INTRODUCTION

This chapter gives an overview of the evidence on long-run trends in the distribution of personal income and wealth. Our focus will be on empirical estimates of the respective distributions, but we will also cover some ideas that aim to explain the observed patterns.

The kinds of questions we try to answer in this chapter are: What do we know (and how do we know) about the distribution of income and wealth over time? Are there common trends across countries or over the path of development? How do the facts relate to proposed theories about changes in inequality? We will mainly draw on the advances made in the field over the past decade, but before outlining the contents of the chapter and its limitations, we want to recall some points in the development of evi­dence on long-run inequality leading up to the recent research.^[212]

7.1.1 From the Kuznets Series, to Household Surveys, and Back Again

In the very beginning of his famous presidential address to the American Economic Asso­ciation in 1954, Simon Kuznets outlined some ideals concerning the data required to study long-term changes in inequality (Kuznets, 1955). The “economist’s pipe dream” that he described roughly corresponds to what we today would call a detailed, individual panel data set, preferably spanning several generations.

In many ways the development of inequality data for a long time after Kuznets' well- known speech focused on this “wish list.” Even though important advances were, of course, made with respect to historical data, it is fair to say that the focus was on the con­struction of contemporary national household surveys and individual micropanel data sets.^[213] Eventually much effort also went into making such data comparable across countries in projects such as the Luxembourg Income Study (LIS) and its more recent companion the Luxembourg Wealth Study. Building on these and other similar projects, compilations of data such as the World Income Inequality Database (WIID) have also been put together.^[214] This development has indeed changed empirical inequality research for the better and made it possible to address a number of new and important questions. But the relative focus on microdata shifted attention away from some issues, and in particular questions regarding long-run developments. Given the relatively recent nature of most household survey data and microdata in general, “the long run” based to these sources naturally becomes quite limited, typically not covering more than the last couple of decades.^[215] Such a relatively short time span is unfortunate because several issues concerned with economic development and structural change require a much longer time horizon.

However, recent research has changed things dramatically. Starting with the path­breaking work of Piketty (2001a, 2003), which extends the methods first used in the sem­inal work by Kuznets (1953) to generate a series of top income shares spanning the entire twentieth century in France using income tax data, similar efforts have followed for many countries.

Most of the series span the whole of the twentieth century, sometimes even longer, making the resulting data set unique in its ability to address long-run issues. There are also other features of the data such as their relatively high frequency (often yearly), the pos­sibility to decompose income by source, and the possibilities to study changes within the top of the distribution that have proven to be of importance and, as we shall discuss in more detail later, have led to new insights about inequality developments over the long run. This renewed interest in the long run and the reevaluation of historical sources has also led to new studies on the historical trends in the wealth distribution (e.g., Dell et al., 2007; Kopczuk and Saez, 2004; Piketty et al., 2006; see further Section 7.3).

The body of work extending and generalizing Kuznets' pioneering research is often labeled according to its focus on the top of the distribution. The “top income literature” is of course a correct description in the sense that it is based on observing only high- income fractions of the population (typically roughly the top decile and sometimes an even smaller share) and then relating their incomes to estimates of total income. From this it follows that top income shares cannot say anything about changes within a large share of the total population. But it does not follow that this data is only about the rich. As we will outline in more detail later, the top income literature is a contribution to both our understanding of long-run changes in overall inequality, as well as a more detailed understanding of developments within the top.

Finally, one should remember that it is not always a matter of choosing the right inequality measure for the question at hand. In fact, when it comes to the study of long-run inequality, the availability of any data at all is often the binding constraint. In such a situation the relationship between different measures becomes important, and we want to know things like: “What are the relationships between different inequality measures?” and “To what extent can this measure serve as a proxy for what we would ideally like to observe?” In the end, the approach to what we know and can know about inequality over the long run will have to be pragmatic. Such an approach calls for cautious interpretation, but not for resignation. We believe, using the words of Kuznets (1955, p. 4), that even “if the trends in the income structure can be discerned but dimly,” we should continue to improve on our informed guesses. This is far from saying that the best we can do is to patch together scattered observations over time, using different sources and methods. In fact, many recent insights points exactly to the opposite. In the end we need to combine an understanding about what we are, in fact, observing, how different measures relate to each other, and an understanding of how they relate to the model or theory we wish to test.

7.1.2 Outline of the Chapter

This chapter has three parts in addition to this introduction: one on the trends in long-run income inequality, a second on trends in long-run wealth inequality, and a third on potential explanations of these trends and how they relate to some of the theories about what determines inequality.

7.1.2.1 Top Income Shares and Other Measures of Long-Run Income Inequality

In Section 7.2 we focus mainly on the new evidence on long-run income inequality that has come out of the top incomes project, including some new data points.^[217] This means that income inequality is generally in terms of total income, that is, income from all sources, before taxes and most transfers.

Thereafter we move on to relating the results based on top income shares with results based on other sources and measures of inequality. We consider both top share measures using somewhat different sources and methods, as well as other estimates of historical inequality based on other measures (wage dispersion across occupations, factor price dif­ferentials and differences in life prospects). In particular, we discuss and update the evi­dence on the issue of how good a proxy top income shares are for other measures of inequality. Putting everything together, we attempt to summarize the overall picture of income inequality developments for the period 1750—2010.

7.1.2.2 Long-Run Trends in the Wealth Distribution

In Section 7.3 we present the evidence on long-run developments of wealth inequality. Similar to the discussion of income inequality trends, we begin by reviewing the different data sources and empirical methods used to estimate the distribution of wealth over time.

