Surging Business Formation in the Pandemic: Causes and Consequences?

ABSTRACT

Applications for new businesses surprisingly surged during the COVID-19 pandemic, rising the most in industries rooted in pandemic-era changes to work, lifestyle, and business. The unexpected surge in applications raised questions about whether a surge in actual new employer businesses would follow. Evidence now shows increased employer business entry with notable associated job creation; and industries and locations with the largest increase in applications have had accompanying large increases in employer business entry. We also observe a tight connection between the surge in applications and quits—or close proxies for quits—both at the national and the local level. Within major cities, applications, net establishment entry, and our quits proxy each exhibit a "donut pattern," with less growth in city centers than in the surrounding areas, and these patterns are closely related to patterns of work-from-home activity. Reallocation of jobs across firm age, firm size, industry, and geography groupings increased significantly. Relatedly, there is evidence of a pause of the pre-pandemic trend toward greater economic activity being concentrated at large and mature firms, but this development is quite modest in magnitude.

The US economic experience during the COVID-19 pandemic featured a surprising surge in applications for new businesses, shown in figure 1. After dropping in March and April of 2020, applications rose sharply, reaching an all-time high in July 2020; the series declined through the rest of 2020, then surged again in 2021, and have remained historically elevated through September 2023. These data received widespread attention amid high unemployment and broader economic volatility, in part because the surge was apparent even among "likely employers," that is, applications with characteristics that predict the hiring of workers and growth.¹ Monthly applications for likely employer businesses in September 2023 were more than 30 percent higher than the 2019 pace. Historically, there has been a tight relationship between likely employer business applications and true employer business formation, but questions have remained about whether the pandemic's surging applications would translate into actual employer businesses with broader macroeconomic implications.

In this paper, we describe noteworthy aspects of the surprising surge in applications that point to its genuine economic content. We then draw on a range of data sources to show that the surge in applications was followed—after some lag—by a surge in employer business creation: quarterly data on establishment entry rose substantially starting in the second quarter of 2021, while annual data on firm entry jumped in the year ending March 2022 (figure 2). Moreover, we document a close empirical relationship between applications and employer business entry across industry and geography, with hallmark patterns from the application data appearing in employer entry data. We relate the surge in business formation to pandemic labor market stories such as the Great Resignation, that is, the rise in worker quit rates starting in early 2021 (Rosenberg 2022). Finally, we describe the striking resilience of small and young firms through the pandemic period, and we highlight modest hints of a reversal of pre-pandemic trends in business dynamism—though we note that it is too early to declare an end to those trends.

This set of facts lends itself to a compelling narrative of pandemic business and labor market dynamics. The pandemic sparked rapid, dramatic changes to the composition of consumer demand and to preferences for work, lifestyle, and business; and these patterns continued to evolve into 2023. From the standpoint of potential entrepreneurs, these dramatic

Figure 1.

New Business Applications

Source: Census Bureau Business Formation Statistics.

Note: Seasonally adjusted. Total applications = BA series; likely employers = HBA (high-propensity applications) series; likely nonemployers is residual. Shaded areas indicate NBER recession dates.

Figure 2.

New Business Entry and New Business Applications

Source: Business Dynamics Statistics (BDS); Business Employment Dynamics (BED); and Business Formation Statistics (BFS).

Note: BDS and BED annual firm births are age zero firms as of March. BFS applications are likely employers (the HBA series). All series expressed as rates except BFS. Quarterly series are seasonally adjusted. Gray bars indicate NBER recession dates (2001:Q1–2001:Q4, 2007:Q4–2009:Q2, 2020:Q1–2020:Q2).

changes presented opportunities—both to meet newly formed consumer and business needs and to change the career trajectories of the entrepreneurs themselves. Entrepreneurs made plans and applied to start businesses both early on and through the fall of 2023; some of these plans have resulted in new firms and establishments that hired workers in large numbers. Entrepreneurial opportunities and the demand for employees at these new firms appear to have played an important role in the Great Resignation, as some quitting workers likely flowed toward new businesses (as either entrepreneurs or new hires). Taken together, these patterns imply significant economic restructuring across industry, geography, and the firm size and age distribution. The extent to which these changes will be long-lasting has yet to be seen.

The surge in applications started in the second half of 2020, but it has taken time to determine the implications for new employer (and non-employer) businesses. One reason for the delay may be that the initial surge in the summer of 2020 was relatively short-lived, with the more sustained surge in applications commencing later—in early 2021. Moreover, likely employer applications take up to eight quarters to yield the first hire—even conditional on making that transition. And in the United States, data on the creation of actual employer businesses—that is, businesses with paid workers—are published with a lag since such measures derive from administrative data with long processing time. The timeliest data on new employer businesses are for establishment births from the Bureau of Labor Statistics (BLS) Business Employment Dynamics (BED); as of September 2023, BED data on establishment births are available through 2023:Q1, while BED data on (annual) firm births are available through March 2022.² The gold standard annual firm birth data from the Census Bureau Business Dynamic Statistics (BDS) are available through March 2021 for all firms, while quarterly data on single-establishment firms go through 2020:Q4. Between these and other sources, we now have sufficient data to characterize patterns of employer business formation and related job and worker flows in the pandemic.

We observe strong sectoral and geographic correlations between business applications and employer business entry (we measure the latter by either firms or establishments, and in either gross or net terms, depending on data availability). The rise in applications and employer entry is highly concentrated in a few industries that are conducive to pandemic patterns of work, lifestyle, and business (such as online retail and other high-tech industries), consistent with the changing sectoral structure of the economy. We also observe substantial spatial variation in the surge in applications and business entry, consistent with geographic restructuring. The surge in applications and business entry is especially notable in the South, with states such as Georgia standing out. Within large cities we observe a "donut effect" with applications surging more in the suburbs of metropolitan areas than in central business districts.

The pandemic and its aftermath have been associated with increased churn of workers as found in (initially) elevated layoffs, many of them temporary (Cajner and others 2020), and, through much of the pandemic, elevated quits. We find a tight spatial correlation—at the state and county level—between surging business applications and quits (or excess separations, a close proxy for quits), with a much weaker correlation between applications and layoffs (or job destruction, a close proxy for layoffs). Among other possible explanations, these results are consistent with workers quitting their jobs to start or join new businesses—and much less consistent with job loss being a key driver of business formation.

This pandemic surge in entry occurred after decades of declining business dynamism in the United States. The pace of job reallocation had fallen by about 25 percent from the 1990s to just before the pandemic.³ This decline in the pace of job reallocation was driven in part by the decline in employer business entry over this same time period, which can be seen in figure 2 or, for a longer view, online appendix figure E1; closely related is the shift of the firm distribution toward large and mature firms. While the sources of this decline have been widely debated in the literature, there is evidence that it has been associated with a decline in productivity-enhancing reallocation and is likely one of the factors underlying sluggish productivity growth in the United States since the early 2000s.⁴

We show that the pandemic featured a surge in job reallocation, including reallocation between cells defined by industry, geography, firm size, and—especially—firm age. We also document a pandemic pause—and modest reversal—of the longer-run shift in activity toward large, mature businesses. The share of activity accounted for by young and small firms has ticked up; young and small firms exhibit a higher pace of dynamism than large and mature firms, so one might anticipate an ongoing increase in the pace of dynamism. In other words, we find early hints of a revival of business dynamism; but in many respects it is too early to ascertain whether a durable reversal of pre-pandemic trends is occurring. Such a reversal—that is, a persistent rise in the pace of reallocation and a substantial shift of activity away from large, mature firms—will require a long-lasting continuation of elevated business entry as well as substantial growth among at least a subset of the pandemic entrants.

It is useful to state our view of our contribution—and the limits to that contribution. A key contribution of our work is that we draw on a wide range of data sources: the Bureau of Labor Statistics' BED, Quarterly Census of Employment and Wages (QCEW), and Job Openings and Labor Turnover Survey (JOLTS), and the Census Bureau's Business Formation Statistics (BFS), BDS, and Quarterly Workforce Indicators (QWI). While none of these data sources alone can tell a comprehensive story of pandemic business entry, each contributes a different perspective in terms of timeliness, industry and geography detail, or measurement concept. We provide an initial assessment of the potential causes and consequences of the surge in business applications by supplying a rich set of empirical facts pointing to substantive pandemic economic stories, but we do not provide identified causal empirical results or new formal theory; rather, we hope our results can direct and discipline future causal analysis. We also hope our approach of exploiting an eclectic combination of data sets can help other researchers better understand the range of available business dynamics and labor market data that can inform timely analysis.

A study of actual application-to-employer transitions, post-entry dynamics, and job-to-job flows of workers must wait for the availability of administrative micro data.⁵ Such micro data can also facilitate rigorous causal analysis and provide additional empirical moments of relevance to theoretical investigations. Separately, while we focus on new employer businesses, the likely surge in new nonemployer businesses appears important and interesting as well; unfortunately, the nonemployer economy is measured with less detail and timeliness than the employer economy, so we leave that investigation for future work (but we provide some additional discussion near the end of this paper and in online appendix A).

Our work complements that of Fazio and others (2021), which documents similar aggregate patterns using zip code–level data on business registrations in eight states from the Startup Cartography Project; Fazio and others (2021) report striking time series relationships between pandemic fiscal stimulus and the registration surge and find that the surge was concentrated in zip codes with relatively high African American population and above-median income. They also find that the surge is apparent outside city centers within large cities; we show that this within-city pattern is apparent in county-level applications data for the United States as a whole, and we build on their earlier work by studying outcomes for net establishment entry and excess worker flows as well. Duguid and others (2023) document similar within-city patterns for retail establishments using credit card merchant data and relate these patterns to population flows and remote work considerations. We also expand on Decker and Haltiwanger (2022), in which we provided a first look at the relationships between business applications and establishment births (and exits) in official data and initially documented the increase in small firms' share of activity during the pandemic.⁶

In section I we briefly describe our main data sources, with much more detail in online appendix A. We review and document patterns of business applications in section II, then explore employer establishment and firm entry and their empirical relationship with applications in section III. We examine the relationship between worker churning—especially quits—and applications in section IV. In section V we document changes in the firm size and age distribution and consider implications for business dynamism. We take stock in section VI, then speculate about potential implications for the future in section VII.

I. Data

We exploit a variety of data sources, all of which are publicly available tabulations. Online appendix A describes each source in detail; here we simply list our main sources with brief descriptions.

Business Formation Statistics (BFS), US Census Bureau: monthly data on IRS employer identification number (EIN) applications. All employer businesses and nonemployer corporations and partnerships must have an EIN, and many nonemployer sole proprietors choose to obtain one for business reasons. The total applications series (called "BA" in the BFS files) counts all EIN applications that are potential employer or nonemployer (zero-employee) businesses (this implies excluding applications for trusts, estates, and financial instruments). Our main interest is employer businesses; therefore, where possible we focus on what we call likely employer applications (high-propensity applications or "HBA" in the BFS files). This subset of the total applications series is based on Census Bureau modeling using application characteristics that have a high propensity for transitioning into an actual employer business with paid workers; these characteristics include planned hiring and corporate legal form, among others. However, at narrow levels of industry (three-digit NAICS) or geography (county) detail, only total applications are publicly available, so we use the total applications series as a proxy for our preferred likely employer series. As shown in figure 1 (and below at more disaggregated levels by industry and geography), total and likely employer applications have tracked each other closely in the pandemic, which mitigates concerns about using the total series as a proxy for likely employers where necessary. We use BFS series through September 2023.

The BFS also includes series that report, in any given time period, the number of applications that actually transition to genuinely new employer firms within four or eight quarters. These series use micro data linkages tying applications to actual employer firm births; the four-quarter and eight-quarter series are currently populated through 2019:Q4 and 2018:Q4, respectively, and relate to new employer firm micro data available through 2020:Q4. Since these transition series end relatively early (constrained by actual employer firm data timing in Census data), the BFS also features series for projected transitions at four- and eight-quarter horizons, where projections are based on application characteristics and include all applications (not just those labeled as likely employers). The motivation for the four- and eight-quarter horizons for actual and predicted transitions is that, as discussed further below, there is often a lag between applications and transitions. An advantage of the projected series is that they take into account the full range of application characteristics (e.g., reason for application and detailed industry).⁷

Quarterly Census of Employment and Wages (QCEW), Bureau of Labor Statistics: quarterly establishment and employment counts by detailed industry and geography. The QCEW is derived from the main business register of the BLS and is based on state unemployment insurance administrative data. We use the QCEW to measure net establishment growth at the national, industry, and local (county) level. The QCEW micro data also underly the Business Employment Dynamics.

