressional Negotiations Resume. With the expected Expectations depletion of program funds, Congressional negotiations over PPP appropriations resume. Dem. Caucus wants appropriations for hospitals and state and local governments, with the GOP against. Werner et al. [2020]

Week of April 12th to 18th (continued) Date Topic Reporting Summary 4/16/2020 Loan PPP Funding Officially Depleted. Some banks stop taking Availability new applications in anticipation of funding depletion with as many as 700,000 businesses awaiting funding. Others continue receiving applications but stop making new loans. Peterson [2020], Rubio [2020b] Loan National Federation of Independent Businesses (NFIB) reports Availability only 4% of surveyed members have been approved for a loan, while none of those approved have yet to receive funding. Niquette and Jacobs [2020] Loan SBA excludes tiered franchises from receiving loans. Bykowicz Availability [2020] Program Ad- There is a growing concern that banks will “authorization hoard,” ministration collecting applications that they do not intend to fulfill which they can chose among when issued more funds. Warmbrodt [2020] 4/18/2020 Funding Pres. Trump Signals Support for Hospital Funding. Expectations President Trump signals support for appropriating funds for hospitals, with the GOP Caucus still unsupportive. Wasson [2020], House et al. [2020] Week of April 19th to 25th Date Topic Reporting Summary 4/20/2020 Funding Democrats Offer More Appropriations for Swing States. Expectations Democratic Caucus outlines compromise offer appropriating funds to local governments, with favorable apportionments to swing states. Flatley et al. [2020a] 4/20/2020 Funding White House Outlines Negotiating Position. White House Expectations and Sec. Mnuchin report a willingness to appropriate $300 Billion for the PPP. Flatley et al. [2020b] 4/20/2020 Loan SEC 8-K Filings Released. SEC 8-K filings reveal 70 publically Recipients traded companies received PPP loans. The White House and the GOP Caucus face sustained public criticism over recipients. Nix et al. [2020] Loan Shake Shack announces they will return their $10 Million PPP Recepients loan to the SBA. O’Connell [2020a] Loan Receipt NFIB reports only about 20% of their surveyed members who applied for a PPP loan had received money by April 17. Further, about 75% had applied with 26% in the process of applying when funding ran out. Niquette [2020]

Week of April 19th to 25th (continued) Date Topic Reporting Summary 4/21/2020 Funding $484 Billion in relief funding passes Senate, with $310 Billion for Expectations PPP loans and $11 Billion for related fees. Werner and Kim [2020] Loan Regional and community lenders are more successful than large Applications banks at making modest loans to a large number of small businesses. Loan seekers report subsituting towards applying for PPP loans at smaller banks. Levitt [2020] Loan JP Morgan Chase faces criticism after disclosing data that reveals Allocation high rejection rates among small business clients and high acceptance rates among large clients. Only 6% of JP Morgan’s 300,000 small business customers received loans. McLaughlin and Davis [2020] 4/23/2020 Funding Congress Passes New PPP Funding. $321 Billion for PPP Expectations funding and related fees 4/23/2020 Loan SBA Bars Firms with Other Cash Sources. The SBA issues Recipients new guidance suggesting companies with other sources of cash would not qualify for PPP loans, and that the firms who have already borrowed under this standard should return their loan by May 7th. O’Connell and Gregg [2020] Loan Kura Sushi and Sweetgreen agree to return millions in PPP loans. Recipients O’Connell [2020b] Program Ad- Small business associations express dismay over new funding ministration numbers. Each expects that replenishment will be exhausted quickly, and criticizes the program’s rollout as insufficient. Brody et al. [2020] Loan Autonation, a national network of auto dealerships, returns more Recipients than $77 Million in PPP funds after two employees whistleblow to the Washington Post. O’Connell [2020c] Loan There is a growing concern that minority owned businesses will be Allocation shut out of PPP funds, as funding sources more readily available to them, like Community Development Financial Institutions, are not PPP approved lenders. Knowles [2020] Program Ad- Small business owners who were denied PPP loans are becoming ministration increasingly dissatisfied with the program. There is significant confusion among applicants and potential applicants over applications, approvals, where to get funds, and if they were accepted, whether they would get funds. There is a growing impression that large businesses had sizable advantages in securing loans combined with a waning belief in the good faith of the program. The self employed also feel shut out of the progam’s rollout. Basak et al. [2020]

