Life-Cycle Wage Growth Across Countries∗ David Lagakos, Benjamin Moll, Tommaso Porzio, Nancy Qian and Todd Schoellman May 22, 2016

Abstract This paper documents how life-cycle wage growth varies across countries. We harmonize repeated cross-sectional surveys from a set of countries of all income levels and then use these data to measure how wages rise with potential experience. Our main finding is that experience-wage profiles are on average twice as steep in rich countries as in poor countries. In addition, more educated workers have steeper experience-wage profiles on average than those with less education; this accounts for around one-third of cross-country differences in aggregate profiles. Our findings are consistent with theories in which workers in poor countries accumulate less human capital over the life cycle and theories in which more severe search frictions in poor countries prevent workers from climbing the job ladder.

∗ Lagakos: UCSD and NBER, [email protected]; Moll: Princeton University and NBER, [email protected]; Porzio: Yale University, [email protected]; Qian: Yale University, NBER, CEPR and BREAD, [email protected]. Schoellman: Arizona State University, [email protected]. This version supersedes an earlier version of the paper entitled “Experience Matters: Human Capital and Development Accounting.” We thank four anonymous referees and the editor, Erik Hurst, for numerous helpful comments. We also thank Daron Acemoglu, Mark Aguiar, Paco Buera, Francesco Caselli, Thomas Chaney, Sylvain Chassang, Angus Deaton, Mike Golosov, Fatih Guvenen, Lutz Hendricks, Erik Hurst, Joe Kaboski, Nobu Kiyotaki, Pete Klenow, Jonathan Parker, Richard Rogerson, Paul Romer, Sam Schulhofer-Wohl, David Sraer, David Weil and Fabrizio Zillibotti for helpful suggestions and criticisms, plus seminar participants at Columbia, Chicago, CUNY, EIEF, EUI, Harvard, Laval, LSE, MIT, Northwestern, Princeton, Rochester, UQAM, USC, Warwick, the World Bank and conference participants at BREAD, DFG, NEUDC, the NBER Summer Institute Growth Workshop, the NBER Summer Institute EFG Meeting, tbe SED Annual Meetings and the Human Capital conference at Washington University in St Louis.

1

Introduction This paper documents how life-cycle wage growth varies across countries. It is well known that wages

grow substantially over the life cycle in the United States and other advanced economies. However, there is little comparable evidence from less developed countries. This is unfortunate, as cross-country differences in life-cycle wage growth are key for addressing questions such as the importance of human capital and labor market frictions for explaining cross-country income differences (Manuelli and Seshadri, 2014; Klenow and Rodriguez-Clare, 1997; Bils and Klenow, 2000; Caselli, 2005; Burdett, 1978; Jovanovic, 1984). We fill this gap by measuring life-cycle wage growth in both low- and high-income countries. We use representative large-sample household surveys from eighteen countries with individual-level data on educational attainment, labor earnings and the number of hours worked. These data allow us to construct similar measures of hourly wages and potential experience for all countries in our sample. Our main finding is that wages increase substantially more over the life cycle in rich countries than in poor countries. We take three alternative approaches to measuring life-cycle wage growth. The first and simplest approach is to construct cross-sectional experience-wage profiles where experience is measured as years of potential experience, i.e. years elapsed since finishing school. To do this, we compute mean wages for each five-year experience bin relative to the bin with the least experience. We show that profiles are steeper in rich countries than in poor countries, with differences that are statistically and economically significant: wages almost double over the life cycle in rich countries whereas they increase by only around fifty percent in poor countries. Put differently, wages rise almost twice as much in rich countries as in poor ones. Our second approach follows Mincer (1974), which allows us to control for years of schooling in the standard way. It also provides a framework for addressing the well-known challenge to estimating life-cycle profiles in age (or potential experience), which is that age is collinear with time and birth cohort (i.e., calendar year and birth year). This means that one cannot separately identify the effect of age from the effect of time or birth cohort without further restrictions, a point that has not been addressed in the existing cross-country literature. We begin by following the standard approach outlined by Hall (1968) and Deaton (1997). They show that experience or age profiles can be estimated if one assumption about the source of aggregate income growth is imposed. We find that if time effects explain half or more of growth, then wages rise more over the life cycle in rich countries. However, there are two challenges: one does not know in general what fraction of growth is due to time effects, and this fraction could differ across countries. Our third and preferred approach draws on economic theory to address this challenge. We draw on a common prediction of theories of life-cycle wage growth that there should be little or no effect of experience

1

on wages near the end of the life cycle.1 For example, human capital theory predicts that the incentive to invest in human capital formation declines at the end of the life cycle, while search and matching theory predicts that the incentive to search for better matches declines similarly. Our insight, based on the work of Heckman, Lochner and Taber (1998), is that this theoretical prediction is sufficient to disentangle experience, time and cohort effects. Intuitively, if we follow a fixed cohort across multiple cross-sections for the last years of their working life, then we rule out both cohort effects (by construction) and experience effects (by the theoretical result above), allowing us to attribute any wage changes to time effects. Once we have recovered the aggregate time effects, it is straightforward to estimate the experience and cohort effects of workers who are not near the end of the life cycle. Applying this method, we again find that estimated experience-wage profiles are substantially steeper in rich countries than in poor countries. We also experiment with variants of this idea in which wage profiles are assumed to decrease at the end of the life cycle, e.g. due to human capital depreciation, and find similar results. We provide evidence that our findings are robust to a number of alternative measurement assumptions and sample restrictions. While our benchmark results focus on full-time male wage workers, we show that experience-wage profiles are steeper in rich countries when we include women, part-time workers and the self-employed. To address concerns that our findings are driven by mismeasurement of experience, we show that our results are similar when using an alternative measure of experience based on age-and-educationspecific employment rates. Furthermore, adding plausible amounts of measurement error to the age and education variables in rich countries does not cause the profiles of rich countries to look like those of poor ones. Another natural concern is that our results are driven by differential selection into (or out of) our sample across countries. We use panel data from the United States and Mexico to show that this is unlikely. Specifically, we show that selection into or out of wage employment, private-sector employment and full-time employment have negligible effects on the estimated relationship between experience and wages in the United States and Mexico. We next explore one natural hypothesis for why experience-wage profiles are steeper in richer countries, which is that richer countries have a greater fraction of educated workers. While Mincer (1974) found that U.S. experience-wage profiles were similar for different education groups, more recent work has tended to find that more educated workers have steeper experience-wage profiles (Lemieux, 2006). Overall, we find that across our eighteen countries, more educated workers have steeper experience-wage profiles on average than less educated workers, and that cross-country differences in the distribution of educational attainment account for around one-third of the flatter aggregate experience profiles in poor countries. This implies 1 For example, Rubinstein and Weiss (2006) review the literature on life-cycle wage growth and explain in detail the three main mechanisms emphasized in this literature (human capital investment, search and learning), noting that all three have “similar implications with respect to the behavior of mean wages, implying rising and concave wage profiles” (p.4).

2

that education is likely to be an important factor for explaining cross-country differences in life-cycle wage growth, but also suggests that other factors play important roles. We conclude by returning to the interpretation and broader implications of our findings. Three popular theories of life-cycle earnings patterns and wage dynamics are human capital accumulation, on-the-job search and long-term contracts. While it is hard to provide a definitive conclusions about which theory best explains our findings, several pieces of evidence point to human capital and search frictions as playing important roles. First, when we look at profiles by broad occupation category, we find robust evidence that manual occupations have flatter profiles than cognitive occupations. Since manual occupations likely have less scope for lifetime learning, and since around half of workers in poor countries are in manual occupations, this suggests a human-capital interpretation. Second, we find that wage variances generally increase over the life cycle, with some evidence of a dip for early experience levels in rich countries. We note that this is predicted by several classes of theories of human capital and search-and-matching frictions. Finally, we look at wage profiles for day laborers, who are not engaged in long-term wage contracts, and find that, in the poor countries for which we have data, these are again flatter than in rich countries. Both human capital and search theories suggest that our findings may help explain cross-country income differences. Through the lens of human capital theory, our findings point to a much greater role for human capital in accounting for cross-country income differences than suggested by previous studies, in particular those of Klenow and Rodriguez-Clare (1997), Bils and Klenow (2000) and Caselli (2005). Specifically, our findings are consistent with workers in rich countries accumulating more human capital over the life cycle than workers in poor countries. This is exactly the theoretical prediction of Manuelli and Seshadri (2014). Through the lens of search and matching theory, our findings suggest less labor market fluidity in poor countries, which prevents workers from climbing the job ladder and may act as a form of misallocation: workers are less able to move to better jobs that fit their skills in poor countries. This misallocation could once again be an important contributor to cross-country income differences, in the spirit of Hsieh et al. (2013). We are not the first to examine the relationship between wages and experience across countries. Our findings contrast with those of earlier work, in particular Psacharopoulos (1994) and Bils and Klenow (2000), who found no relationship between returns to experience and GDP per capita. Our conclusion differs for three main reasons. First, previous studies focus on earnings, which conflates growth in hourly wages and growth in hours worked. Second, some of the earlier estimates draw on small, non-representative samples and the cross-country comparisons combine estimates from underlying studies with different specifications and sampling frames. In contrast, we restrict our attention to comparable nationally representative samples of 5,000 or more full-time, male, private-sector workers. Third, the previous literature focuses exclusively on 3

cross-sectional estimates – often a single cross section – and does not address the potentially confounding influences of cohort and time effects. This paper is organized as follows. Section 2 describes our household-survey data. Section 3 documents that simple cross-sectional experience-wage profiles are flatter in poorer countries. Section 4 measures experience-wage profiles using the Deaton-Hall and Heckman-Lochner-Taber methods. Section 5 investigates the robustness of our estimated experience-wage profiles. Section 6 considers interactions between schooling and experience, and the role of schooling in accounting for aggregate experience profiles. Section 7 discusses broader implications and interpretations of our findings. Section 8 concludes.