Business Employment Dynamics (BED), Bureau of Labor Statistics: quarterly data on establishment openings, closings, births, exits, expansions, and contractions, with associated job flows. The BED also features a research product with annual employment, firm, and establishment counts by firm age, where a firm is defined by an EIN. We use quarterly BED data extending through 2023:Q1 and annual firm age data through 2022:Q1. Importantly, in the BED, an establishment (firm) birth represents an establishment (firm) that did not previously exist; a new firm requires a new business application, while a new establishment of an existing firm does not require but may obtain a new EIN. Notably, new EINs acquired by existing firms would not count as employer firm transitions in the BFS four-quarter and eight-quarter transition series mentioned above but may appear as new establishments (or firms) in BED data.

Quarterly Workforce Indicators (QWI), US Census Bureau: quarterly data on employment and job and worker flows (i.e., hires and separations) by firm age with detailed industry (four-digit NAICS) and geography (county) tabulations. The QWI is the public-use version of the Longitudinal Employer-Household Dynamics (LEHD) data based on state unemployment insurance records and collected on a state-by-state basis; we use a balanced panel of forty-five states that covers just over 80 percent of private employment as of 2020. These data extend through 2022:Q2.

Job Openings and Labor Turnover Survey (JOLTS), Bureau of Labor Statistics: monthly survey-based estimates of hires, separations, quits, and layoffs with state-level detail. We use JOLTS data through September 2023 with a focus on quits and layoffs.

American Community Survey (ACS), US Census Bureau: annual survey-based data on work-from-home (WFH) prevalence for large counties. ACS data are available in two samples: five-year samples including the entire United States, and one-year samples including large counties. We use the one-year sample for 2019–2021 and focus on changes in WFH prevalence across counties within large cities. ACS WFH measures are based on location of worker residence; we discuss existing literature on WFH using other data (Hansen and others 2023) in online appendix A.

Additionally, we use data from the Census Bureau's Business Dynamics Statistics (BDS) in certain figures (e.g., figure 2); these data do not currently cover the pandemic period, so we do not use them in most of the exercises that follow. In online appendix A, we provide a discussion of the BDS and its relation to the BLS data sources listed above.

II. Business Application Patterns

II.A. The Early Pandemic Period

At the onset of the pandemic, plummeting weekly business application and registration data received widespread attention (Fazio, Guzman, and Stern 2020; Haltiwanger 2020; Federal Reserve System Board of Governors 2020).⁸ But, as shown in figure 1, applications quickly recovered and surged to historic levels in July 2020. The surge is apparent in every application series, including total applications and likely employer applications (both shown in figure 1) as well as applications with planned wages and applications for corporations.⁹ Applications did fall off in August 2020 through December 2020 (albeit still higher in December 2020 than prior to the pandemic) but then surged again in early 2021. This second wave has been more resilient, with monthly likely employer applications in 2023 so far averaging about 30 percent higher than the 2019 pace. Total applications are about 40 percent higher in 2023 relative to 2019, reflecting the even larger surge of likely nonemployers.

The sharp rise in the likely employers series is in stark contrast to the previous recession. Dinlersoz and others (2021) and Haltiwanger (2022) explore this comparison in detail; here we note that the decline in total applications seen in the Great Recession was driven by the likely employer series, while the likely nonemployer series was roughly flat in that episode.¹⁰ Flat or even rising nonemployer entrepreneurship during a recession can easily be rationalized in light of lack of opportunities for wage and salary employment, which may push many individuals into self-employment activities out of necessity; and, indeed, one plausible explanation for the pandemic surge in applications was that unemployment was elevated in the wake of spring 2020 shutdowns. But rising employer entrepreneurship is more difficult to understand, as businesses hiring employees are more likely to be pursuing genuine entrepreneurial opportunities; hence, the stark difference in likely employer behavior between the pandemic recession and the prior recession is all the more striking. And the pandemic surge in applications has persisted even as unemployment has fallen toward historic lows.

A number of factors could help account for the surge in applications for likely employers in the pandemic compared to the drop of likely employer applications and employer start-ups in the Great Recession. The pandemic provided new market opportunities given the changing nature of consumer demand and of work and lifestyles, and financial conditions—including house prices—were robust compared to the Great Recession (at least through early 2022). The potentially supportive role of stimulus programs—which included sizable support for aggregate demand and household balance sheets—is an open question. The US federal and state governments implemented a wide range of fiscal support programs which could have had myriad effects on business formation; one example is the expansion of unemployment insurance benefits, which Choi and others (2023) find had a positive effect on business applications. On the other hand, programs like the Paycheck Protection Program (PPP)—along with other business support facilities—may have dampened new business formation since they provided support for incumbents and thus deterred exit.¹¹

Even though some factors have been more favorable for business formation in the pandemic than in the Great Recession, an open question has been whether genuine employer business creation would result. Historically, likely employer applications have been strongly predictive of actual firm entry, with a national correlation of 0.9 and an elasticity roughly centered on one at the aggregate level, within states, and within industries.¹² But one might fear that the transition rate from applications to actual businesses could change in the pandemic. Perhaps especially in the early months of the pandemic, maybe there was a surge in nascent entrepreneurship—individuals thinking about starting a business—without necessarily making the transition to an actual new business. This is a core question we address by providing available evidence on actual employer business formation below, but first we delve further into the applications themselves.

II.B. Sectoral Patterns of Applications

One clue about the economic substance of surging applications is the pattern across industries. For likely employer applications, data are only available at the broad sector level; while interesting (and discussed below), this level of industry detail misses important stories. For more detail, we use total applications, which are available at the three-digit NAICS industry group level (published as a special tabulation after the end of each calendar year—currently these data are available through the end of 2022). We use the total applications series with some caution given our focus on employer business entry, but we note that there has been a coincident surge in likely employer and likely nonemployer applications at observable national, state, and industry levels.¹³

Figure 3.

New Business Applications, Selected Three-Digit Industries

Source: Census Bureau Business Formation Statistics.

Note: All applications. Average weekly pace by quarter (seasonally adjusted).

The surge in total applications was highly concentrated among three-digit industries; a Herfindahl-Hirschman index of industry-level applications jumped by more than 10 percent in 2020 versus 2019 and remained historically elevated through 2022 (online appendix figure E5). Indeed, more than 20 percent of the jump in applications from 2019 to 2022 was accounted for by nonstore retailers (NAICS 454), which includes online retail; and more than half of the overall surge was accounted for by just five three-digit industries, shown in figure 3.

The industries making large contributions to overall application growth can plausibly be related to pandemic patterns of work, lifestyle, and business models. Nonstore retailers (NAICS 454) include online retail businesses facilitating shopping from home. Professional, scientific, and technical services (541) is a tech-intensive sector, with about half of its employment in STEM-intensive industries such as architectural, engineering, and related services (5413), computer systems design (5415), and scientific research and development services (5417); business formation in these industries may be related to helping other businesses facilitate pandemic work and lifestyle changes and may also relate to recent technological developments like artificial intelligence (AI).¹⁴ The sector also includes industries such as building inspectors and interior designers potentially associated with the pandemic surge in home sales or rearrangement of home office environments. Personal and laundry services (812) include some industries that were likely harmed by the pandemic (e.g., nail salons) but also industries that enhanced work-from-home environments or facilitated pandemic hobbies, such as pet care. Administrative and support services (561) includes employment services that are sometimes important during recessions (e.g., temporary help agencies); industries that may facilitate changes in business models such as document preparation, call centers, and mail carriers; and businesses facilitating work-from-home transitions such as landscaping services and carpet cleaners. Truck transportation (484) includes both general and specialized freight trucking (an example of the latter is NAICS 484210, used household and office goods moving); such businesses likely benefited from changes to the use of commercial real estate, the shift toward online shopping, and the rotation of consumer spending away from services and toward goods.

The patterns in figure 3 also hint at interesting changes over the course of the pandemic and its aftermath. Applications for nonstore retailers exhibited the most dramatic surge early in the pandemic; and while this remained elevated at the end of 2022, it has declined substantially from its 2020:Q3 peak. By mid-2022 the highest industry was professional, scientific, and technical services; this tech-intensive industry has exhibited a sustained surge since the beginning, with 2022:Q4 being at about the same pace as 2020:Q3. Truck transportation had a smaller initial surge, peaked in mid-2021, then declined gradually, a pattern consistent with new businesses entering to address supply chain constraints along with the surge in goods consumption, both of which have receded somewhat in recent quarters.

We find similar patterns for likely employer applications at the broad sector level (online appendix figure E2); in particular, we observe strong increases in likely employer applications in the retail trade sector and in "Tech"—a proxy for the high-tech sector that combines professional, scientific, and technical services with the information sector. Interestingly, when we use the projected firm births series from BFS, the two-digit sector that has the highest level of applications during the pandemic is the high-tech sector (online appendix figure E3). As discussed in online appendix B, the predicted start-up series (PBF4Q and PBF8Q) is a better predictor than HBA of actual start-ups, particularly at the sector level.

II.C. Geographic Patterns of Applications

We next analyze spatial variation in applications, and we introduce a simple measure of growth in applications per capita in the pandemic

Figure 4.

Growth in Likely Employer Applications per Capita, 2020–2022 versus 2010–2019

Source: Census Bureau Business Formation Statistics and population estimates.

Note: Difference of average (log) likely employer applications per capita, 2020–2022 versus 2010–2019.

relative to the pre-pandemic norm, which we denote as g. We define g as follows, using annual data at various levels of geography:

where x_t is applications per capita in year t.¹⁵ That is, we study the difference between the average of (log) applications per capita in 2020–2022 and the average of (log) applications per capita during 2010–2019.

Using likely employer applications, figure 4 shows substantial variation across states, with the highest-growth states having growth rates of between 34 and 73 log points while the lowest-growth states exhibit little or no growth. Growth was particularly strong in the South and also parts of the West (e.g., California).

Figure 5.

Growth in Total Applications per Capita, 2020–2022 versus 2010–2019

Source: Census Bureau Business Formation Statistics and population estimates.

Note: Difference of average (log) all applications per capita, 2020–2022 versus 2010–2019.

More variation can be seen at the county level, though at this level we must use total applications rather than likely employer applications.¹⁶ Growth in business applications has been widespread across US counties; more than 95 percent of counties saw a higher pace of applications during 2020–2022 than during 2010–2019, on average. Figure 5 provides the county analog to figure 4; the rapid growth in the South is evident in the county map as well, but there are pockets of rapid growth throughout the country.

While a small number of counties actually saw declines in applications per capita, the median county saw an increase of 40 log points, and the highest quintile saw growth of between 61 and 289 log points. The variation in county-level growth suggests material geographic restructuring, with some counties experiencing dramatically more business applications per capita than in pre-pandemic times.

Much of the variation across counties reflects larger geographic shifts: variation between Census Bureau divisions accounts for 25 percent, variation between states accounts for almost 50 percent, and variation between commuting zones accounts for 70 percent of the between-county variation

Figure 6.

New York City: Growth in Applications per Capita, 2020–2022 versus 2010–2019

Source: Census Bureau Business Formation Statistics and population estimates.

Note: Difference of average (log) all applications per capita, 2020–2022 versus 2010–2019.

in total application growth (reported in online appendix table F1). However, counties vary considerably in scale, and even though we are examining growth in applications per capita, the latter is increasing in initial county population (and population density). Among counties that are part of large core-based statistical areas (CBSAs), those with population above 1 million, about 50 percent of the between-county variation is accounted for by between-CBSA effects; over half of the US population is in these large CBSAs, so exploring the variation within large CBSAs is of independent interest.

As an example of within-city variation, figure 6 zooms in on the counties of the New York City area (which includes counties in New York State, New Jersey, and Pennsylvania), again reporting growth in (total) applications per capita as calculated in equation (1).

Growth of applications per capita in New York City counties ranges from 19 to 64 log points. We also observe a striking "donut" pattern: growth is stronger outside New York County (i.e., Manhattan—the central business district of the city) than inside it.¹⁷ These patterns are broadly consistent with zip code–level patterns documented earlier by Fazio and others (2021) using state business registrations; those authors find that, after the widespread initial registration decline early in the pandemic, Manhattan registrations returned to their 2019 pace while the Bronx, Harlem, and parts of Brooklyn saw historic registration growth.¹⁸ Duguid and others (2023) find similar results for retail establishments based on credit card transaction data for the country as a whole; the authors report relatively weak (or negative) establishment growth in core downtown areas, with stronger growth in inner suburbs (though not in outer suburbs).