Week of April 19th to 25th (continued) Date Topic Reporting Summary 4/24/2020 Funding President Signs New PPP Funding into Law. $321 Billion Expectations for PPP funding and related fees Week of April 26th to May 2nd Date Topic Reporting Summary 4/26/2020 Program Ad- SBA Institutes Cap on Per-Bank Applications. The SBA ministration sets a cap on the number of relief loans a single bank can process. Banks can issue at maximum 10% of the funding authority of PPP. Niquette and Levitt [2020] 4/27/2020 Loan Applications for New PPP Funds Open. Small Businesses Availability encouraged to apply as soon as possible to access funds. Davis et al. [2020] Loan PPP loans mandate that 75% of the funds go towards salaries and Requirements 25% go towards rents to receive forgiveness. Businesses in high rent metros feel the squeeze, and either shut down or worry their loans will not be forgiven for spending more on rent than the loans require. Hayashi [2020b] Week of May 3rd to 9th Date Topic Reporting Summary Program Ad- Demand for PPP loans falls, but not demand for loans in general. ministration Borrowers look elsewhere for funds, as they worry they will not get PPP loans forgiven and are unconvinced they can apply for and get funds quickly. Of $310 Billion appropriated, 40% remains available as of May 8th. Banks finding about 10% of their loan applications are duplicates. An illustrative quote from a business owner: “It’s basically a large loan that we are going to be stuck with.” Hayashi [2020c] 5/8/2020 Program Ad- Inspector General Statement. The SBA inspector general ministration finds that the SBA did not issue guidance to prioritize rural and underserved communities following congressional guidance on PPP administration. Omeokwe [2020b]

Week of May 10th to 16th Date Topic Reporting Summary 5/14/2020 Program Ad- Census Bureau Small Business Survey. Census Bureau ministration survey finds that nationally, about 75% of small businesses sought some sort of Federal aid, with 38% of respondents reporting they had gotten the loan money. Of those surveyed, 30% sought EIDL loans, with 10% having received payment. The respondents believe it will take a significant period of time to return to pre-COVID-19 levels of demand. Omeokwe [2020c] Week of May 17th to 23rd Date Topic Reporting Summary 5/17/2020 Loan Congress Plans Changes to PPP Loan Requirements. Requirements Congress expresses intent to change loan requirements to give businesses both more time and more flexibility as to how to disburse PPP funds. Hayashi [2020d] 5/21/2020 Loan Senate Adjourns Without PPP Deal. Senate adjourns for Requirements recess, failing to pass legislation to extend the term of PPP loans. Adjournment delays resolution of the problem until at least next month. Hayashi and Andrews [2020] Program Ad- SBA’s administration of the program increasingly perceived as ministration opaque and ignoring legislative priorities. Omeokwe [2020d] 5/23/2020 Program Ad- Treasury Rule Change. Treasury Department revises PPP loan ministration rules allowing the SBA to review program loans “of any size at any time of SBA’s discretion” signaling intent to monitor how businesses spend funds. Borrowers must retain documentation for 6 years after initial loan and alert state unemployment offices if workers refuse requests to return to work. Hayashi [2020e] Program Ad- The Treasury guidance is seen as partly insufficient, as it leaves ministration business’s uncertaint about eligibility for forgiveness and loan terms untouched. However, the guidance does provide clairification on how to calculate payroll and non-payroll expenses. Hayashi [2020e]

Week of May 31st to June 6th Date Topic Reporting Summary 6/3/2020 Loan Senate Passes PPP Extension. Senate passes bill extending Requirements loan forgiveness timeline from eight to twenty four weeks. The bill also reduces payroll requirements to allow 60% of the loan to go towards payroll, compared to the previous requirement of 75%. Andrews [2020] 6/5/2020 Loan Senate bill is signed into law, largely unchanged. Deadline to Requirements apply for new PPP loans Aug. 8th. Hayashi [2020f] Week of June 7th to 13th Date Topic Reporting Summary Loan Changes to program complicate the process of obtaining loan Forgiveness forgiveness. Uncertainty over the loan spending requirements and the prospect of increased fraud enforcement scares some businesses from applying for forgiveness. Lenders determine whether a loan satisfies forgiveness guidelines and fear being stuck with unprofitable loans, but the SBA has not yet said how lenders should submit forgiveness applications. The NFIB reports nearly one-third of businesses who received PPP loans passed the original eight week deadline by June 14. Multiple interviewees express that the loan could end their business if not forgiven instead of saving it, believing they would be unable to service a debt even if economic conditions returned to normal. Simon and Rudegeair [2020a] Week of June 14th to 20th Date Topic Reporting Summary Program Ad- The PPP widely seen as not living up to expectations. Only 4.6 ministration million loans were allocated to 31.7 million US small businesses. The application process is widely seen as confusing and onerous in retrospect, with many small businesses report either returning money or not applying because they did not think they met the requirements. Simon et al. [2020]

