Recent Patterns of New Business Creation

Note: Without implication, this talk relied primarily on research done jointly with John Haltiwanger (University of Maryland and the National Bureau of Economic Research).¹ The analysis and conclusions set forth are those of the author and do not indicate concurrence by other members of the research staff or the Board of Governors of the Federal Reserve System.

Let’s start with this chart from a Census Bureau product called the Business Formation Statistics that shows new business applications. The I.R.S. gives data to the Census Bureau on applications for new Employer Identification Numbers (E.I.N.). If you want to start a business, and you plan to become a corporation, or a partnership, or if you are going to be hiring workers, you will need an I.R.S.-issued E.I.N. Even some sole proprietorships obtain E.I.N.s for various reasons. The I.R.S. compiles these data, ships them over to the Census Bureau, then the Census Bureau takes the dataset, throws out non businesses like trusts and estates, and keeps everything else. Then they report these applications. Now importantly, some of these applications are aspirational. These businesses might not actually do or build anything, but they are all people who have applied for E.I.N.s.

The black line shows total applications. Those are monthly data. The green line shows likely employer applications. The Census Bureau flags some application characteristics, including indicated industry, the legal organizational form, or the intention to hire employees. Likely employer applications is a subset of the total series that have one or more characteristics that tend to be predictive of actual employer business creation. Importantly, this is not a model-based prediction of business entry, but a subset of the applications that tends to move closely with actual firm entry.

There are some very striking and surprising patterns. If you think back to how you felt in March or April 2020, very few of us expected a historic surge in business entry or really anything positive at all, but that is what we got. Initially, there was a dip in these applications, but then they surged to the highest pace on record. The sharp point on the chart is July 2020. Then the first wave eased down throughout the rest of the year. By 2021 however, all these series picked up again, and have remained elevated, even into March 2024.

A striking thing about this is that the likely-employer series moved closely with the likely-non-employers series (the difference between the total and the likely employers). This is surprising because it runs contrary to historic patterns. There are many reasons why “nonemployer entrepreneurship”—i.e., just doing your own thing and not hiring anybody/self-employment—might increase in a weak economic environment. People looking for income in a bad labor environment might try self-employment. Indeed, in the Great Recession, we saw an increase in non-employer entrepreneurship. But the likely employer series is surprising because employer entrepreneurship during recessions tends to plummet. But here we have it moving closely with the likely non-employers.

This surprise led to many questions, some of which we will answer today. First, are these applications serious? Initially many economists were skeptical about what this surge meant. Applying for an E.I.N is easy and low-cost. Maybe people who were working from home or temporarily laid off in summer 2020 with nothing to do were filing for E.I.N.s. It was possible that this surge in applications was meaningless. The concern was that even though these applications series have historically predicted actual employer entry successfully, for the reasons I just said, the pandemic surge could be different.

The next question is even if these are real employer entrants, what happened and why? I do not have rigorous causal analysis on this question today but what I am going to do is use these data to tell pandemic stories. We can make sense of these data in terms of geographic patterns and industry patterns by comparing them to other things we know about the pandemic. I will argue that this surge makes sense, given everything else we know.

I will also touch on productivity. Historically business entry is important for productivity because of how new and innovative ideas spread and accumulate productive resources. But could this time be different? Maybe, for example, these new businesses were lifestyle entrepreneurs who did not have an idea so much as they were looking for a change of pace. And if that is the case, then we might not see the productivity boost from this entry surge that we have seen from past entry surges. And finally, I will talk about longer run trends. My co-author and I and others have done a lot of research in pre-pandemic trends in business entry, I will put us in that context.

Let’s start by answering the first question. Actual employer business creation did, in fact, follow these E.I.N. applications. The chart’s green line is the likely employer application series. The orange line is a measure of employer establishment births (quarterly). These are actual business locations with actual formal W-2 employees. Let me clarify a bit of technical jargon first. The word “establishment” has a very precise meaning in U.S. statistical agencies. An “establishment” is any employer business operating location. What does that mean? It means that an “establishment birth” could be the start of a brand-new firm, meaning a new company starting up and hiring people, or alternatively, it could be a new Starbucks location. That distinction is important. We get those data at a quarterly frequency. Starting in the second quarter of 2021, establishment births increased. And they have remained very high, though recently they have started to come down again. But from the second quarter of 2021 through the most recent data in the third quarter of 2023, new establishments were creating a million jobs per quarter. That is an enormous number. And it has played a significant role in labor market stories.