The donut pattern is apparent in other major cities as well; for example, online appendix figure E7 shows the state of Washington, where King County—the central business district for Seattle—shows less application growth than surrounding counties.¹⁹ In unreported results, we visually observe a similar donut pattern in other cities, in the sense that a number of surrounding (close in and outlying) counties within CBSAs exhibit higher growth in applications per capita than the county that contains the central business district.²⁰

The donut pattern we observe for applications appears related to popular pandemic themes about high-density downtown areas and the transition of many workers to work-from-home (WFH) activity. We more formally explore the relationship between the growth of applications, density, and WFH within cities using regressions reported in online appendix table F9. In particular, at the county level we regress growth of total applications per capita on population density, establishment density (from QCEW data), and growth of WFH activity (from ACS data, where the fraction of workers working from home is based on location of residence). We find highly nonlinear, statistically significant empirical relationships for all three covariates. There is an alternating negative linear effect, positive quadratic effect, and negative cubic effect with magnitudes implying the linear negative term dominates for low values (of density and change in WFH share), the positive quadratic becomes relatively more important for larger values,

Figure 7.

New York City: Predicted Growth in Applications per Capita, 2020–2022 versus 2010–2019 (Spatial Model)

Source: Census Bureau Business Formation Statistics and population estimates; author modeling.

Note: Predicted difference of average (log) all applications per capita, 2020–2022 versus 2010–2019.

then the negative term kicks in for very large values. In considering these patterns, it is useful to observe that within New York City, Manhattan has the highest population density and establishment density and a mid-range growth of WFH.²¹

We also consider a more complex spatial regression specification where we include a cubic of all of these terms for both own and adjacent counties (online appendix table F10). We find that each of these covariates have significant own- and adjacent-county effects in this multivariate specification. Given the complexity of this specification, we focus on the overall predictive power; the R² of this specification is 0.77, compared to 0.49 in a specification with only CBSA fixed effects. Figure 7 shows that the predicted

Figure 8.

High-Propensity Business Applications and Start-ups Eight Quarters Ahead

Source: Census Bureau Business Formation Statistics.

Note: Start-ups within eight quarters. Seasonally adjusted. Normalized by average 2006 levels. Shaded areas indicate NBER recession dates.

variation in counties in the New York City CBSA from this spatial model closely corresponds to the actual application pattern (compare to figure 6). Put simply, we are able to approximately replicate the within–New York City donut pattern using population density, establishment density, and growth of WFH activity in own and adjacent counties—consistent with the broader high model fit for all cities suggested by the R² of 0.77.

II.D. Applications and Actual Firm Births in the BFS

There is typically a lag between EIN application and new employer firm entry, even conditional on a successful transition. In much of our analysis of employer entry from other sources we use quarterly data, annual data, or growth rates based on the difference between pre-pandemic and pandemic averages, mitigating this lag.²² Figure 8 shows a tight relationship between likely employer applications and employer firm births within eight quarters. The solid line provides an index of actual employer firm births within eight quarters (through 2018:Q4), and the dotted line is an index of projected employer firm births within eight quarters. The surge in likely employer applications in the pandemic is accompanied by a surge in projected business formations.²³

Online appendix tables F2 and F3 provide more detail about this tight relationship between applications and actual employer start-ups. As a rough approximation, the (pre-pandemic) elasticity of new employer firms within eight quarters with respect to likely employer applications is centered on one in both the aggregate time series and in state-by-time pooled data. These historical relationships as well as the projected series suggest strongly elevated employer entry during the pandemic as well—but with some lag relative to the timing of applications. The lag between application and new employer entry was increasing prior to the pandemic (see online appendix figure E4). While the actual transitions are not yet available beyond 2020, we explore these relationships below using a variety of available employer entry rates.

III. New Employer Businesses in the Pandemic

We now turn to data on actual employer business formations during the pandemic, expanding on the data first shown in figure 2. Here we draw on several sources: we use BED quarterly establishment births and openings data through 2023:Q1, BED annual firm births data through March 2022, and QCEW quarterly net establishment births data through 2023:Q1 (which permit finer geographic and industry detail than BED data). Importantly, the gold standard data set for tracking true employer firm births is the Census Bureau's BDS, which features a more comprehensive firm identifier than the BED (see discussion in online appendix A); we report two different BDS series in figure 2, but these data do not currently cover a significant portion of the pandemic period.

The BED and QCEW have the key advantage of timeliness, though the most timely data are on establishment entry (gross entry in BED and net entry in QCEW), which include not only new firms but also new establishments of incumbent firms (e.g., new Starbucks locations). While our primary focus is on new firms, new establishments opened as expansions of existing firms are of independent interest, since such establishments are important components of the reallocation of activity across business locations. Moreover, it is likely that new establishments of existing businesses reflect similar incentives of new firms to take advantage of the market opportunities that arose in the pandemic and its aftermath.

III.A. Aggregate Establishment and Firm Entry: Gross and Net

Figure 9 shows quarterly data on high-propensity business applications (panel A), BED establishment births and exits (panel B), and jobs created (destroyed) by births (exits) (panel C).

Figure 9.

Business Applications, Establishment Births, and Exits

Source: Census Bureau Business Formation Statistics (BFS) and BLS Business Employment Dynamics.

Note: Seasonally adjusted. Shaded areas indicate NBER recession dates. High-propensity applications. Panel C shows jobs created by births and jobs destroyed by exits.

The surge in establishment births is especially pronounced starting in 2021:Q2—several quarters after the initial surge in applications in July 2020 but also after the second wave of the surge in applications in early 2021. It is not surprising that there is some lag since, as discussed above, it can take up to eight quarters for applications to transit to employer businesses—conditional on transiting at all. Like business applications, establishment births have reached record levels during the pandemic. Note also that births have been well in excess of exits, aside from the initial exit surge in 2020:Q2.

As shown in panel C, job creation from establishment births has been above one million per quarter, on average, during 2021:Q2–2023:Q1—a historically high pace. Establishment birth has played a significant role in the pandemic job recovery, accounting for more than 10 percent of gross private job creation from 2020:Q3 through 2023:Q1; at a quarterly frequency, in 2022:Q4 establishment births' share of gross job creation reached 12.9 percent for the first time since 2007. While this increase in job creation from births is striking, the surge in the number of establishment births (panel B) is proportionally greater than the surge in birth employment; the average size of a new establishment birth declined from about 3.3 jobs in 2019 to 2.9 jobs in 2022.²⁴ As we discuss in section V.B below, average firm entrant size also stepped down in the pandemic—though incumbent size declined as well.

Figure 10.

Net Growth of Establishments and Firms

Source: BDS; BED; and QCEW.

Note: Annual Davis-Haltiwanger-Schuh denominator (DHS) growth rate of unit counts, first quarter versus one year earlier.

The elasticity of establishment births with respect to likely employer applications has, if anything, strengthened in the pandemic—at least at the national level; we obtain this evidence with simple regressions of establishment births per capita on applications (online appendix table F2). For the aggregate series we actually find a higher elasticity of establishment births when we include the pandemic period than if we end the sample in 2019. Online appendix table F4 reports state-by-quarter regressions and table F6 reports sector-by-quarter results, in which the pre-pandemic elasticities (shown on the top panel of each table) are generally similar to those estimated on pandemic-inclusive data (bottom panel). We also examine the relationships between firm births, establishment births, and the projected start-up series from the BFS. Online appendix tables F7 and F8 illustrate three findings. First, there is a strong positive historical (prepandemic) relationship between BFS predicted firm births and actual firm and establishment births. Second, this relationship remains strong during the pandemic for establishment births. Third, especially for the sector-based results, the elasticities are substantially higher using the BFS projected start-ups series compared to those using the likely employer applications. We discuss these analyses more in online appendix B.

It is clear from the BED data in figure 9 that net establishment entry surged in the pandemic; this fact is corroborated in other data sources and for firms as well. Figure 10 shows annual net growth of firm and establishment counts from the BDS, BED, and QCEW. Reassuringly, the various series track each other well through March 2020, after which the BDS becomes unavailable. Net establishment growth was strong in 2021 and, especially, 2022 and 2023.²⁵ Firm growth was similarly impressive, as the total number of firms (in BED data) increased by more than 250,000 from March 2020 through March 2022, from under 5.3 million to more than 5.5 million. The largest surge is from March 2021 to March 2022—broadly consistent with the finding that the increase in establishment births is especially pronounced starting in 2021:Q2. In online appendix figure E9 we report similar results if growth is calculated on a per capita basis.

Here we have focused on true establishment birth and exit; temporary closings and reopenings of establishments also played a large role in early pandemic labor market dynamics. In online appendix figure E12 we report total establishment openings and closings, and figure E13 shows reopenings (i.e., openings minus births) and temporary closings (i.e., closings minus exits). In 2020:Q2, more than 400,000 establishments closed temporarily, with nearly 1.8 million associated jobs. Reopenings jumped in the following quarter, accounting for 1.2 million jobs in 2020:Q3 and nearly 800,000 jobs in 2020:Q4. These patterns imply a need for caution in the use of establishment openings out of context—especially in 2020:Q3; the patterns also highlight the large role of temporary job dislocation in the early pandemic labor market.

While establishment reopening and temporary closure was a significant feature of the pandemic—particularly in early quarters—the cumulative job reallocation associated with births and exits is even a bit larger.²⁶ Over the 2020:Q2–2023:Q1 period, job reallocation from establishment births and exits cumulated to 20.6 million jobs, with births contributing 11.4 million and exits 9.2 million. Reallocation from births and exits necessarily reflects permanent job reallocation. During the same period, temporary closings and reopenings cumulated to 17.5 million jobs, with temporary closings contributing 9.1 million and reopenings 8.4 million. In contrast to births and deaths, these job flows associated with temporary closings and reopenings reflect transitory reallocation—although it may be that some workers who lost their jobs to temporary closings did not return to the same employer, since reopenings took some time. We discuss the implications of these dynamics for job and worker reallocation further below.

III.B. Sectoral Patterns of Employer Business Entry

As noted in section II.B, the industry pattern of business applications is consistent with broader economic restructuring in the pandemic. We next ask whether these industry patterns are reflected in data on actual employer business formation. Annual firm births by broad sector are available from the BED through March 2022; the scatterplots in figure 11 compare pandemic firm births with likely employer applications by sector, where we focus on pandemic growth relative to pre-pandemic norms as described in equation (1).

Panel A in figure 11 gives insight into the contribution of different sectors to the aggregate surge in firm births and likely employer applications by measuring the average level of births or applications—in thousands—during the pandemic versus the pre-pandemic pace. Educational and health services, professional and business services, and construction are sectors with large increases in both firm births and likely employer applications, accounting for a large share of the aggregate surges in both.

Panel B is more informative about growth within sectors, as it is based on the log difference between pandemic and pre-pandemic norms. Sector-level growth in firm births and business applications is strongly positively related, with most sectors lining up reasonably close to the 45-degree line. Transportation and warehousing, information, education and health services, financial services, construction, and professional and business services are all sectors with large growth (approximately 20 percent or larger) of both applications and firm births. The retail trade sector is notable, however, for having a smaller surge in firm births than in applications; this could reflect the differing nature of the 2020 application surge (which, as discussed above, was led by online retail) versus the later pandemic surge, where other sectors became more important.²⁷ It may be that the early surge in applications, especially in sectors like online retail, saw lower rates of transition to employer business formation. Indeed, the BFS itself suggests this; in online appendix figure E11, we find that the sector-level relationship

Figure 11.

Firm Births and Business Applications, Industry Detail

Source: Business Employment Dynamics (BED) and Business Formation Statistics (BFS).

Note: Average pace during 2021–2022 versus average pace during 2011–2020. Panel A expressed as average annual pace. Solid line is the 45-degree line. "T&W" is transportation and warehousing. Years end in March. High-propensity applications.

Figure 12.

Net Establishment Growth from Pre-pandemic to Pandemic

Source: QCEW and Census Bureau population estimates.

Note: Difference of average (log) establishments per capita, 2020–2022 versus 2010–2019.

between firm births and BFS projected firm births is even stronger, with the retail sector being less of an outlier.