Week of June 21st to 27th Date Topic Reporting Summary Loan Firms who received PPP loans early quickly spent the money to Forgiveness comply with the original rules, but now, cannot reapply to receive more PPP funding under the PPP’s “one business, one loan” policy. Simon and Rudegeair [2020b] References Kate Davidson. Small-Business Loans Expected to Start Friday, Mnuchin Says. Wall Street Journal, March 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ small-business-loans-expected-to-start-friday-mnuchin-says-11585577658. Ryan Omeokwe, Amara and Tracy. WSJ News Exclusive Treasury Department Encour- | aged Banks to Prioritize Existing Customers for PPP Loans, House Panel Says. Wall Street Journal, October 2020a. ISSN 0099-9660. URL https://www.wsj.com/articles/ treasury-department-encouraged-banks-to-prioritize-existing-customers-for-ppp-loans-democratic-report-says-11602861336. Erik Larson. N.Y. Targets Loan Website So Confusing Mnuchin Got It Wrong. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/business/on-small-business/ ny-targets-loan-website-so-confusing-mnuchin-got-it-wrong/2020/04/16/ f0e9c7d2-7fe7-11ea-84c2-0792d8591911_story.html. Peter Rudegeair, Ruth Simon, and Bob Davis. Small Business Loan Program Makes Bumpy Start. Wall Street Journal, April 2020a. ISSN 0099-9660. URL https://www.wsj.com/ articles/small-business-loan-program-makes-bumpy-start-11585939613. Ella Nilsen. What we know about the fourth coronavirus relief bill. Vox, April 2020. URL https://www.vox.com/2020/4/3/21206931/ what-we-know-about-congress-fourth-coronavirus-bill. Peter Rudegeair and Ruth Simon. Big Banks Favor Certain Customers in $350 Billion Small-Business Loan Program. Wall Street Journal, April 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ big-banks-favor-certain-customers-in-350-billion-small-business-loan-program-11586174401. Marco Rubio. “eighth issue is money. based on the demand we saw on day one it is clear we will need more money for #ppp by late may. our rough estimate is that if 2/3 of the lenders that will participate in #ppp loans arent up & running yet, the $349 billion will run out around 6 june. 12/13”. twitter.com, April 2020a. URL https://twitter.com/ marcorubio/status/1246396191623057409. Kate Davidson and Nick Timiraos. Fed Preparing to Finance New Small-Business Payroll Loans. Wall Street Journal, April

2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ fed-preparing-to-purchase-new-small-business-payroll-loans-11586194588. Bob Davis and Heather Haddon. Big Restaurant, Hotel Chains Won Exemption to Get Small Business Loans. Wall Street Journal, April 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ big-restaurant-hotel-chains-won-exemption-to-get-small-business-loans-11586167200. Peter Rudegeair, Bob Davis, and Andrew Duehren. Lawmakers, White House Vow to Quickly Provide More Small-Business Loans. Wall Street Journal, April 2020b. ISSN 0099-9660. URL https://www.wsj.com/articles/ mcconnell-says-small-business-loan-program-needs-more-funding-11586275489. Li Zhou. Many small businesses are being shut out of a new loan program by major banks. Vox, April 2020. URL https:/www.vox.com/2020/4/7/21209584/ paycheck-protection-program-banks-access. Sarah Hansen. Most small businesses applied to the paycheck protection programand most are still waiting for the money. Forbes, April 2020. URL https://www.forbes.com/sites/sarahhansen/2020/04/10/ most-small-businesses-applied-to-the-paycheck-protection-program-and-most-are-still-waiting-for-the-money/ ?sh=2d1563b254fb. Yuka Hayashi. Small-Business Loans Face Delays Even as Coronavirus Program Expands. Wall Street Journal, April 2020a. ISSN 0099-9660. URL https://www.wsj.com/articles/ small-business-loans-face-delays-even-as-coronavirus-program-expands-11586511002. Mark Niquette and Jennifer Jacobs. Small Business Relief Funds Drained Fast With Many Shut Out. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ small-business-relief-funds-drained-fast-with-many-shut-out/2020/04/16/ b91a291e-7ff1-11ea-84c2-0792d8591911_story.html. Zachary Warmbrodt. Small business loan program stumbles as sba system crashes. Politico, April 2020. URL https://www.politico.com/news/2020/04/06/ small-business-loan-program-system-crashes-169614. Senate Democratic Caucus. President trump and his administration must fix these problems immediately: Small business owners across the country say thay aren’t getting needed funds from the paycheck protection program. Press Release, April 2020. URL https://www.democrats.senate.gov/imo/media/doc/ SmallBusinessOwnersAreNotReceivingPromisedFundsfromPPPCoronavirusProgram. pdf. The Editorial Board. Small Business Loan Snafus. Wall Street Journal, April 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ small-business-loan-hold-up-11586907564.