2

Data Our analysis uses large-sample household survey data from eighteen countries. The surveys we use satisfy

three criteria: (i) they are nationally representative and have at least 5,000 observations on full-time males in the private sector; (ii) they contain individual labor earnings; and (iii) they contain individual data on the number of hours worked. The large sample size in (i) is important for estimates which require us to cut the sample into multiple groups, such as our estimates of life-cycle wage growth by educational attainment later in the paper. Restrictions (ii) and (iii) are important because they allow us to compute individual-level wages. Note that all of our data have demographic as well as educational attainment information on all individuals. We focus much of our analysis on a sample of eight core countries that satisfy restrictions (i)(iii) and additionally have repeated cross-sections spanning fifteen or more years. This additional restriction is necessary for our method to disentangle experience, time, and cohort effects in Section 4.2 Table 1 lists the countries in our sample, the income level of each country, the data source, the years of coverage and whether each country is in the core sample. The countries in both the full and core samples comprise a wide range of income levels, from the United States and Germany to Bangladesh (in the extended sample) or Jamaica (in the core sample). Please see Table 1 and Appendix A.1 for the source of each survey. The main limitation in terms of data coverage is that we do not observe the poorest countries in the world, such as those in Sub-Saharan Africa, since data from these countries do not satisfy the criteria described earlier. By “rich countries” we mean those with greater than $20,000 per capita income in 2011 at PPP (from the World Bank’s World Development Indicators), and by “poor countries” we mean those with per capita income below this threshold. The main outcome variable is an individual’s wage, which we define to be his labor earnings divided 2 An earlier version of our paper (Lagakos et al., 2012) used data from thirty-five countries. For fourteen of these countries, the data did not satisfy all of the criteria (i) to (iii) listed above. An additional three countries were removed because they reported income in a way that was inconsistent with all of the other countries. Details are available upon request. However, note that our main finding that experience-wage profiles are steeper in rich countries is still present in this expanded set of countries.

4

by the number of hours that he worked. In most countries, we observe earnings during the month prior to the survey and hours worked during the week prior to the survey. For the United States, Canada, Brazil and Jamaica, we observe labor income and hours worked at an annual frequency. We restrict attention to individuals with zero to forty years of experience who have positive labor income and non-missing age and schooling information. In all surveys, we impute the years of schooling using educational attainment data. For all countries, we examine earnings and wages in local currency units of the most recent year for which we have a survey, using the price deflators provided by the International Monetary Fund’s International Financial Statistics. In our main analysis we use sample selection criteria that are standard in the labor and development literature on returns to education and experience (Murphy and Welch, 1990; Duflo, 2001; Lemieux, 2006). We restrict our attention to male, full-time workers who earn wages. These restrictions are motivated by the fact that potential experience is a better proxy of actual experience for male and full-time workers than for female and part-time workers. The restriction to wage workers is motivated by the observation that earnings of self-employed workers can reflect payments to both capital and labor, making it difficult to accurately measure wages of the self-employed (see Deaton, 1997; Hurst et al., 2014; Gollin, 2002, for example). In addition to these standard restrictions, we focus our analysis on private-sector workers, which is motivated by the concern that public sector workers may receive non-wage compensation such that their wages do not reflect the full payment for their labor. In the main analysis, we follow the literature and define potential experience as experience = age – schooling – 6 for individuals with twelve or more years of schooling and as experience = age – 18 for individuals with fewer than twelve years of schooling. This definition implies that individuals begin to work at age eighteen or after they finish school, whichever comes later. The cutoff at age eighteen is motivated by the fact that few individuals have positive wage income before the age of eighteen in the data. Although each of these sample restrictions and the definition of potential experience are fairly standard in the literature, we re-consider each of them in Section 5.

3

Life-Cycle Wage Growth Across Countries: Cross-Sectional Evidence In this section, we present cross-sectional evidence on life-cycle wage growth. We focus first on our core

eight countries, where we have the most data, and compute experience-wage profiles, a simple measure of life-cycle wage growth that has been studied in the literature. We find that profiles are steeper in the rich countries than in the poor countries. We then turn to our full set of countries and document the same pattern.

5

3.1

Experience-Wage Profiles for Core Countries

We begin by presenting experience-wage profiles for our eight core countries. We focus on experience-wage profiles as our measure of life-cycle wage growth rather than age-wage profiles. This is because experiencewage profiles allow us to summarize the evolution of wages over the life cycle for groups with different educational attainment and hence different ages of entry into the labor market. Relatedly, age-wage profiles typically differ by education groups, while experience profiles tend to be much more parallel. We discuss these issues in detail in Section 6.1 where we present age- and experience-wage profiles separately by educational attainment. For each country, we calculate an experience-wage profile for each survey year by computing the average wage by five-year experience bin and expressing it as a percent difference from the average wage of the lowest experience bin (0-4 years of experience). We then compute each country’s experience-wage profile as the average profile across calendar years. Note that this is conceptually similar to estimating experience-wage profiles with repeated cross sections while controlling for time (i.e., the year of each survey) fixed effects. The reason is that, by normalizing the average wages of workers in each experience group by the average wage of the lowest experience bin in each year, the profiles are made comparable over time for countries with different time trends. Figure 1 plots experience-wage profiles for our core countries.3 For expositional purposes we plot the profiles for rich countries on the left-hand panel and for poor countries on the right-hand panel. In all countries, profiles are increasing until at least twenty years of potential experience, and then flatten or decline afterwards. Among the rich countries, Germany has the steepest profile, at above 100 percent higher wages by twenty years of experience. The profiles for the United States, Canada and the United Kingdom are similar and somewhat flatter than that of Germany, with around seventy-five percent higher wages by twenty years of experience. Among the poor countries, Brazil is the steepest, reaching a height of just above seventy percent, followed by Chile, Mexico and then Jamaica. To summarize these findings and more formally compare experience-wage profiles across countries, we compute four summary statistics for each country. The first is the height of the profile at 20-24 years of experience, or twenty years more experience than the least experienced bin. The second is the height of the profile at 35-39 years of experience, which is the highest experience bin. The third is the average height of the profile, computed as the average across all experience bins other than the lowest. The fourth is the average height when discounting each year at four percent per year, which is meant to be a simple measure of the discounted value of lifetime income gains.4 3 See 4A

Appendix Figure A.1 for the same figure with the 95 percent confidence intervals. convenient property of the discounted average height is that it appropriately trades off wage gains that occur early versus

6

Panel A of Table 2 reports the summary statistics for each country. The reported heights are relative to the least experienced group, which comprises of workers with zero to four years of experience. Germany’s profile is the steepest, reaching 105 percent by 20-24 years of experience. This is followed by the United States (90 percent), the United Kingdom (85 percent) and Canada (80 percent).5 Brazil’s profile is the steepest amongst poor countries, at approximately seventy percent. This is followed by Chile (45 percent), Mexico (40 percent) and Jamaica (33 percent). The heights at 35-39 years of experience paint a similar picture, as do the average and discounted heights. Panel B of Table 2 presents permutation tests of the null hypothesis that experience-wage profiles are the same in the rich and poor countries. The logic of the permutation test is that under the null, one can resample the data many times to compute the probability that one would observe a difference as extreme as the actual difference in the data by chance. Permutation tests have better properties for small samples than other commonly used tests, such as t-tests (Lehmann and Romano, 2005). The differences between the mean for rich countries and poor countries are large and statistically significant for all four of the summary statistics. In the rich countries, the wages of workers with 20-24 years of experience are 89.3 percent higher than those with less than five years of experience. In contrast, in the poor countries the wages of workers with 20-24 years of potential experience are just 47.6 percent higher than those with less than five years of experience. The difference is 41.7 percentage points, which means that experience-wage profiles are roughly twice as steep on average in rich countries by twenty years of experience. The profiles are also roughly twice as steep in rich countries according to the other summary statistics. By the highest experience level, 35-39 years of experience, wages in the rich countries are 81.6 percent higher on average than for the least experienced workers, compared to 36.9 percent in the poor countries. The average height of the profile is 68.3 percent in the rich countries and 36.0 percent in the poor countries, for a difference of 32.3 percentage points. The discounted average height is 31.5 percent in the rich countries and 16.8 percent in the poor countries, for a difference of 14.7 percentage points. The p-values for these differences is below five percent in all cases, meaning that these differences are unlikely to have occurred by coincidence. Thus, experience-wage profiles are on average steeper in the rich countries. Finally, a point worth emphasizing is that virtually the entire difference in steepness between rich and poor countries occurs over the first 20 years of workers’ potential experience. For instance, Panel B of 2 late in life, and it therefore can, for example, be used to compare the profiles of two countries that cross. This summary statistic is also related to a statistic commonly used to compute returns to education: the difference in the present discounted value of lifetime earnings across different education groups (see e.g. Todaro and Smith, 2012, Section 8.2 and references cited there). 5 Our estimated experience-wages profiles for the rich countries are largely in line with previous estimates in the literature. In the United States, for example, Lemieux (2006) uses CPS data to estimate an increase in wages of 0.7 log points, or roughly one hundred percent, between zero and twenty years of experience. Our estimates of other measures of life-cycle income growth, for example age-earnings profiles, also line up well with previous estimates in the literature. Guvenen et al. (2014b) use administrative data to estimate 127 percent higher average earnings for those aged 51 than those aged 25. Using our data, we calculate 116 percent higher average earnings for those aged 51 than those aged 25.

7

shows that 41.7 of the 44.7 percentage point mean difference between rich and poor countries in the height of the experience profiles is due to potential experience increasing from 0-4 to 20-24 years, and only an additional 3 percentage points are due to potential experience increasing further to 35-39 years. This fact is also apparent visually from Figure 1. In Appendix A.3 we explore in more detail at what point of the life cycle the differences in returns to experience between rich and poor countries occur and show that about half of the difference in profiles at 20-24 years of experience is realized after five years only.

3.2

Experience-Wage Profiles for All Countries

We now turn to a broader set of countries for which cross-sectional wage profiles can be constructed. This offers a more comprehensive examination of life-cycle wage growth across countries, simply by way of covering more countries, though the non-core countries cover fewer individuals and years than the core countries. Figure 2 presents experience-wage profiles for all eighteen countries in our sample. Countries are sorted in descending order of GDP per capita from the top left to the bottom right panel. The top left panel adds the profile for Australia to the those of the four rich core countries (Figure 1). The top right panel includes the second richest group of countries: France, South Korea, Uruguay and Chile; the bottom left panels include the second poorest group of countries: Indonesia, Brazil, Peru and Mexico; and the bottom right panel includes the poorest countries in our sample: Bangladesh, Guatemala, India, Jamaica and Vietnam. A comparison of the profiles across panels shows clearly that experience-wage profiles become flatter as the country’s income falls. The difference between the richest countries (upper left) and poorest (lower right) is readily apparent, with all of the richest countries having 75-100 percent higher wages by twenty years of experience, and all of the poorest countries having less than 50 percent higher wages. Life-cycle wage growth in countries with more intermediate income levels (in the upper right and lower left panels) is roughly in between that of the richest and poorest countries. While Figure 2 confirms that experience-wage profiles are steeper in richer countries, it also highlights that there is a fair amount of dispersion in steepness particularly in the intermediate income groups (upper right and lower left panels). For example, the experience-wage profile for Indonesia ultimately rises as much as that for some of the richest countries. Nevertheless, on average, the overall pattern that experience-wage profiles are steeper in richer countries remains: taking an average across all rich countries (in the core and full sample), the average height at 20-24 years of experience is 83.5 percent. For the poor countries, the average is 45.9 percent, which results in a difference between rich and poor countries of 37.5 percentage points. This difference is statistically significant at the one-percent level and comparable in magnitude to the difference in the core sample. Thus, Figure 2 shows that the finding that experience-wage profiles are steeper in richer countries is true in the full sample as well as the

8

sample of core countries. Finally, as already noted in the previous section, the majority of the differences in profiles between rich and poor countries occur over the first 20 years of workers’ life-cycle.

