Earnings Business Cycles: The Covid Recession, Recovery, and Policy Response

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Earnings Business Cycles: The Covid Recession, Recovery, and Policy Response Jeff Larrimore, Jacob Mortenson, and David Splinter 2023-004 Please cite this paper as: Larrimore, Jeff, Jacob Mortenson, and David Splinter (2023). “Earnings Business Cycles: The Covid Recession, Recovery, and Policy Response,” Finance and Economics Discussion Series 2023-004. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.004. 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.

Earnings Business Cycles: The Covid Recession, Recovery, and Policy Response Jeff Larrimorea Federal Reserve Board Jacob Mortensonb Joint Committee on Taxation David Splinterc Joint Committee on Taxation December 2022 Abstract Using a panel of tax data, we follow the earnings of individuals over business cycles. Compared to prior recessions, the Covid policy response and recovery were far more progressive. Among workers starting in the bottom quintile, median real earnings including fiscal relief increased 66 percent in 2020 and earnings increases offset relief decreases in the 2021 recovery. After the prior two recessions, this measure had decreased by 24 percent. Among those starting in the top quintile, median and average real earnings were approximately unchanged. This difference from prior recessions is largely attributable to larger Covid-era stimulus payments and unemployment insurance. Keywords: Covid-19, wages, earnings, stimulus checks, unemployment insurance, countercyclical policy, government transfers JEL: D31, E24, H53, J30, J65 ________________________________________________________________________ The online appendix and data are available at david.splinter.com. For helpful comments, we thank Tom Barthold, Yonatan Berman, William Boning, Jim Cilke, Connor Dowd, Michael Love, Bert Lue, Robert Moffitt, Rachel Moore, Ryan Nunn, John Sabelhaus, Daniel Waldenström, James Ziliak, and Eric Zwick. We thank Eric Heiser for assistance with the late-data imputation procedure. Larrimore: The results and opinions expressed in this paper reflect the views of the authors and should not be attributed to the Federal Reserve Board. Mortenson and Splinter: This paper embodies work undertaken for the staff of the Joint Committee on Taxation to improve individual income tax modeling, but as members of both parties and both houses of Congress comprise the Joint Committee on Taxation, this work should not be construed to represent the position of any member of the Committee. a Corresponding Author. 20th St. and Constitution Ave. N.W., Washington, DC 20551. Jeff.Larrimore@frb.gov b 502 Ford House Office Building, Washington, DC 20515. Jake.Mortenson@jct.gov c 502 Ford House Office Building, Washington, DC 20515. David.Splinter@jct.gov 1

1. Introduction The Covid recession was historically short, lasting just two months. Nevertheless, the severity of economic disruptions caused many low-earning workers’ annual earnings to decline substantially in 2020. While employment rebounded in 2021, fiscal relief declined and inflation increased, offsetting some economic gains. Using a panel of administrative data including all wage earners, we measure the evolution of individual-level earnings over recent recessions and through the first year and a half of the Covid recovery, both with and without fiscal relief. Compared to prior recessions, the Covid recession was more regressive, but the Covid recovery and policy response were far more progressive. The data we use offer several advantages relative to other sources. First, we follow the same individuals over time, contrasting with widely available cross-sectional estimates, which compare different individuals over time.1 Second, our estimates use extremely large samples—5 percent of workers—that exceed those of available survey data. Third, using independent reporting by employers and governments, we precisely measure earnings and direct fiscal relief (i.e., unemployment insurance benefits, stimulus payments, tax holidays, and various tax credits). Using these data, we find that after declining markedly in 2020, earnings among lowearning workers increased in 2021 (after accounting for inflation). Despite these improvements, real earnings among those who were in the bottom quintile before the recession typically remained below their 2019 levels. In contrast, among top-quintile workers, real earnings increased in 2020, but declined in 2021 as inflation accelerated, leaving their median earnings slightly below 2019 levels. Incorporating the substantial and progressive fiscal relief during this period paints a different picture. For bottom-quintile workers, market earnings gains in 2021 mostly offset the partial withdrawal of fiscal relief. Consequently, earnings after fiscal relief for the bottom quintile remained high in 2021—on average, more than double pre-recession earnings. We perform a similar analysis on the 2001 and 2008 recessions and find substantially different patterns. We consider four measures of the distribution of real earnings changes from the year before to the years after a recession: the share of workers whose earnings increased, the median earnings change by quintile, the share of workers with large changes in earnings, and the distribution-wide progressivity of earnings changes. Results are consistent across these measures: the initial distributional impacts of market earnings changes in the Covid recession were more regressive than in prior recessions, whereas both the Covid recovery and policy response have been far more progressive. 1 See Semega and Kollar (2022) and Blanchet, Saez, and Zucman (2022) for cross-sectional income trends. Relative to these cross-sectional discussions, panel data control for significant re-ranking due to earnings mobility and are valuable for following the trajectory of the same individuals. See online appendix Figure A5 for a comparison of panel and cross-sectional results. 2