Erica Werner, Aaron Gregg, and Renae Merle. Trump says negotiations underway to revive small business loan program, as firms impacted by virus grasp for new loans. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/us-policy/ 2020/04/16/congress-coronavirus-small-business-trump/. Kristina Peterson. Small-Business Aid Program Set to Run Out of Money. Wall Street Journal, April 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ small-business-aid-program-set-to-run-out-of-money-later-today-11586971294. Marco Rubio. Senator marco rubio announces ppp program to run out of funding, with 700,000 applications in limbo. twitter.com, April 2020b. URL https://twitter.com/ marcorubio/status/1250516086136754176. JulieBykowicz. Small-BusinessAidPackageExcludesManyFranchisesinCoronavirusCrisis. Wall Street Journal, April2020. ISSN0099-9660. URLhttps://www.wsj.com/articles/ small-business-aid-package-excludes-many-franchises-in-coronavirus-crisis-11587029401. Erik Wasson. Trump Suggests He Could Back Deal to Fund Hospitals in SBA Bill. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/business/on-small-business/ trump-suggests-he-could-back-deal-to-fund-hospitals-in-sba-bill/2020/ 04/17/167f12d2-8109-11ea-84c2-0792d8591911_story.html. Billy House, Erik Wasson, and Steven T. Dennis. Mnuchin, Democrats Remain Far Apart on Bolstering Virus Stimulus. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ mnuchin-democrats-remain-far-apart-on-bolstering-virus-stimulus/2020/04/ 16/f15bcbde-7fe7-11ea-84c2-0792d8591911_story.html. Daniel Flatley, Hailey Waller, and Billy House. Mnuchin, Democrats Say Aid Package May Be Passed in Days. Washington Post, April 2020a. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ mnuchin-democrats-say-aid-package-may-be-passed-in-days/2020/04/19/ 0765d366-823b-11ea-81a3-9690c9881111_story.html. Daniel Flatley, Hailey Waller, and Billy House. Trump Joins Chorus Saying Agreement on New Virus Aid is Close. Washington Post, April 2020b. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ trump-joins-chorus-saying-agreement-on-new-virus-aid-is-close/2020/04/19/ 9f05b31c-829f-11ea-81a3-9690c9881111_story.html. Naomi Nix, Mark Niquette, and David McLaughlin. Shake Shack, Potbelly Among Chains Tapping Small-Biz Funds. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ shake-shack-potbelly-among-chains-tapping-small-biz-funds/2020/04/19/ 005514d4-8249-11ea-81a3-9690c9881111_story.html.

Jonathan O’Connell. White House, GOP face heat after hotel and restaurant chains helped run small business program dry. Washington Post, April 2020a. ISSN 0190-8286. URL https://www.washingtonpost.com/business/2020/04/20/ white-house-gop-face-heat-after-hotel-restaurant-chains-helped-run-small-business-program-dry/. Mark Niquette. Most Small Firms Report Not Yet Getting Money From Virus Funds. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/business/on-small-business/ most-small-firms-report-not-yet-getting-money-from-virus-funds/2020/ 04/20/7419af4e-834b-11ea-81a3-9690c9881111_story.html. Erica Werner and Seung Min Kim. Senate passes $484 billion bill that would expand small-business aid, boost money for hospitals and testing. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/us-policy/2020/04/ 21/congress-coronavirus-small-business/. Hannah Levitt. Giant U.S. Lenders Outpaced by Rivals in Small- Business Rescue. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/business/on-small-business/ giant-us-lenders-outpaced-by-rivals-in-small-business-rescue/2020/04/ 17/74004e76-80fa-11ea-84c2-0792d8591911_story.html. David McLaughlin and Michelle F. Davis. JPMorgans Small Business Loans Instead Went to Its Biggest Customers. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ jpmorgans-small-business-loans-instead-went-to-its-biggest-customers/ 2020/04/22/d440d5a0-8477-11ea-81a3-9690c9881111_story.html. Jonathan O’Connell and Aaron Gregg. New SBA guidelines ask big companies to consider returning coronavirus loans. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/business/2020/04/23/ new-sba-guidelines-say-big-companies-that-got-coronavirus-loans-should-return-money/. Jonathan O’Connell. Two more restaurant chains are giving back small-business funds. Washington Post, April 2020b. ISSN 0190-8286. URL https://www.washingtonpost. com/business/2020/04/23/kura-sushi-sba/. Ben Brody, Naomi Nix, and Mark Niquette. Small Businesses Fear Fresh Loan Funding Wont Be Nearly Enough. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ small-businesses-fear-fresh-loan-funding-wont-be-nearly-enough/2020/04/ 23/47ed6fc2-859f-11ea-81a3-9690c9881111_story.html. Jonathan O’Connell. AutoNation, a retailer worth billions, says it received nearly $80 million in SBA funds. Washington Post, April 2020c. ISSN 0190-8286. URL https://www.washingtonpost.com/business/2020/04/24/ autonation-an-auto-retailer-worth-billions-received-nearly-95-million-sba-funds/.

