Capital in the Twenty-First Century

by Thomas Piketty

  Is it possible to imagine societies in which the concentration of income is much greater? Probably not. If, for example, the top decile appropriates 90 percent of each year’s output (and the top centile took 50 percent just for itself, as in the case of wealth), a revolution will likely occur, unless some peculiarly effective repressive apparatus exists to keep it from happening. When it comes to the ownership of capital, such a high degree of concentration is already a source of powerful political tensions, which are often difficult to reconcile with universal suffrage. Yet such capital concentration might be tenable if the income from capital accounts for only a small part of national income: perhaps one-fourth to one-third, or sometimes a bit more, as in the Ancien Régime (which made the extreme concentration of wealth at that time particularly oppressive). But if the same level of inequality applies to the totality of national income, it is hard to imagine that those at the bottom will accept the situation permanently.

  That said, there are no grounds for asserting that the upper decile can never claim more than 50 percent of national income or that a country’s economy would collapse if this symbolic threshold were crossed. In fact, the available historical data are far from perfect, and it is not out of the question that this symbolic limit has already been exceeded. In particular, it is possible that under the Ancien Régime, right up to the eve of the French Revolution, the top decile did take more than 50 percent and even as much as 60 percent or perhaps slightly more of national income. More generally, this may have been the case in other traditional rural societies. Indeed, whether such extreme inequality is or is not sustainable depends not only on the effectiveness of the repressive apparatus but also, and perhaps primarily, on the effectiveness of the apparatus of justification. If inequalities are seen as justified, say because they seem to be a consequence of a choice by the rich to work harder or more efficiently than the poor, or because preventing the rich from earning more would inevitably harm the worst-off members of society, then it is perfectly possible for the concentration of income to set new historical records. That is why I indicate in Table 7.3 that the United States may set a new record around 2030 if inequality of income from labor—and to a lesser extent inequality of ownership of capital—continue to increase as they have done in recent decades. The top decile would them claim about 60 percent of national income, while the bottom half would get barely 15 percent.

  I want to insist on this point: the key issue is the justification of inequalities rather than their magnitude as such. That is why it is essential to analyze the structure of inequality. In this respect, the principal message of Tables 7.1–3 is surely that there are two different ways for a society to achieve a very unequal distribution of total income (around 50 percent for the top decile and 20 percent for the top centile).

  The first of these two ways of achieving such high inequality is through a “hyperpatrimonial society” (or “society of rentiers”): a society in which inherited wealth is very important and where the concentration of wealth attains extreme levels (with the upper decile owning typically 90 percent of all wealth, with 50 percent belonging to the upper centile alone). The total income hierarchy is then dominated by very high incomes from capital, especially inherited capital. This is the pattern we see in Ancien Régime France and in Europe during the Belle Époque, with on the whole minor variations. We need to understand how such structures of ownership and inequality emerged and persisted and to what extent they belong to the past—unless of course they are also pertinent to the future.

  The second way of achieving such high inequality is relatively new. It was largely created by the United States over the past few decades. Here we see that a very high level of total income inequality can be the result of a “hypermeritocratic society” (or at any rate a society that the people at the top like to describe as hypermeritocratic). One might also call this a “society of superstars” (or perhaps “supermanagers,” a somewhat different characterization). In other words, this is a very inegalitarian society, but one in which the peak of the income hierarchy is dominated by very high incomes from labor rather than by inherited wealth. I want to be clear that at this stage I am not making a judgment about whether a society of this kind really deserves to be characterized as “hypermeritocratic.” It is hardly surprising that the winners in such a society would wish to describe the social hierarchy in this way, and sometimes they succeed in convincing some of the losers. For present purposes, however, hypermeritocracy is not a hypothesis but one possible conclusion of the analysis—bearing in mind that the opposite conclusion is equally possible. I will analyze in what follows how far the rise of labor income inequality in the United States has obeyed a “meritocratic” logic (insofar as it is possible to answer such a complex normative question).

  At this point it will suffice to note that the stark contrast I have drawn here between two types of hyperinegalitarian society—a society of rentiers and a society of supermanagers—is naïve and overdrawn. The two types of inequality can coexist: there is no reason why a person can’t be both a supermanager and a rentier—and the fact that the concentration of wealth is currently much higher in the United States than in Europe suggests that this may well be the case in the United States today. And of course there is nothing to prevent the children of supermanagers from becoming rentiers. In practice, we find both logics at work in every society. Nevertheless, there is more than one way of achieving the same level of inequality, and what primarily characterizes the United States at the moment is a record level of inequality of income from labor (probably higher than in any other society at any time in the past, anywhere in the world, including societies in which skill disparities were extremely large) together with a level of inequality of wealth less extreme than the levels observed in traditional societies or in Europe in the period 1900–1910. It is therefore essential to understand the conditions under which each of these two logics could develop, while keeping in mind that they may complement each other in the century ahead and combine their effects. If this happens, the future could hold in store a new world of inequality more extreme than any that preceded it.21

