Chapter 3

Toward a Pareto Economy

The Gaussian distribution is by no means the only distribution of economic outcomes imaginable. At the turn of the twentieth century, Italian economist Vilfredo Pareto noted that, at the time, 20 percent of Italian families owned 80 percent of Italy’s land.¹ Most of the remaining 80 percent, who owned no land, farmed the land owned by their rich and often oppressive landlords. The Pareto distribution named in his honor—or Power Law distribution to most statisticians—takes the shape of the curve seen in figure 3-1.

On this curve the many poor Italians with little to no land are on the left side and the very few superrich, landowning families are in the long tapered end to the right.

Along with a very different shape, a Pareto distribution has markedly different characteristics than a Gaussian distribution. There is no meaningful mean or median in a Pareto distribution. The median is often nil—for example, the median Italian family in Pareto’s time owned no land. To say the median landownership in nineteenth-century Italy was zero doesn’t tell one much about the overall distribution, while saying the median height of an American male is 5' 9.5" tells one quite a bit more about that overall distribution.

Also, a Pareto distribution is not stable. As discussed in chapter 1, when additional observations are added to a Gaussian distribution, the overall pattern of distribution becomes only more pronounced thanks to the central limit theorem. The addition of new data points has no such predictable effect on a Pareto distribution. They might make it less extreme, make no difference, or make it even more extreme. Often, for reasons explained below, it is the latter, but it is not necessarily so.

The key difference between Gaussian and Pareto distributions is that while independence of one observation from another is a central characteristic of a Gaussian distribution, the observations in a Pareto distribution are connected to one another. In the Gaussian distribution of American adult male height, for example, no one person’s height has any impact on the height of any other male. This creates a distribution in which the tallest American male—at 7' 8"—is less than 1.5 times as tall as the aforementioned median.²

Now consider social media—Instagram, for example. If the distribution of the number of followers per Instagram user were a Gaussian distribution, it would be because when Instagram users decided whom to follow, they did not take into consideration the number of followers a candidate for followership already had. This, of course, is not the case by any stretch of the imagination. The number of existing followers is a key, if not the key, consideration in deciding whether to follow a person. If a person has few followers, the inference is that they are not worth following. If a person has many followers, she is not only worthy in general, she is probably worthy of following.

That is why the median user of Instagram has 100–150 followers while soccer star Cristiano Ronaldo, according to a recent count, has 218 million followers.³ The phenomenon is called “preferential attachment.”⁴ I see it as attractive for me to attach to Ronaldo because Ronaldo already has many followers. In the Pareto distribution, an important dynamic is that the effect (having more followers) is the cause of still more of the effect (having still more followers), which causes yet more effect, and so on.

Since Pareto’s time, many scholars have studied the dynamics and implications of Pareto distributions, including those who work in complexity theory, who note that Gaussian distributions aren’t the only ones that appear in nature. Pareto distributions do as well. A favorite example for complexity theorists is the collapsing sand pile. Imagine on the above Pareto graph, the number of grains of sand dropped on a pile is measured on the vertical axis and the effect of each sand grain on the horizontal axis. If the grain of sand has no noticeable impact on the shape of the sand pile, it shows up on the extreme left side of the chart. If it causes the pile to collapse entirely, it is on the extreme right. When grains of sand are dropped on top of an existing sand pile, nothing much happens to the shape of the sand pile for thousands and thousands of drops of sand grains—hence the very tall bar on the extreme left of the Pareto chart. But after enough grains are dropped, the dropping of one additional grain will cause the sand pile to collapse entirely—that grain is marked way out on the long, tapered end of the Pareto chart. This experiment can be run with a thousand sand piles, and the distribution will always be Pareto. The overwhelming majority of sand grains are dropped with no evident effect, and then one becomes the proverbial straw that breaks the camel’s back—another great metaphor—and the pile collapses.

From Gaussian Distributions to Pareto Distributions

This matters to the fate of American democratic capitalism, because complexity scholars, including Bill McKelvey of the Anderson School at UCLA, from whom I learned this, have identified several factors that systematically push Gaussian outcomes toward Pareto distributions. Chief among them are two things: (1) pressure on the system in question and (2) ease of connection among the participants in the system.⁵

The sand pile can be used to illustrate. First, if the sand pile existed in a low-gravity context, it wouldn’t collapse. It collapses only as the pressure applied by earth’s relatively strong gravity accelerates that final grain down with enough force to jar enough other grains out of position to drive a total collapse. Second, if the grains of sand were not connected with one another but rather independently placed in a superstructure, the additional sand grains would not have impacts on the others and would not start a collapse. For this reason, a Gaussian distribution facing a relatively low level of pressure and relatively high costs of connection between participants in the distribution would stay Gaussian for a long period of time—so much so that we think of it as naturally Gaussian. But if more pressure is applied to the participants in the distribution and the cost of connection falls, the distribution will more quickly turn Pareto (see figure 3-2).