Hannah Knowles. Covid-19 could devastate minority-owned businesses, amid fears theyll also miss out on aid. Washington Post, April 2020. ISSN 0190-8286. URL https://www.washingtonpost.com/business/2020/04/24/ covid-19-could-devastate-minority-owned-businesses-amid-fears-theyll-also-miss-out-aid/. Sonali Basak, Emma Kinery, and Bob Ivry. Resentment Grows on Main Street Over Bailout Winners and Losers. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ resentment-grows-on-main-street-over-bailout-winners-and-losers/2020/04/ 24/21f93dbc-860a-11ea-81a3-9690c9881111_story.html. Mark Niquette and Hannah Levitt. SBA Sets Cap on Amount of Relief Loans That Banks Can Process. Washington Post, April 2020. ISSN 0190- 8286. URL https://www.washingtonpost.com/business/on-small-business/ sba-sets-cap-on-amount-of-relief-loans-that-banks-can-process/2020/04/26/ 55632164-8807-11ea-80df-d24b35a568ae_story.html. Susan Davis, Claudia Grisales, Kelsey Snell, and Alana Wise. Trump signs latest coronavirus economic relief package, aimed at small businesses. National Public Radio, April 2020. URL https://www.npr.org/2020/04/23/842706630/ congress-passes-latest-economic-relief-package-to-confront-coronavirus. Yuka Hayashi. For Small Businesses With High Rents, Coronavirus Aid Falls Short. Wall Street Journal, May 2020b. ISSN 0099-9660. URL https://www.wsj.com/articles/ for-many-small-businesses-u-s-coronavirus-aid-falls-short-11588325404. Yuka Hayashi. Demand for Small-Business Loans Cools. Wall Street Journal, May 2020c. ISSN 0099-9660. URL https://www.wsj.com/articles/ demand-for-small-business-loans-cools-11588930201. AmaraOmeokwe. SBAVeeredFromGuidelinesonSmall-BusinessLoans, ReportSays. Wall Street Journal, May 2020b. ISSN 0099-9660. URL https://www.wsj.com/articles/ sba-veered-from-guidelines-on-small-business-loans-report-says-11588971101. Amara Omeokwe. Coronavirus Drives 75% of Small Businesses to Seek Federal Aid. Wall Street Journal, May 2020c. ISSN 0099-9660. URL https://www.wsj.com/articles/ coronavirus-drives-75-of-small-businesses-to-seek-federal-aid-11589476452. Yuka Hayashi. U.S. Expected to Revise Small-Business Aid Program. Wall Street Journal, May 2020d. ISSN 0099-9660. URL https://www.wsj.com/articles/ u-s-expected-to-revise-small-business-aid-program-11589707801. Yuka Hayashi and Natalie Andrews. Senate Adjourns Without Approving Deal to Extend PPP Spending Window. Wall Street Journal, May 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ lawmakers-seek-to-double-duration-of-small-business-loans-11590058811.

Amara Omeokwe. SBA Under Spotlight as Frustrations With Small-Business Aid Programs Grow. Wall Street Journal, May 2020d. ISSN 0099-9660. URL https://www.wsj.com/articles/ sba-under-spotlight-as-frustrations-with-aid-programs-grow-11590053400. Yuka Hayashi. Questions Persist After Release of New Small-Business Loan Rules. Wall Street Journal, May 2020e. ISSN 0099-9660. URL https://www.wsj.com/articles/ questions-persist-after-release-of-new-small-business-loan-rules-11590269875. Natalie Andrews. Senate Approves Bill Extending Paycheck Protection Program. Wall Street Journal, June 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ senate-approves-bill-extending-paycheck-protection-program-11591226261. Yuka Hayashi. Major Fixes Made to Small-Business Loan Program. Wall Street Journal, June 2020f. ISSN 0099-9660. URL https://www.wsj.com/articles/ major-fixes-made-to-small-business-loan-program-11591390344. Ruth Simon and Peter Rudegeair. Small Businesses Tackle New PPP Puzzle: Forgiveness. Wall Street Journal, June 2020a. ISSN 0099-9660. URL https://www.wsj.com/ articles/small-businesses-tackle-new-ppp-puzzle-forgiveness-11592136025. Ruth Simon, Peter Rudegeair, and Yuka Hayashi. PPP Small-Business Loans Left Behind Many of Americas Neediest Firms. Wall Street Journal, June 2020. ISSN 0099-9660. URL https://www.wsj.com/articles/ ppp-small-business-loans-left-behind-many-of-americas-neediest-firms-11592407677. Ruth Simon and Peter Rudegeair. Small Businesses Raced to Spend PPP Funds but Covid-19 Pandemic Drags On. Wall Street Journal, June 2020b. ISSN 0099-9660. URL https://www.wsj.com/articles/ small-businesses-raced-to-spend-ppp-funds-but-covid-19-pandemic-drags-on-11593163801.