III.C. Geographic Patterns of Employer Business Entry

Given the striking geographic pattern of business applications described in section II.C, we next explore county-level correlations. Data limitations continue to bind, however, as BED establishment birth (or opening) data are not available at the county level, so we focus on net establishment entry (i.e., change in the number of establishments) in QCEW data. To start, we consider the spatial variation in growth of establishments per capita between the pre-pandemic and pandemic periods using the same measure as implied by equation (1). Figure 12 highlights substantial variation in the growth of establishments per capita across counties (this figure can be compared usefully with figure 5, the analogous map for business applications). In the top quintile, establishments per capita increased between 13 and 52 log points while in the bottom quintile establishments per capita declined.

The spatial patterns in figures 12 and 5 are broadly similar, with the South and parts of the West standing out as having especially high growth in both applications per capita and establishments per capita. We can see this more formally in panel A of figure 13, which is a binscatter relating county-level growth in total establishments per capita to growth in applications per capita, 2020–2022 versus 2010–2019, following equation (1).

Figure 13.

Net Establishments Growth versus Applications Growth

Source: QCEW and BFS.

Note: County-level log differences of 2020–2022 versus 2010–2019 levels. Straight line is a regression line with reported slope and standard error. Total applications. Panel A is a binscatter with one hundred bins.

We observe a tight, highly statistically significant relationship between establishment growth and applications. Of course, net establishment growth conflates establishment birth and exit, and the latter has likely been an important margin of local economic adjustment during the pandemic period; see Decker and Haltiwanger (2022) and Crane and others (2022) for discussion (though recall that figure 9 shows that establishment death was not materially elevated after its initial spike in 2020:Q2, with the exception of 2022:Q2).²⁸ Moreover, as in our three-digit industry scatter-plots above, at the county level we have total business applications, not the narrower category of high-propensity applications, though recall that total applications and high-propensity applications have moved together in the pandemic. The strong spatial relationship between net establishment entry and total applications suggests that surging business applications are related to growth in net entry in the geographic cross section.²⁹

Figure 14.

Net Establishment Growth, New York City

Source: QCEW and Census Bureau population estimates.

Note: Difference of average (log) establishments per capita, 2020–2022 versus 2010–2019.

We provide some concrete perspective into our county maps and the binscatter just mentioned by focusing on the counties in two states: Georgia and Washington. Panel B of figure 13 depicts the growth in applications and establishments for counties in just these two states; Georgia (crosses) is a state with high growth on both margins, while Washington (squares) is not. Interestingly, this between-state pattern holds pervasively across counties within these respective states.

As another specific example, figure 14 shows net establishment growth for counties of New York City in the same manner as figure 6. While not identical to the pattern of application growth, we still observe a donut pattern of strong growth in establishments per capita in the city suburbs, with less growth in the city center of Manhattan.

We provide further perspective on these geographic patterns in online appendix C. We find, for example, that the high-growth counties in terms of net establishment growth in the NYC area have higher growth rates than Manhattan across a wide variety of industry sectors. Some of this reflects sectors that are apparently supporting the change in the habits of the daytime population (e.g., large increases in sectors such as leisure and hospitality—NAICS codes 71 and 72). However, we also observe the high-growth counties having higher growth in high-tech sectors like information (51) and professional, scientific, and technical services (54). Similar observations apply to high-growth states such as Georgia relative to low-growth states such as Washington.

Our geographic exercises, like our industry exercises, suggest a strong relationship between business applications and actual employer business growth. Moreover, these patterns are consistent with thriving business creation in industries that complement pandemic changes in work and lifestyles as well as movement of some forms of economic activity from city centers to outer areas. Notably, our geographic analysis is all done on a per capita basis, so these flows of businesses do not simply reflect underlying population flows.

IV. Worker Flows and Business Formation

The pandemic labor market has featured several notable patterns, including mass layoffs followed by rapid job growth, migration, and a large number of workers quitting their jobs (which has been called the Great Resignation). A natural question is whether these labor market patterns have any relation to the surge in business formation. In section III.A we described the significant role of firm and establishment birth in gross and net job growth in the pandemic; and in section III.C we reported striking geographic patterns consistent with popular stories about migration flows (north to south, inner cities to outer cores) but that reflect flows above and beyond simple population moves.

In this section, we focus specifically on quits and layoffs or, where necessary, close proxies for quits and layoffs. The early pandemic period was characterized by a massive spike in layoffs; while many of these proved temporary (Cajner and others 2020), the 2020:Q2 spike in establishment deaths (figure 9) indicates that there was also considerable permanent job destruction. Separately, the pace of quits rose to record levels—and well above its pre-pandemic trend—in late 2021 and early 2022.

Workers who experience a permanent separation through either quits or layoffs could be joining a new business either as the entrepreneur or as an early employee. Indeed, since quits are thought to be dominated by job-to-job flows, workers who quit likely had a job to go to at the time of the quit.³⁰ But the administrative micro data required to track these flows on a comprehensive basis are not yet available. Instead, we examine patterns at the aggregate and spatial levels as we have in previous sections.

For this purpose, we exploit data from the Census Bureau's Quarterly Workforce Indicators (QWI) and other sources. The QWI provide information on hires (i.e., new worker-firm matches), separations (broken worker-firm matches), job creation (growth in firm employment), and job destruction (contraction of firm employment) in various granular tabulations.³¹ We take advantage of that granularity to decompose separations into job destruction and what we denote—following Davis and Haltiwanger (1992)—as excess separations (the difference between separations and job destruction).

It is important to grasp the intuition of excess separations. Separations include both layoffs and quits. Workers may be separated from jobs because those jobs are being destroyed as a firm contracts; for example, a firm may be eliminating a position entirely as part of a downsizing or restructuring plan. In these cases, there is no excess separation, and worker and job flows are equal. But many workers are separated from jobs while those jobs continue to exist and will be filled by another worker. A likely reason for such a separation is that the worker is quitting the job to start a new job elsewhere. Both conceptually and historically, job destruction and layoffs track each other well, and excess separations and quits track each other well (Davis, Faberman, and Haltiwanger 2012).

Figure 15 reports worker flows (i.e., quits and layoffs and their proxies), establishment births, and business applications. Panel A shows excess separations from QWI and the standard quits series from the BLS Job Openings and Labor Turnover Survey (JOLTS), along with BED establishment births and BFS high-propensity business applications (all series indexed to 2019 rates). Prior to the pandemic, quits and excess separations moved in similar patterns (albeit with some level shift), consistent with their close conceptual relationship. This co-movement continued in the pandemic, with an initial drop in quits and excess separations followed by a recovery to historic levels (admittedly more dramatic for quits). Over the same period, business applications and actual establishment births surged as well. Panel A shows one other series as well: job-to-job separations from

Figure 15.

Worker Flows and Applications

Source: QWI; JOLTS; BED; BFS; and the US Census Bureau J2J.

Note: Index of series expressed relative to employment or, for births, to establishments; seasonally adjusted. Applications are likely employers (HBA). Shaded areas indicate NBER recession dates.

the Census Bureau's Job-to-Job Flows (J2J), which is closely related to the QWI. This series measures separations of workers in which the worker quickly starts a new job with a different firm; as suggested by the discussion, excess separations closely track job-to-job separations in figure 15, as both are closely related to quits.

Panel B of figure 15 shows the spike in job destruction and layoffs in the second quarter of 2020. Both spikes are short-lived and, as noted previously, the layoffs in particular reflect a surge in temporary layoffs. Using data from the Current Population Survey (CPS) on inflows to unemployment from employment (using those entering unemployment in a month based upon duration data), about 85 percent of the massive surge

Figure 16.

Quits, Layoffs, and Applications, 2020–2023 versus 2010–2019

Source: JOLTS and Business Formation Statistics (BFS).

Note: State-level log differences of 2020–2023 versus 2010–2019 seasonally adjusted pace. The straight line is a regression line with reported slope and standard error. Data through August 2023.

in unemployment inflows in 2020:Q2 was due to temporary layoffs (see online appendix figure E17). Both series drop to low levels after mid-2020, even while business applications surged.

The two panels of figure 15, taken together, are suggestive of a relationship between quits (or their proxy, excess separations) and business formation, consistent with a theory in which workers quit their jobs to start, or join, new businesses. On the other hand, such a relationship between layoffs and business formation is not obviously apparent, as if the surge in business creation does not simply reflect laid-off workers starting businesses due to weak labor market opportunities. Still, these are simply aggregate series.

We therefore turn to spatial variation. We start at the state level, where JOLTS data on quits and layoffs as well as BFS likely employer applications are available; we employ the same approach as prior analyses to study the pandemic relative to pre-pandemic norms. As shown in panel A of figure 16, states with especially large surges in likely employer applications also saw especially large surges in quits during the 2020–2023 period; while there is much variation in both series, there is a substantive positive relationship that is statistically significant.³² As seen in panel B, there is

Figure 17.

Excess Separations, Layoffs, and Applications, 2020–2022 versus 2010–2019

Source: Quarterly Workforce Indicators (QWI) and Business Formation Statistics (BFS).

Note: County-level log differences of 2020–2022:Q2 versus 2010–2019 seasonally adjusted pace. The straight line is a regression line with reported slope and standard error. Binscatter with one hundred bins.

no apparent association between layoffs and high-propensity applications across states, consistent with the aggregate data in figure 15.

We next drill down to the county level, where we can examine related patterns using excess separations and job destruction from the QWI (our proxies for quits and layoffs) and total applications from the BFS. In figure 17, panel A shows a binscatter of county-level growth in the excess separations rate and county-level growth in (total) business applications per capita, where growth is again constructed as in equation (1). We observe a tight, statistically significant spatial relationship between growth in excess separations and growth in business applications. In panel B, though, we observe a much weaker (albeit positive) relationship between job destruction and applications.

While we might imagine multiple mechanisms underlying the observed spatial relationships, one possible explanation is that surging business creation and resulting labor demand is an important component of the overall story of worker flows in the pandemic, including quits. New businesses aggressively poach workers from other firms (Haltiwanger and others 2018) and, therefore, likely contributed to the pandemic reallocation of workers by providing new opportunities in pandemic-friendly industries. We know from figure 9 that job creation by establishment births during 2021 was substantial; with new establishments creating roughly one million jobs per quarter, some job-to-job flows—arising from excess separations—would likely result.

Interestingly, within cities we find a donut pattern of excess separation growth similar to the pattern for applications (and net establishment births); online appendix figure E19 shows that county-level growth in excess separations for New York City has been greater in the counties surrounding Manhattan than in Manhattan itself.

V. Business Dynamism Revived?

A large body of literature explores declining business dynamism, or the slowing of job and business flows in recent decades, including a decline in the firm entry rate and the share of activity accounted for by young and small firms. The evidence above suggests that the pandemic has been a period of increased dynamism relative to the 2010–2019 period. In this section, we consider the possibility of a return of the higher dynamism pace of the past (pre-2000). While we find noteworthy evidence of substantial economic restructuring during the pandemic—including reallocation of jobs and changes in the firm age and size distribution—we conclude that more time (and data) is needed for a material reversal of pre-pandemic trends.

V.A. Job Reallocation

Following literature that goes back a long way (Davis and Haltiwanger 1992), we define the job reallocation rate as:

where jc_t is gross job creation (total jobs created by entering and expanding establishments), jd_t is gross job destruction (total jobs destroyed by downsizing and exiting establishments), e_t is employment, and t indexes time (quarters, for our purposes). Job reallocation is a summary measure of the reallocation of jobs across expanding, opening, contracting, and closing establishments and is often used as a measure of business dynamism. The denominator in equation (2) is the Davis-Haltiwanger-Schuh (DHS) denominator after Davis, Haltiwanger, and Schuh (1996). Panels A and B of figure 18 show gross job creation, gross job destruction, and job reallocation; panel A zooms in on the pandemic period, while panel B shows a longer view.

Figure 18.

Perspectives on Job Reallocation

Source: Business Employment Dynamics (BED).

Note: Reallocation is jc + jd, from equation (2). Excess reallocation is jc + jd − |jc − jd|, with jc and jd averaged over indicated horizon. Seasonally adjusted. Shaded areas indicate NBER recession dates.

As has been extensively documented in the literature, job reallocation exhibits a downward trend over the last few decades and especially since the early 2000s. More recently, job reallocation spiked early in the pandemic; as shown in panel D, the pandemic spike was historic. The 2020:Q2 spike in reallocation was driven by the surge of job destruction. In the following quarter, reallocation moved down some but remained elevated; initially this reflected the surge of job creation as temporarily destroyed jobs returned.

