Last week, the Kauffman Foundation released a research brief with the Census Bureau asking, "where have all the young firms gone?" The pace of new firm formation, job creation by startups (age 0 firms), and the share of young firms in the economy have been on a secular downward trend for many years. During the Great Recession, new firm formation fell off a cliff.
On the face of it, according to these data, the United States "is becoming less entrepreneurial." But maybe there is more to the story.
First, let's just take the basic math. The startup rate is calculated by taking the number of brand new companies (age 0 firms or "startups") and dividing it into the overall number of firms in the economy. In the 1980s, this hovered between 12 and 13 percent; in the 1990s, around 11 percent. In the late 1990s it fell below 10 percent and, from 2000 to 2006, it rebounded slightly, to between 10 and 11 percent. The last few years have seen only single digits.
Part of this is basic arithmetic. The absolute number of new firms has stayed more or less between 450,000 and 500,000 per year since 1977 (although this fell to 394,000 in 2010). Because older firms (of any age) don't exit the economy at the same rate, in most years the United States enjoys net positive firm entry: more firms are added to the overall number than are subtracted. Thus, the denominator grows. If the numerator stays relatively constant while the denominator grows more or less steadily, that quotient should shrink, which is precisely what we see with the startup rate. So the finding that the startup rate is falling proves that math works, which is good news.
Beneath that net entry is the survival rate, which moves more slowly than the entry rate--quite naturally. The half-life of any given cohort of new firms is five years: it takes five years for roughly 50 percent of any year's cohort to fail to survive. (It is important to note here, and will be discussed further below, that "survival" here is a value-laden term that is mismatched with the dataset. What we take as survival or "failure" is really just an exit from the dataset. We don't know what that exit is--failure, closure, acquisition, whatever--just that a specific company no longer exists in this particular dataset.) We also have little insight, in the aggregate, into the reasons for "failure": Noam Wasserman points out that, in some data, after 48 months of a company's existence, half of the founders have left their company. Wasserman's research corroborates that of others showing that companies fail due to "people problems" more often than business problems.
The aggregate but rough cohort survival rate is basically the same for most of the new firm cohorts since 1977, although it has fallen slightly in recent years. After five years, the survival line moves downward more slowly: each age to which a company survives increases the probability that it will survive to the next age. As we said in our Neutralism paper: "a slowing slope of failure acts upon a shrinking pool of survivors." The resulting dynamics are such that the overall population of (employer) firms in the United States has grown rather steadily.
Yet, why is this automatically a bad thing? Doesn't it mean that something is going right in the economy if, as the result of competition, more companies than not are surviving? Is the implicit counterpoint that we should wish for a shrinking population of firms overall? In an economy at the very frontier of economic complexity, isn't one natural--and good--development that we have more firms to fill all those nodes on the network? I'm asking this as a genuine question, not a rhetorical one: is this a good or bad thing? In this sense, the falling startup rate is not entirely a bad thing.
It's true that, when the number of new startups is plotted as a per capita rate (rather than against the population of companies), we also see a steady downward trend. But this isn't exactly the same thing and shouldn't be taken as corollary evidence for the entrepreneurship apocalypse. The growth, documented by Census data, in the annual number of new firms in the 1960s and 1970s occurred at a very different time demographically. A country has different sectoral and business needs when it is approaching the demographic sweet spot and when it is in the latterly phases of it.
Let's move beyond the statistics. One puzzle that we are persisently presented with is why the numbers in the BDS data appear so different from the Census data (from the Current Population Survey) that comprise the Kauffman Index of Entrepreneurial Activity. In the former, roughly 500,000 new firms have entered the economy each year; in the latter, approximately 550,000 new businesses are said to start each month. How can this be? For one thing, the KIEA/CPS data include the self-employed and non-employer companies, while the BDS data only count employer firms. The stock of non-employer firms in the United States is roughly four times the size of the population of employer firms, so we might expect that gross inflows are also much larger.
Second, the annual-monthly disparity highlights something important. Many people, when looking at charts showing more or less constant annual levels of firm entry, have wondered whether there is some demographic cap on the number of people who can, or are willing to, start new businesses each year. Maybe ~500,000 is just the "natural" ceiling. But these are employer firms--those with employees from the get go. We know from prior work that a "significant fraction" of supposedly new employer firms in each year's cohort emerge from the ranks of non-employer firms. So they've already been in business to some extent prior to entering the BDS dataset. That ~500,000 number is already the result of a winnowing process and reflects those people or companies that made it from idea to garage to non-employer to employer firm (in gross, simplified, metaphorical terms). If CPS data tell us that 550,000 people report spending significant time on a "new" business each month (though there are likely repeat appearances in there), clearly a large number of people either give up, fail rather quickly, or just remain in the self-employed population instead of moving to the employer firm category.