B Data Appendix B.1 Outbreak Severity Because local economic conditions and the resulting employment effects of the PPP are affected by the severity of the local Covid-19 outbreak and the non-pharmaceutical interventions (NPIs) used to fight it, we use Covid-19 cases, deaths, and NPIs as controls in some of our specifications. The Covid-19 confirmed cases and deaths are county level data from JHU CSSE (2020), which are the most prominent and frequently used source of case and death data. We use only their US data, which are cleaned and organized data provided by the US Centers for Disease Control and Prevention (CDC). B.2 Non-pharmaceutical Interventions NPI data are from the Oxford COVID-19 Government Response Tracker Hale et al. (2020). TheirrepositorycontainsNPIsattheStatelevelforall50statesandtheDistrictofColumbia. We assess the county NPIs during the week that contains the 12th of the month to align with the monthly timing of our labor market data.38 B.3 Occupational Exposure to the Covid-19 Shock We rank occupations according to their exposure to the Covid-19 shock on three dimensions, needforproximitytoothers, essentialness, andabilitytoworkfromhome. Todosowefollow the metrics generated by Leibovici et al. (2020) and Jackson (2020). B.4 Social Safety Net The CARES Act and subsequent executive action significantly increased eligibility for and assistancefromthesocialsafetynet,inparticularunemploymentinsurance. TheCARESAct mandatedanexpansionofunemploymentinsuranceeligibility–PandemicUnemploymentAs- $ sistance (PUA)–and an additional 600 per week, through the end of July, in unemployment insurance for those whose claims were approved–Pandemic Unemployment Compensation (PUC). States took variable amounts of time to implement the new policies resulting in one to two month differences in the start date of payments under the two programs (Nunn et al. (2020)). After these PUC lapsed, President Trump signed an executive order mandat- $ $ ing 300 to 400 extra per week in unemployment insurance under a program called “Lost 38As noted in section 3.2 the Current Population Survey queries about labor market activities during the week of the month that contains the 12th. 66

Wages Assistance” (LWA). These funds were supported by grants from the Federal Emergency Management Agency (FEMA). Due to variable time to approval for FEMA funds states began paying LWA and exhausted Federal funding or the program at different times (Singh (2020)). B.5 Additional details about the Current Population Survey In the fourth and the eighth monthly surveys, respondents report their weekly earnings if they are public or private employees. In addition, each March respondents also report their wageandsalaryincomeinthepastyearregardlessoftheircurrentemploymentstatusorclass of worker. Thus, for each respondent, we can calculate one to two proxies for the unemployment insurance that she would receive were she to claim and be eligible for unemployment insurance, given state level unemployment insurance laws, state level variation in the PUC, PUA, and LWA, and her past reported labor income.of Labor (2020) In our baseline specification we control for the state-level timing of PUC, PUA, and LWA participation interacted with class of worker (self-employed, private, public). Results are robust to controlling for unemployment insurance replacement rates under these programs inferred from the Earner Study or March Supplement income data and to replacement rates estimated using imputed earnings based on either of these sources of earnings data. The COVID-19 outbreak complicated the collection of CPS data in 2020, as it complicated virtually all other government, business, and household activities. The CPS is typically collected via a mixture of telephone and in-person interviews with in-person interviews primarily being used for the first and fifth wave of data collection. In March 2020, CPS suspended in-person interviews to protect its employees. In person interviews resumed in some parts of the country in July. These changes in operating protocol resulted in large increases in non-response and there is evidence that non-response is non-random. There is some evidence that the COVID-19 pandemic and related policy interventions had a nonneutral impact on the CPS’ representation of demographic subgroups (Ward and Edwards, 2020). To ameliorate the effects of these survey challenges, we control for interview type in all of our specifications. 67