2. Relation to Existing Research on Earnings During the Pandemic This paper expands upon the research examining earnings trends in the pandemic. Research based on micro-level survey data (Moffitt and Ziliak 2020; Montenovo et al. 2020; Cortes and Forsythe 2022), macro-level administrative data (Berman 2020; Blanchet, Saez, and Zucman 2022), and data from state governments and private companies (Bartik et al. 2020; Cajner et al. 2020; Chetty et al. 2022) consistently show disproportionate initial losses among low-wage occupations in 2020. See Cortes and Forsythe (2022) for a review of the extensive literature on the 2020 downturn. Additionally, Cortes and Forsythe (2021) and Larrimore, Mortenson, and Splinter (2022a) found that low-earning workers received the bulk of the direct fiscal relief to households and families, offsetting increases in market earnings inequality. This existing research, however, has primarily focused on 2020 rather than medium-term trends into 2021. The tax data used here track individuals over long time-periods and measure individual earnings changes before, during, and after the pandemic shock. These data allow all wage earners (whether they file a tax return or not) to be followed for multiple years, providing precise microlevel earnings changes. In contrast, the Current Population Survey (CPS) can only track individuals for one year, and even then, only for individuals who do not change residence. Major panel surveys, such as the Panel Survey of Income Dynamics, can track people over time but have delayed reporting, smaller samples, and must contend with sample attrition (Fitzgerald, Gottschalk, and Moffit 1988). Consequently, we are unaware of other research showing how the same workers from before the pandemic fared two-years later in 2021. In addition to their ability to accurately track individual earnings over time, tax data also capture unemployment benefits and tax credits that are underreported in survey data (Larrimore, Mortenson, and Splinter forthcoming; Meyer et al. 2020). While underreporting of fiscal relief is a regular concern, survey errors were larger than usual during the Covid pandemic (Rothbaum and Bee 2021). Hence, this research using administrative data can also fill gaps in our understanding of recent trends that resulted from the challenges that the pandemic caused for economic surveys. Although we are unaware of other work documenting the extent to which individual workers saw earnings gains in 2021, our findings are consistent with Greig, Deadman, and Sonthalia’s (2022) findings on checking account balances. They observed that at the end of 2021, balances were well above their 2019 levels, and these increases were most pronounced among low-income families with bank accounts. Similarly, Meyer, Murphy, and Sullivan (2022) observed that lower-income groups did not substantially reduce consumption since the pandemic began, suggesting that fiscal relief’s income stabilization flowed through to consumption. Our estimates on market earnings patterns in prior recessions are also consistent with earlier findings using administrative earnings data. Guvenen, Ozkan, and Song (2014) estimated that during recessions, earnings increases among high-earning worker become less frequent and earnings losses among low-earning workers become more frequent. McKinney, Abowd, and Janicki (forthcoming) also observed procyclical skewness of annual earnings changes. Our 3

findings on the effects of tax and transfer policies are also consistent with findings from other countries (Busch et al. 2022). 3. Administrative Tax Data Panel This paper builds on the data and methods from Larrimore, Mortenson, and Splinter (2022a), which used a random 5-percent sample of all individuals appearing in IRS tax data (Internal Revenue Service and Joint Committee on Taxation 2022). We extend the analysis forward to 2021 to consider the Covid recovery and back to 1999 to allow for comparisons with the 2001 recession. A. Tax Data Sources Annual wages and salaries (“earnings”) are measured from Form W-2, and unemployment insurance benefits are measured using Form 1099-G and Form 1040 tax returns (the latter is used if larger than the former). Estimates from IRS audit studies suggest that these information returns are extremely comprehensive and only miss an estimated one percent of earnings. To incorporate measures of stimulus payments and other fiscal relief observed in tax data, we use Form 1040 tax returns and other population-level data, as discussed below. B. Panel of Individual-Level Earnings Measures All estimates are based on multi-year panels of earnings and fiscal relief. These panels include anyone in the initial year with wages or unemployment benefits, even if they have no income from either source in subsequent years. We focus on those with this labor market attachment in years just prior to recessions (2000, 2007, and 2019). While we allow for exits (earnings of zero dollars), we do not consider changes in entry rates into the labor force during each recession.2 The sample is limited to adults aged 24 or older in the initial year, excluding those who are deceased at the end of the year. The age range avoids large earnings fluctuations among young workers and retains earnings declines among retirees who were working before the pandemic.3 All data are at the individual level. Individual-level totals for earnings and unemployment insurance benefits are obtained by aggregating all W-2 forms and 1099-G forms.4 Stimulus payments and other tax credits, as well as the self-employment portion of the payroll tax holiday, are determined at the tax-unit level and then divided equally between spouses for tax units filing 2 Entry and exit rates are largely similar each year from 1999 through 2021 and therefore do not appear to substantially affect our results. For example, entry rates range from 4.3 to 5.9 percent with an average of 5.4 percent. For workers at least 60 years-old, entry rates are lower (3.3 percent in the last two recessions) and exit rates are higher (11 and 12 percent in the last two recessions). See online appendix Figures A2 and A3 and online data Tables B2 and B3. 3 This captures accelerated retirements resulting from the Covid recession. Domash and Summers (2022) estimated that 1.3 million workers left the workforce due to aging reasons unrelated to the pandemic and another 1.3 million left due to accelerated retirements. Results for working-age adults (ages 25–59) in 2020 were similar to those for all adults over age 24 (Larrimore, Mortenson, and Splinter 2022a). 4 We focus on Medicare Wages (Box 5) on Form W-2, which is the broadest wage measure on the form. We retain the most recent Form W-2 with a non-missing amount for each individual from each employer in each year. These forms do not include self-employment income. Expanded unemployment insurance included Pandemic Unemployment Assistance payments to independent contractors, although our sample definition means these recipients are only included if they had earnings or unemployment benefits in the initial year. 4