There are two critical points to make about the early pandemic spike in reallocation. First, as just noted, the 2020:Q2 spike was driven entirely by surging job destruction and therefore simply reflects net (negative) job growth in that quarter rather than a dynamism phenomenon of simultaneous job creation and destruction across establishments; the 2022:Q3 elevation is similar but driven by job creation. Second, the pandemic was peculiar in that many of the jobs created in 2020:Q3 (and the immediately following quarters) were the same jobs—in the same establishments—that had been destroyed in 2020:Q2, as pandemic business restrictions or voluntary social distancing causing initial business closures and temporary layoffs were followed by quick resumption of business activities and recalls (Cajner and others 2020). As a result, quarterly excess job reallocation (job reallocation in excess of absolute net employment growth, or jr_t − | jc_t − jd_t|) actually moved down in 2020:Q2 and has not generally been significantly elevated during the pandemic (this can be seen in the one-quarter line in panel C of the figure, which we discuss more below).

Readers should carefully note that excess reallocation measures can be misleading in quarterly data, as noted in Davis and Haltiwanger (1992) and related work, especially when creation and destruction are decoupled or staggered in terms of timing. A clearer perspective emerges from measuring excess job reallocation using multi-quarter averages of job creation and destruction. Excess reallocation measured at two-, four-, or six-quarter horizons did indeed surge to a pace not seen in more than a decade, as can be seen in panels C and D of figure 18 (which also shows the dip in one-quarter excess reallocation).³³ Excess reallocation measured at multiquarter horizons (e.g., the six-quarter line in figure 18) was elevated for an extended period in the pandemic, though it came down again in 2022.

Without access to the micro data, we still cannot be certain that this multi-quarter horizon increase in excess job reallocation does not simply reflect job destruction in one quarter followed by job creation in the same establishment in subsequent quarters. To explore this question more, we return to the rich QWI data and focus on between-cell excess job reallocation, where cells are categories that can be defined in terms of firm age groups, firm size groups, geographic divisions, or industries; details are provided in online appendix D, but we provide an overview here. We find that between-cell excess job reallocation increased substantially in the pandemic, especially for cells defined in terms of firm age or firm size by themselves as well as when interacted with spatial or sectoral cells. In other words, we observe a substantial rise in the flow of jobs across these cell boundaries, which implies genuine job reallocation across businesses. The dominant role of reallocation across firm age and firm size groups leads us to explore changes in the firm age and size distribution in the next section.

V.B. Changes in the Firm Age and Size Distribution

The evidence on reallocation—and especially between-cell excess reallocation—implies an increase in the reallocation of activity across businesses in the pandemic. While the changes in the magnitudes of between-cell excess reallocation are large in percentage terms, they are relatively small in terms of absolute flows of jobs. We know from Decker and others (2016), Decker, Haltiwanger, and others (2020), and Karahan, Pugsley, and Şahin (2019) that an important source of the decline in indicators of business dynamism is the shift in activity toward large, mature firms: young and small firms are inherently more dynamic, so the decline in the share of the economy accounted for by young and small firms underlies a significant fraction (albeit far from all) of the decline in the pace of reallocation. In this context, it is instructive to explore changes in the age and size distribution of activity that occurred in the pandemic; we use annual BED data on activity by firm age and size through March 2022.

Figure 19 reports the change in the firm age distribution from March 2020—the very beginning of the pandemic—through March 2022. Panel A shows the percentage point change in the share of firms (solid bars) and employment (hollow bars) accounted for by each firm age group. Young firms' share of activity has risen a bit during the pandemic (after decades of trend decline); the shift in the share of firms is greater than the shift in employment, which is not surprising since pandemic entrants have been smaller than before the pandemic and because the effect of the surge of business entry on employment shares will inherently take time depending on survival rates and post-entry growth patterns of the new firms. The surge in entry has clearly left a mark on the firm age distribution, but even the share of firms five to nine years old increased; these are not pandemic births but are instead relatively young firms that were born before the pandemic. While the activity share changes in panel A of figure 19 must sum

Figure 19.

Changing Firm Age Distribution, March 2020 to March 2022

Source: BLS Business Employment Dynamics (BED).

Note: Firms and firm age defined by EIN.

to zero, panel B of the figure shows the percentage growth in the number of firms (solid bars) and employment over this period; for the 2020–2022 period as a whole, all firm age groups saw absolute growth, but the rate of increase was much higher for younger firms (though the growth rates are not quite monotonic). Again, even the oldest young firm category—those age five to nine years—saw rapid growth, with 5 percent more firms and 2 percent more employment than at the beginning of the pandemic (do not forget, though, that firms naturally progress through the age distribution via the process of aging).

We also examine changes in the firm size distribution. This is more challenging since firms can move both directions through the size distribution; firms with net job destruction may move into smaller size bins, while growing firms may move into larger bins. With this caution in mind, figure 20 reports changes in the size distribution in a manner analogous to figure 19. Panel A shows a shift in the share of firms and employment accounted for by small firms with fewer than 20 employees; but this shift has not been monotonic—firms with between 50 and 499 workers have seen large declines in their share of employment and, especially, firms. In contrast, firms with at least 500 employees have exhibited a modest decline in their share of firms—possibly reflecting firm exit but more likely reflecting firms downsizing into lower bins—but actually saw an increase in their share of employment, as some large firms likely benefited from the pandemic. Panel B, which reports growth in the level of firms and employment, tells a somewhat similar story, with all but the smallest size class seeing a decline in the number of firms but with the largest size class adding jobs. It is important to note that the 1 percent employment growth rate among large firms is substantial given that these firms account for roughly half of all employment, compared with the smallest size class whose share of employment is closer to one-sixth; at the same time, the smallest size class accounts for roughly 90 percent of all firms, so its 3 percent firm count growth rate reflects a large gain in the number of small firms.

As just noted, a challenge associated with firm size distribution analysis is that firms may move either direction across the distribution. But an attractive feature of the BED is that statistics on what BLS denotes as "dynamic sizing" are provided. Dynamic sizing assigns firm job growth to the size bin in which it occurred. For example, if a firm increases from zero employees (i.e., is a firm birth) to thirty-five over a window of time, the first nineteen jobs added are attributed to the 1−19 size class, and the increase from twenty to thirty-five jobs is attributed to the 20−49 size class. Thus, dynamic sizing provides insights into how much of the change in employment observed by size class is due to firms moving across size classes relative to changes within size classes. The BED provides dynamic sizing–based job growth by firm size bin on a quarterly basis.³⁴

Panel C of figure 20 reports both the actual change in the level of employment associated with each size bin (hollow bars), which is based on comparing employment levels in March 2022 and March 2020, and

Figure 20.

Changing Firm Size Distribution, March 2020 to March 2022

Source: BLS Business Employment Dynamics (BED).

Note: Firms defined by EIN. Dynamic method distributes net growth across size categories in which it occurs.

the cumulative dynamic sizing–based employment change (solid bars, constructed by summing quarterly dynamic job flows, by size class, from March 2020 through March 2022). Consider the smallest size class: since the solid bar (dynamic change) is larger than the hollow bar (change in levels), we can infer that there was net movement of firms up and out of this size bin; job growth of firms that graduate out of the size class is (partly) attributed to that size class under dynamic sizing (solid bar) but is not attributed to that class when we simply measure the change in static employment levels (hollow bar). This result for the smallest class is consistent with the surge in firm births, which are typically small, and suggests that some of these firm births—and perhaps also some preexisting small firms—grew out of this size bin. In contrast, for the largest size class, the hollow bar is larger than the solid bar, from which we can infer that there was net movement of firms downward out of this size bin; this is consistent with the net decline in the number of firms in this size class shown in panel B.

Additional perspective on firm size can be gained by studying the average size of new firm entrants; online appendix figure E14 shows that the average size of new firm entrants in BED data stepped down in the pandemic, consistent with our earlier discussion about unit counts versus employment from entrants. But figure E14 also shows that average entrant size relative to average incumbent size has remained on its pre-pandemic trend; that is, the drop in average entrant size is similar—relative to trend—to the change in average incumbent size. In other words, the relative small size of entrants in the pandemic is not unique to entrants.³⁵

These shifts in the firm age and size distribution are remarkable, particularly in a recessionary environment; small and young firms—including young firms born before the pandemic and its dramatic business formation surge—appear to have fared remarkably well during the pandemic. But how much have these shifts reversed the pre-pandemic trends toward mature and large firms? The answer is "not much." Figure 21 depicts the evolution of indexes of employment (panels A and C) and firm counts (panels B and D) by firm age (panels A and B) and firm size (panels C and D). Focusing on panels A and B reporting firm age data, the pre-pandemic shift in activity toward mature firms is evident as the indexes of mature-firm employment (panel A) and firm counts (panel B) rise dramatically during 2000–2020.³⁶

Figure 21.

Evolution of Firms and Employment by Age and Size

Source: Business Employment Dynamics (BED).

Note: March snapshots. For age classes above zero, employment measured as implied quarterly DHS denominator. Series indexed to their 2019 value.

In contrast, consistent with the decline in employer business entry, the indexes for young firms (especially firm births at age zero) decline, on net, over this twenty-year period. In the pandemic, these trends begin to reverse—but the decline for mature firms is very modest.

Turning to the evolution of the size distribution (panels C and D of figure 21), the activity shift toward larger firms in recent decades is evident.³⁷ Again, in the pandemic we have seen some reversal of earlier trends—especially for small firm counts, but not so much on the employment side. Large firms have more employment in 2022 than in 2019, consistent with our evidence above.

Our reading of the data is that there is potentially a beginning of a reversal of the shift in activity to large and mature firms; this is noteworthy and suggests young and small firms weathered the pandemic reasonably well (and, of course, entry has been remarkable). But so far the reversal relative to previous trends is quite modest. A related way to see that the impact has been modest is to compare firm- and employment-weighted entry rates, which we do in online appendix figure E1; there is a notable increase in the firm-weighted start-up rate, but the increase in the employment-weighted start-up rate is less noteworthy.

It is too early to declare an end to the multi-decade decline in business dynamism; such an end will require a sustained increase in employer business entry with, in turn, robust post-entry dynamics (i.e., not a decline in survival rates and post-entry growth conditional on survival). A onetime increase of entry and job reallocation—even if spanning a few years—is different from a persistent elevation of dynamism flows. Still, the striking rise in young and small firm activity in the pandemic is noteworthy.

VI. Taking Stock

Using several official data sources, we document close relationships between business applications, business entry, and job and worker flows during the pandemic. Our findings indicate that the surprising surge in business applications and registrations seen during the pandemic represented genuine entrepreneurial activity and resulted in considerable job creation and reallocation of jobs and workers. This surge in employer entrepreneurship is remarkable given the weakness in broader economic conditions from which it emerged, and it stands in sharp contrast with the plunge in employer entrepreneurship seen during the Great Recession. The increase in entrepreneurial activity left its mark on the firm age and size distributions, with a higher share of activity accounted for by young and small firms.

Our findings are consistent with the surging applications yielding increasing new employer businesses. However, it is still too early to study these transitions directly, a task that will require micro data not currently available: the micro data will permit studying applications that transitioned into employer start-ups with a focus on characteristics like industry, location, and entrepreneur demographics, along with post-entry life cycle dynamics. Investigating the demographic patterns of pandemic entrepreneurship looks to be of considerable interest; for example, Fazio and others (2021) find that at the zip code level African American population is strongly predictive of business registrations, so the pandemic may have provided entrepreneurial opportunities to minority groups that have historically faced challenges to business entry.³⁸

A related issue that warrants further attention is the high-frequency dynamics of applications and business entry over the course of the pandemic. As we have noted, the surge in applications came in two waves: an initial short-lived wave in the summer (especially July) of 2020, then a second, still ongoing wave commencing in early 2021. It may be that these two waves reflect different incentives and dynamics. The first wave may reflect the distinct market opportunities that arose just after the onset of the pandemic (e.g., online retail), but it may also reflect an increase in nascent entrepreneurship or entrepreneurial brainstorming. Many people found themselves with extra free time in the summer of 2020, given avoidance of high-contact leisure activities and time savings from fewer commutes; some may have used that time—along with broader reassessment of career goals—to consider starting a business. In some cases, these early entrepreneurial ideas may have been overtaken by the (partial) return to more normal patterns of work and leisure later in 2020, and, indeed, we find that the BFS projected firm birth series jumps less than simpler application count series and features a smaller surge in the retail sector. In contrast, in 2021, vaccines started becoming available and pathways out of pandemic isolation were becoming increasingly clear as the country gradually transitioned toward a post-pandemic new normal. Potential entrepreneurs had more information to plan and start serious businesses by 2021 and this has continued through 2023. We raise these issues since it may be that the transition dynamics of applications to new businesses are very different across these waves. We still lack the data to rigorously discern this distinction, but we do find preliminary evidence for lower transition rates in the early wave, which we discuss in online appendix B.