4

Life-Cycle Wage Growth: Controlling for Education, Time, and Cohort In the previous section, we presented cross-country evidence on experience-wage profiles by simply plot-

ting average wages within age- or experience bins in the cross-section of individuals. While we view this as a useful starting point because it imposes minimal structure and assumptions on the data, there are a number of important issues that such a simple exercise does not address. First, our cross-sectional profiles ignore the role of schooling. Second, cross-sectional estimates leave open the possibility that experience-wage profiles are driven by cohort effects, such as improvements in the health of subsequent birth cohorts. In this section we address both of these issues. Throughout this section, we estimate flexible versions of Mincer regressions of individuals’ wages on their years of schooling and potential experience. That is, we estimate equations of the form

log wict = α + θsict + f (xict ) + γt + ψc + εict .

(1)

wict is the wage of individual i, who is a member of birth cohort c and observed at time t. sict and xict are her years of schooling and experience. γt is a vector of time-period dummy variables, ψc is a vector of cohort dummy variables and εict is a mean-zero error term. We follow the textbook specification and assume that schooling and experience enter in an additively separable fashion. This assumption is relaxed in Section 6.1, where we allow the returns to experience to differ between more and less educated workers. In what follows, we estimate equation (1) separately for each country under various assumptions on cohort and time effects, and then assess how the function f (·) varies across countries. Equation (1) differs from the traditional Mincer regression in two ways. First, we allow the relationship between experience and wages to be flexible and do not restrict the functional form to be linear. Second, we allow for cohort and time effects, as we describe below.

4.1

Deaton-Hall Approach

The main challenge to estimating returns to experience (or age) is that one cannot separately identify the effects of experience, birth cohort and time, due to collinearity. In this section, we consider the effects of cohort and time controls following the approach proposed by Hall (1968) and Deaton (1997) for estimating returns to experience using repeated cross sections. The main purpose of the Deaton-Hall approach is to illustrate the mechanics of the econometric difficulty. The next section then provides a theoretically motivated method for disciplining time and cohort effects. Before proceeding, we note that panel data would

9

not solve this identification problem. The reason is that even when following specific individuals (rather than cohorts) over time, one cannot separate how much of their wage growth is due to aging or the passing of time. In either cross-sectional or panel data these effects can only be identified with additional assumptions, which, as is well-known in the literature, are identical for both types of data.6 To implement (1), we regress the logarithm of wages on schooling and a set of dummy variables for five-year experience groups

log wict = α + θsict +

X

x φx Dict + γt + χc + εict ,

(2)

x∈X

in combination with one additional linear restriction on the set of cohort and time effects corresponding to x different versions of the Deaton-Hall approach. Dict is a dummy variable that takes the value of one if a

worker is in experience group x ∈ X = {5-9, 10-14, ...}; the omitted category is experience less than five years. This specification allows us to capture non-linearities in a flexible way. The coefficient φx estimates the average wage of workers in experience group x relative to the average wage of workers with less than five years of experience. In terms of our notation of equation (1), the φx terms represent f (x) such that the coefficient estimate corresponding to each experience level, x, identifies the experience-wage profile evaluated at point x. To resolve the difficulty of collinearity, Hall (1968) and Deaton (1997) impose one additional linear restriction on the set of cohort and time effects in equation (2). We consider three different versions of the Deaton-Hall approach. The first version attributes all labor productivity growth to cohort effects and uses year dummies to capture only cyclical fluctuations. This is the assumption made in Deaton’s (1997) original analysis and more recently by Aguiar and Hurst (2013). We implement this by estimating equation (2) with birth-cohort dummies and time dummies, with the restriction that the time dummies are orthogonal to a time trend. See Appendix A.2 for a more formal description of our methodology. The second version takes the opposite extreme and attributes all labor productivity growth to time effects. We implement this by estimating equation (2) with cohort and time dummies, but now we restrict the cohort effects to be orthogonal to a time trend. The third takes the intermediate view that productivity growth is attributed in equal parts to cohort and time effects. While we are agnostic on the most natural split between time and cohort effects, the case of an equal split is nonetheless useful for illustrating how the estimated returns to experience across countries depend on the relative importance of the two effects. Figure 3a plots the estimates from the first version, in which all income growth is attributed to cohort 6 See for example Heckman and Robb (1985, p.140) who note that “it is by now well known (Cagan, 1965) that [panel] data do not solve the identification problem,” and that “panel data and a time series of cross sections of unrelated individuals are equally informative.”

10

effects.7 The left-hand panel shows that Germany and the United Kingdom have the steepest profiles, with more than 100 percent growth by twenty years of experience, while the United States and Canada have around sixty percent growth by twenty years of experience. The right-hand panel shows that all of the poor countries have steep and linear (or close to linear) experience profiles, with Brazil being the steepest, followed by Jamaica, Chile and then Mexico. The reason that this version has such steep profiles is that, with time effects shut down, all wage growth by individual cohorts over their lifetimes is attributed to their increased experience. In countries like Brazil and Jamaica that have experienced high rates of aggregate growth over this period, the size of the effects attributed to experience is large.8 Figure 3b plots the estimates from the second version, in which all labor productivity growth is attributed to time effects. The left-hand panel shows that Germany is still the highest, at more than 100 percent growth, while Canada, the United Kingdom and the United States are close behind at between 75 percent and ninety percent growth. The right-hand panel shows that the poor countries have flatter profiles than the rich countries, with Brazil still highest at around seventy percent growth, followed by Chile at 65 percent growth and Mexico and Jamaica at just under fifty percent growth. These profiles are very similar to the crosssectional profiles in Section 3 because both sets of profiles attribute wage growth over time to changes in aggregate economic conditions rather than to improvements across cohorts. Panel A of Table 3 reports the four summary statistics when all growth is explained by cohort effects. By 20-24 years of experience, profiles are on average 10.8 percentage points higher in the rich countries, though the difference is statistically insignificant. By 35-39 years of experience, profiles are on average higher in the poor countries by 38.7 percentage points, though again the difference is statistically insignificant. The average and discounted heights are slightly higher in the rich countries, but the magnitudes are small and statistically insignificant. Panel B of Table 3 shows the intermediate case when growth is explained equally by cohort and time effects. By 20-24 years of experience, the rich mean is 27.4 percentage points higher than the poor-country mean, which is significant at the ten percent level. By 35-39 years of experience, rich and poor countries have similar means. The average height is 16.1 percentage points higher among the rich, while the discounted height is 9.0 percentage points higher among the rich, with the latter being statistically significant at the ten percent level. Panel C of Table 3 reports the results when all growth is explained by time effects. The mean for rich countries is 44.4 percentage points higher by 20-24 years of experience and 31.1 percentage points higher by 7 The

confidence intervals tend to be narrow for most countries, so we omit them for brevity. and Jamaica had wage growth of 3.5 percent per year and 2.1 percent per year on average, with Chile and Mexico had growth of 1.6 percent and 1.1 percent. Among the rich countries, the United Kingdom and Germany had wage growth of 2.0 and 1.9 percent, while Canada and the United States had average wage growth of 0.5 percent. 8 Brazil

11

35-39 years of experience. The average height is 31.6 percentage points higher for the rich countries, while the discounted height is 15.0 percentage points higher. All differences are statistically significant at the five percent level except for the height at 35-39 years. We conclude that if cohort effects explain all of growth, the profiles of the rich countries are marginally steeper than those of the poor countries we observe. If, however, time effects explain half or more of growth, then experience-wage profiles are steeper in rich countries than in poor countries, with differences that are statistically and economically significant. Thus, we next ask whether economic theory can help us further discipline these profiles.

4.2

Heckman-Lochner-Taber Approach: No Growth at End of Life Cycle

The insight from the previous illustration is that the interpretation of the cross-sectional results depends on the extent to which aggregate growth is attributable to time or cohort effects. In this section, we propose a theoretically motivated method for disentangling the relative importance of time and cohort effects. In particular, we draw on the basic prediction of a large number of theories of life-cycle wage growth that there should be little or no growth in the final years of a worker’s career. This prediction is shared by the three basic mechanisms for explaining life-cycle wage profiles emphasized in the literature, namely human capital investment, search and learning.9 The basic idea of our approach is to use the assumption that there are no experience effects in the final working years as a restriction to identify time effects and cohort effects. A similar reasoning has been used by Heckman, Lochner and Taber (1998), so we refer to this as the HeckmanLochner-Taber (HLT) approach, though credit is due more broadly, as variants of this idea have appeared in the works of McKenzie (2006), Huggett et al. (2011), Bowlus and Robinson (2012) and Schulhofer-Wohl (2013).10 A simple example helps motivate how this method identifies the effect of wages due to experience (or age) rather than time or cohort. Imagine we follow the wages of two cohorts: a “young cohort” that has 0-4 years of experience in the year 2000 and an “old cohort” that has 30-34 years of experience in the year 2000. Say we observe that the young cohort has wage growth of five percent between 2000 and 2005, while the old cohort has growth of only one percent over the same period. Under the assumption that the old cohort has no wage growth coming through experience, the difference in the time effects between 2000 and 2005 must be one percent. Thus, we infer that the young cohort had wage increases of four percent (five minus one) 9 See

for example the review by Rubinstein and Weiss (2006). Lochner and Taber (1998) and Bowlus and Robinson (2012) have used a similar insight in models of human capital to separate prices and quantities of human capital, and Huggett et al. (2011) have used the assumption of no human capital investment at the end of the life cycle to identify shocks to human capital. McKenzie (2006) shows that when using repeated cross sectional data, second differences of age, cohort and time effects are identified without any assumptions, and that first differences can be identified as well with a restriction on one first difference. Our method selects one such restriction using economic theory. Similarly, Schulhofer-Wohl (2013) argues that one should use the curvature of wage profiles to identify parameters of structural models. 10 Heckman,