For the US economy, efficiency is gravity’s equivalent. The obsession with ever greater efficiency, amplified by the pursuit of surrogated proxies for efficiency, has converted a Gaussian distribution of outcomes into a Pareto one. Relentless pressure on labor and procurement costs, the near elimination of slack, and the opening of markets by the General Agreement on Tariffs and Trade and the World Trade Organization have increased competition across industries. If companies fail to become more efficient in their operations, they will be driven out of business by local or foreign competitors. To this pressure is added, thanks to the efficiency defense for mergers, increasing consolidation as more and more industries turn into oligopolies or monopolies.

Meanwhile, the internet has slashed the costs of connection. Everything is more easily and inexpensively connected to anything and everything else. The internet is not the only driver of falling connection costs in the economy. The mergers of the capital-market exchanges, for example, have lowered the cost of connecting one equity pool to the next, while regulations such as the Financial Industry Regulatory Authority (FINRA) Rule 5310, which requires all trades to be routed to the market that has the best current price for the security, are forcing markets to interconnect seamlessly, thereby reducing the costs of connection to zero. With microwave towers now connecting the Chicago Board of Trade to the New York Stock Exchange, arbitrage between the two markets has become virtually nonexistent. Forced salary disclosure by the Securities and Exchange Commission of the five top-paid executives of publicly traded companies means that, for example, the compensation of our company’s CEO is seamlessly connected to and comparable with the compensation of every public-company CEO in the country, meaning that if those salaries go up, there will be instant upward pressure on the compensation of our CEO.

To illustrate the impact of pressure and connectedness, let’s return to the fame game introduced above. Fame has always been a competitive game. It was difficult to achieve singing stardom like that of Frank Sinatra in the 1960s or supermodel stardom like that of Cindy Crawford in the 1980s. But these were arguably smaller and more fragmented games. We didn’t routinely ask who was more famous: a singer or a supermodel. It was hard to measure and compare—like apples and oranges. And the competition wasn’t as intense, because the payoff for achieving standout fame in your particular discipline was lower. For example, as the most famous crooner of his generation, Sinatra amassed a lifetime fortune of $100 million. And as the most famous supermodel of her generation, Crawford also amassed a lifetime fortune of $100 million.⁶

Like much of the US economy, fame has gotten much more competitively intense and much more tightly connected. Ronaldo has ascended the global soccer mountain, which is probably somewhat more competitively intense and tightly connected than it was twenty-five years ago, but not orders of magnitude more so. But parlaying that soccer fame into ubiquitous celebrity fame is a much more pressure-filled game. If you aren’t positioning yourself cleverly on social media, you are going to be left behind—and once you fall behind, it is ever less possible to catch up. Thanks to the zero cost of connection and the utterly standardized measuring stick (Instagram followers), soccer-star Ronaldo is pitted against female singer Selena Gomez, male singer Justin Bieber, wrestler-turned-movie-star Dwayne “the Rock” Johnson, celebrity Kim Kardashian, and celebrity-turned-cosmetics-mogul Kylie Jenner.

The dynamic has very tangible consequences. With his lofty standing in Instagram followers, Ronaldo earns a reported $750,000 for every sponsored post. But he has to take a back seat to twenty-one-year-old Jenner. She trails Ronaldo with (a mere!) 175 million followers, but the advertisers like her followers better than Ronaldo’s, so she earns a reported $1 million per sponsored post.⁷ Imagine: in a single Instagram post, she can make 1 percent of the lifetime wealth of Cindy Crawford. Two posts per week for two years would generate upward of $200 million in revenues. Given that the cost of doing so is minimal, she would—after taxes—accumulate wealth of over $100 million in less than two years on Instagram posts, while probably spending, over those two years, only about as much time as Sinatra did in one concert or Crawford did in one modeling shoot. Thanks to her fame, and her skincare and cosmetics line based entirely on that fame, Jenner is the world’s youngest billionaire.

