The Atlanta Fed's macroblog provides commentary and analysis on economic topics including monetary policy, macroeconomic developments, inflation, labor economics, and financial issues.

Authors for macroblog are Dave Altig, John Robertson, and other Atlanta Fed economists and researchers.

« January 2014 | Main | March 2014 »

February 26, 2014

The Pattern of Job Creation and Destruction by Firm Age and Size

A recent Wall Street Journal blog post caught our attention. In particular, the following claim:

It’s not size that matters—at least when it comes to job creation. The age of the company is a bigger factor.

This observation is something we have also been thinking a lot about over the past few years (see for example, here, here, and here).

The following chart shows the average job-creation rate of expanding firms and the average job-destruction rates of shrinking firms from 1987 to 2011, broken out by various age and size categories:

In the chart, the colors represent age categories, and the sizes of the dot represent size categories. So, for example, the biggest blue dot in the far northeast quadrant shows the average rate of job creation and destruction for firms that are very young and very large. The tiny blue dot in the far east region of the chart represents the average rate of job creation and destruction for firms that are very young and very small. If an age-size dot is above the 45-degree line, then average net job creation of that firm size-age combination is positive—that is, more jobs are created than destroyed at those firms. (Note that the chart excludes firms less than one year old because, by definition in the data, they can have only job creation.)

The chart shows two things. First, the rate of job creation and destruction tends to decline with firm age. Younger firms of all sizes tend to have higher job-creation (and job-destruction) rates than their older counterparts. That is, the blue dots tend to lie above the green dots, and the green dots tend to be above the orange dots.

The second feature is that the rate of job creation at larger firms of all ages tends to exceed the rate of job destruction, whereas small firms tend to destroy more jobs than they create, on net. That is, the larger dots tend to lie above the 45-degree line, but the smaller dots are below the 45-degree line.

As pointed out in the WSJ blog post and by others (see, for example, work by the Kauffman Foundation here and here), once you control for firm size, firm age is the more important factor when measuring the rate of job creation. However, young firms are more dynamic in general, with rapid net growth balanced against a very high failure rate. (See this paper by John Haltiwanger for more on this up-or-out dynamic.) Apart from new firms, it seems that the combination of youth (between one and ten years old) and size (more than 250 employees) has tended to yield the highest rate of net job creation.

John RobertsonBy John Robertson, a vice president and senior economist in the Atlanta Fed’s research department, and

Ellyn TerryEllyn Terry, a senior economic analyst in the Atlanta Fed's research department

February 26, 2014 in Economic conditions, Employment, Labor Markets, Small Business | Permalink


TrackBack URL for this entry:

Listed below are links to blogs that reference The Pattern of Job Creation and Destruction by Firm Age and Size:


So young high growth firms that succeed create the most jobs. WOW!

Large successful old firms are stable and neither create nor destroy large amounts of new jobs. The analytical insight is unbelievable!

and to top if off small, risky start-ups destroy the most jobs as they constantly fail. OMG, Nobel Price of Economics right there!

Americans please send even more of your tax dollars to the Atlanta Federal Reserve given the amazing level of research and analytical insights they are capable of.

Posted by: Alex | February 27, 2014 at 01:46 AM

Are the highlighted features of the chart -- that "rates of job creation and destruction tend to decline with firm age" and that "the larger dots tend to lie above the 45-degree line, but the smaller dots are below the 45-degree line" -- stable over the time period examined, or do they come and go from year to year? And if the latter, can that change be correlated in a useful way with events in the economy as a whole (eg the "dot bomb" of 2001, the financial crisis of 2007-8, or the Great Recession to which that crisis gave rise)?

Posted by: Ed Blachman | February 27, 2014 at 09:29 AM

I keep thinking about the pattern and I wonder what it would look like if it was in motion through time. I wonder if it would follow patterns in nature. I also wonder if the large industries 4 years or less move around much since it is so unusual for a company to hire that many employees in such a short time because of outliers. I also wonder the implications for future employment levels as there are fewer entrepreneurs. Thank you so much for posting this. Great article. I've read other stuff you've done that you linked out to. Great job guys. You've found a lot of the critical data points that really matter. I've been looking to find something like this. I will bookmark this page.