The point is that the number reflected in the BDS dataset, and counting as a falling startup rate and thus the mark of a "less entrepreneurial" country, says nothing about the overall population of potential entrepreneurs or entrepreneurial ferment. It only reflects the winnowing process between idea and transition to employer firm. The former vice president of entrepreneurship at Kauffman, Bo Fishback, now founder of Zaarly, at one point drew a large X-Y graph on his whiteboard and marked the 0,0 point as t=0, to demarcate the time at which an employer firm was officially founded. Before that, to the left of the Y axis, was an amorphous population of ideas, halting attempts at firms, self-employed, home-based businesses testing out a venture, and so on. There is a huge amount of activity out there that, for whatever reason, doesn't make it to point 0,0; the activity at the Inspiration and Discovery and Action phases is vitally important. We need to look at the structural context of firm formation--sector by sector, region by region--to understand why.
But let's take all the numbers at face value: so entrepreneurship, as measured by new firm entry, is declining in the United States. What might this say about economic activity? Rob Fairlie has found that, during the late 1990s in Silicon Valley, the entrepreneurship rate in the high-tech sector actually fell compared to the rest of the United States. After the 2000-01 bust, it was higher than in the rest of the country, completely contrary to the popular narrative about the dotcom bubble and bust. Fairlie's explanation is that, at a time of strong high-tech growth in Silicon Valley, the opportunity cost of starting a new business was simply too high because there were so many good opportunities at companies such as Cisco, Oracle, and (at the time) Yahoo!.
Something like this clearly lurks beneath the BDS data in any given year. It highlights, moreover, the dangers of dealing exclusively with aggregate data--they can give you a quick and easy snapshot of firm formation, but they gloss over all the variation underneath, which is where the real action is. In particular, differential sectoral trends in firm formation probably explain a great deal about the trends we think we perceive in firm formation and job creation. (For instance, the housing boom in the early to mid-2000s likely accounted for at least part of the rebound in firm formation during that time.) And, recently, the apparent startup boom in the tech sector is evidently being swamped by low firm formation rates in other sectors. (Or, to follow Fairlie's solid line of logic, some would-be entrepreneurs are choosing to join startups rather than start their own, a career choice that is recommended by some knowledgeable folks.)
Recall the difference above between a given company's survival and a statistical exit from the dataset. We have no idea what share of exits were actually failures or closures (but not considered failures) or acquisitions. In all likelihood, failures (or, at least, non-survival) constituted the largest share of them, and acquisitions a (much) smaller share. We know that, when it comes to VC-backed exits in the last decade, a far larger share (roughly 80 percent per year) have been acquisitions rather than IPOs, an almost complete statistical reversal from the 1990s. As Jay Ritter and colleagues have noted, this is due less to regulatory changes than to changing market incentives. And, as either cause or effect or some combination of both, Clay Christensen, progenitor of the "disruptive innovation" concept, has apparently found that big, established companies have accounted for a growing share of disruptive innovations in the past decade, compared to the previous century. The methodological details are unclear, but this at least falls in line with some other research findings.
So, one can construct an economic narrative that sees the opportunity cost calculus shifting such that opportunities at existing companies (of whatever age, in whatever sector) are more attractive, for a larger share of individuals, than starting a new company. Or, than pursuing that idea far enough to take it to the employer firm stage. (I am talking here about the long-term decline in startup rates, not the recession-driven decline.) In some ways, this reflects greater efficiency and lower transaction costs in overall economic activity and, at least by one measure, perhaps a growing share of disruptive innovation in larger and older companies. So maybe a falling startup rate, if it reflects this narrative to any degree, is, again, not entirely a bad thing.
I have no idea if this is right. But we should try to dig beneath the aggregate data, rather than treating them as the complete picture, otherwise we are painting not with broad brush strokes but with buckets of paint. The pace of new firm formation in an economy reflects a huge variety of factors and developments, and a fall in that pace should be treated as such, not as a monolithic disease plaguing the economy.
You might want to take a look at reasons for "exiting" from a startup, according to a number of sources. First, John C. Maxwell in his book "Failing Forward" says that most entrepreneurs fail around 4-5 times before they're successful. This is anecdotally reflected in the saying, "You know what you call a failed Silicon Valley entrepreneur? Experienced. Then there is merging of solopreneurs into partnerships that combine businesses. Lastly, take a look at the book "Getting to Plan B" describing how companies pivot with changes that occur.
This has been all briefly discussed in my book, "How to Start a Business: Mac Version" as I began to find info about startups.
Posted by: Kevin Cullis | May 07, 2012 at 08:48 PM
Thanks, Kevin. Great suggestions. These also seem to underscore the point that what we see as a dataset "exit" could reflect numerous possible scenarios. And you're right: within those aggregate numbers is lots of repetition in terms of multiple foundings, exits, and more foundings. It's exactly that variation on which we need to get a better grasp. Thank you!
Posted by: Dane Stangler | May 08, 2012 at 05:47 AM