  Problems of Synthetic Indices

  Before turning to a country-by-country examination of the historical evolution of inequality in order to answer the questions posed above, several methodological issues remain to be discussed. In particular, Tables 7.1–3 include indications of the Gini coefficients of the various distributions considered. The Gini coefficient—named for the Italian statistician Corrado Gini (1884–1965)—is one of the more commonly used synthetic indices of inequality, frequently found in official reports and public debate. By construction, it ranges from 0 to 1: it is equal to 0 in case of complete equality and to 1 when inequality is absolute, that is, when a very tiny group owns all available resources.

  In practice, the Gini coefficient varies from roughly 0.2 to 0.4 in the distributions of labor income observed in actual societies, from 0.6 to 0.9 for observed distributions of capital ownership, and from 0.3 to 0.5 for total income inequality. In Scandinavia in the 1970s and 1980s, the Gini coefficient of the labor income distribution was 0.19, not far from absolute equality. Conversely, the wealth distribution in Belle Époque Europe exhibited a Gini coefficient of 0.85, not far from absolute inequality.22

  These coefficients—and there are others, such as the Theil index—are sometimes useful, but they raise many problems. They claim to summarize in a single numerical index all that a distribution can tell us about inequality—the inequality between the bottom and the middle of the hierarchy as well as between the middle and the top or between the top and the very top. This is very simple and appealing at first glance but inevitably somewhat misleading. Indeed, it is impossible to summarize a multidimensional reality with a unidimensional index without unduly simplifying matters and mixing up things that should not be treated together. The social reality and economic and political significance of inequality are very different at different levels of the distribution, and it
is important to analyze these separately. In addition, Gini coefficients and other synthetic indices tend to confuse inequality in regard to labor with inequality in regard to capital, even though the economic mechanisms at work, as well as the normative justifications of inequality, are very different in the two cases. For all these reasons, it seemed to me far better to analyze inequalities in terms of distribution tables indicating the shares of various deciles and centiles in total income and total wealth rather than using synthetic indices such as the Gini coefficient.

  Distribution tables are also valuable because they force everyone to take note of the income and wealth levels of the various social groups that make up the existing hierarchy. These levels are expressed in cash terms (or as a percentage of average income and wealth levels in the country concerned) rather than by way of artificial statistical measures that can be difficult to interpret. Distribution tables allow us to have a more concrete and visceral understanding of social inequality, as well as an appreciation of the data available to study these issues and the limits of those data. By contrast, statistical indices such as the Gini coefficient give an abstract and sterile view of inequality, which makes it difficult for people to grasp their position in the contemporary hierarchy (always a useful exercise, particularly when one belongs to the upper centiles of the distribution and tends to forget it, as is often the case with economists). Indices often obscure the fact that there are anomalies or inconsistencies in the underlying data, or that data from other countries or other periods are not directly comparable (because, for example, the tops of the distribution have been truncated or because income from capital is omitted for some countries but not others). Working with distribution tables forces us to be more consistent and transparent.

  The Chaste Veil of Official Publications

  For similar reasons, caution is in order when using indices such as the interdecile ratios often cited in official reports on inequality from the OECD or national statistical agencies. The most frequently used interdecile ratio is the P90/P10, that is, the ratio between the ninetieth percentile of the income distribution and the tenth percentile.23 For example, if one needs to earn more than 5,000 euros a month to belong to the top 10 percent of the income distribution and less than 1,000 euros a month to belong to the bottom 10 percent, then the P90/P10 ratio is 5.

  Such indices can be useful. It is always valuable to have more information about the complete shape of the distribution in question. One should bear in mind, however, that by construction these ratios totally ignore the evolution of the distribution beyond the ninetieth percentile. Concretely, no matter what the P90/P10 ratio may be, the top decile of the income or wealth distribution may have 20 percent of the total (as in the case of Scandinavian incomes in the 1970s and 1980s) or 50 percent (as in the case of US incomes in the 2010s) or 90 percent (as in the case of European wealth in the Belle Époque). We will not learn any of this by consulting the publications of the international organizations or national statistical agencies who compile these statistics, however, because they usually focus on indices that deliberately ignore the top end of the distribution and give no indication of income or wealth beyond the ninetieth percentile.