Posted by: buttmunch1 | February 28, 2014 at 12:17 AM

One question: Would the chart look much different if you stopped at 2006? The idea is that large numbers of small firm jobs were likely destroyed by the financial crisis. Further, after the crisis small firms have lacked access to start-up funds, thus diminishing gross new firm creation (and its attendant jobs). One can think of the GFC as a seminal event in small firm birth/death dynamics, much more so than for large firms. While it may look from the data that small firms don't have much impact on net job creation, this may be because of the lack of small firm "births" in the past five years.

Posted by: Diego Espinosa | March 07, 2014 at 12:17 PM

The chart is better art than economics.

The obvious correlation between firm age and firm size means that it is impossible to separate the effects of the two factors (on job creation) by assigning colors and sizes to firms and graphing them against each other. The "cure" for multicollinearity is not changing the color or size of data points -- but recognizing that both change simultaneously.

A successful startup firm in 1987 moved along a path from then to 2011 that took them from tiny blue dots in the direction of giant orange balls. What does that PATH suggest about job creation and destruction? We can be certain that companies traveling the same path in the opposite direction have a far higher rates of job destruction to creation. Shall we paint both of them green, and assign both medium-size dots?

Finally, I was confused by the artist's practice of measuring dependent variables along both principal axes, then graphing observations for independent variables as points in the x-y plane. This is perfectly fine if the sole purpose is to describe the data in a compact way. But if one's purpose is to guide the reader's mind toward cause-effect relationships, this is a poor practice.

Posted by: Thomas Wyrick | March 08, 2014 at 10:05 AM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

February 21, 2014

What Is the Stance of Monetary Policy?

Will the Federal Open Market Committee's (FOMC) current large-scale asset purchase program, so-called QE3, continue to melt away as spring arrives? The release of the minutes from the January meeting of the FOMC, along with commentary from various participants in that meeting (noted in rapid succession here, here, and here, for example) have left the distinct impression that the answer is most probably yes.

The anticipated winding down of asset purchases almost inevitably invokes a habit of language concerning what it all means for the stance of monetary policy. From the New York Times, for example, we have this (emphasis added):

When Federal Reserve officials last met at the end of January, they were surprised by the strength of the economy, cheered by the optimism of consumers and convinced they should continue to dismantle the Fed's economic stimulus campaign, according to an account the Fed released Wednesday.

The sentiment expressed in that highlighted passage is front and center at the G-20 meetings, currently taking place in Australia (again, emphasis added):

Setting the scene for this weekend's Group of 20 meetings, Australian Treasurer Joe Hockey's main challenge was to avoid appearing partial in the escalating blame-game between the U.S. and developing countries over the recent exodus of capital from emerging markets….

Emerging market countries like India and Brazil have blamed the wide-scale selloff in local stocks, bonds and currencies on the Federal Reserve's plan to exit gradually from monetary-stimulus policies, which last year began sending investors into a panic.

Here's the thing. It is not at all clear that winding down asset purchases means an exit from or dismantling of monetary stimulus, gradual or otherwise. In Atlanta Fed President Dennis Lockhart's words yesterday:

In our public remarks over much of last year, my colleagues and I stressed a couple of very important messages. First, even with the phase-out of asset purchases, the basic stance of policy remains highly accommodative. To translate, the Committee intends to keep interest rates very low. The second message was that the QE program and the Fed's policy interest-rate target are two separate tools of policy. Consequently, we can wind down the asset purchases—a program that was meant to provide temporary, supplemental "oomph" to the low interest-rate policy—and preserve the accommodative positioning of policy appropriate for the reality of our economic situation.

But those are not just words. Several months back, Jim Hamilton publicized the work of Cynthia Wu and Dora Xia (former and current students of his), who have developed a method of using term structure data to infer the "shadow," or implicit, monetary policy rate. (Follow-up posts appeared thereafter at Econbrowser—here and here—and from the crew here at macroblog.)

Just recently, the Wu-Xia data has been updated, giving us a first glance at the post-taper shadow policy rate (see the chart):

Both Treasury yields and the shadow policy rate did in fact spike last June following then-Chairman Ben Bernanke's post-FOMC press conference, wherein he signaled that the asset-purchase taper was indeed on the table. But he also made the point that bringing down the QE pillar of the Fed's policy mix is decidedly not the same thing as bringing monetary stimulus to an end, a message that was subsequently emphasized by Fed officials many times, in many forums.