But if these are new Starbucks locations rather than new firms, the implications would be quite different. The black line on the chart is our annual data on firm births. There is another, subtle technical point here: annual data end in March, meaning the data cover the twelve months leading up to March of a given year. So that first jump in the black line means that in the twelve months leading to March 2022, we saw a big jump in actual firm births, not just new Starbuck locations. They jumped in 2022, then moved down a bit in 2023, but stayed elevated. And these firm entrants created nearly two million jobs per year over the last couple of years. That is the highest pace of job creation from new firms since around 2007.

Let’s talk about some pandemic themes, starting with geography. I am going to show how this surge makes sense. Now, I am an economist, and we are often good at explaining things after the fact that we did not see coming. I am going to do that today: take advantage of hindsight. Here is the surge in business applications by state. Let’s take the pandemic-pace of business applications and compare it to the pre-pandemic-pace by state. We will take it on a per capita basis, so that we are not just picking up population flows. Which states saw the biggest increase in applications during the pandemic compared to pre-pandemic? There are some familiar patterns here: economic activity moved away from the Northeast towards the South, the Sunbelt, and the West.

Within cities, there is also an interesting geography story. This map on the right is showing you New York City’s counties. If you look at Manhattan, the little white county in the middle, business applications did not surge. Manhattan’s new business entry rates have tended historically to be high, so this is a break from the pre-pandemic norm. Our statistical analysis shows that this was generally the pattern in large cities: when compared with the suburbs, application increases in city centers were very low, and these patterns were correlated with remote work activity. This is a familiar pattern: new gyms and restaurants opened in the suburbs probably because more people were working from home. We are all familiar with what has happened in downtown areas, especially in major cities, and more in some parts of the country than in others. This movement from downtown areas to the surrounding counties is consistent with other pandemic stories.

It turns out that we see the same thing in the actual establishment entry as we did for E.I.N. applications. Let’s look at the net establishment entry at the county level in all U.S. counties. The chart on the left is a simple scatter plot. The horizontal axis is the surge in applications at the county level (pandemic versus pre-pandemic pace). And the vertical axis is the surge in net establishment entry at the county level. The counties that saw a surge in applications also saw a surge in actual employer establishment entry. There is a positive, strong correlation between the two. The story is similar in New York City. We find that, in general, the actual employer establishment entry followed a similar pattern within cities.

Now we will address some big industry stories during the pandemic. I am going into a relatively narrow level of industry detail. I have identified the top five industries that gave us these applications. The biggest one is the blue line, which shows non-store retail including online retail. There was a huge application surge in online retail during the pandemic. The chart shows a big blip in 2023. That is likely spurious: there were some changes to reporting requirements for marketplace third party sellers. But if you ignore that blip, you see an initial huge surge in non-store, then online retail that has eased off throughout the pandemic.

The red line is professional, scientific, and technical services. This is a large sector that includes industries ranging from home appraisers and architects, to various important high-tech services. It also includes some of the high-tech services that in survey data have been associated with A.I. activities within firms. The red line also includes computer systems design and research and development services. It has had a solid surge, and if anything, it has picked up over the last year or two.

Personal and laundry includes pet care. I’m not sure what exactly is happening in this group of industries, but maybe many people bought pets in 2020. Let’s skip to truck transportation, which includes last mile delivery. This makes sense. Early during the pandemic, there was a surge in people having purchases delivered to their houses. That required a lot more delivery vehicles. You can see it rise in 2021-2022, which is when the U.S. had tremendous supply chain challenges, including in freight markets and trucking markets. These industries are conducive to pandemic work, lifestyle, and business. The application surge starts to make sense.