Our strongest evidence on the surge in business entry is from data on gross and net establishment entry, which includes both new firms and new establishments (new operating locations) of incumbent firms. We find a large and sustained increase in aggregate gross and net establishment entry through 2023:Q1, and the industries and locations with the largest increases in gross and net establishment entry tend to have the largest increases in new business applications. Our evidence on firm entry is consistent with these patterns but is only available through 2022:Q1 and with less industry and spatial detail.

The incentives for new business opportunities induced by the pandemic and its aftermath apply to both new and existing firms, but is the distinction important? Both types of establishment entry are inherent components of reallocation of business activity across the economy, but historically, rapid post-entry growth and innovation are more associated with new firms than with new establishments of existing firms.³⁹

Our findings also raise questions about the role of pandemic policies that strongly supported aggregate demand and eased credit conditions—which may be expected to boost firm entry—while also subsidizing incumbent firms via the PPP, the Main Street Lending Program, and the Federal Reserve's corporate credit facilities; Decker, Kurtzman, and others (2020) find that these business support policies included virtually the entire (incumbent) business distribution in their nominal scope for firm size, industry, and legal form. We must leave these and related questions for future research, which we hope will be informed by the large collection of facts we have assembled. In the meantime, our existing results suggest that entrepreneurship has played a key role in pandemic-era labor market dynamics.

One topic that is conspicuously missing from our analysis is an investigation of the surge in business applications that are likely nonemployers. Per Bayard and others (2018), likely nonemployer applications have a very low probability of becoming employer businesses (about 3 percent), and prior to the pandemic these applications tracked nonemployer activity reasonably well (Haltiwanger 2022).⁴⁰ Given the very large increase in likely nonemployer applications, the increase in entrepreneurship may be substantially greater than we have characterized via the potential increase in new nonemployer businesses. But the Nonemployer Statistics (NES) from the Census Bureau are currently available only through 2020. An alternative path is to use the Current Population Survey (CPS) or other household surveys that track self-employment activity; but there has been a growing discrepancy between self-employment activity tracked by the administrative data, such as the NES, and household data (Abraham and others 2021). Relatedly, the nonemployers of relevance to the BFS are those with an EIN, but most nonemployers do not have an EIN. Nonemployers with EINs are substantially larger than those without an EIN; only 15 percent of sole proprietors have EINs, and the small sole proprietors without EINs are dominated by individuals for whom nonemployer activity is supplemental (often to a wage and salary income) or reflects stopgap activity.⁴¹ Published NES data do not separately tabulate sole proprietors with and without EINs, and the CPS only distinguishes between incorporated and non-incorporated self-employed. In short, there are challenges to investigating the implied dynamics of the surge in likely nonemployers. But given the magnitude of the increase in likely nonemployer applications (see figure 1), exploring this topic is of considerable interest; moreover, there has been much discussion of the pandemic changing attitudes toward work, including the recognition that important tasks can be done remotely. And an argument could be made that the nonpecuniary benefits of being one's own boss—as discussed in Hurst and Pugsley (2011)—may have risen. A potential implication is that individuals have increasingly decided to go out on their own as nonemployers, but at this point nonemployer measurement is limited.

VII. Implications for the Future?

Given that we are only beginning to observe the real activity effects connected to the surge in new business applications, discussion of the implications of this surge for the future of US economic activity can only be highly speculative. Thus, here we provide some discussion about what potential patterns are worth contemplating in the coming months and years.

First, we emphasize that the full implications of the pandemic start-up surge will take several years to unfold. This reflects the highly volatile nature of start-ups, especially over their first five to ten years. Most startups fail or, at least, do not grow (Decker and others 2014). A small fraction grow rapidly, and this small subset of entrants is disproportionately important for the contribution of start-ups to job creation, innovation, and productivity growth (Decker and others 2014; Guzman and Stern 2020; Sterk, Sedláček, and Pugsley 2021). Theory and evidence suggest that start-ups are a core part of the experimentation that accompanies the development and adoption of new technologies and production processes, though this experimentation necessarily involves many business failures (Foster and others 2021).

Second, this increase in start-ups has occurred in spite of factors that were dampening the pace of business entry—and business dynamism more generally—in the decades leading up to the pandemic (Decker, Haltiwanger, and others 2020). It is unlikely that those factors, while still not completely understood, have disappeared entirely. Whether the countervailing forces driving the pandemic surge are sufficient to change the pre-pandemic trend decline is unclear; as we discuss in section V, the shock to entry and reallocation seen during the pandemic would have to be very persistent, and the new cohorts of entrants would have to feature a sufficient number of high-growth firms, for past trends to be substantially reversed.

Third, it may be important to consider the dynamics of aggregate productivity prior to the pandemic. In online appendix figures E21 and E22, the well-known productivity slowdown in the post-2005 period, and especially since 2014, is evident even in the innovative high-tech sectors of the economy. Many factors have been proposed as underlying this slowdown—including the decline in dynamism and entrepreneurship (Decker, Haltiwanger, and others 2020)—so the pandemic-era pattern of business formation may have implications for how productivity evolves going forward.

This discussion suggests some possible implications of the pandemic business entry surge. One possibility is that this surge is associated with a burst of innovation, with start-ups being an important component of the experimentation leading to that innovation. Hints of this possibility may be seen in the industry composition of surging applications and establishment openings (online appendix figure E10), with high-tech industries like nonstore retail, software publishing, computer systems design, scientific research and development services (e.g., AI businesses), and data processing apparently seeing especially elevated entry. While the evidence on actual new employer businesses in high-tech industries is still emerging, high-tech industries have the highest pace of projected start-ups of any broad sector through September 2023. Tracking the potential for surging entrepreneurship to spark economic growth and technological progress should be a high priority; eventually we would hope to see such progress reflected in productivity statistics, and a productivity boost from surging start-ups could mean stronger growth of potential output for the economy overall. Again, it will take some time for these dynamics to unfold, but early signals of the nature and composition of this surge might be detected, for example, using the nowcasting methodology of Guzman and Stern (2017).

Alternatively, this surge may reflect the type of spatial and sectoral restructuring that we have detected—but only insofar as such restructuring is necessary for providing basic support activities for the changing nature of work and lifestyle, with no broader spillovers in terms of innovation, productivity, and growth. In other words, the surge in start-ups suggested by the data we have reviewed could reflect a reshuffling of economic activity without leading to additional technological progress or growth. The surge of entrants in the service industries (e.g., restaurants and gyms) is consistent with this perspective. And the within-city donut effects we (and others) observe in the spatial patterns of applications and actual increases in net establishment growth may reflect business formation to support the increased fraction of working hours spent at home, and little else. Such support activity is likely very important to enable the changing nature of work—to the extent that the change is persistent—but it is unclear that such reallocation would herald a burst of innovation and productivity growth. A related possibility is that the pandemic presented a shock to entrepreneurial preferences, as in Hurst and Pugsley (2011); this is consistent with the drop in average entrant size. Whether persistent or not, such a shock is also unlikely to be associated with a burst of innovation and productivity growth.

Finally, we acknowledge the widely speculated upon possibility of an economic slowdown. Since early 2022, US monetary policy has tightened materially in response to elevated inflation, and financial condition measures are now much more restrictive than they were in the early pandemic period (Ajello and others 2023). While business applications have remained reasonably stable at their elevated pandemic level through September 2023 (see figure 1), monetary policy is typically thought to operate with long and variable lags. Existing literature—for example, Davis and Haltiwanger (2021)—finds that start-ups and young businesses are particularly sensitive to business cycle fluctuations, particularly those associated with tight financial conditions (e.g., falling house prices, rising interest rates, or declining business lending activity). The young businesses started during the pandemic, and the continued elevated trend of business applications, may be at risk in the event of a broad economic slowdown.

Comment and Discussion

Jorge Guzman

comment by JORGE GUZMAN

Ryan Decker and John Haltiwanger bring to this issue of BPEA a thought-provoking piece on the evolution of US entrepreneurship after the COVID-19 pandemic. Using multiple US Census Bureau data sets they present systematic evidence that the level of firm formation for both employer and nonemployer firms increased after COVID-19. This increase is large and, at least up to the time of writing, persistent. Importantly, the rise in entrepreneurship comes as a much needed respite to the long drop in the quantity of young firms previously documented by the authors (Decker and others 2014). At least within their census data, it is the first substantial increase in the number of new firms since 1977, the earliest year available. Other data sets unrelated to the census have also documented an increase in entrepreneurship after COVID-19, most notably business registration statistics using state-level registries (Fazio and others 2021), suggesting that the increase documented by Decker and Haltiwanger is real.

The bulk of my discussion focuses on two questions. First, what is causing this boom in new firm formation? Second, what does such a large increase in new firms imply for the economy? Neither question has a clear answer, but the gap is particularly salient for the latter one. The inability to answer these questions emphasizes how nascent our understanding of the role of entrepreneurship in the economy is, making it fertile ground for future research.

Why did entrepreneurship increase after covid-19?

Is entrepreneurship rising due to higher business dynamism and creative destruction?

For economists, the most valuable benefit of entrepreneurship to the economy is its crucial role in productivity growth. This occurs through two channels: business dynamism, or the reallocation of labor and capital from less productive to more productive firms even within narrowly similar product categories (Decker and others 2014); and creative destruction, or the process through which the desire for profits leads to process, product, and organizational inventions that incorporate a de novo way of doing economic activities (Schumpeter 1943; Akcigit and Kerr 2018; Acemoglu and Robinson 2013). The line between these two activities is not clearly defined. Many cases may imply both, and some economists have used the terms business dynamism and creative destruction interchangeably. However, they refer to different sources of variation on the nature of productivity growth. Business dynamism more closely relates to the efficient allocation of capital and labor across existing projects in the economy, while creative destruction, even if resulting in business dynamism, focuses more on the way profit motives promote investment for the development of new technologies, organizations, and business models.

Both within and outside the paper, evidence is consistent with both effects being partly responsible for the changes in US entrepreneurship after COVID-19.

Consider creative destruction first. By looking at changes within individual industries, Decker and Haltiwanger show in figure 3 that changes in industry composition related to technological innovation are taking place. Some of these industry changes are temporary (e.g., the need for more personal and health care services in 2020 or the supply chain struggles of 2021). However, by 2021, we observe what appears to be a partial reorganization of the economy: the founding of new nonstore retailers (e-commerce) has more than doubled, and new firm start-ups in the professional, scientific, and technical category, which includes the majority of those typically called tech firms, has also increased.

Other evidence outside the paper also supports this hypothesis. In particular, there was a boom in venture capital financing during the COVID-19 years, which in 2022 reached its highest levels since the year 2000. Research has documented clearly that venture capital booms lead to the financing and growth of more innovative ideas (Howell and others 2020; Nanda and Rhodes-Kropf 2013), making it possible that the current wave of new innovations, such as artificial intelligence or commercial space travel, creates a more productive organization of the economy.

Next, consider business dynamism. Beyond innovation incentives, do we observe economic activity reallocating from less productive to more productive firms?

Here, it is useful to remind ourselves of the details of the economic moment in which the boom in entrepreneurship occurred. In the period after COVID-19, employee quit rates increased despite strong economic fundamentals, leading to a phenomenon sometimes called the Great Resignation. By 2022, for example, employee quit rates were 50 percent higher than would have been predicted by models based on economic fundamentals (Gittleman 2022). At the same time, existing firm sales dropped precipitously, by up to 40 percent in 2021 (Barrero and others 2021), while labor force participation appears, if anything, to have increased (Sheiner and Salwati 2022). Put simply, the economy is robust and there are a substantial number of jobs, but incumbent firms are not doing well and workers are leaving them quickly. Where is all this labor to flow? The most likely possibility is new entrants, that is, entrepreneurship.

Decker and Haltiwanger present evidence that appears consistent with this story. In figure 9, for example, they show that establishment exits have been increasing concurrently with entry. In figure 11, they show that excess entry has occurred in virtually all sectors of the economy.