joint returns. This equal split accords with the per-person basis of most of these policies, which are described below. Our individual-level earnings definition excludes other income sources and generally ignores tax-unit sharing, which requires tax return data that is not yet sufficiently available for 2021 (for estimates accounting for tax-unit sharing in 2020, see Larrimore, Mortenson, and Splinter 2022b). Inflation-adjusted values are based on the chained-CPI. Once the IRS data are complete, they represent a population-level panel. To provide timely estimates, we use current data even though some 2021 forms are yet to be processed by the IRS. In these data, some workers with 2021 earnings appear to have no earnings and some with multiple jobs have just one processed Form W-2, resulting in artificially lower earnings. We account for this by estimating the number of people with not-yet-processed 2021 W-2 forms and the earnings on those forms. This imputation is based on historical patterns of late-processed forms based on age, prior-year wages, and W-2 presence in the current tax year (see online appendix B for details). Since our data (from August 2022) are nearly complete, the imputation has relatively small effects, increasing 2021 median earnings by $100 within the bottom-quintile and about half that for higher earnings. In our final data, aggregate real wage earnings increased by 4.9 percent between 2019 and 2021, which is close to the 4.7 percent increase in aggregate real earnings in the National Income and Products Accounts over this period. C. Relief Considered During Recent Recessions In addition to unemployment benefits, we include other direct fiscal relief observed in tax data in our post-relief earnings measure. However, many sources of direct fiscal relief are not reported in tax data (e.g., SNAP, indirect relief via Federal Reserve policies, and indirect relief through local programs or charitable organizations). Note that to the extent macroeconomic policies or Payment Protection Plan forgiven loans resulted in higher wages, they are included in our earnings measures, although not broken out, as this would require a counterfactual estimate of lost earnings (Autor et al. 2022; Dalton 2022). Total fiscal relief increased across the three recessions considered. Fiscal relief was about $200 billion in 2001–2003, $500 billion in 2008–2010, and $1,800 billion in 2020–2021.5 Larger Covid-era fiscal relief mostly resulted from expanded unemployment insurance benefits and stimulus payments. Unemployment insurance benefits increased from $140 billion and $350 billion in the earlier periods to $880 billion in the Covid era. Stimulus payments also increased over the last three recessions. The 2001 stimulus payments totaled $38 billion and approximately two-thirds of tax units received a payment of $300 per adult (Kaplan and Violante 2014). The 2008 payments totaled $96 billion and maximum amounts were generally $600 per taxpayer and $300 per child younger than 17 years old (Parker et al. 2013). The 2020 payments were far larger. They totaled $413 billion, including $280 billion for the first round of payments. Most individuals received $1,800 per non-dependent filer and $1,100 per qualifying 5 Note that our panel focuses on employees and therefore excludes relief going to non-employees, such as Covid-era unemployment benefits going to the self-employed and stimulus payments to retirees. Among workers in our data, fiscal relief was about $170 billion in 2001–2003, $490 billion in 2008–2010, and $1,380 billion in 2020–2021 (in 2021 dollars: $240 billion, $600 billion, and $1,420 billion). 5

child. In 2021, additional stimulus payments were distributed, including $408 billion from thirdround payments that were $1,400 per non-dependent filer and qualifying child.6 Additionally, we account for stimulus check “true ups” on 2001, 2008, and 2020 tax returns in the year when the payment is received (Splinter forthcoming). Several other provisions that provided relief during the Covid Recession and Great Recession are also included. The making work pay credit for 2010 and 2011 totaled $60 billion each year and was $400 per adult worker (subject to an income phase out). The payroll tax holiday for 2012 and 2013 reduced employee payroll taxes by $100 billion each year and consisted of a two-percentage point tax rate reduction.7 In 2021, most parents received advance child tax credits (CTCs) of $1,800 per qualifying child under age six and $1,500 per qualifying child under age 18. 4. Estimates of Earnings Changes and Relief Across Business Cycles To show the evolution of earnings over recent business cycles, we follow individual workers over time. For each recession, workers are grouped into quintiles or percentiles of the prerecession earnings distribution (including wages plus unemployment insurance, to more closely reflect their usual pre-recession earnings). These rankings into quintiles and percentiles are consistent across all figures, maintaining the same rankings when looking at earnings changes with and without relief. Importantly, our panel approach differs from evaluating trends using repeated cross-sections because individuals are always classified into quintiles or percentiles based on their pre-recession earnings, thereby allowing us to focus on individual-level earnings mobility. Relative to prior recessions, market earnings changes in the Covid recession were more regressive, but both the policy response and the Covid recovery have been more progressive. This is seen with each of our four measures: (a) the share of workers whose earnings increased; (b) median earnings changes within earnings groups; (c) the share of workers with large (10 percent or more) changes in earnings; and (d) Gini-based measures summarize the progressivity of changes across the distribution. Finally, we show how specific parts of fiscal relief, especially unemployment insurance and stimulus payments, offset earnings declines in the Covid recession and recovery. A. Frequency of Earnings Increases and Declines During the Covid Recession Across the entire population, market earnings for most workers were resilient through the Covid recession, as 51 percent had real earnings increases in 2020. With rising inflation, the share with real earnings above pre-recession levels ticked down slightly in 2021, but just over half (50.4 6 Most second-round payments were made in 2020 and are allocated to 2020 based on actual payment dates. The 2008, 2020, and 2021 stimulus checks were subject to income phase outs starting at $75,000 for single filers ($150,000 for married filers). For details, see Splinter (forthcoming). 7 We estimate the individual-specific amount of this tax holiday using individual measures of earnings from Form W-2. 6