In the discussions of these Pareto outcomes in income and wealth, the two causes pointed to most often are globalization and technology. While both have indeed played a role, it is important to understand that globalization and technology are not in and of themselves the causes of the Pareto shift. While globalization has contributed to increased pressure, it is hard to see globalization as the central cause when the shift toward a Pareto distribution of outcomes was well under way by the 1970s yet most of the meaningful increase in globalization took shape much later. In terms of truly free trade, America’s first consequential free-trade agreement wasn’t until 1988, with Canada, and then 1994 when it was expanded to include Mexico (NAFTA).⁸ There wasn’t another free-trade agreement until after 2000. (The only prior one was a more symbolic than economic agreement with minor trading partner Israel in 1985.) Arguably the meaningful impacts of globalization took effect close to a quarter-century after the Pareto shift took shape. The same timing caveat has to be issued for the impact of technology on lowering the cost of connection. The first commercially successful personal computer—the Apple Macintosh—didn’t make its appearance until 1984. The internet didn’t come into play until the early 1990s, and smartphones until the late 1990s. E-commerce was still a new thing as of the first (and largely unsuccessful) dot-com boom at the end of the 1990s. This is a quarter-century into the Pareto shift. To be sure, globalization and technology have deepened and intensified the Pareto shift. But we can’t argue that two forces that arrived at the party in question a quarter-century late are, together, responsible for the party’s outcomes. The obsession with efficiency began much earlier.⁹

At a broader and more fundamental level, increases in pressure fueled by an obsession with efficiency, measured through surrogated proxies, in a context of increased connectedness, are making the distributions of outcomes in almost all spheres of economic activity increasingly Pareto, crowding out the historically Gaussian patterns that we have always assumed. It is most obvious in wealth, where the distribution is far beyond the 80–20 of Vilfredo Pareto’s Italy. The richest 1 percent of Americans own almost 40 percent of the country’s wealth, while the bottom 90 percent own just 23 percent. The richest American is 10 million times richer than the median American.¹⁰ Beyond income and wealth overall, we’re also seeing the distributions of job-market rewards and company profits moving smartly from Gaussian to Pareto, with fewer bigger winners and plenty left behind. To be sure, we are not there yet. But as the next few pages will show, we are systematically heading in that direction. Let’s start with jobs.

The Rewards for Your Labor

It is becoming increasingly apparent that certain kinds of jobs in certain kinds of industries offer disproportionately more rewards for the people doing them. This can be seen by combining the findings of two scholars of work, one who studies the industries in which Americans work, the other who studies the kind of tasks American workers do.

Let’s take industry first. Michael Porter, in The Competitive Advantage of Nations, demonstrated that it really matters in what kind of industry an American worker holds a job.¹¹ It could be an industry that sells its output in markets outside its local region (as with steel or semiconductors or pharmaceuticals) or one that does not (like hairdressing or landscaping). It turns out that the former end up clustered in only a few regions (or maybe only one) in America, because companies in these industries have the capacity to grow to great scale and they must be highly competitive to triumph over players from other regions. Hence, these companies invest more in plant and equipment, research and development, and marketing in order to prevail. In contrast, the latter are found in every region of the country, because their goods and services are needed everywhere but it is difficult if not impossible to sell outside their region. For example, with few exceptions, a hair salon will serve only its local market. About a third of US jobs are in the former, clustered industries, and the rest are in the latter, dispersed industries.

From the job-type perspective, another scholar, Richard Florida, in The Rise of the Creative Class, demonstrated that the content of an American worker’s job really matters.¹² In particular, it matters whether the content of the job is creativity-intensive or routine-intensive. In a creativity-intensive job, the worker needs to exercise meaningful, independent judgment and decision making in order to fulfill the requirements of the job. Such workers include doctors, business executives, teachers, researchers, and police. In a routine-intensive job, the opposite holds. The worker should not exercise meaningful judgment or decision making and should instead follow a prescribed routine. Such workers include assembly-line workers, hospital orderlies, and retail clerks. Creativity-intensive jobs account for just under 40 percent of US jobs. The rest (60 percent) are routine-intensive jobs.

Unsurprisingly, jobs in clustered industries, on average, pay considerably more than jobs in dispersed industries, because the employer invests much, much more capital behind each employee, and that opens up the possibility of that employee being much more productive and earning a salary that is consistent with that higher productivity—though, as we have seen, the historic relationship between increased productivity and wage growth has weakened. Also, unsurprisingly, jobs that are creativity-intensive, on average, pay considerably more than jobs that are routine-intensive. The creativity-intensive jobs require much more formal education and higher skill levels than the routine-intensive jobs, and hence pay consistently higher wages.