Though financial market participants may have been convinced that the taper meant tightening initially, it does appear that communications and forward guidance have done the trick of more than reversing that initial impression. At the very least, the Wu-Xia calculations are consistent with that interpretation.

David Altig By Dave Altig, executive vice president and research director at the Atlanta Fed

February 21, 2014 | Permalink


TrackBack URL for this entry:

Listed below are links to blogs that reference What Is the Stance of Monetary Policy?:


Could you explain the economic meaning of these shadow rates? I do not understand the concept.



Posted by: François | February 22, 2014 at 08:23 AM

So how is one to interpret the Wu-Xia policy rate chart... what does it mean that since June there has been apparently what appears to be a 150 bps "decrease" in the rate?

Hamilton proposes that it’s going to be helpful to have an alternative to the fed funds rate to summarize the stance of monetary policy.

If the point is... that since June 2013 the market has re evaluated Fed policy... surely the simplest thing to look at is the January 2015 FED Funds futures... or any of the OIS curve... those have rallied more than 50 bp's since June of 2013...

Posted by: stan jonas | February 24, 2014 at 04:55 PM

The Fed's stance is restrictive. Money growth has declined from 10% in 2012 to 6% today. Inflation is far below target, and nominal growth is 2-3% below where it was pre-crash. QE has had no observable impact on money growth.

Posted by: Christopher Mahoney | March 28, 2014 at 05:41 PM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

February 11, 2014

A Second Look at the Employment-to-Population Ratio

This analysis is a companion piece to my Atlanta Fed colleague John Robertson's recent macroblog post. John's blog highlighted some findings of a recent New York Fed study by Samuel Kapon and Joseph Tracy on the employment-to-population (E/P) ratio. Their work has received considerable attention in the media and blogosphere (for example, here, here, and here). Kapon and Tracy's final chart (reproduced below) has received particular scrutiny.

The blue line represents the authors' estimate of the demographically adjusted E/P ratio purged of business-cycle effects. This line can be thought of as "trend." The chart shows that as of November 2013, the E/P ratio was only –0.7 percentage point below trend. Was the "gap" between actual and trend E/P really this small?

Attempting to answer this question requires digging into the details of Kapon and Tracy's method for estimating trend. One key excerpt is the following:

To overlay our demographically adjusted E/P ratio with the actual E/P ratio, we need to adopt a normalization… [W]e adopt the normalization that over the thirty-one years in our data sample [1982–2013] any business-cycle deviations between the actual and the adjusted E/P ratios will average to zero.

This methodology seems reasonable since one might typically expect business cycle effects to average out over 30 years. However, the 1982–2013 sample period is somewhat unusual in that the unemployment rate was elevated at both the starting and ending points.

The chart below shows estimates of three labor market gaps derived from the Congressional Budget Office's (CBO) estimates—released on February 4, 2014—of the potential labor force and the long-term natural rate of unemployment. (This rate is often referred to as the nonaccelerating inflation rate of unemployment, or NAIRU, and refers to the level of unemployment below which inflation rises.)

On average, the trend E/P ratio is below the actual rate by 0.86 percentage point. If one were to normalize the Kapon and Tracy E/P trend so that its average value was equal to CBO's trend, then the November 2013 E/P gap is about 1.5 percentage points. Whether or not the CBO estimate is the right benchmark is a matter of taste. CBO's recent estimate of NAIRU in the fourth quarter of 2013—5.5 percent—is lower than the 6 percent median estimate from the Survey of Professional Forecasters in the third quarter of 2013.

A second, more subtle issue in the Kapon and Tracy analysis is their treatment of cohorts:

We divide these individuals into 280 different cohorts defined by each individual's decade of birth, sex, race/ethnicity, and educational attainment. We assume that individuals within a specific cohort have similar career employment rate profiles. We use the 10.2 million observations [of CPS microdata] to estimate these 280 career employment rate profiles.

