The Silver Lining to the Drop in Startups

Photograph by Gregory Kramer/Getty Images

A long-term trend toward fewer business failures. That’s what you could have titled the recently released Kauffman Foundation-Census Bureau study (PDF) on U.S. entrepreneurship, entitled “Where Have All the Young Firms Gone?” While the report focuses on the decline in business creation, implicit in the data it discusses is a long-term decline in the failure rate of businesses.

The study discusses the dramatic drop in startup activity in the United States over the past three decades: U.S. Census data reveal that the share of new businesses shrank, from 13 percent of U.S. employers in 1980 to 8 percent in 2010 (when measured as the number of new ones formed as a percentage of all those in operation in the prior year). While the press release that accompanies the report implies that the decline is problematic because new businesses have historically been an important source of job creation, the interpretation need not be so dire.

Created from Census Bureau data
The data also suggest that existing business owners have gotten better at running their companies, thereby reducing the business failure rate and destroying fewer jobs via shut-downs. While the number of new employers dropped by half between 1977 and 2010, the number of employers in operation grew 5 percent when the two are measured on a per capita basis. The only way to explain this: Failure rates must’ve dropped, too.

While the decline in the formation rate has accelerated during the Great Recession and that of the failure rate hasn’t, both display long-term downward trends from 1978 to 2010. This pattern means the U.S. economy has become less dynamic, with fewer new companies replacing old ones.

That might not be a bad thing. Most new businesses replace existing small businesses and serve the same customers with similar products, researchers have shown.

Perhaps fewer people are starting new companies to meet old needs by replacing existing companies because the guys already doing it have learned a thing or two. With existing business owners are doing a better job, their companies have become less likely to fail, so fewer people are starting businesses to replace them.

What about the report’s observation that new companies are a less-important source of employment than they used to be? It’s true that the number of startup jobs has declined over the past three decades: In 2010, new businesses employed 2.1 percent of private-sector workers, vs. 5.9 percent in 1977.

But the Census data show that this decline in employment by new businesses has occurred without any reduction in the ability of the private sector to provide jobs. From 1977 to 2010, American industry increased its employment, from 30 percent to 36 percent of the population. Therefore, the decline in hiring by new businesses was countered with a rise in hiring by existing businesses—and then some.

That’s because the decline in new-business job creation has been offset by a decline in existing-business job destruction, particularly job destruction from business failure. While the job-creation rate of new ventures dropped, from 9.2 percent of employment in 1977 to 4.3 percent in 2010, the job destruction rate also fell, from 15.2 percent in 1977 to 14 percent in 2010.  Moreover, job destruction from businesses that shut down declined, from 6.1 percent of employment in 1977 to 4.1 percent of employment in 2010.

In addition, existing businesses have gotten bigger over the past three decades. The Census data show that their average size has grown, from 19.3 employees in 1977 to 22.2 in 2010. That means we need fewer businesses to provide the same number of jobs.

We should pay attention to the decline in business-formation rates, but we need to think about the different reasons why this might have occurred. Where the pessimist sees a loss of job-creating startups, the optimist sees better management of existing businesses. Both are plausible hypotheses. Before policy makers try to counteract the fall in startup rates, they first need to figure out which one is most consistent with the data.