12

coming from their increased experience. Repeating this idea for many cohorts, we can build up a full series of time effects. Given time effects, we can then estimate the remaining cohort and life-cycle age or experience effects. This method is easily extended to allow for depreciation of skills or of match quality at the end of life. In this case, we replace the assumption that age/experience effects are zero with the assumption that they are −d percent where d is the depreciation rate. The rest of the method proceeds as above. This approach requires assumptions about two main parameters: first, the number of years at the end of the life cycle for which there are no experience effects, and second, a number for the depreciation rate. We follow Huggett et al. (2011) and consider either five or ten years with no experience effects. We consider two alternative depreciation rates of either zero or one percent per year. Given the assumptions about the number of years without experience effects, y, and a depreciation rate, d, this approach to estimating the experience-wage profile in a particular country works as follows. First, we guess an initial trend in the time effects. We then deflate wages for each individual in each year by the wage growth rate implied by the time effect. Next, we estimate equation (1) with experience effects and cohort effects, and we check whether the estimated experience effects have on average declined by d percent in the last y years. If they have, we stop. Otherwise we adjust the trend in the time effects and repeat. Once the process has converged, it produces separate estimates of cohort effects, time effects and experience effects for a given country and for given values of y and d. For the purposes of our paper, there are two main benefits to this HLT approach. First, it uses economic theory to motivate restrictions on time and cohort effects. Second, it allows the sources of growth to be country-specific, which is useful when comparing countries with very different income levels and growth rates.11 Figure 4 plots the experience-wage profiles estimated using the HLT method under the assumption that there are no experience effects in the last ten years of the life cycle and no depreciation. In the rich countries, the experience-wage profiles are concave and grow by seventy to one hundred percent by twenty years of experience. Profiles for the poor countries are also concave, but are flatter, and wage growth ranges between forty and sixty percent.12 Table 4 reports summary statistics of the profiles in rich and poor countries for the experience-wage profiles estimated using the HLT approach. Panel (a) summarizes estimates for the case with no experience effects over the last ten years and zero depreciation (as in Figure 4), panel (b) summarizes the case with no experience effects over the last five years and zero depreciation, panel (c) summarizes the case with no 11 The most widely applied alternative theoretical restriction, proposed by Deaton (1997), restricts time effects to sum to zero, as in Figure 3a above. The theoretical rationale for this was “to use the year effects to capture cyclical fluctuations or business-cycle effects that average to zero over the long run” (p.126). This restriction is less relevant for our analysis given that our sample includes many fast-growing countries. 12 Appendix Figure A.2 presents the same profiles with their 95 percent confidence intervals.

13

experience effects over the last ten years and one percent depreciation, and panel (d) summarizes the case with no experience effects in the last five years and one percent depreciation.13 In all four panels the rich-poor country differences in heights at 20-24 are large and statistically significant. The same is true for the heights at 35-39 years of experience, the average heights and discounted average heights. The largest differences are estimated under the assumption that there are no experience effects in the last five years and no depreciation (Panel b), while the differences are smallest when depreciation is one percent and there are no experience effects in the last ten years (Panel c). The reason is that when there is depreciation, the profiles themselves are flatter in all countries, hence cross-country differences become smaller. In summary, the results in Table 4 show that the heights of the profiles can be sensitive to the depreciation rate or the length of time with no gains from experience. However, our main result that there are more life-cycle wage gains in rich countries is present in all cases. Similarly to the cross-sectional profiles in Section 3.1, most of the difference in steepness between rich and poor countries occurs over the first 20 years of workers’ potential experience. For instance, panel (a) shows that with no experience effects over the last ten years and zero depreciation, experience-wage profiles at 20-24 years are 40.1 higher in rich countries, and at 35-39 years they are 37.5 percent higher. That is, toward the end of the life cycle poor countries actually make up for a small part of the gap in the height of the profiles. Also see Appendix A.3 where we explore this point in greater detail and show that, similarly to our cross-sectional results, about half of the difference in profiles at 20-24 years of experience is realized after five years only. Note that the HLT results in Table 4 are quite similar to the cross-sectional estimates shown earlier in Table 2. In light of the discussion in the previous section, this is consistent with most of the growth experienced by the countries in our core sample being attributable to time effects.

5

Robustness This section considers the robustness of our main finding that life-cycle wage profiles are steeper in richer

than in poorer countries. In particular, we demonstrate that our main result that experience-wage profiles are steeper in rich countries is unlikely to be an artifact of how we measure experience or restrict the sample. Unless otherwise stated, we focus on our preferred estimates that use the Heckman-Lochner-Taber (HLT) method to decompose age, time, and cohort effects and restrict our attention to the core sample of countries. Most of our results are summarized in Table 5. Each row corresponds to an alternative sample selection criterion or variable construction. We focus on the heights of the profiles at 20-24 years of experience for 13 Note that assuming a depreciation rate of one percent is rather extreme as it causes poor countries to experience almost no growth over the life cycle. This is because assuming that there is no growth over the last few years of the life cycle mechanically rotates the experience-wage profiles clockwise and adding depreciation further rotates the tail end of the life cycle in the same direction.

14

brevity. The columns contain the average height across the four rich countries, the average height across the four poor countries, and the difference. We conducted similar analyses to verify that our cross-sectional results from Section 3 are also robust. See Appendix Table A.2, where we present the results for both the core sample of eight countries and the full sample of eighteen countries.

5.1

Measurement of Experience

Our benchmark measure of potential experience is constructed as years since the expected date of graduation or age 18, whichever comes last. This could introduce measurement error into our main explanatory variable for several reasons, which we discuss in detail in this section. Since measurement error (if classical) can cause attenuation bias, a natural concern is that there is more measurement error of experience in poor countries, which biases the difference between the experience-wage profiles of rich and poor countries upwards. 5.1.1

Alternative Measure of Experience

One potential concern with our main measure of experience is that we may mis-date the start of work, either because we mis-date graduation or because some less-educated workers undertake meaningful work before graduation or age 18. We consider two alternatives. First, we simply allow experience to start at the expected date of graduation or age 16, which may be more appropriate for the poorer countries in our sample. Doing so raises poor country profiles modestly but does little to rich countries. Row (1) of Table 5 contains our baseline results, for comparison. Row (2) shows that lowering the age at which individuals start accumulating experience has little effect on our results. Second, we can construct an alternative measure of experience, which we refer to as “constructed experience.” 14 The idea behind this measure is to use the cross-sectional relationship between employment and age by education group to infer the life-cycle relationship between experience and age. Mechanically, we divide workers into three broad education groups (less than high school, high school, and more than high school) and calculate the percentage of individuals who are engaged in wage employment for each age and education group.15 We then normalize this employment rate by dividing it by the employment rate of an arbitrary group, which we choose to be forty-year olds. To calculate the years of experience for an individual we sum the normalized employment rates over all prior ages. For example, if for high school graduates, the employment rate was 70, 35 and 50 percent for forty-, eighteen- and nineteen-year olds, then we infer that 14 This follows an older labor literature that uses the cross-sectional relationship between age, school attendance, and school enrollment/work to estimate the age at graduation rather than simply equating it with years of schooling plus six (Hanoch, 1967; Gould and Welch, 1976). Smith and Welch (1978) and Welch (1979) estimate returns to experience that build on this alternative measured age at graduation, emphasizing as we do that it allows them to incorporate variation in age at graduation (in their case, by race and cohort within the U.S.). 15 We use the three groups to be consistent with the exercises later in Section 6.1. The robustness results presented here are similar when we use more disaggregated groups (e.g., each year of education). These are available upon request.

15

the average high school graduate who is 20 years old has 35/70 + 50/70 = 1.21 of constructed experience. We calculate constructed experience for each country. This allows us to test whether our results are sensitive, for example, to differences in post-graduation employment patterns between poor and rich countries. Row (3) of Table 5 shows that the rich-poor differences in heights at 20-24 years of experience using constructed experience are, if anything, slightly larger than our baseline results.16 5.1.2

Measurement Error in Age or Education

Since potential experience is constructed using reported age and estimated years of schooling, mismeasurement of either variable that is more pronounced in poor countries could cause the experience-wage profiles of poor countries to attenuate more than that of rich countries. In particular, one may worry that survey respondents in poor countries are more likely to round their ages or provide a noisy estimate of their actual educational attainment. This could in principle lead to a spurious finding of flatter experience-wage profiles. When looking at reported age distributions in the poor core countries, we do observe that there is some age heaping in Mexico and Chile, where there are small spikes in population frequency at every ten years of age (see Appendix Figure A.6). To examine whether age heaping drives the difference between poor and rich country experience-wage profiles, we artificially distort the age distributions of rich countries to match the age heaping observed in the Chilean data.17 We then re-construct potential experience using the distorted age data in each country, and re-estimate the experience-wage profiles using our HLT approach. We find that with these distorted age data, the experience-wage profiles of the rich countries are very similar to the actual profiles. Row (4) in Table 5 shows that with distorted ages, the rich countries have an average height of 81.1 compared to the 79.3 in the baseline. Artificially distorting the age distribution to replicate the same level of age heaping observed in Chile makes the profile of Germany slightly steeper and the profiles of the United States, Canada and the United Kingdom marginally flatter (see Figure A.3). We conclude that it is unlikely that mismeasurement of age plays an important role in explaining our findings. To address concerns about measurement error in education, we turn to the Chilean data, where respondents were asked to report both the number of years they attended school and the highest level of attainment.18 The data show that there is indeed variation in the number of years of actual schooling for a 16 Note that interpreting the results using constructed experience relies on the assumption that patterns of work are consistent over time within a country - that if the average high school graduate gains 0.5 years of experience at age eighteen in 2001, the same was true for earlier cohorts. We do not expect this assumption to hold exactly, but nevertheless find it a useful robustness check as it allows us to present results that do not depend on assumptions about the expected graduation date or the earliest possible age of work. 17 We estimate a smooth version of the age distribution in Chile using a quintic regression and define age heaping as the difference between the actual age distribution and the smoothed one. Equipped with the estimated age heaping level for each age, we turn back to the micro data from the rich countries and artifically distort it. For example, if we observe that in Chile there are 5 percent fewer individuals aged 19 than expected according to the smoothed age distribution, and 5 percent more individuals aged 20, then we randomly assign 5% of 19-year olds to be 20 years old instead in each rich country. This exercise replicates, by construction, the same amount of age heaping as in Chile. 18 We focus on the Chilean data from 2009 since it is the most recent year, though similar data are available in 2000, 2003 and 2006. The Jamaican data from 1991 and 2001 also asked these questions. However, the quality of the data for years of

16

given level of attainment (see Appendix Table A.1). For example, in Chile, the years of schooling for someone who completed “some primary” ranges from three to eight years. For those who complete “college”, the number of years only vary between sixteen to eighteen years. Thus, those who report having “some primary” range between 33 percent fewer and 100 percent more than the imputed number of years of schooling. And those who complete college range between 0 percent less and 12.5 percent more than the imputed number of years of schooling. To investigate whether this variability drives the steeper profiles in rich countries, we impose the same dispersion onto the four rich countries in our core sample. Since the categories of educational attainment differ across surveys, we divide the data into three groups: less than high school, high school, and more than high school. For each group, we use the data from Chile to calculate the average percentage deviation from the imputed years of schooling for each percentile. We then distort the data for the rich countries such that the dispersion in the years of schooling for each attainment level follows the Chilean distribution of the group that the level belongs to, and re-estimate the experience-wage profiles with the distorted data. We find that the data with distorted education levels yield experience-wage profiles that are modestly flatter than the actual profiles. Table 5 row (5) shows the mean for rich countries is 71.7 compared to 79.3 in the benchmark. The difference with the poor countries is still large, at 32.5, and statistically significant at the 5 percent level. Therefore, at least with the measurement error in the Chilean data as a guide, mismeasurement of education is not likely explain much of our findings.19