percent) still had higher earnings than in 2019. In the Great Recession a slightly smaller share, 47 percent, had earnings above their pre-recession levels in both 2008 and 2009. Yet earnings trends differ through the distribution. When considering market earnings by quintile of pre-recession earnings in Figure 1 (dashed lines), low-earning workers were typically making less in 2021 than they were in 2019 before the recession. Of these bottom-quintile workers, 44 percent had higher real earnings in 2021 than two years earlier, while the rest had lower earnings. This reflects frequent earnings declines in 2020, followed by a partial recovery in 2021. Just 41 percent of bottom-quintile workers had increases in real market earnings in 2020. For comparison, in the first year of the Great Recession, 47 percent had earnings increases in 2008. But the Covid recession was also historically short, and the share of bottom-quintile workers with real earnings above their pre-recession level increased slightly in 2021. In contrast, during the Great Recession, the share of bottom-quintile workers with earnings above pre-recession levels fell in the second and third year after the start of the recession—past the official end date of the recession. Hence, a larger share of low-earning workers made more in 2021 than before the recession than was the case in 2009 or 2010 following the Great Recession. The effects of the Covid recession on market earnings were also particularly regressive. Contrast the bottom-quintile results in the left panel with those for the top quintile on the right, still focusing on the dashed lines (results for all quintiles are in the online data). In 2020, a majority (57 percent) of those who started in the top quintile had market earnings gains despite the economic downturn. In 2021, the share with real earnings gains among this group fell, although slightly over half of these workers still had earnings above pre-recession levels. 7

Figure 1. Share of workers with real earnings at pre-recession level or higher, by income quintile 75% 50% 25% 0% 2001 2002 2003 2008 2009 2010 2020 2021 Note: Among workers aged 25 to 99 with earnings or unemployment benefits in pre-recession year and alive at end of specific year. Quintiles (with and without relief) are based on wages plus unemployment benefits in pre-recession years (2000, 2007, and 2019). Earnings are indexed with the chained CPI-U. Source: Authors' calculations using tax data. Source: Author’s calculations using Internal Revenue Service and Joint Committee on Taxation (2022) tax data. Once incorporating fiscal relief (solid bars), however, the pattern is quite different. Comparing the solid and dashed lines in Figure 1 highlights the progressive effect of relief in offsetting earnings losses. Among workers who were in the bottom quintile before the Covid recession, relief increased the share with earnings increases in 2020 by 31 percentage points— from 41 percent to 72 percent. For the middle quintile (not shown) and the top quintile, this increase is only 19 and 5 percentage points, respectively—an indication of progressive fiscal relief. In 2021, the total amount of direct fiscal relief fell by about one-third relative to 2020. Nevertheless, most low-earning workers had higher earnings with relief than before the recession. From 2019 to 2021, earnings with relief increased for 69 percent of the bottom quintile. B. Magnitudes of Earnings Changes Figure 2 goes beyond the share with earnings increases to also consider the magnitude of earnings changes. It displays median real earnings of workers in each quintile relative to their median earnings in the year before each recession. Once again, the dotted bars only include market earnings, and the solid bars add fiscal relief. The Covid recession stands out for its severe effects on market earnings of low-earning workers as well as its progressive recovery. For workers starting in the bottom-quintile in 2019, real median market earnings in 2021 were 26 percent below 2019 levels (see dashed lines). This reflected a decline of nearly one-third in 2020 followed by a slight increase in 2021. Among those who started in the middle quintile, median earnings remained relatively flat, and among the top quintile, real median earnings declined slightly in 2021. Hence, despite a progressive recovery in 2021, the combined effect on market earnings since 2019 was regressive. 8 raey noissecer-erp evoba sgninrae htiw erahS Bottom Quintile 2001 recession 2008 recession 2020 recession 75% with relief (solid) without relief (dashed) 2020 relief 50% 25% 0% 2001 2002 2003 2008 2009 2010 2020 2021 Top Quintile 2001 recession 2008 recession 2020 recession with relief (solid) without relief (dashed) 2020 relief