But there are also some important differences between studying income and wealth concentration. Personal wealth is more difficult both to define and to measure, and the nature of wealth data is also different from income data. Even though information on the distribution of wealth has been collected throughout history (the Doomsday Book from 1086 in England being an early and well-known example), wealth holdings have not typ­ically been taxed directly in a systematic way. Assets have instead mostly been taxed on their transfer and in particular at the time of death. Indeed, most of the information we have on the distribution of distribution comes from inheritance or estate tax data, some­times complemented by wealth data collected in connection to population surveys. The section describes how researchers have handled these challenges in estimating the wealth distribution and to what extent meaningful cross-country comparisons can be made.

After having discussed methodology, we move on to presenting the broad results emerging from this work covering 10 of today’s industrialized economies from their respective eras of industrialization until the present. For a few countries (Finland, the Netherlands, Norway, and Sweden) the chapter also presents some new estimates of wealth concentration.^[219]

7.1.2.3 Searching for Explanations

In Section 7.4, we then discuss the possible explanations behind the observed facts. How should we relate the shifts in the income and wealth distributions over time to other developments in society? To what extent are there global forces and events that affect all countries in similar ways (possibly with some time lag between countries)? What the­ories can shed light on shifts in capital incomes, what theories could explain increasing top wages? How should we think about the development of total income stemming from both wages and capital? What evidence do we have from regression analysis?

We begin by discussing some broad topics often suggested as a cause (and sometimes consequence) of inequality and sketch how the developments of these relate to our evi­dence. In particular, we will look at how our series correspond to broad global develop­ments such as globalization, technological revolutions, wars and economic shocks, and patterns of economic growth. We then focus on more some specific aspects. First, we look at theories emphasizing capital incomes and also the interactions between earned income and capital as well as the cumulative effects of taxation. These things were all of key importance for the decline of top shares in the first half of the twentieth century and for the lack of recovery after the wars. We then consider some mechanisms that have been suggested to explain increased top wages such as skill-biased technological change, the rise in executive pay and related so-called super-star theories. These have all been suggested as important factors in the recent rise in top shares in many countries. Finally, we review some insights from econometric studies trying to use the new long-run inequality data to shed light on the developments.

Clearly our coverage of possible theories will be both selective and incomplete. In the end it is based on our subjective reading of which aspects we think are key for under­standing the long-run developments of inequality, especially in light of the new evidence produced in the past decade. Furthermore, much of what we write about has been cov­ered in previous overviews and surveys of the top incomes literature (Atkinson and Piketty, 2007, 2010; Atkinson et al., 2010, 2011; Leigh, 2009; Piketty, 2005; Piketty and Saez, 2006), overviews of the changing earnings distribution (Atkinson, 2008a) and in overviews on wealth concentration trends (Atkinson, 2008b; Davies and Shorrocks, 2000; Ohlsson et al., 2008; Wolff, 1996). In general, our aim is to focus on the most recent work in the field building on previous surveys such as Lindert (2000) and Morrison (2000).

7.1.3 What Is This Chapter Not About?

There is a lot of work and several issues regarding inequality over the long run that this chapter does not cover. As we see it, there are four major themes that we do not address but that are still closely related to what we discuss. Two of these omitted themes concern the descriptive scope of our chapter, whereas the other two relate more to the attempts to understand the developments.

First, we will not deal with issues of mobility but instead focus on repeated cross­sections of data.^[220] A distribution where individuals constantly move in and out of the top (or bottom) of the distribution and where an individual’s position 1 year says nothing of his or her position the next year is clearly very different to one where every individual keeps his or her place over time. Reality is obviously characterized by something in between the two extremes, but importantly the few studies that have been able to directly address this question (or aspects of it) conclude that trends in cross-sectional data are not driven by changes in mobility and do capture actual inequality.^[221] In short, even if repeated cross sections of inequality, in theory, could be misleading when discussing changes in inequality over time, this does not seem to be the case in practice.

Second, we will restrict our study in time to a period starting roughly at the beginning of the British Industrial Revolution (with data this far back being limited to a few data points for a small number of countries only), and with more comprehensive data starting in the beginning of the twentieth century. Recently there has been a lot of interesting work devising ingenious methods of estimating distributional outcomes in premodern societies.^[222] All of this work certainly adds to our understanding of inequality in historical episodes as well as its long-run evolution. However, because these earlier figures are mostly based on occupational groupings or social class, we think one should be cautious when connecting our series to the estimates in earlier periods.

Third, we will not review theories about long-run inequality developments in any detail or with any attempt at fully coverage. We will instead outline some ideas and sug­gested mechanisms in a highly selective way to outline aspects that can help explain the key developments we find in the data.^[223]

Fourth, we primarily discuss inequality as a left-hand side variable in an econometric sense. This means that our discussion will mainly be one about how we can understand the developments of inequality and its determinants and not so much about the conse­quences of inequality on other developments such as, for example, economic growth, political outcomes, or health.^[224] Of course, such a distinction is somewhat artificial in the sense that the distribution of resources at any point forms the basis for economic and political decisions, resulting in outcomes that then create the distribution for the next period.^[225] Many questions are, thus, ultimately not about one causing the other, but rather about the dynamic interplay over time. Nevertheless, it is often useful to separate ques­tions in terms of how we think about the causality. In this separation we focus on under­standing how and why inequality has changed, not on the consequences of inequality on other developments in society.

7.2.