C Online Appendix Tables 68

Table 12: Unemployment Effects. ShareDelayedX January -0.00912 -0.0105 -0.0109 -0.00906 -0.00917 -0.00877 -0.00930 -0.0221 -0.0108 (0.0154) (0.0153) (0.0154) (0.0154) (0.0154) (0.0155) (0.0154) (0.0219) (0.0154) February -0.0193 -0.0225 -0.0224 -0.0192 -0.0195 -0.0180 -0.0200 -0.0367 -0.0224 (0.0191) (0.0188) (0.0189) (0.0191) (0.0191) (0.0192) (0.0191) (0.0254) (0.0189) March 0.00459 0.000297 -0.0000610 0.00499 -0.000699 0.00450 0.00407 -0.0434 -0.00432 (0.0231) (0.0225) (0.0225) (0.0231) (0.0231) (0.0231) (0.0231) (0.0313) (0.0226) April -0.00208 -0.0129 -0.0126 -0.0206 -0.00165 -0.00136 -0.00383 -0.0456 -0.0399 (0.0541) (0.0433) (0.0432) (0.0502) (0.0541) (0.0540) (0.0540) (0.0572) (0.0429) May 0.116** 0.106*** 0.108*** 0.100** 0.120** 0.117** 0.131** 0.0883* 0.105*** (0.0503) (0.0373) (0.0376) (0.0471) (0.0506) (0.0503) (0.0511) (0.0513) (0.0387) June 0.115** 0.102*** 0.102*** 0.114** 0.112** 0.117** 0.110** 0.0622 0.0932** (0.0499) (0.0387) (0.0389) (0.0454) (0.0493) (0.0499) (0.0496) (0.0535) (0.0372) July 0.113*** 0.0983*** 0.0973*** 0.111*** 0.103** 0.114*** 0.107** 0.0670 0.0827** (0.0439) (0.0353) (0.0354) (0.0423) (0.0429) (0.0437) (0.0437) (0.0468) (0.0347) August 0.109*** 0.0953*** 0.0953*** 0.105*** 0.102*** 0.110*** 0.106*** 0.0698* 0.0842*** (0.0379) (0.0307) (0.0305) (0.0380) (0.0372) (0.0378) (0.0378) (0.0420) (0.0307) September 0.126*** 0.117*** 0.117*** 0.120*** 0.119*** 0.127*** 0.118*** 0.0925** 0.103*** (0.0363) (0.0305) (0.0305) (0.0362) (0.0357) (0.0363) (0.0362) (0.0406) (0.0305) October 0.0799** 0.0705** 0.0721*** 0.0839*** 0.0720** 0.0800** 0.0775** 0.0517 0.0662** (0.0316) (0.0275) (0.0274) (0.0314) (0.0308) (0.0315) (0.0315) (0.0373) (0.0273) R2 0.665 0.675 0.674 0.666 0.666 0.666 0.666 0.675 0.675 N 792,197 792,197 792,197 792,197 792,197 792,197 792,197 792,197 792,197 Individual YES YES YES YES YES YES YES YES YES Month YES YES YES YES YES YES YES YES YES Month-in-2020 YES YES YES YES YES YES YES YES YES Controls IndustryandOccupation A B B B CasesandDeaths C C Non-pharmaceuticalInterventions D D CovidInducedMeasurementChanges E E E UnemploymentInsurance F F State×MonthFE H Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. A:Fixedeffectsfor2-DigitNAICSand22BroadIndustriesfullyinteractedwithMonth-in-2020fixedeffects. B: Fixed effects for 2-Digit NAICS and controls for occupational exposure to the Covid shock fully interacted with Month-in-2020 fixed effects. Occupational exposure is measuredbyabilitytoworkfromhome,requiredproximitytoothers,andessentialness. CodingfollowsCITE,CITE,andCITE,respectively. C:NewcasesandnewdeathsintheCBSA.NewYorkCityisanoutlier,asaresultweallowforthecoefficientoncasesanddeathstodifferforthisCBSA.Datacomefrom theCOVID-19DataRepositorybytheCenterforSystemsScienceandEngineering(CSSE)atJohnsHopkinsUniversity,2020,https://github.com/CSSEGISandData/COVID-19. D:StatelevelstringencyIndex: OxfordCOVID-19GovernmentResponseTracker. Haleetal.(2020). E:FixedeffectsformonthinwhichtherespondentwasfirstinterviewedbytheCPSandforinterviewtypefullyinteractedwithMonth-in-2020fixedeffects. F: Indicator for whether PUC, PUA, and/or LWA were made in the respondent’s state in the reference week interacted with class of worker. Data come from https://www.brookings.edu/blog/up-front/2020/05/13/incomes-have-crashed-how-much-has-unemployment-insurance-helped/ and https://www.savingtoinvest.com/ when-will-300-lwa-unemployment-start-being-paid-in-my-state/.\ 69