Figure 2. Median real earnings relative to pre-recession year, by income quintile 70% 50% 30% 10% -10% -30% -50% 2001 2002 2003 2008 2009 2010 2020 2021 Note: Among workers aged 25 to 99 with earnings or unemployment benefits in the pre-recession year and alive at end of specific year. Earnings are indexed with the chained CPI-U. Quintiles (with and without relief) are based on wages plus unemployment benefits in pre-recession years (2000, 2007, and 2019). Median real earnings without relief (2021 dollars) for pre-recession years (2000, 2007, and 2019)—bottom quintile: $5,800, $5,910, $7,560; middle quintile: $38,390, $39,480, $42,770; top quintile: $100,350, $106,850, $120,450. Relief increases median real earnings in 2021 by $6,980 for the bottom quintile, $3,120 for the middle quintile, and $1,860 for the top quintile. Source: Authors' calculations using tax data. Source: Author’s calculations using Internal Revenue Service and Joint Committee on Taxation (2022) tax data. Low-wage workers benefited greatly from progressive policy responses in the Covid recession and recovery—especially expanded unemployment benefits and stimulus checks. When including fiscal relief, as shown in the solid bars in Figure 2, bottom-quintile workers saw their real median earnings with relief increase by 66 percent in 2020 and then remain up 62 percent in 2021 as market earnings gains largely offset the withdrawal of fiscal relief. Middle-quintile real 9 sgninrae laer naidem ni egnahC raey noissecer-erp ot derapmoc A. Bottom Quintile 2001 recession 2008 recession 2020 recession with relief (solid) without relief (dashed) 2020 relief 20% 15% 10% 5% 0% -5% 2001 2002 2003 2008 2009 2010 2020 2021 sgninrae laer naidem ni egnahC raey noissecer-erp ot derapmoc B. Middle Quintile 20% 2001 recession 2008 recession 2020 recession 15% 10% with relief (solid) without relief (dashed) 2020 5% relief 0% -5% 2001 2002 2003 2008 2009 2010 2020 2021 C. Top Quintile 2001 recession 2008 recession 2020 recession with relief (solid) without relief (dashed) 2020 relief

median earnings with relief increased by 9 percent in 2020 and remained 8 percent above 2019 levels in 2021 (note the change in scale in Figure 2 for the middle and top quintiles). Top quintile median earnings with relief increased by only 2 percent in 2020, and then returned to near-2019 levels in 2021.8 The 62 percent median increase after fiscal relief from 2019 to 2021 for this group was remarkably large and far exceeded that seen even during recent non-recessionary periods, highlighting the magnitude of the progressive policy response. Since 1999, the median increase in bottom-quintile real earnings (after relief) over a two-year period never exceeded 11 percent prior to the Covid recession. Frequently, the median two-year change for this group was slightly negative.9 C. Frequency of Large Earnings Changes during Covid For some workers, increases or decreases in earnings may be small. To consider the more dramatic effects of recessions and recoveries on workers, we also consider the share of workers with large real annual earnings changes, excluding small changes. Large changes are defined as 10 percent or more, where large declines include those exiting the workforce (i.e., going from positive to no earnings). Consistent with the earlier discussion, those entering the workforce are excluded. Over the last two decades, an average of 28 percent of workers had large increases and 28 percent had large declines each year. Other estimates using administrative tax data find similarly high shares of workers with large short-term earnings changes (Congressional Budget Office 2008).10 Expansions often coincide with more workers having large increases. In the 2012–2019 expansion, the share with large increases exceeded the share with large declines by an average of 3 percentage points. In 2021, the first full year of the Covid recovery, the share of workers with a large earnings increase was 3 percentage points above the share with a large earnings decline. 8 Small business owners benefitted from forgiven Payment Protection Plan loans. Since this paper focuses on employee wages and these loans are not captured on individual tax records, they are not included here. Autor et al. (2022) estimated that up to one third of forgiven loans flowed through to 2020 wages (hence included in our earnings measures), and that the rest increased top-quintile household incomes about five percent. This implies these loans would increase our average top-quintile annual earnings changes with relief to near the fourth-quintile increase but well below the bottom quintile’s 90 percent average increase. 9 As is standard in panel-based studies, average earnings growth was larger for low-wage workers than their median earnings growth. This is because the panel includes people who are just starting in the workforce and because there is mean reversion among those with temporary earnings declines. This earnings growth, however, is offset somewhat by workers aged 60 or older with large declines (online data Table B1). When removing initial-year workers with earnings below $5,000, bottom-quintile median earnings decreases were less common and smaller (online data Tables B4 and B5). However, even when looking at average earnings, the increases in bottom-quintile earnings after relief in the Covid recession exceeded other two-year periods since 1999. Bottom-quintile workers saw their real average earnings with relief increase by 90 percent in 2020 and then increase to 107 percent above 2019 levels in 2021 (online appendix Figure A1). 10 After controlling for age, large earnings losses are associated with unemployment spells, interstate moves, and divorce; while large increases are associated with adding dependents and being married (Larrimore, Mortenson, and Splinter 2016). 10