5.2

Sample Selection

Our baseline analysis focused on a sample that is designed to maximize comparability between countries and minimize measurement concerns: full-time, private-sector male wage workers. This raises two questions. The first and most important for our study is the concern that the main result that experience-wage profiles are steeper for the sample of interest in richer countries is driven by differential selection into the sample. For example, if less productive workers select out of wage employment in rich countries as they age, while such workers select into wage employment in poor countries as they age, our finding of steeper profiles for wage workers could be driven by differential selection. The second question is whether the profiles will still be steeper once we relax the sample restrictions and include other types of workers.20 In this section, we provide evidence against the concern that selection is the main driving force of our results, and suggestive education is poor: there are many missing values and implausible responses (e.g., the number of years for those who report “no education” as their highest level of attainment range from zero to fifteen years). Thus, we do not use the data from Jamaica. 19 Appendix Figure A.4 presents the experience-wage profiles using the distorted and actual education data. We also estimated the profiles using distorted age and education data, finding profiles that were again only modestly flatter than those of the baseline analysis; see Appendix Figure A.5 and row (6) of Table 5 for the rich-poor differences in profile heights. 20 Note that for the eight core countries, the size of our sample as a percentage of total male workers is 63% (USA), 66% (U.K.), 65% (Mexico), 45% (Jamaica), 67% (Germany), 70% (Chile), 67% (Canada) and 61% (Brazil).

17

evidence that the profiles will still be steeper when we expand the sample. We explain our approach in detail for self-employed workers, then briefly overview the parallel results for public sector workers, women, and part-time workers. 5.2.1

Self-Employed Workers

An important sample restriction is that we focus on wage earners because wage income is a direct payment for labor services that is generally considered to be accurately reported. In contrast, the income of the selfemployed presents two challenges. First, it can represent payments for both labor and capital services, implying that it is less directly related to life-cycle theories of human capital accumulation or search and matching. Second, it is well-known that the reported income of the self-employed suffers from substantial underreporting (Hurst et al., 2014). A concern with using only wage workers in repeated cross-sectional data is that there may be selection into or out of self-employment over time. We address this concern in several ways. One is to simply include these workers in our estimates. In the row (7) of Table 5, we show the result from including the self-employed, taking their reported income to be their wage and salary income. Doing so has little effect on our results. The caveat for interpreting this result is that proxying for wages this way introduces measurement error for the reasons discussed earlier. To address this, we use panel data. Since panel data are not widely available, we choose one rich country, the United States (Panel Study of Income Dynamics, PSID, annually 1975-1997, bi-annually 1999-2013), and one poor country, Mexico (Mexican Family Life Survey, FLS, 2002, 2005 and 2009). In the main analysis, U.S. workers have very steep profiles, while Mexican workers have flat profiles. Thus, they are useful for understanding whether selection into or out of self-employment causes profiles in rich countries to be steeper. Moreover, to the best of our knowledge, Mexico is the only poor country within our core sample to have panel data. As with the main exercise, we examine male full-time workers in the private sector. Experience-wage profiles following the same individuals First, we show that the estimated experiencewage profiles from panel data are similar to those from repeated cross sections. To be transparent, we employ a non-parametric approach and simply follow individuals over time without controlling for time fixed effects. Since the Mexican FLS data are only available for the years 2002, 2005 and 2009, we use waves of the PSID from a comparable time period, 2003-2013. The sample is restricted to individuals who were present during all of the specified waves for the surveys. We divide the sample into cohorts based on the level of potential experience in the first year of the data (i.e., 2002 for the FLS and 2003 for the PSID). As with the main exercise, a cohort group comprises five years of experience levels (e.g., the youngest cohort in Mexico had 0-4 years of experience in 2002). We then

18

calculate the average wage for each bin and normalize it by dividing it by the average wage of the youngest cohort that year. Figure 5a shows that the U.S. profile is higher than the Mexican profile. Each line segment in figure is the normalized wage of a cohort over time. Figure 5b shows the analogous profiles from the repeated cross-sectional data (these are identical to those in Figure 1). A comparison of the two figures show that the panel and repeated cross-sectional data are broadly similar. Selection into and out of wage employment Next, we use the panel data to explicitly estimate the wages of workers who switch from wage employment to self-employment. For selection into self-employment to cause experience-wage profiles to be steeper in rich countries, it must be the case that workers who exit wage work for self-employment in rich countries are less productive than workers who make the same switch in poor countries. To test this idea, we compare the year t-1 wages of workers who are engaged in wage work in consecutive periods t-1 and t with the wages of workers who are engaged in wage work in period t-1 but subsequently switch to self-employment in t. This approach allows us to characterize any differences in wages between those who remain in and exit our baseline sample and has the virtue of not requiring us to use the reported incomes of the self-employed. We implement this idea using all the available waves of the Mexican FLS and the U.S. PSID. We focus on the subsample of workers who appear in at least two consecutive survey waves. We estimate the relative wage for each experience bin by regressing wages earned in the previous year on dummy variables for the experience bins and the interactions of each experience-bin dummy variable with a dummy variable for whether the worker is self-employed during the current year, controlling for the years of educational attainment and year fixed effects. The reference group is the 0-4 years of experience bin for workers who remain in wage work. From this regression, we can predict the relative wage residuals for each experience bin for workers who remain in wage employment and workers who exit. Note that our estimates control for time effects and we do not implement the HLT method because the panel data (particularly for Mexico) have smaller sample sizes such that there are very few workers with high levels of experience. Note also that we ignore attrition from panels when we estimate these profiles. In Appendix Section A.4, we show that our estimates are very similar using bounded samples where we add those who exit the panel into the sample. Figures A.7a and A.7b plot differences in residualized wages between the two types of workers at each point of the life cycle (i.e., the wage residuals of workers who remain in wage employment minus those of workers who move to self employment) and their 95 percent confidence intervals.21 They show that U.S. and Mexican workers who remain in wage employment earn similar wages except during 15-20 years of 21 The

estimates are more precise for the United States because the PSID has a much larger sample size than the FLS.

19

experience in the United States, and 25-30 and 35-40 years of experience in Mexico.22 Since the wage gap is small or not significantly different from zero throughout the life cycle, the exit of workers to self-employment is unlikely to significantly affect our estimated wage profiles. Moreover, such moves are relatively infrequent; only 3 percent of workers in the United States and 5-11 percent in Mexico exit in a given year (see the thin dashed line in Figures A.7a and A.7b). What matters for our profiles is the product of these two variables: the frequency of exiting multiplied by the selectivity of those who exit, measured as their wage gap. We plot this product as well as the solid line in Figures A.7a and A.7b. The line is positive but small in both countries at approximately 1-2 percent. This indicates that in a typical year, the exit of low-wage workers to self-employment pushes wages up by 1-2 percent as compared to keeping a constant sample in both countries. This small and balanced effect has little scope to affect our cross-country patterns. We can also address the concern that workers select from self-employment into wage employment. An example of how this can cause steeper profiles for richer countries is if workers who enter wage work from self-employment in rich countries are more productive than workers who make the same switch in poor countries. To investigate this, we conduct an exercise very similar to before and compare wages in year t + 1 of workers who were self-employed in year t to wages of workers who worked for wages in year t. Figures A.7c and A.7d plot the difference in wage residuals between workers who are in wage employment in both years and workers who moved from self-employment into wage employment and their 95 percent confidence intervals. They show that workers who are in wage employment earn higher wages in the United States throughout the life cycle. In Mexico, such workers earn less when they have 25-30 years of experience. As in the earlier results, moves are uncommon; roughly 3 percent per year in the United States and 10 percent in Mexico. The net effect of entry is given by the product of the two and is shown with the solid line in Figures A.7c and A.7d. These effects are small and positive, meaning that workers who switch into wage employment earn less and drag down wage profiles by roughly 1-2 percent per year in both the United States and Mexico. The examination of workers who move into and out of self-employment in the United States and Mexico shows that such selection is very unlikely to cause the estimated experience-wage profiles of rich countries to be steeper. Since these results are specific to workers who move between wage- and self-employment, it is natural to wonder whether we will find steeper profiles in richer countries if we include workers who remain in self-employment for their entire life cycle (and had accurate wage data for these workers). Note that for their inclusion to drive our results, self-employed workers would need to have relatively steeper profiles than wage workers in poor countries.23 However, this goes against the conventional wisdom that self-employment 22 During

these points of the life cycle, workers who remain in wage employment earn approximately 30 percent more. self-employed workers can have steeper profiles in both rich and poor countries, but the difference between is larger in magnitude in poor countries. 23 Alternatively,

20

in poor countries is largely disguised unemployment.24 If self-employed workers in rich countries are more likely to include successful entrepreneurs while self-employed workers in poor countries are more likely to disguise unemployment, then our exclusion of self-employed workers will, if anything, cause us to understate the difference in the steepness of profiles between rich and poor countries. 5.2.2

Other Sample Selection

In this section we quantify the importance of the remaining sample selection criteria, using many of the same techniques introduced at length in the previous section. We start with the restriction to private-sector workers. Row (8) of Table 5 shows that our results are very similar if we include public-sector employees. We then repeat the same analysis using panel data to compare workers who remain in private sector workers to those who switch to or from public sector employment. We find that there is no obvious pattern over the life cycle and the flow of workers from one sector to the other is very low such that selection into and out of the private sector can only have negligible effects on our estimates; the details and figures are in Appendix A.5. Finally, we find that public sector employment is unlikely to drive our results because very few workers are employed in the public sector – on average, 8 percent in the rich countries in our sample, and 3 percent in the poor countries in our sample. We next turn to female workers, who were excluded from the baseline analysis. Table 5 row (9) shows the results if we include female full-time workers in the private sector. The results are very similar to the baseline in row (1). We do not use the panel data approach employed elsewhere since gender is a fixed characteristic and we therefore are not worried about women transitioning in and out of our sample as we are for the selfemployed. Instead, we consider robustness to measuring women’s experience using constructed experience as in section section 5.1.1 rather than potential experience; this helps address the concern that potential experience may not accurately reflect the women’s actual experience if they are more likely to experience career interruptions. The results in Row (10) show that this actually increases the gap in the heights of experience profiles between rich and poor countries. Finally, we investigate the importance of the exclusion of part-time workers (those who work less than thirty hours a week). Table 5 rows (11) and (12) show that the estimates change little when we include workers who work at least twenty hours a week or all part-time workers (i.e., workers who report any wage income). Row (13) shows the results when we use constructed experience, where this measure is constructed for full-time and part-time workers separately. The results are very similar to the baseline. Finally in 24 For survey evidence, see Lerner and Schoar, eds (2010) and Poschke (2013). Also see recent studies by Schoar (2010), which documents that operations run by self-employed individuals have very different growth dynamics from conventional entrepreneurs, and La Porta and Shleifer (2008), which divide subsistence and transformational entrepreneurship. Similarly, the notion that much self-employment in developing countries is disguised unemployment is supported by evidence that selfemployment and small-scale entrepreneurship increases during economic downturns (Adhvaryu et al., 2014; Gunther and Launov, 2012; Paulson and Townsend, 2005).

