Why Do Cities Matter? Local Growth and Aggregate Growth

This is a working paper by Chang-Tai Hsieh and Enrico Moretti that examines how growth in cities relates to the growth of the nation writ large. The upshot is that, according to their model, if workers were allowed to move more freely to high productivity cities, such as New York and San Francisco, the US could have a substantially higher level of GDP. Let’s walk through the argument.

To start, we can safely assume that there is variation in the productivity of individuals, which is to say we have differing marginal products of labor (MPL). But this productivity does not exist in a vacuum; ultimately a worker’s output also depends on the environment into which they are inputting labor. A mathematics PhD may have a very high potential MPL, but if they are inputting that human capital into a retail cashier production function, there won’t be much difference in output from someone with only a high school diploma.

The second part is that there is geographic, or spatial, variation in productivity as well. It’s well documented that high density urban areas have far higher productivity than rural areas. Now, this could be due to sorting: high productivity workers tend to move to cities so of course cities have higher average productivity. But there is research that suggests density itself leads to higher productivity, as explained in this paper, through phenomenon like knowledge spillovers.

So we have variation in productivity across individuals and across space. In equilibrium, we would see workers distributed spatially such that the marginal product of labor would be equal across the country. Now of course we will never actually reach equilibrium because the economy is made of so many moving parts, but this gives us a frame of reference to look at data and see what direction things are moving in. If the US were moving closer to equilibrium, we would expect to generally see convergence in the spatial distribution of wages; recall that it is assumed that wages and the MPL are linked.

What the authors find is the opposite, however: after controlling for factors like worker characteristics, the geographic variation in wages across cities in the US has increased since 1964. This suggests that differences in worker productivity across cities are growing.

The natural follow up question is, why? Land use regulations seem to be the culprit. Over the past several decades, there has been a huge increase in demand for finance and tech skills, which are concentrated in New York City and the Silicon Valley areas. In other words, these local labor markets have undergone a large positive shift in demand. However, stringent land use regulations have made the labor supply curve in these areas highly inelastic, meaning the demand has primarily resulted in an increase in wages instead of employment.

The key here is to realize that the consequences of this are lot merely local; this affects the country as a whole. There are workers with high potential MPLs that are prevented from fully realizing them, and are stuck in lower productivity jobs in lower productivity areas simply due to the fact they cannot physically move to the region that would best use their talents. This means that the US has a lower level of output, and social welfare, than in the counterfactual reality where housing policy was more flexible.

I’m increasingly convinced that these stringent land use regulations and related NIMBYism are responsible for much of the economic pain of the past few decades. Mobility in the US has declined significantly since the 1980s, and the increased social sorting such policies encourage has only exacerbated racial and class differences, leading to a horrible generational feedback cycle and overall loss of opportunity.

Moving the locus of land use policy away from the local level is low hanging fruit with huge potential upside, and yet the national conversation on economic growth pays essentially zero attention to this. This needs to change.



Fabula delenda est

In the coming days and weeks there are going to be many words about ‘the meaning’ of this election. But in the spirit of the name of this blog, we should be cautious in believing any of them. The fact is that the margin in this election was so close that interpreting what was essentially a coin flip as portending a seismic shift in American politics is folly. Nate Silver makes the point here that if one in one hundred voters, just 2% of the voting electorate, had shifted to Clinton, we would be crafting an entirely different narrative.

Furthermore, political science models that were based on fundamentals, such as the state of the economy and length of incumbent party control, predicted a Trump victory (why people didn’t take such models seriously is a separate issue). As Larry Bartels argues, this should suggest to us that in many ways this election was entirely normal.

That being said, of course this election still raises many questions:

Continue reading “Fabula delenda est”

Wages and Productivity

One of the major assumptions of many economic models is that workers’ pay and productivity are linked: wages are determined by the marginal product of labor. On the other hand, some economists have pointed out a growing divergence between wages and productivity in the US economy, which suggests the link is not so close. This chart is a common example:


Who’s right? Well, this very topic is the subject of one section of a new AEI book, simply titled ‘The US Labor Market: Questions and Challenges for Public Policy’ (free pdf available). The contributor list is a veritable who’s who of economists whose analyses I find invaluable, and it looks to hit on many themes I’m interested in, so I’m excited to dig into this. But for today I’ll stick with wages and productivity.