Table 13: Nonemployment Effects. ShareDelayedX January -0.0138 -0.0166 -0.0169 -0.0136 -0.0138 -0.0120 -0.0135 -0.00901 -0.0150 (0.0217) (0.0217) (0.0217) (0.0217) (0.0217) (0.0215) (0.0217) (0.0304) (0.0215) February -0.0155 -0.0186 -0.0188 -0.0153 -0.0158 -0.0139 -0.0158 -0.0474 -0.0179 (0.0251) (0.0245) (0.0246) (0.0251) (0.0251) (0.0250) (0.0251) (0.0344) (0.0245) March 0.0227 0.0180 0.0171 0.0234 0.0152 0.0233 0.0227 -0.0358 0.0121 (0.0286) (0.0279) (0.0277) (0.0286) (0.0283) (0.0286) (0.0287) (0.0392) (0.0276) April 0.0112 0.00244 -0.000630 -0.0245 0.0117 0.0115 0.00791 -0.0261 -0.0453 (0.0604) (0.0479) (0.0480) (0.0545) (0.0604) (0.0602) (0.0600) (0.0629) (0.0461) May 0.157*** 0.147*** 0.148*** 0.135*** 0.162*** 0.158*** 0.171*** 0.125** 0.141*** (0.0553) (0.0412) (0.0412) (0.0514) (0.0555) (0.0552) (0.0560) (0.0570) (0.0421) June 0.123** 0.112*** 0.112*** 0.126** 0.118** 0.124** 0.118** 0.0541 0.108*** (0.0545) (0.0423) (0.0425) (0.0497) (0.0537) (0.0544) (0.0542) (0.0581) (0.0405) July 0.115** 0.102*** 0.102*** 0.112** 0.101** 0.114** 0.110** 0.0487 0.0827** (0.0481) (0.0395) (0.0396) (0.0465) (0.0470) (0.0481) (0.0480) (0.0543) (0.0389) August 0.140*** 0.127*** 0.128*** 0.132*** 0.130*** 0.136*** 0.137*** 0.109** 0.106*** (0.0428) (0.0358) (0.0356) (0.0430) (0.0421) (0.0429) (0.0428) (0.0511) (0.0359) September 0.134*** 0.118*** 0.122*** 0.125*** 0.124*** 0.133*** 0.130*** 0.0962* 0.103*** (0.0435) (0.0357) (0.0357) (0.0433) (0.0426) (0.0433) (0.0443) (0.0501) (0.0361) October 0.0850** 0.0708** 0.0745** 0.0851** 0.0742* 0.0782* 0.0810** 0.0363 0.0541 (0.0411) (0.0351) (0.0352) (0.0408) (0.0400) (0.0407) (0.0409) (0.0490) (0.0348) R2 0.639 0.645 0.645 0.639 0.639 0.642 0.639 0.649 0.649 N 831,737 831,737 831,737 831,737 831,737 831,737 831,737 831,737 831,737 Individual YES YES YES YES YES YES YES YES YES Month YES YES YES YES YES YES YES YES YES Month-in-2020 YES YES YES YES YES YES YES YES YES Controls IndustryandOccupation A B B B CasesandDeaths C C Non-pharmaceuticalInterventions D D CovidInducedMeasurementChanges E E E nonemploymentInsurance F F State×MonthFE H Standarderrors,clusteredattheCBSAXIndustry,inparentheses. *p<0.10,**p<0.05,***p<0.01. A:Fixedeffectsfor2-DigitNAICSand22BroadIndustriesfullyinteractedwithMonth-in-2020fixedeffects. B:Fixedeffectsfor2-DigitNAICSandcontrolsforoccupationalexposuretotheCovidshockfullyinteractedwithMonth-in-2020fixedeffects. Occupationalexposureis measuredbyabilitytoworkfromhome,requiredproximitytoothers,andessentialness. CodingfollowsCITE,CITE,andCITE,respectively. C:NewcasesandnewdeathsintheCBSA.NewYorkCityisanoutlier,asaresultweallowforthecoefficientoncasesanddeathstodifferforthisCBSA.Datacomefrom theCOVID-19DataRepositorybytheCenterforSystemsScienceandEngineering(CSSE)atJohnsHopkinsUniversity,2020,https://github.com/CSSEGISandData/COVID-19. D:StatelevelstringencyIndex: OxfordCOVID-19GovernmentResponseTracker. Haleetal.(2020). E:FixedeffectsformonthinwhichtherespondentwasfirstinterviewedbytheCPSandforinterviewtypefullyinteractedwithMonth-in-2020fixedeffects. F: Indicator for whether PUC, PUA, and/or LWA were made in the respondent’s state in the reference week interacted with class of worker. Data come from https://www.brookings.edu/blog/up-front/2020/05/13/incomes-have-crashed-how-much-has-unemployment-insurance-helped/ and https://www.savingtoinvest.com/ when-will-300-lwa-unemployment-start-being-paid-in-my-state/. 70