Let me show another application-based series that uses an actual predictive model. The Census Bureau receives applications and then builds a model that looks at an application’s characteristics and predicts whether it will turn into a true employer firm birth. They can do this because they have the data on both sides of the transition. The blue line is retail trade at the broad sector level. That little jump I told you about in 2023 does not show up here, meaning the predictive model was not fooled. Let me highlight the professional scientific and technical services line—the red line—which has been picking up recently. It is still an application-based series. It is predictive of the employer entry, but it is application based.

Here we have some actual employer entry information. This scatterplot compares actual employer entry to the application surge at the broad sector level. The horizontal axis shows the surge in applications. And then I look at the surge in actual firm births. The 45-degree line is growth rates. Sectors that land on that 45-degree line saw the same surge in applications and in firm births. Quite a few sectors are on the line. Sectors that saw a big application surge did in fact see a surge in firm births.

There are a couple of noteworthy exceptions. Retail is the most glaring: the surge in applications in the retail sector was much larger than what actual employer data has shown. There is also an exception in the information sector. That is a tech intensive sector that includes things like software publishing and data hosting. Very interestingly, the fact that we saw a bigger increase in actual firm births than we did in applications might suggest that the transition rates—transitions from application to employer—in that industry were very high. Now we can slice these data a bunch of different ways—we can do it with establishment births or in narrower levels of industry detail. In general, we see that industries with a lot of applications saw a lot of employer entry. Now, let’s segue into high tech’s role.

There is a long research literature in economics suggesting that business entry, among other sources of dynamism, is historically important for aggregate productivity. America’s last big productivity boom, from the late 90s to the early 2000s, was associated with high tech using and producing sectors that had seen an earlier surge in business entry. So we hope that this pandemic surge in entry is going to lead to stronger U.S. productivity growth going forward. But there might be reasons to be skeptical of that hope.

One possibility is that the business entry surge is a simple restructuring. If we move the restaurants and gyms from the downtown area to the suburbs, the effects on aggregate economy-wide productivity are unlikely to be very substantial. It is great for those entrepreneurs and for the people who want to work from home and have those businesses nearby, but it is probably not going to drive aggregate productivity and G.D.P. much higher. And that kind of entry is part of the story, as the evidence I just showed suggests. But there is an alternative theory: there could have been a burst of innovation.

Let’s investigate the industry composition of the surge in employer establishment entry. I am going to focus on narrower industry detail where we can think about how high-tech intensive a specific narrow industry is and what kind of establishment entry patterns we saw in those industries. This chart shows annual net establishment gains in growth rates. The horizontal axis represents percentage points, roughly speaking. It is normalized to 2019, so 2019 is zero. And you can see this surge relative to the pre-pandemic pace.

I've divided these narrow industries into high-tech or non-tech. In this case, we are looking at how STEM intensive this industry’s employment is. Is this an industry that employs a lot of high-tech type people? If it is, we will call them tech. These high-tech industries saw an enormous increase in net establishment creation during the pandemic. In 2022, the net establishment entry rate was about eight percentage points higher than the pre-pandemic norm. And the pre-pandemic norm for high-tech was something like four percent a year. It is quite striking. Now, if I had shown you this chart without the tech line and only the blue line—the non-tech industries—you would have found that striking too. In fact, we have seen a large increase in establishment formation in non-tech industries as well. But the tech surge is much larger.

So there does seem to be something here. We have pushed on this question and found the result to be pretty robust. I made a list of the top twenty narrow industries in terms of the number of establishments that they gained since the pandemic started. I have also listed the four-digit industry codes. Tech is a relatively small share of the economy, depending how you define it, only about five to fifteen percent, but it is overrepresented among the top twenty industries.

Here are the tech industries that fall within the top twenty industries for pandemic net establishment entry. There are some interesting industries here, including computer systems design and related services. There are many reasons one might need such services, not only to help companies transition to work from home, but also A.I. investment requires an enormous amount of computer system design. There's also technical consulting, software publishers, and research and development services.