However, one-year improvements include mean reversion of prior-year losses, which is why we also consider two-year changes. Over the two-year period from 2019–2021, large earnings increases before fiscal relief were 1 percentage point less common than large earnings declines (34 percent vs. 35 percent). For comparison, in the pre-Covid expansion years from 2017 to 2019, large increases were 7 percentage points more frequent than large decreases. Next, we consider the frequency of large earnings gains and declines across the earnings distribution. Figure 3 displays the share of workers with at least a 10 percent real earnings decline or increase by earnings percentile. Panel A displays the share with large annual decreases for 2019, before the Covid recession, and for 2020 and 2021. As previously illustrated by Larrimore, Mortenson, and Splinter (2022a), the share with large market earnings declines during the 2020 recession was elevated throughout the distribution relative to 2019, but especially so among the bottom half of the distribution. In 2021, the share of workers with large earnings declines for the bottom quintile is above the 2019 analog for each earnings percentile, although the shape is similar. Panel B shows the share of workers with large earnings increases. In 2020, the share experiencing a large earnings increase was similar to that in 2019 for the top of the distribution but depressed for the bottom half. The 2021 pattern was also similar for those in the top half of the distribution, but there was a notable surge in earnings increases for those with earnings between the 10th and 40th percentiles. Shifting to two-year changes in Panels C and D, the progressive recovery in market earnings somewhat offset the regressive nature of the recession. Between 2019 and 2021, 51 percent of workers who were in the bottom quintile had a large earnings decline. This is 7 percentage points above the 44 percent with large two-year earnings declines from 2017 to 2019. Among the top quintile, the 27 percent of workers with large earnings declines between 2019 and 2021 was 3 percentage points above the share with large earnings declines from 2017 to 2019. Similarly, large earnings increases among the top quintile were 2 percentage points less likely between 2019 and 2021 than in the two preceding years. However, despite being smaller than in 2020, fiscal relief continued to be extremely progressive into 2021. Once incorporating fiscal relief, large earnings declines were far less common among the bottom half of the distribution from 2019 through 2021 than during the most recent expansion, and large income gains were far more common for everyone outside of the top decile.11 Among the bottom quintile, relief decreased the share with large declines from 51 percent to 29 percent and increased the share with large increases from 41 percent to 65 percent. 11 Note that the progressivity of relief in 2020 is not contributing to this observed progressivity in 2021, since it only compares calendar years 2019 and 2021. 11

Figure 3: Share of workers with large real earnings decline or increase 60 50 40 30 20 10 0 0 20 40 60 80 Note: Among workers aged 25 to 99 with earnings or unemployment income in the initial year (t–1 or t–2) and alive at end of final year. Percentiles are based on wages plus unemployment benefits in initial years. Earnings are indexed with the chained CPI-U. Source: Authors' calculations using tax data. Source: Author’s calculations using Internal Revenue Service and Joint Committee on Taxation (2022) tax data. D. Progressivity of Earnings Changes and Fiscal Relief To summarize the progressivity of earnings changes across the entire distribution, we estimate percentage decreases in inequality for the same population. This new measure extends Gini coefficients to a panel framework. We define the progressivity of earnings changes as the initial-year earnings Gini coefficient less the final-year concentration index (the final-year Gini modified to rank workers by initial-year earnings) divided by the initial-year Gini coefficient. In the years for which data is available, this measure is always positive because earnings changes among initial-year workers are always sufficiently progressive to overcome any increases in cross- 12 srekrow fo tnecreP A: 10% real earnings decline: 1-year 100 2019-20 80 2020-21 60 2018-19 40 20 0 0 20 40 60 80 Prior-year earnings centile (year t–1) srekrow fo tnecreP B: 10% real earnings increase: 1-year 2018-19 2020-21 2019-20 Prior-year earnings centile (year t–1) 60 50 40 30 20 10 0 0 20 40 60 80 srekrow fo tnecreP C: 10% real earnings decline: 2-year 100 80 2019-21 2017-19 60 40 2019-21(+relief) 20 0 0 20 40 60 80 Prior-year earnings centile (year t–2) srekrow fo tnecreP D: 10% real earnings increase: 2-year 2019-21(+relief) 2017-19 2019-21 Prior-year earnings centile (year t–2)

sectional inequality.12 Progressive changes partly result from mean reversion but persist even when controlling for observable differences such as age (Berman, forthcoming). The experience during the Covid recession stands in contrast to recent years and recent recessions. From 1999 through 2020, the average progressivity of one-year earnings changes was 4 percent, meaning that earnings inequality fell by this percentage when following workers for one year. Yet between 2019 and 2020, at the start of the Covid recession, progressivity fell to a twodecade low of 1 percent. In contrast, at the start of the 2001 recession, annual progressivity was 7 percent; and at the start of the Great recession, annual progressivity was 5 percent. Hence, the Covid recession was more regressive than these prior recessions and more regressive than all other years since 1999 (two-year changes show similar patterns, see online appendix Figure A4). As discussed above, however, the regressive market earnings changes in the Covid recession were offset by progressive fiscal relief. Between 2019 and 2021, two-year progressivity with relief was 10 percent, or 7 percentage points above the two-year progressivity without relief. Hence, once again we observe the extent to which the fiscal relief during the Covid recession and recovery reduced inequality by reaching those lower in the distribution. E. Which Fiscal Relief Measures Mattered Most? Table 1 shows how Covid public policies reduced the share of workers with large income declines across different years (annual and two-year changes) and between those with and without dependents. Panel A considers annual earnings declines in the 2020 recession. Relief reduced the share with large (10 percent or more) annual declines by 14 percentage points, from 33 percent for earnings without relief to 19 percent with relief. Unemployment insurance benefits explain twothirds of this distribution-wide stabilization effect and stimulus checks explain one-third. Among the bottom quintile, the share with large declines fell much more from fiscal relief—by 27 percentage points (from 53 to 26 percent)—where unemployment insurance and stimulus checks each explain half of the change. Unemployment insurance was generally more progressive than stimulus payments in recessions. In both 2009 and 2020, the bottom quintile of adults received about one-third of unemployment benefits, while the middle received about one-fifth and the top quintile less than one-tenth (Larrimore, Mortenson, and Splinter forthcoming). Stimulus payment amounts, in contrast, were about the same across the distribution except for a phase out for top income groups. Unemployment benefits also target workers with earnings declines, whereas stimulus payments are largely insensitive to earnings declines. 12 This measure differs from cross-sectional Gini coefficients in two ways. First, it follows the same individuals over time, rather than incorporating population changes due to death or entry into the labor market. Second, it ranks individuals based on their initial-year earnings rather than current-year earnings. Hence, this progressivity measure can be thought of as the negative of the change in cross-sectional Gini coefficients if holding the population constant, plus the progressivity from individuals changing their rank in the distribution. The reranking effect is positive by definition (because it reduces the correlation between the ranking variable and earnings), so the progressivity measure will be positive if either the (population constant) cross-sectional Gini falls or if the reranking effect exceeds any cross-sectional Gini increase. 13