21

Appendix A.5 we study the experience profiles of workers who switch between part-time and full-time work. We find that such switches are common but have little effect on the profiles, suggesting that they are unlikely to drive our results. In summary, we find that life-cycle wage profiles are consistently steeper in rich than in poor countries. Our results are not sensitive to the sample selection criteria we impose or how we measure potential experience.

5.3

Alternative Assumptions about Depreciation at End of Life-Cycle

Our implementation of the HLT method relies on the assumption that wage movements near the end of the working life can be attributed to depreciation rates. We started with the natural assumption that the depreciation rate was common across countries. In this section we consider relaxing that assumption and allowing the depreciation rate of human capital to differ across countries. In order to discipline this analysis, we link possible variation in the depreciation rate to the type of jobs done. As we will document below in Section 7, most workers in rich countries are engaged in occupations that use their learned knowledge, whereas around half of workers in poor countries are engaged in occupations that use physical strength and stamina. This leads us to discipline the analysis through the age-related decline in mental and physical performance. For the former, we draw on a large literature that documents the life-cycle performance of athletes who compete in track and field events on an age-adjusted basis (Tanaka and Seals, 2008). Consistent with the model, we study the decline in record performance (measured as distance or time) between ages 60–64 and ages 65–69, downloaded from World Masters Athletics Current Records (2010). Peak performance declines in all categories, with the median and mean depreciation agreeing closely at 1.2 and 1.3 percent. For the depreciation of knowledge we turn to the corresponding psychological literature, which generally focuses on vocabulary knowledge or other forms of related, easy to measure knowledge. The stylized finding in this literature is that learned knowledge grows until roughly age 60, with no large changes for the subsequent ten years, implying that knowledge depreciation is zero percent at least through age 70 (Salthouse, 2003, 2013). We consider four additional robustness checks based on these findings. In each, we fix the depreciation rate in rich countries at 0 percent. We consider allowing the depreciation rate in poor countries to be 1 or 2 percent for 5 or 10 years, motivated by the depreciation of physical strength and stamina. The results are give in rows (14)-(18) of Table 5. The main result is that these robustness checks flatten out experience profiles in poor countries, which only serves to increase the gap between poor and rich countries. We conclude that restricting depreciation rates to be the same in poor and rich countries is probably a conservative assumption.

22

6

Interactions Between Schooling and Experience In this section, we allow for interactions between schooling and age or experience and ask what fraction of

cross-country differences in aggregate wage profiles is accounted for by cross-country differences in education levels.

6.1

Experience-Wage Profiles by Schooling Level

We have so far presented experience-wage profiles under the standard assumption that returns to experience do not vary by educational attainment. We now relax this assumption. That is, we generalize equation (1) to allow schooling and experience to enter in a non-separable fashion, and estimate returns to experience separately for different education groups. That age profiles typically differ across education groups, with more educated individuals having steeper age-earnings or age-wage profiles in developed countries, is well-known from earlier studies such as Mincer (1974), Carroll and Summers (1991), Kambourov and Manovskii (2009) or Guvenen (2007). Thus, we first check whether these patterns also exist in our data. Figure 6 plots age-wage profiles separately for three education groups: “college” (more than 12 years of school), “high school” (9-12 years of school), and “less than high school” (less than 9 years of school). This choice of categories is motivated by a desire to have sufficiently many observations in each group, particularly for poor countries, but we show in Appendix Figure A.12 that similar results apply if we look at finer education categories. For each country, we keep only education groups for which there are at least ten observations in each education-experience bin. We find that profiles are substantially steeper for more educated workers in every single country in our data. The similarity between the rich countries of our sample and the findings from the existing literature reassures us of the integrity of our sample. Interestingly, we find that similar patterns also exist in poor countries.25 Mincer (1974) observed that while age profiles typically differ by education groups, experience profiles tend to be much more parallel. In other words, one would expect age profiles to be mechanically steeper for more educated workers even if experience profiles do not in fact depend on educational attainment. This is the reason that Mincer (1974) controls for experience instead of age in the wage regression. To see this, assume that individuals’ wages satisfy the additively separable Mincer equation (1) and consider the relationship between wages and age a given schooling s, log w = α + θs + f (a − s − 6) + ε. Then the returns to age are mechanically increasing in educational attainment if the experience-wage profile is concave: if f 00 < 0, then ∂ 2 log w = −f 00 (a − s − 6) > 0. ∂a∂s 25 For completeness, Appendix A.6 presents the aggregate cross-sectional age-wage profiles (i.e. not disaggregated by education groups). Analogously to the experience-wage profiles in Figure 1 and Table 2, age-wage profiles are flatter in poor countries.

23

Exploring interactions between schooling and experience and how these differ across countries is therefore also more meaningful than exploring interactions between schooling and age, because the latter exercise would pick up interactions mechanically even if the true experience profiles do not, in fact, depend on educational attainment. We therefore concentrate on interactions between schooling and experience in the remainder of the paper. Figure 7 plots experience-wage profiles separately for our three education groups.26 As expected, we find that experience-wage profiles are much more similar across education groups than age-wage profiles. Nevertheless, experience-wage profiles are moderately steeper for more educated workers in some countries. Amongst poor countries, the differential returns to experience for different education groups are particularly pronounced in Mexico and Brazil.27 This finding is not obvious a priori. In the next section, we explore its implications in more detail.

6.2

Accounting for Experience-Wage Profiles: The Role of Schooling

The finding that experience-wage profiles are steeper for more educated workers in some countries suggests that part of the cross-country differences in average experience-wage profiles may be due to a simple composition effect: in rich countries, a larger share of the workforce is educated and since more educated workers have steeper profiles, this mechanically results in a steeper average profile. To assess the quantitative importance of the cross-country differences in the distribution of educational attainment, we conduct a counterfactual exercise where we ask: what would a country’s experience-wage profile look like if that country had the United States’ distribution of educational attainment as measured by the number of workers in our three education groups.28 If all of the cross-country differences in experience-wage profiles were due to differences in educational attainment, then this counterfactual would eliminate all such differences. Figure 8 plots the average height (the integral under the profiles) of the counterfactual profile against the average height of the actual profile for each country in our sample. If composition effects explained all cross-country differences in the returns to experience, the counterfactual heights for all countries would lie on a straight horizontal line, marked “100%”, at the level of the United States. If they explained none of 26 Because the next section requires estimates by education groups for all our countries and not just our core countries, the figure presents simple cross-sectional profiles as in Figures 1 and 2. For our core countries, we have also computed profiles using our HLT methodology. These are similar and available upon request. 27 One caveat to interpreting these cross-sectional experience-wage profiles by education groups is that there may be differential selection into these education groups across cohorts, particularly in some of the poor countries that have seen a rapid rise in educational attainment across cohorts. For example, the group of workers in older cohorts that have acquired a college education may be more positively selected than that in younger cohorts. This would result in mechanically steep cross-sectional experiencewage profiles of college-educated workers, and mechanically flat cross-sectional profiles of less-educated workers. This logic may partly explain the stark differences in profiles across education groups in Mexico and Brazil. Note that this issue concerns the experience-wage profiles by education group in the present section but not our baseline results that aggregate across these education groups. 28 We also conduct a similar exercise for age-wage profiles, and find modestly higher explanatory power of education composition. Though, given the mechanical composition effect for age profiles discussed in the preceding subsection, we view this exercise as less informative and therefore did not include it in the paper. They are available upon request.

24

the differences, all countries would lie on the 45-degree line marked “0%”. For exposition we also added lines at 25% and 50%. We find that most of our countries lie between the zero- and the fifty-percent lines. For example, for Chile, Mexico and Jamaica, differences in the distribution of educational attainment account for around thirty to forty percent of the difference of each country’s profile relative to the United States. A few countries lie close to or even above the horizontal line, which means that composition effects explain more than the entire gap. However, for all of these countries, the actual profiles are quite similar to the United States to begin with, which means that the gap is small in the first place. Overall, we find that for countries with substantially different experience-wage profiles from the United States, differences in the composition of educational attainment account for twenty-five to forty percent of the difference in experience-wage profiles with the United States. It is important to emphasize that this is an accounting result and not a causal relationship. For example, it could well be that an omitted factor jointly causes low educational attainment and low returns to schooling. On the other hand, it is possible that we have understated the role for education in explaining the patterns in experience profiles because the differences in human capital from education may exceed the differences in years of schooling given the large cross-country differences in education quality (Schoellman, 2012).29 Nonetheless, we view this finding as progress because it implies education is likely an important factor for explaining cross-country differences in life-cycle wage growth. At the same time, the finding suggests that other factors also play important roles.30

7

Potential Explanations In this section, we discuss several potential explanations of lower life-cycle wage growth in poor countries.

We focus on three main candidate theories suggested by the literature: human capital accumulation, searchand-matching frictions and long-term wage contracts. It is beyond the scope of this paper to conclusively rule in or rule out any particular theory. However, we can shed light on the likely relevance of theories by computing additional moments with our data. First, we compute life-cycle wage profiles by broad occupation groups, and find that manual occupations have substantially flatter profiles than cognitive occupations. Second, we compute life-cycle wage variance profiles, finding some evidence of U-shaped profiles, at least in 29 It is useful to note that the returns to experience for college-educated workers in our poor countries are similar to the returns to experience for entirely uneducated workers in rich countries (see Figure A.12 in the Appendix). Hence, education quality can entirely explain cross-country differences in life-cycle wage profiles is if the education quality is so low in poor countries that students learn essentially nothing. 30 We also explored the importance of cross-country differences along dimensions other than education in accounting for age-wage and experience-wage profiles (e.g., share of agriculture, manufacturing, services, public sector employment, etc.). In Section 7.2.1 below we find that manual occupations have flatter profiles than cognitive occupations. Another key difference between rich and poor countries is that poor countries tend to have a much larger share of workers that work in agriculture than rich countries. This could affect our estimates of average experience-wage profiles for each country if profiles are flatter for agricultural workers, which has been found to be true in the United States (Herrendorf and Schoellman, 2015). We therefore conducted an analogous exercise to the one discussed in the text, except that we estimate profiles separately for agriculture and non-agriculture rather than for different education groups. We found that such sectoral differences account for a relatively small fraction of differences in age-wage and experience-wage profiles. These and other results on compositional effects are available upon request.