There is another important bit of technical detail you need to know: industry codes are set at the establishment level. Walmart is a retailer. But if Walmart opens a data center somewhere, it will register in the data in the appropriate industry category, such as a data center type industry. The scientific research and development services could be any number of services attached to incumbent firms. The establishment births here could be incumbent firms or new firms. But these numbers are striking. Computer systems design added 120,000 establishments since the pandemic started. That is enormous; the total increase in establishments is about 1.7 million, and 120,000 of them are coming from that one little industry. Now, you might say this is a little unfair to the tiny industries. If you are a small but important industry, you cannot add that many establishments and get high in this ranking.

So let’s do it another way and just look at the industry level growth rate in the number of establishments. And we get a similar list here that ends up reordered a bit. Scientific R&D comes in a bit higher; it is the sixth highest industry in terms of its within-industry growth. The number of establishments in the software publishers industry more than doubled during the pandemic. You even see some manufacturing. There is some very important, high-tech activity in the manufacturing sector we see here. The tech story seems legitimate.

But we do not know yet whether existing firms or entrepreneurs are opening these new tech sector establishments. We do not have the data yet, so I cannot answer this exactly, but I can get close. First, at the broad sector level, I can look at establishment births, and I can separate incumbent firm establishment births from entirely new firms. If Walmart opens a data center, for example, that is an incumbent firm establishment birth, and I can separate that from the genuine new firms. I am going to do it in growth terms relative to pre-pandemic. On the horizontal axis, I am looking at the growth in establishment births coming from new firms. And on the vertical axis, I am doing it for incumbent firms. If you are on the 45-degree line, you are a sector that saw the same increase in establishment births from both new firms and incumbents. If you are above the 45-degree line, incumbents played a larger role. And you can see there are some some sectors that are above the line. In green in particular, some are particularly tech intensive—the information sector, professional and business services, and manufacturing. These industries all have relatively high STEM shares. In fact, these are the top three industries, depending what year you measure it. Each of these saw bigger surge in establishment births from incumbent firms than from new firms. Let me emphasize that a couple of these did get a significant increase from new firms as well. If you look at the information sector, it is well along that horizontal axis, as is the professional and business sector. Manufacturing is not. So it seems like both new firms and incumbent firms are participating in this tech establishment entry story, though more on the incumbent side than on the new firm side.

But let’s note one other thing: if we think back to that application data from earlier, I told you how the Census Bureau has internal modeling that predicts actual firm births and that these predictions tend to be accurate using the application data. I have their predicted firm births from the application data here for these three big tech sectors. It looks like early in the pandemic, there were a lot of predicted births in manufacturing, but it has eased off. But the professional, scientific, and technical services prediction is still high and indeed it has picked up recently. And again, this is an important sector. It includes a lot of important high-tech services, such as research and development, computer systems design, etc.

Let’s shift gears and talk about one more pandemic item before turning to longer run trends. You might remember hearing the term “great resignation.” During the pandemic and in the years following the pandemic, we saw an incredible increase in the share of workers quitting their jobs in a given month. This chart’s purple dotted line shows the quit rate—the monthly share of workers who quit their jobs. If you look at the pandemic period, you can see the great resignation in the data. It was very elevated in 2021 and 2022. The green line is the likely employer application series from the beginning of my talk. And the orange line is the establishment birth series. Establishment births and quits move together closely. There are obvious theories for why this might be the case: people quit their jobs and flowed to these new establishments, for example, either as early workers or potentially as firm founders. But it is also the case that there are a lot of macroeconomic time series that behave similarly during the pandemic. It was an extreme event; we should not draw too many conclusions from just looking at these lines.

So what I want to do is go down to fine geographic detail. I cannot do that in the quits data because it is not available at the county level, but I can with a proxy called excess separations. To calculate total separations, we look at all the workers who left their firms for any number of reasons over a given period. We are going to calculate excess separations by subtracting off what economists call job destruction. If you leave your firm, and your firm never refills your job, the firm's total head count decreases. That is job destruction. If that happens, it is reasonable to assume your firm laid you off; the firm’s size decreased. If you leave your firm but someone else immediately replaces you and the firm does not destroy the job, then that often represents a quit. Firing is also possible, but quits are much more common.

