Economic Dynamism and Productivity Growth

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

There’s a paper out in the latest Journal of Economic Perspectives by Decker, Haltiwanger, Jarmin, and Miranda (DHJM) on “The Role of Entrepreneurship in US Job Creation and Economic Dynamism“. They document, in more detail than an earlier Brookings report I talked about recently, that the proportion of firms that are “young” has declined over the last 30 years.

DHJM use the number of young firms – those less than 5 years old – as a proxy for entrepreneurship. And therefore the conclusion is that entrepreneurship has declined over the last 30 years. You can see this in their figure 4, below. In 1982, for example, roughly 50 percent of all firms were less than 5 years old, while by 2011 only about 35 percent of firms were under 5 years old. Similarly, the share of total employment in young firms fell from about 18 percent in 1982 to about 13 percent in 2011.

DHJM 2014 Figure 4

Perhaps most important, the share of job creation from young firms has declined over the same period. In the early 1980’s, young firms were responsible for about 40 percent of all new jobs, while by 2011 this was down to about 33 percent. In sum, there are fewer young firms, they employ fewer people, and they create fewer jobs today than they did 30 years ago.

Where is this decline coming from? DHJM show in their figure 5 that for manufacturing, the share of employment in young firms has declined very slightly over the same period, and was never very large to begin with. In contrast, in the service sector the proportion of jobs in young firms was over 25 percent in 1982, and now is around 15 percent. The shift of economic activity from manufacturing to service firms both raised the share of employment in young firms (because of the higher rate in services) and lowered the share of employment in young firms (because of the downward trend within services). On net, the downward trent in services won out, and overall the proportion of jobs in young firms has dropped.

DHJM 2014 Figure 5

There’s nothing to dispute in these numbers, and I don’t think DHJM have done anything to misrepresent what is going on. But the big question is: did this decline in the proportion of young firms lower productivity growth? The short answer is, I don’t see any evidence that it did. [Update 8/1/14: Just to be clear, DHJM are not claiming that it does lower productivity growth. This is a question I have given their data.]

Consider the figure from Fernald’s (2014) recent paper on productivity. It shows the trend of labor productivity from the late 70’s until today. There is no secular slowdown in productivity growth between 1982 and 2011. Productivity growth from 2003-2011 is just as fast as it was in the pre-1995 period. As Fernald points out, 1995-2003 is an outlier, probably associated with the IT revolution. Therefore, if the decline in the number of young firms is bad for productivity, it hasn’t been so bad that it shows up in any aggregate numbers over the last 30 years.

Fernald 2014 Figure 3

So what does the decline in young firms mean? One plausible explanation is mentioned by DHJM, which is the advance of “big box” or national stores relative to mom-and-pop operations. In 1982, if you saw a niche for a coffee shop in your town, you would open up a coffee shop. Now, a Starbucks was there three years ago. National retailers have gotten very good at identifying lucrative retail locations, and are able to move more quickly than individuals.

Note that this doesn’t imply that national retailers are any more productive than mom-and-pop stores (although they do pay higher wages than small retail establishments). If they were, then we should have seen some kind of long-run boost to productivity from 1982-2011. We don’t. My guess is that it just means national retailers have a distinct advantage in identifying and opening lucrative retail locations compared to individuals.

Of course, it could be that the loss of productivity from the drop in young firms is offset almost perfectly by the increase in productivity from having national firms more readily identify and take advantage of new retail opportunities. If so, okay. From a productivity standpoint, though, it’s a wash, and does not necessarily have any implications for future productivity growth.

DHJM 2014 Figure 3

Does it imply anything about employment? Well, as DHJM document in their figure 3, there has been a decline in the job creation and job destruction rates from 1980-2011 (don’t get too worked up about the big dip in the trend line for job creation – HP filters are sensitive to the end points you use). Both rates are declining, meaning that there is less worker churn in the economy, which is consistent with less churn in firms, which is what fewer young firms implies. Again, note that the trend of decline in job creation and destruction occurs over the 80’s, 90’s, and 2000’s consistently, which covers periods in which the employment to population ratio rose pretty consistently before leveling off in the last decade.