Combining the two factors makes the effect more extreme. If type of industry and content of work are combined, the compensation distribution across the four combinations begins to look substantially Pareto. Routine-in-dispersed jobs (e.g., that of retail clerk) make up the largest cohort, at 45 percent of the US job market, and bring an average income that is 37 percent below the national average. Routine-in-clustered jobs (e.g., that of assembly-line worker) make up 16 percent of the labor market and bring an income 18 percent below the national average. Creative-in-dispersed jobs make up 25 percent of the labor market and bring earnings 36 percent above the national average. And in the catbird seat, the 14 percent of workers in the creative-in-clustered category earn 78 percent more than the national average. While not a perfectly Pareto distribution, it is more Pareto than Gaussian, and heading swiftly in a Pareto direction. The data for 2012 shows a markedly more Pareto shape than the data from a mere twelve years earlier, in 2000 (the earliest year for which this data set is available).

The top graph in figure 3-3 compares the overall data for these four categories in 2000, while the four smaller graphs below it break out the changes between 2000 and 2012 for the individual categories. The top of these four graphs shows that in the largest and lowest wage category, the wages have stagnated completely for twelve years, as the 2000 and 2012 boxes overlap completely. For the next category, wages at the higher end of the category (the seventy-fifth percentile) have grown marginally—the solid box (2012) extends a bit past the hatched box (2000)—while the lower bound (the twenty-fifth percentile) stays completely constant. Things get considerably better for the third category, as both the lower bound and the upper bound have increased substantially. But that pales in comparison to the progress of the highest-income category in the bottom graph. In a classic case of the rich getting richer, the lower bound increases nicely and the upper bound increases dramatically. Hence in a mere twelve years, the whole cohort has become especially well-off and much more so for its top end. Again, it must be emphasized that this dramatic divergence between the accelerating creativity-intensive workers and the stagnating routine-intensive workers has taken place in a mere twelve years, less than one-third of the working life of the workers across these four boxes.

Pareto Distribution in Companies

Industry consolidation is increasingly common in the developed world. In more and more industries, profits are concentrated in a handful of companies. In the United States, for instance, 75 percent of industries have become more concentrated in the past twenty years, and profits have followed suit. In 1978 the one hundred most profitable US firms earned 48 percent of the profits of all publicly traded companies combined, but by 2015 the figure had nearly doubled, to 84 percent.¹³ The success stories of the so-called new-economy companies are in some measure responsible. The dynamics of platform businesses, where competitive advantages often derive from network effects, very quickly convert random distributions into Pareto ones, as Instagram did with fame sorting.

But Pareto distributions are also common in traditional industries. Consider the American waste-management industry. At one time there were thousands of little waste-management companies—garbage collectors—across the country. Each had one-to-several trucks serving customers on a particular route. The profitability of those thousands of companies was fairly normally distributed. Most clustered around the mean, with some highly efficient and bigger companies earning higher profits and some weaker ones earning lower profits.

Then along came the late Wayne Huizenga, the founder of Waste Management Inc. (WMI). Looking at the cost structure of the business, he saw that two big costs were truck acquisition (the vehicles were expensive, and because they were used intensively, they needed to be replaced regularly) and maintenance and repair (intensive use made this both critical and costly). Each small player bought trucks one (or maybe a handful) at a time and ran a repair depot to service its small fleet.

Huizenga realized that if he acquired a number of routes in a given region, two things would be possible. First, he would have much greater purchasing leverage with truck manufacturers and could acquire vehicles more cheaply. Second, he could close individual maintenance facilities and build a single, far-more-efficient one at the geographic center of each region. As he proceeded, the effect—greater efficiency—became the cause of more of the effect.

Huizenga generated the resources to keep buying small garbage companies and expanding into new territories, which made WMI bigger and more efficient still. This put competitive pressure on all small operators, because WMI could come into their territories and underbid them. Those smaller firms could either lose money or sell to WMI. Huizenga’s success represented a huge increase in pressure on the system.

Like a collapsing sand pile, the industry quickly consolidated, with WMI as the dominant player, earning the highest profits. Fellow consolidator Republic Services established itself as the second player, earning decent profits. Several considerably smaller would-be consolidators earn little to no returns, and lots of tiny companies mainly operate at subsistence levels. (See figure 3–4.)

The industry today is structured as a Pareto distribution, with WMI as winner-take-most. The company earned more than $14 billion in 2017. Huizenga died a multibillionaire.

That is the fundamental story of the modern American economy. Income and wealth are becoming fundamentally Pareto distributed. The pie is getting bigger, as it is supposed to. But the vast majority of every dollar of economic growth is ending up in the hands of a small minority of Americans—that is, American citizens and American companies. As a consequence, there is not enough left to move the rest of the distribution forward. The middle class (the hump of the bell curve in a Gaussian distribution) is at worst going away and at best stagnating rather than moving forward.