Excess separations are used as a proxy for quits historically and in economic research. The separations in which you were immediately replaced move closely with what economists call job-to-job separations. That is the chart’s blue line. If the worker immediately went to another job, that is likely a quit. At the county level, I can use excess separations as a proxy for quits. What I want to see is if this pandemic-era seeming relationship between business entry and quits holds up across geography and is not just a spurious time series aggregation issue.

This scatterplot uses county level data. Business applications per capita—the surge in business applications relative to pre-pandemic—is plotted on the horizontal axis. I can do this for other measures like establishment entry and get a similar result. The vertical axis is the surge in quits using my proxy of excess separations. What we see is that counties with a high number of business applications also had a lot of quits. To some extent, the great resignation happened in the same counties that this business entry surge happened. This lends some support to the theory that people were quitting their jobs and flowing to these businesses.

I can also calculate this for layoffs. One theory explaining the surge in entry is that if the labor market collapses, people have little choice but to turn to self-employment. The pace of layoffs surged in 2020. The labor market was bad for a time. But that does not explain the business entry surge that we also got. So, if I make this same scatterplot using layoffs or job destruction, we see a very, very weak relationship. Depending on how I specify it, the relationship is negative, zero, or slightly positive, but nothing like what we are seeing here. Layoffs were high in 2020 and parts of 2021, but the layoff pace in aggregate has been very subdued for the last couple of years. So the idea that layoffs are causing the surge and entry is not consistent with the fact that layoffs decreased far earlier than the entry surge. It would seem the great resignation is partly related to this business surge in entry. Entry seems to have played some role in this enormous churning of workers that we have seen over the last three years.

Now we will turn to some pre-pandemic trends. Part of what made this surge so remarkable and surprising is that prior to the pandemic, the U.S. had seen a multi-decade decline in business entry rates. The chart shows this in a couple of ways. On the chart’s left is the entry rate—the number of new firms divided by the total number of firms. New firms as a share of all firms has had a long run trend decline. On the right panel, I show business entry rates as share of employment: jobs that new firms create as a share of all the economy’s jobs. That has also been declining. As the chart shows, these are some large declines. In the economics literature, this is called the secular decline in business dynamism. The literature is extensive. I have a paper with my current co-author and others from 2014, in which we documented a lot of these trends.²

Other dynamism measures were declining as well during this time period. Job reallocation—the pace at which jobs are reallocated across the economy—was declining. Businesses shrink and grow and jobs flow between them. The pace of worker reallocation—quits and hires and the rate at which workers are switching jobs—had also been declining. Even rates of internal U.S. migration had been declining.

There had also been a trend of weaker productivity selection. An important element of a market economy is that productive businesses—the ones with the good ideas—need to grow. They are the ones that need the workers and the capital. Unproductive businesses need to shrink or exit. Productivity selection has been an important contributor to aggregate productivity growth in the past, and it had weakened pre-pandemic. It was still there, and the relationship was still positive at the firm or establishment level, but it had been slowing down. The trends also pointed to rising average firm size and concentration. The decline in measures of dynamism could have big implications for aggregate job creation. There's even evidence that some of the productivity growth slowdown we saw after 2004 might have been related to this decline in dynamism. It even matters for business cycles.

The literature on this is large and explaining this decline was occupying an enormous amount of time. We have not nailed its exact cause. The literature proposes various theories; one is demographics. Historically, and even in standard economic models, business entrants absorb labor force growth. Without labor force growth, it is hard for businesses to enter. The U.S. did see slowing labor force growth pre-pandemic. That might be part of the story. There are also stories about the regulatory and business policy environment: a tax on firing is a tax on hiring. There is some evidence that unlawful discharge regulations and the like may have slowed down dynamism.

There are also possible benign causes for the decline in dynamism. We wrote a paper in which we argued that part of the decline was concentrated in the retail trade sector, where we saw the rise of big box retail. It turns out that big box retail is enormously productive, much more productive than the businesses that they replaced. And they pay better wages. But that did not explain all of it.