One the side of divergence is Dean Baker of the Center for Economic and Policy Research. He argues that the top and bottom of the income distribution exhibit a disconnect between wages and productivity. He starts with the example of CEO pay, which as I noted in my post on inequality in firms, has pulled away from the rest of workers in very large firms. If pay were linked to productivity, this would indicate American CEOs have become an order of magnitude more productive than they were decades ago (as well as more productive than similar CEOs in other rich countries). What this suggests is really going on is that due to poor corporate governance structures, CEOs are commanding wages far out of line with what they would earn in a truly competitive marketplace.

Next Baker turns to the minimum wage, and basically argues that a combination of a wage floor and firms’ desire to have a differentiated wage structure means there is another disconnect between wages and productivity. If the minimum wage increases, there is an incentive to raise other workers’ pay as well to maintain distinctions based on seniority, skills, etc. He also argues this means that these workers are being paid less than their productivity already, using a monopsony model as a base. There’s a decent explanation here, but basically a monopsony is the opposite of a monopoly; i.e. one buyer and many sellers, so the buyer (employer) has power over pricing (wages), and in these models wages are well below what they would be in a competitive marketplace.

I don’t find this particular part of the argument so convincing based on my experience in a city that has raised the minimum wage. Over the last few years, the DC minimum wage has gone from $8.75/hr prior to 2014 to $11.50/hr as of July this year, and will reach $15/hr by 2020. In that time, at the (large) retail company I work for, there have been no raises beyond regular cost of living for workers between cashiers and salaried managers. So now workers employees with decades of experience are at the same level as fresh cashiers, and the wage difference between a supervisor (who is typically hired around $12/hr) and a regular cashier has basically disappeared. It still suggests there is disconnect between wages and productivity, just not the way he envisions it.

Moving on, Robert Lawrence of Harvard argues that looking at charts such as the one above is misleading because the deflators used in measuring compensation and output across time are used improperly. The real compensation series uses the CPI deflator, while the real output series uses an implicit price index for GDP. The CPI has some problems; many believe it overstates the real cost of living, and Lawrence argues that in this context an output price index should be used to deflate compensation. Why? Because using a consumption index includes factors that domestic workers do not produce, such as imports and housing. As such, a ‘product wage’ should be used, and doing so accounts for a significant chunk of the observed gap. Using his estimates of productivity, he find that productivity and compensation track each other closely until about 2001, and it’s not until the Great Recession and its aftermath that a real divergence opens up, though not a massive one.

Something that’s often implicit in arguments in this space is that productivity/pay divergence is the main source of inequality. I think the reason for this is that it implies an easy solution: if we can just control the greed of the top 1% and make companies pay workers more, then everything will be ok. If it turns out that people really are broadly being paid their ‘fair’ share, then solving the problem of inequality becomes much harder (as we’ve seen described in The Wealth of Humans).

Where do I come down on this? While I do think there’s something ‘unfair’ going on with things like CEO compensation, I also just don’t see that playing out at the scale required to explain the levels of inequality in the US. And so we’re left to grapple with far more difficult questions.

Firming Up Inequality

Another day, another NBER paper. ‘Firming Up Inequality’, by Jae Song et al., is about a year old, but the latest version is from this summer (pdf). This paper uses a comprehensive matched employer-employee data set to investigate just how changes in inequality have played out between and within firms.

The fact of dramatic increases in wage inequality in the US is well documented, but there is considerable disagreement on its root causes. As we’ve seen argued in books like The Wealth of Humans, skills biased technical change is one hypothesis, which boils down to changes in markets rewarding a particular subset of workers. Another hypothesis is that changes in inequality are mostly due to the powerful members of society extracting economic rents.  A common example of this is the relative rise in American CEO pay vs other developed countries. If this rent seeking hypothesis were true, we would expect to see increasing inequality within firms, as CEOs and other top managers captured greater shares of firms’ profits.