The fact that the proportion of young firms in the U.S. is declining doesn’t seem to be anything to get worked up about, and it doesn’t imply that U.S. productivity or employment are doomed to stagnate in the future. If there is some “optimal” amount of young firms to have, we have no idea what it is, and we could as easily be over that amount as under it. For now, I’m mentally filing the decline in young firms alongside the secular shift away from manufacturing and towards services. It’s one of those structural changes that occur as economies grow. But evidence from either (a) longer time periods in the U.S., or (b) across countries, could easily change my mind.

The Perils of “Instant” Communication

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

The Atlantic has a short piece about people’s horrified response to the demonic speed of the….telegraph. Classic “Get off my lawn” response to technological innovation. From the New York Times, 1858

So far as the influence of the newspaper upon the mind and morals of the people is concerned, there can be no rational doubt that the telegraph has caused vast injury. Superficial, sudden, unsifted, too fast for the truth, must be all telegraphic intelligence. Does it not render the popular mind too fast for the truth? Ten days bring us the mails from Europe. What need is there for the scraps of news in ten minutes? How trivial and paltry is the telegraphic column? It snowed here, it rained there, one man killed, another hanged. Even the Washington letter has deteriorated since the innovation, and I can conscientiously recommend my own epistles prior to 1844, in preference to those of later years.

That last line is also a great example of spurious correlation. Of course it had to be the telegraph that caused the writers letters to deteriorate over time.

Patents, Bargaining, and Innovation

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

The economics of endogenous growth imply a non-linear relationship between intellectual property rights (IPR) and innovative activity. Very low patent protections (or other types of IPR) and no one will want to bother innovating because they will not be able to reap any benefits before someone comes along and copies their idea. Too much patent protection (or other IPR) and it will be too hard to replace existing companies, so no one will bother to innovate. This means that there is a hump-shaped relationship of patent protection and innovation, with some area of optimal patent protection that maximizes the amount of innovation that gets done – just enough to encourage new innovators, but not so much that we freeze existing companies in place.

Two recent papers have some clues about that patent/relationship curve, but not in the way I was originally expecting. Bhaven Sampat and Heidi Williams have a paper (h/t Joshua Gans) on cumulative or follow-on innovation, meaning patents that are filed that rely on existing patents. They looked specifically at patents for genes, which allows them to clearly identify subsequent patenting activity. Within the patent office. They use a nifty strategy involving the random assignment of patent officers to get variation between genes that are patented and genes that are not. Comparing those with patents to those without, they find no difference in future innovation.

This is not what Alberto Gallaso and Mark Schankerman find (h/t Free Exchange). They use invalidations of patents by US Court of Appeals as the source of variation in intellectual property rights in their study. For some fields, invalidating a patent meant a tripling of the number of citations in future patents for that invalidated patent. So removing the IPR was a boon to future innovation. But this only held in fields where products rely a host of patents, think biotechnology in general or IT. For fields that they classified as simple, there was no effect of invalidating a patent on future innovation – similar to the Sampat and Williams finding.

Perhaps what these papers are telling us is that we should be more worried about shifts in the patent/innovation relationship, rather than movement along that curve. The Sampat and Williams paper shows that cumulative innovation is the same, regardless of whether the gene is patented or not. This suggests that a patent is simply re-distributing the total gains of cumulative innovations towards the current patent holder and away from the new innovator. But, the gains are realized though the cumulative innovation – the market appears to be figuring out for itself how to ensure these innovations get made. But note that this works for a very specific sub-field of research in which the use of an innovation (a gene) is clear cut.

The Galasso and Schankerman paper shows us what happens when the cumulative innovation is more complex. When a patent is struck down in a field that relies on webs of patents, there is a burst of cumulative innovative activity. Why? The many parties involved in the web of patents will find it hard to negotiate and allocate out the rents from the cumulative innovation. Practically speaking, it may be too hard to get everyone in the same room together. The court ruling solves the bargaining problem exogenously. It eliminates the need for the meetings, the e-mails, the lawyer’s fees, etc.. etc..

What Galasso and Schankerman are measuring, then, is the effect on innovation of lowering bargaining costs. This shifts the entire patent/innovation relationship up, meaning more innovation at any level of formal patent protection. In the Sampat and Williams study, the bargaining problem was already straightforward, so the difference in having a patent or not doesn’t affect anything. The costs were already low.