Overall, the evidence suggests an increase in business dynamism and business reallocation. However, it is also fundamental to ask why individuals have increased their preference to become entrepreneurs.

Is there a changing utility value of entrepreneurship, and could there be a role for work-from-home technology?

A different family of explanations does not focus so much on macroeconomic concepts such as creative destruction or dynamism but instead uses a choice-based approach to consider why some individuals would leave wage employment for the opportunity to start a new firm. When one considers the typical US resident's utility function, what is entrepreneurship's role in maximizing utility, and has this changed? Explanations considering this argument focus on two separate shocks through COVID-19. First, they emphasize that the COVID-19 shock and lockdowns, by requiring families to remain at home for extended periods (sometimes making significant changes to their space at home or their living situation), increased the importance individuals placed on being at home or independent. This, in turn, led them to start more firms. Second, the argument also tends to have a technological logic behind it: the advent of work-from-home (WFH) technologies, particularly videoconferencing, enabled many individuals to remain at home and finally do the independent work that is best suited to them.

Under these utility-based explanations, the economic benefits of the rise in entrepreneurship become more nuanced. Even if the choice to start a firm is utility maximizing, it does not lend itself directly to productivity improvements for the economy. While the once-worker-now-entrepreneur is possibly better off (at least based on revealed preferences), the economy may be the same. Indeed, in extreme cases, the additional focus on independence and leisure may lead to a productivity slowdown, in which the economy is composed of too many small firms that do not scale due to utility-driven growth frictions (Hamilton 2000).

For existing workers, the hypothesis that WFH technologies increased entrepreneurship does not appear consistent with research on the impact of information technology (IT). In particular, the presumed role of WFH technologies in enabling a large portion of new home-based businesses seems less likely, because even though WFH technologies certainly increased the possibility of starting a business at home, its most significant impact was in enabling the possibility of working from home as the employee of a company. The main utility benefits of entrepreneurship, such as freedom and time flexibility (Hamilton 2000), being close to home (Rosenthal and Strange 2012), or being away from one's boss, have become relatively much more accessible to company employees. Given evidence that IT typically supports a decentralization of decision authority and an emphasis on subjective incentives, both of which seem complementary to working from home, the most realistic prediction would instead be a reduction in new firm formation and a boom in jobs in big corporations, as a large share of both existing and new workers find a series of jobs (previously inaccessible) that give the freedom they seek.

Yet, this argument is only half the picture. To the extent that worker preferences also changed toward being an entrepreneur by valuing freedom and flexibility more, or that WFH technologies allowed individuals previously out of the labor force to reenter the economy, then the overall incidence of entrepreneurship could increase.

A different potential channel for WFH technologies involves changing the boundary of the firm, allowing some transactions that used to take place in a firm to be done through the market (Forman and McElheran 2019). This is the case, for example, with gig workers on platforms such as Uber and Taskrabbit, both of which created many small-scale entrepreneurs who provide services to the platform or use gig work as a baseline to start firms on their own (Barrios, Hochberg, and Yi 2022). The possibility that these platforms enabled additional online services is still to be investigated.

Finally, bringing back the possibility of changing worker preferences, there may be individual changes in the types of jobs people are willing to accept. Besides the COVID-19 pandemic, the year 2020 was witness to one of the largest social movements since the civil rights era, Black Lives Matter, leading one to ask whether minority groups might have more directly experienced a change in the way they think through or choose their career.

Is there a rise in entrepreneurship for minorities?

Building on the results presented in earlier work by Haltiwanger, in Fazio and others (2021) my coauthors and I use business registration records to document a significant heterogeneity in the changing geography of entrepreneurship after COVID-19.¹ Our key result is that this heterogeneity does not merely reflect the gradual transition of individuals out of central business districts into the suburbs but instead is statistically related to race: zip codes with a high share of Black residents have the highest increases in entrepreneurship. Other variables such as income, population density, or age hold no relationship. The impact is even more striking when one considers contiguous zip codes within a city. For example, in maps of New York City, we can consider changes in entrepreneurship across neighborhoods that are adjacent but have significantly different racial compositions, such as Central Harlem, Morningside Heights, and Washington Heights. There are substantial differences among them, with Harlem clearly having a larger increase in new firms compared to others. This pattern is also apparent in this paper by Decker and Haltiwanger. In their state-level map (figure 4), we observe that the largest increases are in the Deep South, including states typically low on productivity, such as Alabama and Mississippi. These increases surpass other states that saw ample in-migration during COVID-19 by people expecting to work from home, such as Florida, Arizona, Texas, and Tennessee. All of this suggests the possibility that the increase in entrepreneurship after COVID-19 is related to the incidence of Black population across regions.

There are at least three mechanisms for such an increase. The first possibility is a change in local demand. Since the pandemic created a significant movement of people, these new residents would now create local demand in new neighborhoods. Such an explanation does not appear readily consistent with the empirical patterns. While pandemic reallocation happened out of business districts and toward lower-density areas, Black neighborhoods are in more dense locations than white neighborhoods. Zip code population density also does not predict the increase in entrepreneurship rates in our analysis.

A second group of explanations instead relates to more behavioral aspects associated with changes in the demand, jobs, or general expectations for potential Black business owners. Bennett and Robinson (2023) document significant differences in business practices across race, which in turn can be influenced by social movements that co-occurred with the COVID-19 pandemic, such as Black Lives Matter. This appears an important question much in need of empirical evidence.

Finally, a third (and clearer) option is that COVID-19 ultimately brought differences in financial access.

Has there been improved financial access for minorities after COVID-19?

There are at least two mechanisms through which the COVID-19 pandemic could have increased financial access. One is government intervention; the other is technological change through fintechs. To understand the logic of both it is important to recognize the differences that exist in the incidence of financial institutions across neighborhoods and race. As documented by Small and others (2021), predominantly Black neighborhoods tend to be farther away from conventional retail banks, making traditional access to financing harder. Policies that reduce such geographic inequality can be particularly valuable for new investments, including new firms.

Consider the government interventions during the pandemic: the COVID-19 stimulus package was, to a large extent, equally distributed across neighborhoods, ameliorating disparate access to financing due to geography. In Fazio and others (2021), we also show a measurable increase in entrepreneurship in the few weeks after the American Rescue Plan (Biden stimulus).

In the case of fintechs, the key possibility is that because online banking companies are less locally determined, they may be able to access areas that are not typically well banked. Erel and Liebersohn (2022) show that fintech banks are more likely to serve minority households and locations with fewer bank branches. Chernenko and Scharfstein (2022) find there were wide racial disparities in the Paycheck Protection Program, which are at least partially ameliorated by fintechs.² In essence, the transition to more online banking after COVID-19 may have had a positive influence for previously underbanked neighborhoods.

What is the impact of a boom in entrepreneurship on the economy?

Moving beyond the causes of the rise in entrepreneurship after COVID-19 to the consequences, it is only natural to ask what are the economic implications of this massive increase in new firms.

Here, one can't help but be surprised at the level of uncertainty that comes with these predictions. Even though the rise in entrepreneurship during COVID-19 is the largest increase in our lifetimes, the predictions drawn from this increase by the authors and other entrepreneurship economists (myself included) are very cautious. We do not know exactly what it means, and we are not sure whether it implies increases in productivity growth, creative destruction, or social equity.

The fact that we are unable to predict outcomes is a symptom of the incompleteness, and opportunity, of entrepreneurship economic theory. Whereas a macroeconomist knows that productivity numbers of 4 percent, 2 percent, or 1 percent are worlds apart from each other in their implications for the economy, or that inflation at 1 percent versus 5 percent would lead to drastically different paths of investment and business activity, entrepreneurship economists do not yet know what to make of the shifts and flows of new firm formation for the economy or even for our own conclusions. While the mechanisms of entrepreneurship are now somewhat appreciated, the way these come together to have an impact on economic growth is not.

Conclusion

Decker and Haltiwanger present a paper that, like many good papers, opens more questions than it answers. By going through the careful process of simply describing the evolution of measures of new firm formation within the US Census, they leave the reader with the desire to learn a lot more about both the causes and consequences of entrepreneurship in the economy. Large economic shocks, such as the Great Depression, the stagflation of the 1970s, or the collapse of the Soviet Union, have always provided fertile ground for economists to test their theories and, ex post, develop new substantive ones that can better explain the changing economy. The COVID-19 pandemic is likely to be a similar shock, providing much to study regarding the reorganization of the economy, with entrepreneurship being one of the settings in which this takes place.

References for the guzman comment

Acemoglu, Daron, and James A. Robinson. 2013. Why Nations Fail: The Origins of Power, Prosperity, and Poverty. New York: Crown.

Akcigit, Ufuk, and William R. Kerr. 2018. "Growth through Heterogeneous Innovations." Journal of Political Economy 126, no. 4: 1374–443.

Barrero, Jose Maria, Nicholas Bloom, Steven J. Davis, and Brent H. Meyer. 2021. "COVID-19 Is a Persistent Reallocation Shock." American Economic Association Papers and Proceedings 111:287–91.

Barrios, John M., Yael V. Hochberg, and Hanyi Yi. 2022. "Launching with a Parachute: The Gig Economy and New Business Formation." Journal of Financial Economics 144, no. 1: 22–43.

Bennett, Victor Manuel, and David T. Robinson. 2023. "Why Aren't There More Minority Entrepreneurs?" Social Science Research Network, February 21. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4360750.

Chernenko, Sergey, and David S. Scharfstein. 2022. "Racial Disparities in the Paycheck Protection Program." Working Paper 29748. Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.org/papers/w29748.

Decker, Ryan A., John C. Haltiwanger, Ronald S. Jarmin, and Javier Miranda. 2014. "The Role of Entrepreneurship in US Job Creation and Economic Dynamism." Journal of Economic Perspectives 28, no. 3: 3–24.

Erel, Isil, and Jack Liebersohn. 2022. "Can FinTech Reduce Disparities in Access to Finance? Evidence from the Paycheck Protection Program." Journal of Financial Economics 146, no. 1: 90–118.

Fazio, Catherine E., Jorge Guzman, Yupeng Liu, and Scott Stern. 2021. "How Is COVID Changing the Geography of Entrepreneurship? Evidence from the Startup Cartography Project." Working Paper 28787. Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.org/papers/w28787.

Forman, Chris, and Kristina McElheran. 2019. "Production Chain Organization in the Digital Age: I.T. Use and Vertical Integration in U.S. Manufacturing." Social Science Research Network, June 13. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3396116.

Gittleman, Maury. 2022. "The 'Great Resignation' in Perspective." Monthly Labor Review, July. https://www.bls.gov/opub/mlr/2022/article/the-great-resignationin-perspective.htm.

Griffin, John M., Samuel Kruger, and Prateek Mahajan. 2023. "Did FinTech Lenders Facilitate PPP Fraud?" Journal of Finance 78, no. 3: 1777–827.

Haltiwanger, John C. 2022. "Entrepreneurship during the COVID-19 Pandemic: Evidence from the Business Formation Statistics." Entrepreneurship and Innovation Policy and the Economy 1:9–42.

Hamilton, Barton H. 2000. "Does Entrepreneurship Pay? An Empirical Analysis of the Returns to Self-Employment." Journal of Political Economy 108, no. 3: 604–31.

Howell, Sabrina T., Josh Lerner, Ramana Nanda, and Richard R. Townsend. 2020. "How Resilient Is Venture-Backed Innovation? Evidence from Four Decades of U.S. Patenting." Working Paper 27150. Cambridge, Mass.: National Bureau of Economic Research.

Nanda, Ramana, and Matthew Rhodes-Kropf. 2013. "Investment Cycles and Startup Innovation." Journal of Financial Economics 110, no. 2: 403–18.

Rosenthal, Stuart S., and William C. Strange. 2012. "Female Entrepreneurship, Agglomeration, and a New Spatial Mismatch." Review of Economics and Statistics 94, no. 3: 764–88.

Schumpeter, Joseph A. 1943. Capitalism, Socialism and Democracy. Oxford: Taylor and Francis.

Sheiner, Louise, and Nasiha Salwati. 2022. "How Much Is Long COVID Reducing Labor Force Participation? Not Much (So Far)." Working Paper 80. Washington: Hutchins Center on Fiscal and Monetary Policy at Brookings.

Small, Mario L., Armin Akhavan, Mo Torres, and Qi Wang. 2021. "Banks, Alternative Institutions and the Spatial-Temporal Ecology of Racial Inequality in US Cities." Nature Human Behaviour 5:1622–28.