What they find is that the majority of the increase in income inequality is actually explained as a between-firm phenomenon, not a within-firm phenomenon. Furthermore, this finding holds across industry sectors, region, demographics, firm size (with one exception; see below), etc. It’s not the case that there has been any sort of differential growth leading to this.

Finally, they find that there has been one type of firm that has seen increases in within-firm inequality: the ‘mega-firm’, i.e. a firm with more than 10,000 employees. In the US there are around 800 mega-firms that employ about 20% of the workforce, and these have seen both decreases in the lower end and increases in the higher end of the income distribution. In the authors’ estimates, the median worker in such firms saw earnings fall around 7% from 1981 to 2013; in contrast, the top 10% saw increases averaging around 11%. And the managers at the very top in these firms have realized real earnings increases of 137%.

So what’s going on here? Most firms are becoming more equal internally, yet firms as a whole are becoming less equal.  And at the same time, the very largest firms are seeing the opposite pattern.

Song et al. propose two potential explanations for the increased in interfirm inequality: a ‘widening firm premium’ story, or a ‘worker segregation’ story. In the former, firm inequality is driven by the fact that some firms are winners and others losers in the economy; winners may end up distributing their increased gains to their employees, while losers cannot. In the latter story, workers are increasingly being sorted into firms by ability; high-ability and low ability workers are clustering into separate firms, rather than mixing together.

Continue reading “Firming Up Inequality”

Recessions and Skill Biased Technical Change

A very fascinating and topical NBER paper (ungated pdf here) recently came out which investigates the connection between skill biased technical change (or routine-biased technological change, as these authors prefer to call it), which has been a major theme in the posts on The Wealth of Humans, and recessions.

One curious fact about the last several recessions is that their recoveries have been largely jobless. If we eyeball the growth in employment over the last several decades, it’s obvious we never returned to trend even after the 2001 recession.


Furthermore, unemployment continued to rise even after the technical end of the post-1990 recessions.


The reasons for this have not been well understood, but the authors take an adjustment cost theoretical framework as a starting point.

We can imagine that as technology makes certain jobs less relevant while creating others, there are costs to employers in making the appropriate adjustments. Recessions can provide the necessary ‘oomph’ to reallocate labor and capital that wouldn’t happen in normal times. In this understanding, labor market adjustment to RBTC occurs episodically, not gradually.

I’m sure it’s not hard to visualize that person at the office who isn’t really necessary anymore since that new software can do 70% of their work, but is still around since the boss is a human being and can’t find a compelling reason to fire them. In these models, economic downturns provide the excuse to clear out inefficiencies that have built up over time.

While this makes intuitive sense, the question is whether there is data to back it up. Other authors have found that the vast majority of declines in middle skill employment have occurred during recessions, and subsequent recoveries have been jobless in precisely these occupations.

Still, these studies have lacked direct evidence on how firms were restructuring. That’s where this paper comes in: they’ve obtained a data set that contains huge amounts of online job postings in major metropolitan areas for the years 2007 and 2010-15.

They find first that employers in metropolitan statistical areas (MSAs) that were hit harder by the Great Recession had an increase in skill requirements in jobs ads relative to both those same areas before and MSAs less affected by the shock. Now, there are two possible causes of this: one could be that employers simply increased the asking skill levels for all jobs, or happened to be posting mostly high skill levels jobs and less low skill jobs. The authors find the former explains most of this ‘upskilling’: in a weak labor market, employers were able to demand higher skill levels for the same jobs relative to a few years prior or to MSAs with better economic conditions.

This leads to the next question: are these firms actually changing what they do, or are they merely changing the type of people they hire? The former implies Shumpeterian idea of a recession as a cleansing activity (and consistent with the adjustment cost explanation above), whereas the latter is more like taking opportunity of a slack labor market to get the type of people you wouldn’t normally be able to. The latter is temporary; the former, permanent.

Continue reading “Recessions and Skill Biased Technical Change”

The Wealth of Humans, Part VII: Solutions?