25

the rich countries. Third, we look at wage profiles for day laborers, who are not engaged in long-term wage contracts, and find that, in the poor countries for which we have data, these are again flatter than in rich countries. Finally, we discuss the evidence from a companion study to this one that looks at experience-wage profiles for immigrants to the United States. We find that these moments are consistent with theories based on human capital or search frictions and less supportive of models based on long-term contracts.

7.1 7.1.1

Candidate Mechanisms Human Capital

There is a long tradition in economics that interprets experience-wage profiles as reflecting human capital accumulation (Becker, 1964). Under this interpretation, our findings imply lower human capital accumulation over the life cycle in poor countries. The simplest version of this theory is that workers in poor countries use simpler technologies, or engage in simpler tasks at work, for which there is less scope for learning. One may imagine that manual occupations, such as agricultural tasks, for example, have fewer possibilities for learning over the life cycle.31 It is well known that manual occupations are more common in poor countries than in rich countries. Another possibility is that workers in poor countries have fewer incentives to accumulate human capital over their lifetimes. Manuelli and Seshadri (2014) propose a model of this type where human capital accumulation requires inputs of both goods and time, as in Ben-Porath (1967). In their model, low TFP depresses the returns to the accumulation of human capital by raising the price of physical inputs to human capital production, thereby resulting in flat experience-wage profiles. Alternatively, extractive institutions in poor countries may discourage workers from investing in human capital, since their returns can be arbitrarily expropriated (Bhattacharya et al., 2013). This logic is consistent with recent evidence that higher taxation of labor income in Europe can explain a substantial fraction of European-U.S. differences in wage inequality and life-cycle wage growth (Guvenen et al., 2014a). Another class of theories based on human capital accumulation focuses on learning through interactions with other individuals. For example, the models of Lucas (2009), Lucas and Moll (2014) and Perla and Tonetti (2014) posit that human capital is accumulated through social interactions with others; all determinants of the frequency or quality of such interactions are potential determinants of cross-country differences in lifecycle wage growth. As one example, de la Croix et al. (2016) argue that in the industrial revolution, the emergence of institutions such as guilds allowed skills to be disseminated faster, which led to increased 31 One way of capturing the idea that simpler technologies result in fewer learning opportunities is as follows. Assume that P the output of a firm is Leontief in the firm’s technology and the human capital of each of the workers it employs: y= N i=1 min {A, hi } where y denotes the firm’s output, A its technology and hi is the human capital of each of its workers indexed by i = 1, ..., N . Since the human capital of a worker equals zero once her human capital reaches A, no worker has an incentive to invest past this point.

26

lifetime human capital accumulation, and hence economic growth. 7.1.2

Search and Matching Frictions

Another candidate explanation for slow life-cycle wage growth in poor countries are search and matching frictions. If the labor market features search frictions and match-specific productivity, slow life-cycle wage growth in poor countries may partly reflect low labor market turnover. This could work through several mechanisms. Burdett (1978), Jovanovic (1984), Burdett and Mortensen (1998) and Bagger et al. (2014) emphasize on-the-job search as a theory of job shopping. If frictions to search and matching lower the incentives or ability of workers to shop for jobs, they are less likely to climb the job ladder and will forego some of the potential increase in labor productivity over the life cycle.32 Empirically, one would then expect to see workers in poor countries experiencing fewer job-to-job transitions and to receive smaller wage gains during such transitions.33 Although these ideas have been applied to studying aggregate labor productivity over the business cycle or across developed countries (Lise and Robin, 2013; Postel-Vinay and Turon, 2014), the large cross-country differences in life-cycle wage profiles suggests a new avenue for exploration. Alternatively, long-lasting frictions may prevent workers from sorting to the jobs that are most suitable to their heterogeneous skills and tastes (Hsieh et al., 2013). Again, the implication would be that workers forego labor productivity increases as they age. 7.1.3

Long-Term Contracts

Finally, if workers and firms form long-term contracts (e.g., Lazear, 1979), wages may not equal workers’ marginal product of labor, and this may lead to cross-country differences in returns to experience. One version of these theories features back-loaded contracts, where workers get less than their marginal products when young and more when they are older. This is the typical prediction of theories with moral hazard or limited commitment on the part of workers. These theories have the potential to explain our finding that experience-wage profiles are steeper in rich countries if long-term contracts are more prevalent in rich countries or they are equally prevalent but more back-loaded. A second version of these theories features front-loaded contracts, where workers get paid more than their marginal product when young, and less later in the life cycle. Front-loading could arise, for example, because firms implicitly lend to financially constrained workers (Azariadis, 1988; Bernhardt and Timmis, 1990). To explain our findings, one would need a theory with more front-loading in poor countries. In summary, long-term contracts have the 32 Burdett (1978, p.219) puts it succinctly: “In the present study it has been assumed workers do not accumulate human capital while working. Older workers in the present study receive higher wage rates, on average, because they have obtained more job offers. And the more job offers a worker receives, the greater the probability a ‘high’ wage rate job will be found.” 33 There is little work comparing such moments between rich and poor countries or regions. One exception is Heise and Porzio (2015) who use matched employer-employee data from Germany to compare wage dynamics in West Germany with those in the considerably poorer East. They find that, when moving job-to-job, workers in the East experience smaller wage gains than those in the richer West, i.e. they face a flatter job ladder.

27

potential to explain flatter experience-wage profiles in poor countries if either wages are more front-loaded in poor countries or more back-loaded in rich countries or both.

7.2

Distinguishing Between Mechanisms: Additional Moments

We now examine some additional moments in our data to attempt to distinguish between potential mechanisms. We find several pieces of evidence that are consistent with human capital and search frictions contributing to our main findings, and no obvious evidence that long term contracts play an important role. Distinguishing between human capital and search is, in turn, more difficult for three reasons. First, our cross-sectional data do not allow us to construct moments typically thought to be informative about these two classes of theories, e.g. wage changes with job-to-job transitions. Second, search and human capital accumulation are difficult to tell apart even with high-quality panel data or matched employer-employee data. One can think of workers in search models accumulating “search capital” while employed that is destroyed upon job loss (Manning, 2000) and, at an abstract level, certain search theories may be observationally equivalent to theories of human capital accumulation. A number of existing studies assess the relative contribution of human capital and job search in the United States and other high-income countries.34 While these studies generally agree that both human capital and search are important determinants of life-cycle wage growth, there is considerable debate about the precise quantitative importance of the two mechanisms, consistent with the idea that the two are hard to tell apart. Third, another difficulty is that there are likely non-trivial interactions between search and human capital (Bowlus and Liu, 2013). With this in mind, we now present four pieces of evidence that make some progress in distinguishing between potential explanations for cross-country differences in life-cycle wage growth. 7.2.1

Experience-Wage Profiles by Occupation

We first explore differences in wage profiles by occupations. Most of our samples report data on occupation. Although the occupational coding schemes vary by country, we are able to translate occupational codes into the ILO’s International Standard Classification of Occupations (ISCO) at the one-digit level. We then aggregate the ISCO one-digit occupational categories further to two broad categories: manual and cognitive. The reason we choose this split is that manual occupations are relatively intensive in physical tasks, rather than mental ones, and may therefore have less scope for learning. If so, this may help explain our overall findings, since manual occupations are more common in poor countries. In the data, we define manual occupations as elementary occupations, agricultural workers, and plant/machine operators and assemblers (ISCO codes 6, 8 and 9). Cognitive occupations include legislators and managers, professionals, technicians, clerks, service and sales, and craftsmen (ISCO codes 1-5 and 7). We exclude from 34 See for example Topel and Ward (1992), Rubinstein and Weiss (2006), Altonji et al. (2013), Bowlus and Liu (2013) and Bagger et al. (2014).

28

this analysis workers with missing occupations or those in the armed forces. For each country, we compute the life-cycle wage profile of workers in manual and cognitive occupations following our cross-sectional method of Section 3.35 In Figure 9, we plot the resulting profiles by occupation for our core countries. Two facts stand out. First, for all of our core countries, the profiles are flatter for manual occupations than for cognitive occupations. Second, the difference is often substantial. For the average country in our sample the return to 20–24 years of experience is 23 percentage points higher for workers in cognitive than in manual occupations. These differences potentially matter because of the sizable differences in employment composition by country (plotted in Appendix Figure A.13). Rich countries have considerably more workers in cognitive occupations, at roughly 80 percent, while in poor countries the labor force is fairly evenly split. For most countries, the employment composition across occupations changes little in magnitude over the life cycle, consistent with the idea that most workers will not systematically move from manual to cognitive work over the life cycle. The differences in life-cycle wage growth across occupations we find are particularly remarkable given our crude classification of workers into manual and cognitive occupations, and given the likely possibility that a given manual occupation may vary in its manual and cognitive intensity between rich and poor countries. We conjecture that, more fine-grained data that allow for a less noisy proxy for occupations with different learning opportunities, would likely result in even larger differences in employment shares between rich and poor countries. As with education, we use a simple accounting exercise to quantify the importance of occupation for aggregate life-cycle wage profile differences across countries. As a reminder, the accounting exercise holds fixed each country’s experience profiles by occupation and asks how much steeper the country’s experience profile would be if it instead had the U.S. employment shares in cognitive and manual occupations. According to this accounting exercise, for most countries, the distribution of employment across occupations accounts for between zero and twenty-five percent of the difference of each country’s profile relative to the United States. Given the striking differences in experience-wage profiles across occupations in Figure 9, it is surprising that this counterfactual does not have larger predictive power. Mechanically, this is due to the fact that, with our crude occupational classification, even in our poorest countries, the employment share in manual occupations is only around fifty percent. Our findings are consistent with a simple human capital interpretation, namely that there are simply 35 The reason for presenting the cross-sectional estimates rather than our HLT estimates is that the accounting exercise below requires estimates by occupation groups for all our countries and not just our core countries. Moreover, the HLT methodology requires the assumption that there are no occupation transitions in the last five or ten years of the life cycle, which could be more controversial than the analogous assumption about wage growth in our baseline HLT exercise. In any case, for our core countries, we have also computed profiles by occupation using our HLT methodology. These are similar to the cross-sectional estimates and available upon request.