But that’s not the only consequence. Nature shows us that systems in which a small number of highly efficient actors dominate become vulnerable to an external change. And it turns out that this is as true for economic systems as it is for natural ones.

The Rise of Monocultures

To illustrate that point, let’s take the case of the almond-growing industry. Not so very long ago, almonds were grown in a number of places in America and across the world. But some places are better than others for growing almonds, and as with most production contexts, there are economies of scale to consolidation. In this case, the Central Valley of California is perfect—totally perfect—for growing almonds. Consequently, over 80 percent of the world’s almonds are now produced in this one valley.¹⁴ This is what agricultural scientists would call a monoculture, and they are a common outcome in systems that maximize efficiency. A factory produces a single product, a single company dominates an industry, a single piece of software dominates computer systems. We remove unhelpful inefficiencies and get more productive.

But with that high efficiency comes an inherent vulnerability to shocks, with potentially catastrophic results: one extreme local event—a wind-swept fire, say, or a pernicious virus—could wipe out 80 percent of global almond production all at once. And there are knock-on effects. All of the almond blossoms need to be pollinated in the same narrow window of time, because all the almond trees grow in the same soil and experience the same weather. The huge volume of simultaneous pollination necessitates shipping in beehives from all over America for the short pollination window. There is an epidemic of honeybees dying in America, creating concerns about the US honeybee population’s ability to pollinate the wide variety of plants that need the bees’ busy work. One theory for the elevated honeybee mortality rates is that beehives are trucked around America for these monoculture pollinations like never before, and that this is stressful for the bees.

Rather than producing resilient ecosystems, our obsession with efficiency proxies is producing fragile monocultures, potentially vulnerable to catastrophic failure. No doubt the monocultures are efficient in a narrow sense, but that efficiency has a dark side. The problem is, we have become so convinced that efficiency-at-all-costs is a universal good that we have lost sight of its risks. We have stopped seeking dynamic balance entirely.

And we shape our infrastructure to support our monocultures. There is now an entire infrastructure designed to bring bees to the Central Valley of California and take them back to their home locations afterward.¹⁵ Once that infrastructure is in place, it makes it considerably easier to add still more acreage of almond trees. The effect (an efficient infrastructure for almond growing in a single locale) becomes the cause of more of the effect (more almond growing in a single locale).

An almond crisis may not happen. But something like it already has. Irish farmers discovered during the seventeenth century that the South American potato grew well in Ireland’s soil and climate. They also found that one particular variety, the “lumper,” was more productive than any alternative food crop they could grow. By the 1840s, nearly every Irish farm was devoted to growing lumpers, which had come to account for the bulk of daily food consumption for over half of the Irish population. Then, in 1845, Irish potatoes fell prey to a water mold called Phytophthora infestans that decimated the potato crop. It wasn’t a situation in which just some of the potatoes fell prey—virtually all did. And since potatoes were the primary source of food for most poor Irish families, the mold was responsible for the deaths (by malnourishment) of approximately 1 million of the 1844 population of 8.8 million, not to mention the emigration of a further 2 million. The devastation was so great that by the time Ireland gained its independence, in 1921, the population was still less than half its 1844 level.¹⁶

The Fundamental Challenge

A Pareto distribution of outcomes wouldn’t be a big problem if everyone just cared about the overall growth rate. It is, to be sure, lower than it was in America’s first two hundred years, but that is not altogether surprising. We expect economies to grow faster when young and to slow down as they mature. The really big problem lies in the distribution of the benefits of that growth. American capitalism requires the consent of the majority of American citizens. As I’ve pointed out already, if 51 percent or more of citizens experience stagnation rather than growth, they will at some point defect from capitalism. They didn’t in the Great Depression, but that doesn’t mean that they won’t this time. Meanwhile, the grains of sand continue to fall. In a Gaussian distribution, that wouldn’t matter: the independence of the occurrences provides stability—with time and more occurrences, such a distribution becomes only more stable. But economic outcomes are not independent: effects are causes of more of the same effects. So, if you are already in the top 1 percent, that position provides more opportunity to accrue still more benefits—and it is that interdependence among the grains of sand that can transform a Gaussian distribution into a Pareto one. As the effects multiply, the Pareto distribution will become only more extreme. That is the great challenge for the future of democratic capitalism. It is why we need a better model for democratic capitalism—one that recognizes it is a natural system that needs constant tweaking rather than a perfectible machine. That is the subject of the next chapter.