Now I’m not sure that his means we can fundamentally change innovation rates by lowering bargaining costs. The bargaining costs seem to be driven by the nature of the field. If the innovation is complex and relies on lots of other patents, then you cannot legally just decree that it is less complex. But definitive legal rulings are probably a boon to innovation, regardless of which way they decide. From a social point of view, we want to maximize innovations, and the exact distribution of the rents to the various patent-holders is not really our concern (Quick, do you care or know how the Apple/Samsung trial turned out? As long as various versions of smartphones remain available, probably not). So based on this, our goal with the patent system should probably be focused on making it transparent, fast, and cheap to navigate, as opposed to worrying precisely about who gets protected for how long.

Solitary, Poor, Nasty, Brutish, Short…and Happy?

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

In response to my recent post on geography/institutions, Thornton Hall posted some comments that led us to go back and forth regarding whether pre-historic/pre-agricultural humans led “long, happy lives” (Hall) or relatively short, dismal lives (me). Yes, this is tangential to the whole geography/institutions thing. It’s the internet, what do you want?

Hall pointed me towards several sources of evidence regarding longevity among hunter-gatherers to support his contention that these humans lived a relatively long time (70-ish years). The main source is a paper by Gurven and Kaplan (2007, direct link here) that as it turns out I had hiding on my hard drive anyway.

Gurven and Kaplan (GK) survey the collected evidence on life expectancies on early hunter-gatherers (much of which is based on reviews of current indigenous populations of h-g’s like the !Kung). I don’t have any reason to dispute the data presented by GK, as I’m not an anthropologist, nor have I read sufficiently in this literature to have any opinion on the sources. My response to Hall’s claims regarding the “long, happy lives” of pre-historic hunter-gatherers is based entirely on how that evidence is interpreted.

Gurven and Kaplan Figure 3
Let’s take figure 3 from the GK survey. This shows age-specific life-expectancies for different populations of hunter-gatherers, a hypothetical pre-historic population, and a population of wild chimps. What do we see?

First, life expectancy at birth is very low. These are the e_0 terms shown below the graph itself. This is what you typically think of when you hear “life expectancy” – how many years do we expect a newborn to live? This ranges as low as 27 for the Hiwi, as high as 42 for the Tsimane, but the inferred value for pre-historic populations is only 20. A newborn in a pre-historic society would – on average – live about 20 years.

This is mainly due to very severe infant and child mortality. Pre-historic babies were very likely to die before their first birthday, and making it to 5 years old is unlikely. But, conditional on making it to 5, life-spans could be quite long. So in the figure, you see that life expectancy at age 5 is almost 50 for some societies (so they would life to roughly 55) and 25 for the prehistoric society (so the 5-year old would life to 30).

Similarly, if you make it to 20 years old, you could expect to live another 40 years (so you’d be 60) in many of these populations, and another 20 years (so you’d be 40) in the hypothetical prehistoric society. Let’s just focus on the relatively high expectancies for the moment. These life expectancies tell us that a very lucky few hunter-gatherers will live long, happy lives. If you can survive to 20, you can expect to live to 60. If you can make it to 30, you can expect to live to about 70.

My claim that life is short for hunter-gatherers is based on the fact that a huge swath of the population dies by the age of 5. If I ignore them, then sure, average life expectancy is high. But that is like saying average wages in the U.S. are really high if I ignore people who are poor.
Gurven and Kaplan Figure 4
Second, the modal age of death has essentially no information in it. The GK survey, in their figure 4, shows the proportion of all deaths occurring at each age. The dark line is for hunter-gatherers, and it has a modal age of death (the highest point on the curve) of about 70 years.

This does not mean that most people die at age 70. If you look at the y-axis, you’ll see that it implies only about 1.6% of all deaths are at age 70. GK refer to this, and note that even with a model age of death of 70, roughly two-thirds of all deaths in the hunter-gatherer society will occur before age 70. Look at the distribution of deaths for the U.S. in 2002 in the same figure. See how the proportion of deaths at ages 15-35 is almost zero. Deaths do not really start to ramp up in the U.S. until age 55 or 60. For hunter-gatherers, there is a consistent chance of dying of about 1.2-1.5% from age 15 to age 65. You are far more likely to die at an age less than 65 in a hunter-gatherer society than you are in the U.S. today.