Footnotes

. Editors' Note: Benjamin Pugsley provided a thoughtful discussion on the conference version of the paper by Decker and Haltiwanger at the Fall 2023 BPEA Conference. The recording of his discussion can be found at https://www.brookings.edu/events/bpea-fall-2023-conference/.

1. See, for example, Haltiwanger (2022).

2. See also Griffin, Kruger, and Mahajan (2023).

GENERAL DISCUSSION

John Sabelhaus asked the authors about the role of the social safety net in a broad sense, including student loan forgiveness, for understanding business formation during COVID-19. While issues such as financing constraints tend to be front of mind when discussing start-ups, Sabelhaus noted how the risk environment for entrepreneurs changed significantly during COVID-19 and how government intervention played an important role in enabling more people to start businesses.

Related to the active role of policy during this period, Ben Harris pointed to a range of programs specifically designed to route capital to small businesses, including the $800 billion in the Paycheck Protection Plan (PPP), $10 billion passed through the American Rescue Plan for the State Small Business Credit Initiative, and $9 billion through the Emergency Capital Investment Program.¹ He asked the authors to what extent they believed the stimulus programs were part of the story.

Moritz Schularick asked the authors to speculate on the literature on aggregate demand conditions and business formation and how that relates to the fact that this period was one marked by extensive government stimulus and relief efforts.

Şebnem Kalemli-Özcan suggested linking some of the results to the broader macro picture. First, she was curious about how entrepreneurs financed their new businesses and the extent to which personal savings played a role on the backdrop of the PPP. Second, Kalemli-Özcan noted that, during COVID-19, labor allocation was limited and asked how this fact could be reconciled with the authors' findings.²

Ryan Decker agreed that trying to incorporate the effect of policy in thinking about changes in business formation is important. Related to the risk environment, Decker commented that he believed there was perhaps a greater risk appetite given the expansion of the safety net, a change in people's sense that "everything will be all right." At the same time, he was struck by how business applications in the 2021 administrative data were rising despite many of the government stimulus programs coming to an end or having already come to an end.

John Haltiwanger clarified that the PPP was not for new businesses but rather for existing businesses—and theory suggests that this may in fact stifle entry. Haltiwanger said that while there has been concern that individuals set up employer identification numbers (EINs) in order to be eligible for PPP, data from the Census Bureau matched to the Business Formation Statistics (BFS) suggest this is not the case. Thus, fraud wouldn't be able to explain the surge, either. Along the same line of argument, Haltiwanger noted that we have seen a strong labor market for two plus years now, with lots of opportunities for employment, and we still had an enormous surge in business applications.

Pinelopi Goldberg argued that as people relocate, demand for services is expected to increase in these locations, explaining some of the new business entries. Similarly, we would expect to see exit rates increase in other locations. Consequently, Goldberg was interested in exploring what the net entry rate looked like.

Ayşegül Şahin asked the authors to discuss which part of the wage distribution workers who turned self-employed came from, stating that she thought it was of importance to wage dynamics. Following up on Şahin's comment, Gerald Cohen asked if there was a way to link micro data such as the Current Population Survey (CPS) to the authors' findings, to identify educational attainment and other characteristics of the newly self-employed. That would help shed light on the extent to which the rise in new businesses would bring increases in productivity, Cohen suggested.

In terms of gaining a more detailed understanding of who these new entrepreneurs are, Haltiwanger pointed to the possibility of integrating the BFS with the Longitudinal Business Database, and with the Longitudinal Employer-Household Dynamics data, which can provide information on who started a business and who was hired, noting that this is an important avenue for future work. He also mentioned that in joint work with Dinlersoz, Dunne, and Penciakova they found enormous spatial variation, suggesting the propensity for entrepreneurship differs by location.³ He further praised the work of discussant Jorge Guzman focusing on racial disparities in access to finance, which Haltiwanger argued is a first-order issue.⁴

Katharine Abraham questioned the paper's implicit assumption that all employer businesses are a primary activity, arguing this need not be the case. She offered the example of a catering business, which likely would have employees but could be something a person ran on weekends. Abraham also questioned the use of CPS data for drawing conclusions about how multiple job holding has changed over time, citing known issues in those data with undercounting the number of secondary jobholders. Offering advice to the authors, Abraham suggested they could use the data employed in their study to explore what kinds of jobs new businesses have been creating. It would be interesting, for example, to know how intensive these new jobs are, something that could be proxied using payroll per added employee.

Haltiwanger agreed that the CPS data do not track self-employment well in general, as documented by Abraham and others (2018), and that it is interesting to consider both new employer and nonemployer businesses. The focus of the paper is on new employer businesses but there has also been a surge in applications for likely new nonemployers as seen in figure 1. Nonemployer businesses are important; overall there are more than 25 million nonemployer businesses, compared to a little more than 6 million employer businesses.⁵ Most nonemployers are very small, but nonemployers that have an EIN are larger, as discussed in the paper. Nonetheless, Haltiwanger conceded that, to Abraham's point, any new non-employer businesses may still reflect mainly secondary activities.

In light of decreasing self-employment rates, Betsey Stevenson remarked that while we did see labor reallocation and increased entrepreneurship supported by the expansion of the social safety net during COVID-19, we ought to consider the extent to which the ability to form new businesses constitutes part of the safety net as well, helping individuals weather a storm when there are no employers around.

Robert Hall noted that a huge number of workers were placed on layoff in April 2020. Over the next few months, they were recalled to their existing jobs.⁶ Hall suggested that the rapid rate of return to existing jobs is an important fact that should be kept in mind in studying business formation during this period.

To the points of Stevenson and Hall, Haltiwanger thought that there might have been a lot of brainstorming related to entrepreneurship going on, particularly in the first period of the pandemic—people wanted to do things differently, and many were not in their offices.

Martin Baily steered the discussion toward productivity and brought up the ambiguity of projected productivity at the beginning of the pandemic. While the authors suggested there was a sense of general pessimism, several sources deemed positive productivity growth likely: work by Barrero, Bloom, and Davis pointed to reallocation effects which could be positive for productivity; Goldman Sachs expected that there would be a productivity surge following some creative destruction, and McKinsey produced a study that suggested there would be increases in investment and an expansion of new technologies.⁷ Baily continued, saying that in retrospect, while there were some fluctuations in productivity, the trend ultimately did not go anywhere, as documented by John Fernald and Huiyu Li.⁸ He asked the authors whether they believed that the increase in dynamism, which they speculated about, would lead to increases in productivity.

In response to Baily's comment, Haltiwanger made the point that while new small businesses may not all turn into the next big tech firm, they do represent a form of economic mobility not just for themselves but for the workers they hire. To Şahin's point, he noted that such hires are often low-skill labor. Haltiwanger believed that ranking industries to determine where innovation will come from next provides very crude information, emphasizing that every industry has a right tail which provides important contributions to innovation, productivity, and job creation. He also pointed to recent work on the particularly high rates of entrepreneurship documented in neighborhoods with a higher proportion of Black residents as a reason for preliminary optimism and an important avenue for continued research.⁹ Nonetheless, Haltiwanger highlighted that professional, scientific, and technical services have historically been particularly important for innovation—with the last productivity surge in the 1990s—but he noted that the effect from a surge in entry in high-tech sectors comes with a lag: previous research has shown that the productivity response comes six to nine years after an entry surge.¹⁰ Consequently, we should expect that any effect on productivity this time around would also take some time to materialize. Work by Gort and Klepper, as well as by Jovanovic and MacDonald, makes a compelling argument about how entrants induce innovation.¹¹ But innovation also spurs new business formation—the causality goes both ways, Haltiwanger concluded.

On the topic of innovation, Michael Falkenheim wondered whether there might be lessons to be learned from the literature on war for business formation and entrepreneurship, noting that COVID-19 was a similarly destructive event, and as such may also give rise to creativity.

Decker pondered whether the pandemic represented a persistent shock to the pace of entry; he expressed some skepticism but noted that one might need only a few cohorts of really innovative new firms in scientific and technical services in order to see an effect on productivity down the line, noting that recent entries include businesses that are helping other firms undergo technical change, such as IT consulting, engineering consulting, and data centers.

Iván Werning suggested that it may be useful to look at the outcome in other countries to perhaps gain additional insights given that we were all affected by COVID-19. Janice Eberly pointed to the United Kingdom as an example, noting that while workers were paid to stay with their employer through the Coronavirus Job Retention Scheme, many still ended up leaving.

Footnotes

1. US Small Business Administration, "SBA Announces Opening of Paycheck Protection Program Direct Forgiveness Portal," https://www.sba.gov/article/2021/jul/28/sba-announcesopening-paycheck-protection-program-direct-forgiveness-portal; US Department of the Treasury, "State Small Business Credit Initiative (SSBCI)," https://home.treasury.gov/policy-issues/small-business-programs/state-small-business-credit-initiative-ssbci; US Department of the Treasury, "Emergency Capital Investment Program," https://home.treasury.gov/policy-issues/coronavirus/assistance-for-small-businesses/emergency-capital-investment-program.

2. John Fernald and Huiyu Li, "The Impact of COVID on Productivity and Potential Output," in Economic Policy Symposium Proceedings: Reassessing Constraints on the Economy and Policy (Jackson Hole, Wyo.: Federal Reserve Bank of Kansas City, 2022).

3. Emin Dinlersoz, Timothy Dunne, John C. Haltiwanger, and Veronika Penciakova, "The Local Origins of Business Formation," working paper CES-23-34 (Washington: Center for Economic Studies, 2023).

4. Catherine E. Fazio, Jorge Guzman, Yupeng Liu, and Scott Stern, "How Is COVID Changing the Geography of Entrepreneurship? Evidence from the Startup Cartography Project," working paper 28787 (Cambridge, Mass.: National Bureau of Economic Research, 2021), https://www.nber.org/papers/w28787.

5. Katharine G. Abraham, John C. Haltiwanger, Kristin Sandusky, and James R. Spletzer, "Measuring the Gig Economy: Current Knowledge and Open Issues," in Measuring and Accounting for Innovation in the Twenty-First Century, Carol Corrado, Jonathan Haskel, Javier Miranda, and Daniel Sichel, eds. (Chicago: University of Chicago Press, 2021); US Small Business Administration, Office of Advocacy, "Frequently Asked Questions," March 2023, https://advocacy.sba.gov/wp-content/uploads/2023/03/Frequently-Asked-Questions-About-Small-Business-March-2023-508c.pdf.

6. Robert E. Hall and Marianna Kudlyak, "The Unemployed with Jobs and without Jobs," Labour Economics 79 (2022): 102244.

7. Jan Hatzius, Joseph Briggs, Devesh Kodnani, and Giovanni Pierdomenico, "The Potentially Large Effects of Artificial Intelligence on Economic Growth," Goldman Sachs, March 26, 2023, https://www.gspublishing.com/content/research/en/reports/2023/03/27/d64e052b-0f6e-45d7-967b-d7be35fabd16.html; Jose Maria Barrero, Nicholas Bloom, and Steven J. Davis, "COVID-19 Is Also a Reallocation Shock," working paper 27137 (Cambridge, Mass.: National Bureau of Economic Research, 2020); Shaun Collins, Ralf Dreischmeier, Ari Libarikian, and Upasana Unni, "Why Business Building Is the New Priority for Growth," McKinsey Quarterly, December 10, 2020, https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/why-business-building-is-the-new-priority-forgrowth.

8. Fernald and Li, "The Impact of COVID."

9. Fazio, Guzman, Liu, and Stern, "How Is COVID Changing the Geography."

10. Lucia Foster, Cheryl Grim, John C. Haltiwanger, and Zoltan Wolf, "Innovation, Productivity Dispersion, and Productivity Growth," in Measuring and Accounting for Innovation in the Twenty-First Century, Carol Carrado, Jonathan Haskel, Javier Miranda, and Daniel Sichel, eds. (Chicago: University of Chicago Press, 2021).

11. Steven Klepper, "Entry, Exit, Growth, and Innovation over the Product Life Cycle," American Economic Review 86, no. 3 (1996): 562–83, http://www.jstor.org/stable/2118212; Michael Gort and Steven Klepper, "Time Paths in the Diffusion of Product Innovations," Economic Journal 92, no. 367 (1982): 630–53; Boyan Jovanovic and Glenn M. MacDonald, "The Life Cycle of a Competitive Industry," Journal of Political Economy 102, no. 2 (1994): 322–47, http://www.jstor.org/stable/2138664.

Copyright © 2024 The Brookings Institution