<Part I, Part II, Part III, Part IV, Part V, Part VI>

To recap: the Digital Revolution has reduced the bargaining power of the average worker through two channels. The first is by creating an abundance of labor through automation, enhanced productivity, and globalization. The labor supply has outstripped labor demand, forcing workers to accept low wages or no work. The second is by changing the nature of the most productive activities. Serving as a cog in an industrial process was once a high value activity with clear and measurable aspects: how many units did you produce? Are they up to standard? But now the highest value activities occur within firms acting primarily as information processing organisms, and the success of these firms relies on social capital. This social capital is hard to measure, and by its nature cannot be taken or easily replicated. So despite being crucial to the success of their firms, workers don’t have the bargaining power to fully capture their share of gains versus the owners of capital.

This has led to a stagnation of wages and an employment trilemma that suggests that the only sources of mass employment in the future will be low productivity, low wage work. Is there anything we can do?

Avent runs through a few proposals to get the average person more money that I’d like to explore. I think it’s important to keep some questions in mind as we consider each.

  • Is our goal to reduce inequality or simply raise the floor for the bottom end of the income distribution?
  • Is there a trade off between inequality/redistribution and economic growth?
  • What types of redistribution do we want to engage in?

It should be said that the assumption here, and one that has underlain the entire book, is that today’s inequality is chiefly due to skills biased technical change. There are other potential causes, such as growth in economic rent seeking, that should be kept in mind, though I won’t deal with them here.

Continue reading “The Wealth of Humans, Part VII: Solutions?”

The Wealth of Humans, Part VI: Superstars

<Part I, Part II, Part III, Part IV, Part V>

It’s tempting to ask, what’s really different about this time? Isn’t this the same story we’ve heard every time some new advance comes along? It’s the nature of technology to destroy some jobs, and enhance and create others; that’s creative destruction and it’s always been with us.

While I’m generally sympathetic to this sort of ‘Ecclesiastical’ thinking (nothing new under the sun), I do think there are signals that this time really is different in important ways.

I don’t believe Avent touches on this in the book, but what I see as one of the big differences is the prevalence of what’s dubbed The Superstar Effect (pdf). This is most plainly evident in certain industries, like music, film, and televised sports: industries that are based around the transmission of information, or where, to quote from the paper, ‘the costs of production do not rise in proportion with the size of the market’. In the case of information transmission, such as film, television, or recorded music, the marginal cost of distribution is virtually zero. Furthermore, these types of goods have the characteristic of being ‘non-rivalrous’: just because I purchased The Wealth of Humans doesn’t prevent you from buying it also (so you should!).

Another key feature is imperfect substitution in these markets. Buying 10 cheap sponges could be the equivalent of buying 1 good sponge: the good one lasts ten times as long as the cheap sponge, but beyond that there isn’t much difference; they function as perfect substitutes. But watching 10 episodes of Home Improvement doesn’t add up to one episode of Breaking Bad.  Consequently, demand increases more than is proportionate to the degree of quality. To quote again from the paper, ‘If a surgeon is 10 percent more successful in saving lives than his fellows, most people would be willing to pay more than a 10 percent premium for his services.’

When markets have both of these features, non-rivalry and imperfect substitution, the superstar effect comes into play: one or a few players can capture outsize shares of the market. Think of the acting abilities of the average Hollywood star; they aren’t thousands of times better at acting than the people you see on TV, but being just slightly preferable to the average consumer means those stars make huge gains.

Technology, via replication, is what allows non-rivalry. And again, historically this has been most evident in industries based around transmission or communication of information. So it’s quite likely the Digital Revolution will continue to exacerbate this problem by rendering more and more industries subject to replication. Furthermore, technology will continue to reduce the marginal costs of production in many other fields so that they also can also expand more than proportionally. One example could be programmers; in the deep learning market, the good being designed is multi purpose. Once you have your program, you can apply it to all sorts of problems at little additional cost; DeepMind didn’t need to be reworked from the ground up each Go match it played, nor will it have to be totally redesigned for its next endeavor. 

Or consider remote surgery. The best surgeons will be able to service vastly expanded markets, rendering the services of all but the most skilled surgeons largely unnecessary. So I could see superstar effects taking hold in many more industries.

This is one big reason why I think the changes brought by the Digital Revolution aren’t some temporary phase. The very structure of many job markets has the potential to be fundamentally transformed to dramatically favor only the best. And if this comes to pass, what happens to the rest of us?