Panel B considers two-year earnings changes between 2019 and 2021. Relief reduced the share with large two-year declines by 9 percentage points, from 35 percent for earnings without relief to 26 percent with relief. Unemployment insurance benefits explain 39 percent, stimulus checks 53 percent, and advance child tax credits 8 percent of the stabilization effect.13 Among the bottom quintile, the share with large declines fell by 22 percentage points due to fiscal relief (from 50 to 29 percent). Fiscal relief disproportionately benefitted adults with dependents, especially in the bottom quintile. For those without dependents, relief reduced the share with large two-year declines by 8 percentage points (Panel C). For those with dependents, relief reduced it by 12 percentage points (Panel D). Among the bottom quintile, relief reduced the share with large decrease for those without and with dependents by 19 and 28 percentage points, respectively. This difference is largely because each tax unit received an additional $1,400 of stimulus checks for each dependent and $1,200 per child of advance child tax credits. Table 1. Share of workers with at least a 10 percent real earnings decline All Among the Among All Among the Among the working- bottom the top working- bottom top quintile age adults quintile quintile age adults quintile A: 2019–2020 (1-year) B: 2019–2021 (2-year) Earnings 33.4 52.6 22.0 34.6 50.3 26.9 + Unemployment Ins. 24.2 38.4 20.8 31.1 41.7 26.6 + Stimulus checks 19.3 25.5 19.7 26.3 30.0 25.6 + Advanced CTC 19.3 25.5 19.7 25.6 28.8 25.3 (earnings + relief) C: 2019–2021 (2-year): No dependents D: 2019–2021 (2-year): With dependents Earnings 36.9 52.2 30.0 29.8 45.2 21.9 + Unemployment Ins. 33.5 44.4 29.7 26.0 34.4 21.6 + Stimulus checks 29.3 33.4 29.0 20.1 20.9 20.1 + Advanced CTC 29.2 33.2 29.0 18.2 16.9 19.3 (earnings + relief) Note: Among workers ages 25 to 99 with wages or unemployment insurance in the initial year (t–1 or t–2) and alive at end of final year. Quintiles are based on wages plus unemployment benefits in the initial year. Children include dependents younger than 17 years old claimed on tax returns. Earnings are indexed with the chained CPI-U. Source: Author’s calculations using Internal Revenue Service and Joint Committee on Taxation (2022) tax data. 13 When averaging these estimates with the reverse-order of adding types of relief (i.e., unemployment insurance added last), unemployment insurance benefits explain a similar 37 percent, stimulus checks 52 percent, and advance child tax credits 11 percent of the stabilization effect. 14

5. Summary With a panel of tax data, we follow individuals over recent business cycles. Reflecting the rapid pace of the economic recovery, since 2019 about half of workers had higher real earnings (after adjusting for inflation) in both 2020 and 2021. Yet, this overall stability masks trends for low-income workers who experienced substantial earnings declines in 2020. However, earnings gains have been progressive in the recovery, partially offsetting the regressive earnings losses in the 2020 downturn. Among workers in the bottom quintile before the Covid recession, real median earnings fell 31 percent in 2020 and then increased slightly in 2021, making it 26 percent lower than real pre-Covid earnings. The earnings increases in 2021 for this lowest-earning group greatly exceeded that in higher earnings quintiles, although the two-year earnings change still lags behind those who entered the recession with higher earnings. Progressive policy responses, especially from expanded unemployment benefits and stimulus checks, also offset initial market earnings losses. When including fiscal relief, bottomquintile workers saw their earnings increase substantially in 2020 and then stabilize in 2021 as market earnings gains offset the withdrawal of fiscal relief. We contrast these findings with the 2001 and 2008 recessions. In these earlier recessions, earnings changes were more proportional over the distribution and fiscal relief had only modest effects. Consequently, the distributional impacts of the Covid recovery and policy response have been far more progressive than in prior recessions. 15

References Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B. Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz. 2022. “The $800 Billion Paycheck Protection Program: Where Did the Money Go and Why Did It Go There?” Journal of Economic Perspectives 36 (2): 55–80. Bartik, Alexander, Marianne Bertrand, Feng Lin, Jesse Rothstein, and Matt Unrath. 2020. “Measuring the Labor Market at the Onset of the COVID-19 Crisis.” Brookings Papers on Economic Activity 2020(2): 239–268. Berman, Yonatan. 2020. “The Distributional Short-Term Impact of the COVID-19 Crisis on Wages in the United States.” Working Paper. Berman, Yonatan. Forthcoming. “Absolute Intragenerational Mobility in the United States, 1962– 2014.” Journal of Economic Inequality. https://doi.org/10.1007/s10888-022-09529-7. Blanchet, Thomas, Emmanuel Saez, and Gabriel Zucman. 2022. “Real-Time Inequality.” NBER Working Paper 30229. Busch, Christopher, David Domeij, Fatih Guvenen, and Rocio Madera. 2021. “Skewed Idiosyncratic Income Risk over the Business Cycle: Sources and Insurance.” American Economic Journal: Macroeconomics 14(2): 207–242. Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-Puertolas, Erik Hurst, Christopher Kurz, and Ahu Yildirmaz. 2020. “The US Labor Market during the Beginning of the Pandemic Recession.” Brookings Papers on Economic Activity 2020(2): 3–33. Chetty, Raj, John N. Friedman, Nathanial Hendren, Michael Stepner, and the Opportunity Insights Team. 2022. “The Economic Impacts of COVID-19: Evidence from a New Public Database Built Using Private Sector Data.” NBER Working Paper no. 27431. Cortes, Guido Matias, and Eliza C. Forsythe. 2021. “Impacts of the Covid-19 Pandemic and the CARES Act on Earnings and Inequality.” Working Paper. Cortes, Guido Matias, and Eliza Forsythe. 2022. “The Heterogeneous Labor Market Impacts of the COVID-19 Pandemic.” ILR Review. https://doi.org/10.1177/00197939221076856 Dalton, Michael. 2022. “Putting the Paycheck Protection Program into Perspective: An Analysis Using Administrative and Survey Data.” BLS Working paper 542. Fitzgerald, John, Peter Gottschalk, and Robert Moffitt. 1998. “An Analysis of Sample Attrition in Panel Data: The PSID.” Journal of Human Resources 33(2): 251–299. Greig, Fiona, Erica Deadman, and Tanya Sonthalia. 2022. “Household Pulse: The State of Cash Balances at Year End.” www.jpmorganchase.com/institute/research/household-incomespending/household-pulse-cash-balances-at-year-end#finding-1 Guvenen, Fatih, Serdar Ozkan, and Jae Song. 2014. “The Nature of Countercyclical Income Risk.” Journal of Public Economics 122(3): 621–660. Internal Revenue Service and Joint Committee on Taxation. 2022. “5-Percent Individual Sample from IRS tax document population.” Last updated August 18, 2022. (accessed October 5, 2022). Kaplan Greg, and Giovanni L. Violante. 2014. “A Tale of Two Stimulus Payments: 2001 versus 2008.” American Economic Review 104(5): 116–121. 16

Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2016. “Income and Earnings Mobility in U.S. Tax Data.” Federal Reserve Bank of St. Louis and the Board of Governors of the Federal Reserve System (eds.) Economic Mobility: Research & Ideas on Strengthening Families, Communities, & The Economy 481–516. Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2020. “Presence and Persistence of Poverty in US Tax Data.” NBER Working Paper 26966. Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2022a. “Earnings Shocks and Stabilization During COVID-19.” Journal of Public Economics 206: 104597. Larrimore, Jeff, Jacob Mortenson, and David Splinter. 2022b. “Income Declines During COVID-19.” AEA Papers & Proceedings 112: 340–344. Larrimore, Jeff, Jacob Mortenson, and David Splinter. Forthcoming. “Unemployment Insurance in Survey and Administrative Data.” Journal of Policy Analysis and Management. McKinney, Kevin L., John M. Abowd, and Hubert P. Janicki. Forthcoming. "U.S. Long-Term Earnings Outcomes by Sex, Race, Ethnicity, and Place of Birth," Quantitative Economics. Meyer, Bruce D., Connacher Murphy, and James X. Sullivan. 2022. “Changes in the Distribution of Economic Well-Being During the COVID-19 Pandemic: Evidence from Nationally Representative Consumption Data.” NBER Working Paper 29878. Meyer, Bruce D., Derek Wu, Grace Finley, Patrick Langetieg, Carla Medalia, Mark Payne, and Alan Plumley. 2020. “The Accuracy of Tax Imputations: Estimating Tax liabilities and Credits Using linked Survey and Administrative Data.” NBER Working Paper 28229. Moffitt, Robert A. and James P. Ziliak. 2020. “COVID-19 and the US Safety Net.” Fiscal Studies 41(3): 515–548. Montenovo, Laura, Xuan Jiang, Felipe Lozano Rojas, Ian M. Schmutte, Kosali I. Simon, Bruce A. Weinberg, and Coady Wing. 2020. “Determinants of Disparities in Covid-19 Job Losses.” NBER Working Paper 27132. Parker, Jonathan A., Nicholas S. Souleles, David S. Johnson, and Robert McClelland. 2013. “Consumer Spending and the Economic Stimulus Payments of 2008.” American Economic Review 103(6): 2530–2553. Rothbaum, Jonathan and Adam Bee. 2021. “Coronavirus Infects Surveys, Too: Nonresponse Bias During the Pandemic.” Census Bureau Working Paper SEHSD WP2020-10. Semega, Jessica and Melissa Kollar. 2022. “Income in the United States: 2021.” Census Bureau Report P60-276. Splinter, David Splinter. Forthcoming. “Stimulus Checks: True-Up and Safe Harbor Costs.” National Tax Journal. 17