A well-known 2006 Brookings paper by Stephanie Aaronson and other Fed economists modeled trend labor force participation rate(LFPR) using birth-year cohorts. With estimates of trend LFPR and NAIRU, we can back out a trend E/P ratio. The chart below, adapted from Aaronson et al., plots age-group LFPRs against birth year.

We see that successive birth-year cohorts born between 1925 and 1950 had steadily increasing labor force attachment. Attachment for more recently born cohorts has leveled off and even declined slightly. People born in the 1990s have very low labor force attachment by historical standards. The inclusion of the "1990s—decade of birth" dummy variable in the Kapon and Tracy research probably implies that their model is interpreting much of this decline as structural. However, an alternative interpretation is that the decline is cyclical, because persons born after 1990 have been in an environment of high unemployment for most of their short working lives.

To gauge the sensitivity of trend or structural LFPR to how the youngest cohorts are treated, I used a stripped-down version of a model similar to Aaronson et al. Monthly LFPRs are modeled as a function of age, sex, birth date, and the CBO's estimate of the output gap during the January 1981 to January 2014 period. Time series published by the U.S. Bureau of Labor Statistics for 30 different age-sex cells are used so that the regression has 11,550 observations. Structural LFPR is constructed with the fitted values of the regression with a value of 0 percent for the output gap at all points in time. The trend E/P ratio is then backed out with the CBO's estimate of NAIRU.

The model is run with two different assumptions: First, following the approach of Aaronson et al., people born after 1986 have the same birth-year cohort effects as those born in December 1986. Second, no constraints are placed on birth-year cohort effects. Trend values of LFPR and E/P (taking on board the CBO's NAIRU) are plotted in the two charts below:

The January 2014 E/P gap with unconstrained cohort effects, as in Kapon and Tracy, is –1.0 percent, well below the –1.7 percent gap in the model with constrained cohort effects. Ultimately, both models are still very consistent with Kapon and Tracy's bottom line:

It is important to control for changing demographic factors when looking at the behavior of the E/P ratio over time. This step is particularly important today when these demographic factors are exerting downward pressure on the actual E/P rate, suggesting that the recent lack of improvement in the E/P ratio does not imply a lack of progress in the labor market. The adjusted E/P rate corroborates the basic picture from the unemployment rate that the labor market has been recovering over the past few years, but that it still has a ways to go to reach a full recovery.

Photo of Pat HigginsBy Pat Higgins, senior economist in the Atlanta Fed's research department



February 11, 2014 | Permalink


TrackBack URL for this entry:

Listed below are links to blogs that reference A Second Look at the Employment-to-Population Ratio:


In 2000 the employment population ratio was 64.5%. This was the last good economy in the Clinton administration when unemployment was 4% without inflation. It is now at 58.5%. This suggests a much more flexible participation rate. If there are jobs the people will seek them.

It also suggests that the lack of education and job skills, sudden onset of indolence by workers, technological revolution(all those robots), and such arguments are as specious as the propagators making them.

Posted by: Account Deleted | February 12, 2014 at 11:22 AM

Excellent analysis.

But to D-41's comment praising "a much more flexible participation rate" perhaps a second thought should be given to

'sudden onset of indolence by workers, and such arguments are as specious as the propagators making them'

How does this correspond to

1) Really high SS-Disability Rates
2) Really long unemployment benefits periods


both of which are largely determined by political forces promoting 1 and 2 above, and not 'If there are jobs the people will seek them'.

Posted by: John Powers | February 23, 2014 at 08:26 AM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

February 06, 2014

A Prime-Aged Look at the Employment-to-Population Ratio

Trying to interpret changes in labor utilization measures such as the employment-to-population ratio is complicated by the fact that they do not refer to the same set of people over time. The age composition of the population is changing, and behavior can vary across and within age cohorts.

This issue is illustrated in a recent New York Fed study of the employment-to-population ratio by Samuel Kapon and Joseph Tracy. This ratio nosedived during the recent recession by about 4 percentage points and has barely budged since.