  This practice is generally justified on the grounds that the available data are “imperfect.” This is true, but the difficulties can be overcome by using adequate sources, as the historical data collected (with limited means) in the World Top Incomes Database (WTID) show. This work has begun, slowly, to change the way things are done. Indeed, the methodological decision to ignore the top end is hardly neutral: the official reports of national and international agencies are supposed to inform public debate about the distribution of income and wealth, but in practice they often give an artificially rosy picture of inequality. It is as if an official government report on inequalities in France in 1789 deliberately ignored everything above the ninetieth percentile—a group 5 to 10 times larger than the entire aristocracy of the day—on the grounds that it was too complex to say anything about. Such a chaste approach is all the more regrettable in that it inevitably feeds the wildest fantasies and tends to discredit official statistics and statisticians rather than calm social tensions.

  Conversely, interdecile ratios are sometimes quite high for largely artificial reasons. Take the distribution of capital ownership, for example: the bottom 50 percent of the distribution generally own next to nothing. Depending on how small fortunes are measured—for example, whether or not durable goods and debts are counted—one can come up with apparently quite different evaluations of exactly where the tenth percentile of the wealth hierarchy lies: for the same underlying social reality, one might put it at 100 euros, 1,000 euros, or even 10,000 euros, which in the end isn’t all that different but can lead to very different interdecile ratios, depending on the country and the period, even though the bottom half of the wealth distribution owns less than 5 percent of total wealth. The same is only slightly less true of the labor income distribution: depending on how one chooses to treat replacement incomes and pay for short periods of work (for example, depending on whether one uses the average weekly, monthly, annual, or decadal income) one can come up with highly variable P10 thresholds (and therefore interdecile ratios), even though the bottom 50 percent of the labor income distribution actually draws a fairly stable share of the total income from labor.24

  This is perhaps one of the main reasons why it is preferable to study distributions as I have presented them in Tables 7.1–3, that is, by emphasizing the shares of income and wealth claimed by different groups, particularly the bottom half and the top decile in each society, rather than the threshold levels defining given percentiles. The shares give a much more stable picture of reality than the interdecile ratios.

  Back to “Social Tables” and Political Arithmetic

  These, then, are my reasons for believing that the distribution tables I have been examining in this chapter are the best tool for studying the distribution of wealth, far better than synthetic indices and interdecile ratios.

  In addition, I believe that my approach is more consistent with national accounting methods. Now that national accounts for most countries enable us to measure national income and wealth every year (and therefore average income and wealth, since demographic sources provide easy access to population figures), the next step is naturally to break down these total income and wealth figures by decile and centile. Many reports have recommended that national accounts be improved and “humanized” in this way, but little progress has been made to date.25 A useful step in this direction would be a breakdown indicating the poorest 50 percent, the middle 40 percent, and the richest 10 percent. In particular, such an approach would allow any observer to see just how much the growth of domestic output and national income is or is not reflected in the income actually received by these different social groups. For instance, only by knowing the share going to the top decile can we determine the extent to which a disproportionate share of growth has been captured by the top end of the distribution. Neither a Gini coefficient nor an interdecile ratio permits such a clear and precise response to this question.

  I will add, finally, that the distribution tables whose use I am recommending are in some ways fairly similar to the “social tables” that were in vogue in the eighteenth and early nineteenth centuries. First developed in Britain and France in the late seventeenth century, these social tables were widely used, improved, and commented on in France during the Enlightenment: for example, in the celebrated article on “political arithmetic” in Diderot’s Encyclopedia. From the earliest versions established by Gregory King in 1688 to the more elaborate examples compiled by Expilly and Isnard on the eve of the French Revolution or by Peuchet, Colqhoun, and Blodget during the Napoleonic era, social tables always aimed to provide a comprehensive vision of the social structure: they indicated the number of nobles, bourgeois, gentlemen, artisans, farmers, and so on along with their estimated income (and sometimes wealth); the same authors also compiled the ear
liest estimates of national income and wealth. There is, however, one essential difference between these tables and mine: the old social tables used the social categories of their time and did not seek to ascertain the distribution of wealth or income by deciles and centiles.26

  Nevertheless, social tables sought to portray the flesh-and-blood aspects of inequality by emphasizing the shares of national wealth held by different social groups (and, in particular, the various strata of the elite), and in this respect there are clear affinities with the approach I have taken here. At the same time, social tables are remote in spirit from the sterile, atemporal statistical measures of inequality such as those employed by Gini and Pareto, which were all too commonly used in the twentieth century and tend to naturalize the distribution of wealth. The way one tries to measure inequality is never neutral.

  {EIGHT}

  Two Worlds

  I have now precisely defined the notions needed for what follows, and I have introduced the orders of magnitude attained in practice by inequality with respect to labor and capital in various societies. The time has now come to look at the historical evolution of inequality around the world. How and why has the structure of inequality changed since the nineteenth century? The shocks of the period 1914–1945 played an essential role in the compression of inequality, and this compression was in no way a harmonious or spontaneous occurrence. The increase in inequality since 1970 has not been the same everywhere, which again suggests that institutional and political factors played a key role.