There are other stories as well. It is popular recently to talk about rising market power and the idea that the U.S. economy is becoming monopolized. Researchers have tried to calculate market power measures and argue that it is related to declining entry. I'm pretty skeptical. Brian Albrecht and I have a recent paper that looks at industries and we do not find that result. But it is a popular story. There are also stories about knowledge investment. Does the increasing importance of knowledge-based capital, as opposed to physical capital, in the productive process create difficulties for new firms and slow down the diffusion of ideas?

So now the question is, did the pandemic surge end the decline? Are we reversing that trend? It is still too early to tell. The blue line in the chart on declining entry rates shows the last couple of years of firm entry rates. Those are pretty striking jumps relative to historic trends, but they do not get us anywhere near to where we were in the 80s or the 90s in terms of firm entry rates. So that is the first issue: this rebound is small compared to the longer run trend. Additionally, some of our papers show that these pandemic-era entrants have been a little smaller than entrants pre-pandemic. That could suggest that these businesses are going to have a lower survival rate or might not have past entrants’ high quality.

The chart does not show it, but I could show you job reallocation. That is another dynamism story. It is similar to entry in that there was a big jump in job reallocation during the pandemic, but even with that jump we are nowhere close to the 90s-era entry rate. More generally, to get a durable reversal of the pre-pandemic trends, we need continued surging entry. Most importantly, we need those recent entrants to grow; they need to innovate, grow, and survive. It is too early to tell what their fates will be, but it is worth keeping in mind.

The other thing to keep in mind are the theories that explain the pre-pandemic decline. Pick a theory you believe is true and then ask, is that still true? If the demographic headwinds are still there, the business and policy environments have not changed, and all these elements are the same, you would not expect the trend to suddenly reverse and bring us back to where we were decades ago. So I think we will still have to think about these longer run stories.

What happened in the pandemic was a remarkable, unexpected story. At the very beginning of the pandemic, applications dipped for a couple of months. Many people predicted that the pandemic would cause a complete collapse of business entry and that was going to have knock on effects on labor markets for years to come. There was a lot of concern during those first couple months. And then we have this enormous reversal. It is quite a story.

But entrepreneurs did what entrepreneurs do; they saw opportunities and they went after them. There were massive changes to consumption, work, lifestyle, and business patterns. Many businesses needed to build out their IT infrastructure, so that they could facilitate work from home or online retail. These were opportunities and entrepreneurs did what they do, which is go after these opportunities. It is an incredible story and a testament to American dynamism. Despite these longer run trends, American dynamism is really something to behold. And the best we can tell in cross-country comparisons, other countries have not seen this. Some countries have seen some growth in entry, but none have seen what we have seen here. It is a remarkable story. Counting out the American entrepreneur was a mistake.

The entry surge seems to have facilitated or followed the broader pandemic economic restructuring across geography and industry. It might be an important ingredient of the remarkable post-pandemic labor market recovery and some of these other reallocations that had to happen. We do have evidence that high tech industries saw a large and disproportionate entry surge. Some of that might be early pandemic stories about online retail and work from home. But it increasingly looks like it could be related to other stories around big IT investments for things like A.I. And then it seems possible that many quitters flowed to the new businesses. So, we have seen some shifts in the industry and geographic pattern of economic activity, and we have a slightly younger firm age distribution. That is the first time in a long time that has happened, a little bit more activity at small firms, and a pause in pre-pandemic trends.

Readers are asked to make allowances for what was originally an oral presentation delivered on May 8, 2024 at The Austin Symposium.

Ryan Decker is the Chief of the Industrial Output Section at the Federal Reserve Board.

---

1 | See Decker, Ryan and John Haltiwanger (2024), “Surging business formation in the pandemic: Causes and consequences?” Brookings Papers on Economic Activity, Fall 2023; Decker, Ryan and John Haltiwanger (2024), “High tech business entry in the pandemic era,” FEDS Notes, 29 April; and Decker, Ryan and John Haltiwanger (2024), “Surging business formation in the pandemic: A brief update,” working paper.

2 | Decker, Ryan, John Haltiwanger, Ron Jarmin, andJavier Miranda (2014), “The role of entrepreneurship in U.S. job creation and economic dynamism,” Journal of Economic Perspectives 28:3-24.