This measure of labor utilization is the clear laggard on any labor market recovery dashboard. But the authors show that it is not so clear that the employment-to-population ratio is really so far from where it should be, once you control for the fact the employment rates tend to be lower for younger and older people and that the age composition within the population has shifted over time. This idea is similar to the one used to estimate the trend labor force participation rate in this Chicago Fed study by Daniel Aaronson, Jonathan Davis, and Luojia Hu. The issue of controlling for dominant demographic trends is one of the reasons we at the Atlanta Fed decided not to feature either the overall employment-to-population ratio or the overall labor force participation rate in our Labor Market Spider Chart.

A simple, and admittedly crude, alternative to computing the demographically adjusted employment-to-population ratio trend is to look at a segment of the population that is on a relatively flat part of the employment (or participation) rate curve. A common standard for this is the so-called prime-aged population (people aged 25 to 54). These individuals are less likely to be making retirement decisions than older individuals and are less likely to be making schooling decisions than younger people. Of course, this approach doesn't control for within-cohort factors like educational differences.

So what do we find? The prime-aged employment-to-population ratio declined almost 5 percentage points between the end of 2007 and 2009 (versus 4 percentage points overall) and since then has recovered about 25 percent of that decline. Using the end of 2007 as reference, the Kapon and Tracy trend estimate has declined about 1.7 percentage points, which implies the overall employment-to-population ratio, by not continuing to decline, has improved by about 40 percent.

Then what does the analysis say about labor utilization in the wake of the recession? Once demographic factors are controlled for, both aforementioned measures indicate that labor-resource utilization has improved relative to trend. In fact, as Kapon and Tracy note, the relative improvement would be even greater if you believed that employment was above trend before the recession.

Photo of John RobertsonBy John Robertson, a vice president and senior economist in the Atlanta Fed's research department

February 6, 2014 in Employment, Labor Markets, Unemployment | Permalink


TrackBack URL for this entry:

Listed below are links to blogs that reference A Prime-Aged Look at the Employment-to-Population Ratio:


Prime-age recovered about 25 percent of that decline...so what is the overall recovery (if any) if you include under 25 and over 54?

I'm trying determine why some are saying the decline (since 2000) was because of Boomers retiring (even though the first didn't retire until 2008 at age 62).

Many over-50ish workers who lost jobs since 2007 were never rehired again, but were too young to take a pension or early Social Security retirement. Most likely they are among the millions of so-called "discouraged workers".

So if demographics were being used, it should not be said the decline was because "people were retiring", but instead should be noted that employers consider them to be obsolete widgets, and don't want them any longer.

Posted by: Bud Meyers | February 07, 2014 at 11:21 AM

The labor force increase has barely kept pace with new entrants even when you all try to show it in its best light. The country needs to face up to permanent loss of jobs due to the rapid pace of technology. And the more we dumb down the educational system with hair brained Washington schemes like common core we weaker employment figures will become.

Posted by: august mezzetta | February 07, 2014 at 05:40 PM

I have determined that young "non-starters" into the labor force (high school and college graduates) and discouraged workers (who are mostly prime-age workers) make up most of the recent decline in the labor force.

Birth rates are currently historically low, so those who are already within the population are graduating from high school at a faster pace than births, and more so than those who are retiring (now at record highs) or those going of disability (which recently have actually declined).

Note: Although disability "claims" have greatly risen since the last recession, actual "awards" for year-to-year net increases are tiny compared to retirees and high school graduates.

At this time, we can not compare job growth to population growth, as it's not relative to maintaining the labor force participation rate or the employment-to-population ratio.

My post with links to data here:


Another post as a follow-up: "Prime Age Workers: Bulk of Discouraged Workers"


And this: "22% of all U.S. Households had no Earners" ---- Maybe someone with a higher pay-grade can reconcile those numbers ;)


And from another post: "8 Million Jobs Short, 6 Million Missing Workers" to show the numbers from the Economic Policy Institute—but they appear to be far too conservative. (I'd say at least 20 million jobs short and 20 million missing workers.)


* Note: There were a couple of minor discrepancies in the SSA data for year-to-year numbers in gains for disability (from two different links at SSA), and one explanation might be that one is the number of "awards" not yet in payment status. After an award is first granted, there is usually a waiting period before a payment is made. But the discrepancy is only minor.

Posted by: Bud Meyers | February 13, 2014 at 11:26 PM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

Google Search

Recent Posts



Powered by TypePad