A last point on this figure is that it refers to ages of death, conditional on reaching age 15, which goes back to my first point. If you only look at the select population of individuals who make it past infant and child mortality, then yes they have the potential to live long periods of time. But that is ignoring the fact that a big chunk of the population will die by the time they are 15.

Final point, which refers to the “happy” part of “happy, long live”. I have no idea how happy these hunter-gatherers really were. It may have been a joyous life for them, perhaps far happier than we are today. I have no way of telling you otherwise.

But let me suggest two negatives that the early hunter-gatherers would have to overcome on their way to bliss. One, they had to witness a brutal rate of infant and child mortality. Every time they had a child, the likely outcome was that this child would be dead within 1 year. If it made it 1 year old, then there’s a slim chance the kid makes it to 5. You would have buried more kids than you ever saw married off. Oh, and let’s not forget that those kids led unhappy lives, dying early, likely from some kind of infectious disease.

Two, according to table 5 from the GK paper, 17% of all deaths for those under 60 were from violence. A similar 17% of deaths for those under 15 were from violence, either homicide or warfare. Close to one-in-five deaths occurred on the end of a spear, knife, arrow, or whatever weapon was at hand. One in five. For comparison, in the U.S. in 2010 (p. 11) only 0.4%, or 1 in 250, deaths are from homicide.

So I don’t buy that hunter-gatherers had it made compared to modern people. They died at astonishing rates at early ages, and a massive fraction of those deaths were through violence. Hobbes may have been wrong about life being “solitary”, as my guess is that you stuck as close as possible to your trusted family network, but “poor, nasty, brutish, and short” is a good first approximation.

Geography Matters Even if it Doesn’t

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

Acemoglu and Robinson have been running a series of posts on the work of James Scott. Their latest regards the role of geography versus institutions.

Their post is an explanation for why the dispersion of population away from navigable rivers and coasts in Africa and Asia is likely the result of state formation. They cite James Scott pointing out that creating distance from the center/hub of the state is one effective, final way of limiting the influence that state can have on you. So the geographic dispersion we see in these continents may be a response to the centralizing efforts of early states (or colonizers).

And that is all well and good. It certainly seems like a plausible idea to explain geographic dispersion. What I do not understand is why this somehow shows that geography does not matter for development. The fact that the geographic dispersion of the population in Africa and Asia has some origin in the politics of state formation doesn’t mean that this dispersion is somehow meaningless for development.

They explicitly cite the Gallup, Sachs, and Mellinger paper that proposes (among other things) this dispersion as a source of under-development. Rivers and coasts make trade less costly, and boosts incomes. By living far from these places – and in having fewer navigable rivers to begin with – Africa and Asia have relatively high trade costs and hence lower development. How is that invalidated by the idea that the population is dispersed because of a need to escape authority? Short answer, it’s not. The fact that they can explain why a geographic disadvantage arose does not mean that it is not a disadvantage.

Being able to back up one step in history is wonderful for our understanding of economic development, but it does not prove anything. We can do some infinite regress where we bounce back and forth from institutions to geography: populations are disperse because of state formation, state formation was centered around high productivity agricultural areas, high productivity agriculture was due to innovations made possible by property rights, property rights only arise in places with land worth owning because of its fertility, etc. etc.. None of those steps imply that geography or institutions is better at explaining economic development.

Which is why I don’t understand the tendency of the Acemoglu and Robinson body of work to insist that institutions matter to the exclusion of all other factors. Yes, institutions (so broadly defined as to capture really anything you like) seem really important. But they are not the exclusive determinant of relative development, are they? Should I literally think that the poor agricultural land, lack of accessible waterways, and absence of valuable commodities in Ethiopia are meaningless to that country’s poverty? I don’t think so. The fact that improving institutions in Ethiopia would make it wealthier does not imply that it’s geography has not been a drag on its development.

Growth Class Slides

NOTE: The Growth Economics Blog has moved sites. Click here to find this post at the new site.

I’ve posted up slides for use with Introduction to Economic Growth on the Class Materials page.

I am a bit of a slide minimalist, so they are not done with Beamer, but they were created using TEX. They contain all the important equations and figures from the book, but there is not a lot of verbiage. I tend to leave that off the slides to force myself to talk through things more slowly.

Both the PDFs and the original TEX are posted (along with the necessary figure files). So you can use them as-is or feel free to edit them as necessary. Happy to hear back comments on them if you do use them in an undergrad class.