29

less learning opportunities in poor countries over the life cycle. They also parallel our previous results that more educated workers have steeper experience-wage profiles in all countries. In that case, the simple human capital interpretation is that education helps one learn how to acquire human capital later in life.36 7.2.2

Experience-Wage Profiles of Day Laborers

We next explore the importance of labor market contracts for our patterns. We note first that theories of long-term contracting all refer to the returns to tenure (experience at a specific firm) rather than the return to life-time potential experience. Thus, long-term contracts are unlikely to have large quantitative effects on experience-wage profiles unless average worker tenure is reasonably long. The limited data on worker tenure do not support this: for example, in the United States, the median tenure is 4.6 years (Bureau of Labor Statistics, 2012) and for Brazil, Chile, Guatemala, Jamaica, Peru and Uruguay it ranges between 1.5 and 5.5 years (Interamerican Development Bank, 2016).37 To provide additional evidence on this point, we study the life-cycle wage profiles of workers employed without long-term contracts. The rationale is that these profiles speak directly to the question of whether our patterns are likely to be explained by contractual arrangements between workers and firms that introduce a wedge between life-cycle wage and life-cycle productivity profiles. We provide such evidence for three countries for which we can identify workers who are unlikely to be on long-term contracts. For India and Mexico, we can identify a subset of workers who are daily workers: those whose employer varies on a dayto-day basis. Clearly, these workers’ profiles are not driven by employer-specific contracting. For the United States, we draw on the Current Population Survey, where we can identify workers who work part-time and are not interested in working full-time even if it were available. These workers report other commitments (home, family, and so on) that make full-time work undesirable. Since these workers apparently value the flexibility and lack of commitment that come with short-term work arrangements, our interpretation is that they are also unlikely to be taking part in long-term contracts. Figure A.15 presents the experience-wage profiles for workers on short-term contracts and the rest of the workforce in these three countries. Recall from above that long-term contracts can explain flatter experiencewage profiles in poor countries in two scenarios. The first scenario is that wages are more front-loaded in poor countries. In this case we would expect day laborers in poor countries to have steeper profiles than the rest of the work force. The second scenario is that wages are more back-loaded in rich countries. In 36 One possible concern is that we may be replicating the results from the education section to the extent that less-educated workers are employed in manual occupations. We explore this idea by estimating separately life-cycle wage profiles by education and occupation. Appendix Figure A.14 plots the profiles by occupation conditional on workers having a high-school education (i.e. we drop workers in the “less than high-school” and “college” categories). We find that each dimension matters and the accounting results are similar. 37 Interamerican Development Bank (2016) only provides estimates for Central and South American countries and we are not aware of corresponding data for other countries in our sample. The numbers for Brazil, Chile, Guatemala, Jamaica, Peru and Uruguay are median tenure for the most recent year in which this data is available.

30

this case, we would expect day laborers in rich countries to have flatter profiles. We find no evidence of either of these scenarios. In India and the United States the two sets of profiles are quite similar and in Mexico the day laborers have flatter profiles than other workers. An obvious caveat to this exercise is that day laborers are likely a selected group of workers that differ from the rest of the work force in a number of other characteristics, for example their skills. Nevertheless, the comparison of day laborers to other workers together with the relatively short job tenure of the typical worker in both rich and poor countries suggests that long-term contracts are unlikely to explain our cross-country patterns. 7.2.3

Variance Profiles

Thus far, we have focused on mean wages by experience. Now we turn our attention to the variance of wages by experience to see if they can provide any additional information. Both human capital and search models make a wide variety of predictions for how the variance of wages might evolve over the life cycle. Hence, these predictions are useful for discriminating among specific human capital theories or among search theories, but not between the two. For human capital models, the key determinant of the shape of variance profiles is the correlation between learning ability and initial human capital (Huggett et al., 2011). Intuitively, individuals with higher learning ability endogenously choose to invest more at the beginning of the life cycle. These choices lead them to have steeper wage profiles than individuals with low learning ability. The level of initial human capital mostly affects the intercept of the profile. Hence, if the two are weakly correlated, the model predicts that high learning ability individuals have lower levels but steeper slopes of wages than their low learning ability counterparts. Therefore profiles cross at some point, implying a U-shaped pattern for the variance of wages over the life cycle (Mincer, 1974; Rubinstein and Weiss, 2006). On the other hand if the two are sufficiently strongly positively correlated, then high learning ability individuals will have higher levels and steeper slopes of wages, implying that the variance of wages rises continuously over the life cycle. Among search theories, we focus on theories of on-the-job search because these have implications for life-cycle wage profiles. Hornstein et al. (2011) show that some recent theories of on-the-job search have the potential to generate realistic levels of wage dispersion. The implications for the shape of variance profiles over the life cycle are less clear. Among existing theories, Bagger et al. (2014) stands out for making quantitative predictions about the life-cycle profile of wage dispersion, finding that it rises at a decreasing rate.38 The key mechanism is a rise in the heterogeneity of firms that workers are matched with over the life cycle. In contrast, Manning (2000) presents a simple model of on-the-job search that generates a U-shaped life-cycle variance profile. 38 Burdett et al. (2011) also formulate a model with life-cycle patterns of dispersion, but they do not calibrate or estimate the model to provide a full description of what the profile might look like.

31

To compute variances over the life cycle, we take the variance of log wages in each experience bin, by the three education groups discussed above. We then take the weighted average log wage variance across experience bins, where each education group is weighted by their share among all workers. Figure 10 plots profiles for the variance of the logarithm of wages across countries. The empirical results generally follow one of two patterns. In four of the five richest countries (top-left quadrant) as well as in South Korea and Jamaica, variance profiles follow a U-shape: declining at the beginning of the life cycle and then rising toward the end. In the remaining countries (the United States and most poorer countries), the profiles are generally rising throughout the life cycle.39 In summary, the evidence on variance profiles is mixed and appears consistent with different versions of both human capital and search theories. For both classes of theories, our empirical results imply restrictions on the range of plausible parameterizations that will be consistent with the data in different countries. This is an interesting avenue for future research.40 7.2.4

Evidence from Immigrants to the United States

In a companion paper, Lagakos et al. (2016) study how wages in the United States vary for immigrant workers of different experience levels. We document that, for immigrants from rich countries, more experienced immigrants tend to earn substantially more than less experienced immigrants. In contrast, for immigrants from poor countries, more experienced immigrants tend to earn only somewhat more than the less experienced. We show that this is true for new immigrants, who earned all their experience abroad, as well as for all immigrants when we control for the amount of U.S. work experience. Since all individuals are observed in the same labor market, this finding is consistent with the hypothesis that immigrants from poor countries accumulate less life-cycle human capital than immigrants from rich countries before coming to the United States. It is also consistent with the theories based on differential selection and skill loss, though our companion paper provides evidence against these theories. Perhaps the most powerful piece of evidence supporting the human capital interpretation of these findings is that the returns to experience we estimate from non-migrants (e.g. the estimates in the current paper) line up closely with the returns we estimate using U.S immigrants. 39 In some countries, the variance profiles are mildly declining at the end of the life cycle, a pattern that could be interpreted as an inverse U-shape. 40 In Ben-Porath-type models, one form of investment in human capital accumulation may be reflected in part-time work and low hours worked more broadly. To investigate this possibility, in Appendix A.7 we relax the restriction to only full-time workers that we have made so far and examine hours and earnings profiles over the life-cycle in this larger sample. We find that in rich countries, hours rise steeply early on in the life cycle, largely reflecting movements from part-time into full-time work. These cross-country patterns are consistent with theories of human capital accumulation, in particular the prediction that workers in rich countries initially invest a lot of time into human capital accumulation, and therefore work little, but then increase their time working over their life-cycle. The fact that poor countries generally do not see such hours increases is consistent with the prediction of less investment early in the life cycle. We also examine variance profiles in the larger sample including both part-time and full-time workers and find stronger evidence of a U-shape (perhaps unsurprisingly given the evidence on hours profiles).

32

7.2.5

Summary of Evidence

In summary, the four pieces of evidence we presented in this section suggest that long-term contracts are unlikely to be an important driver of cross-country differences in life-cycle wage growth. In contrast, both human capital and search appear consistent with the moments we have presented here. A particularly simple explanation for slow life-cycle wage growth in poor countries is that workers in poor countries may simply have fewer opportunities for learning due to the nature of the occupations or tasks they perform, consistent with the evidence on experience-wage profiles across occupations we presented above. At the same time and as noted in the beginning of this section, it is generally hard to tell apart human capital and search as drivers of life-cycle wage growth, and as such, more severe labor market frictions are an equally promising candidate explanation for flat experience-wage profiles in poor countries.

8

Conclusion This paper documents that experience-wage profiles are steeper in rich countries than in poor countries.

In the rich countries, the wages of the most experienced workers are on average almost one hundred percent larger than the wages of the least experienced workers. In contrast, in the poor countries, the wages of the most experienced workers are only around fifty percent larger than wages of the least experienced workers. We find that some, but not all, of this pattern is accounted for by differences in education levels across countries, with more educated workers having steeper profiles. While it is difficult to provide a definitive explanation of our findings, several additional moments of our data support human capital or search frictions as promising explanations. Providing a more definitive explanation for cross-country differences in life-cycle wage growth is an important task for future research. In particular, doing so could help account for cross-country income differences. Earlier studies in this literature found no relationship between returns to experience and GDP per capita (Psacharopoulos, 1994; Bils and Klenow, 2000), and concluded that it was safe to ignore any cross-country human capital differences arising through experience, rather than schooling. Recent work by Manuelli and Seshadri (2014) predicts that workers in rich countries accumulate more human capital over the life cycle as well, and our evidence offers support for this idea. If true, the importance of human capital in accounting for income differences is substantially higher than previously concluded.41 Turning to the implications of a search-based explanation of flat experience profiles in poor countries, the macro-development literature generally assumes competitive labor markets. The main exceptions are papers 41 Any development accounting exercise along these lines would need to be careful to distinguish between two different forms of human capital accumulation with potentially different quantitative implications: on-the-job training as in Ben-Porath (1967) and learning-by-doing. Through the lens of models of on-the-job training, the steep wage growth in early years of the life-cycle in rich countries may partly reflect a declining time investment into human capital accumulation, so that the underlying human capital profile is flatter than the observed wage profile. Therefore, cross-country differences in life-cycle human capital would be smaller than those implied by simple learning-by-doing models in which a worker’s wage is proportional to her human capital.

33

that focus on distortions to the allocation across sectors or locations, which generate misallocations of labor (Restuccia et al., 2008; Caselli, 2005; Gollin et al., 2014). Viewed through the lens of a search and matching model, our findings suggest that it may be time to incorporate analogous frictions to on-the-job search or job choice that generate misallocation of labor over the life cycle. Explorations along these lines would be particularly interesting given that low labor market turnover in poor countries could have implications for a number of important issues besides aggregate productivity. For example, low turnover could hamper mobility if poor workers escaping poverty involves an element of “job shopping.”

34

References Adhvaryu, Achyuta, Namrata Kala, and Anant Nyshadham, “Booms, Busts, and Household Enterprise: Evidence from Coffee Farmers in Tanzania,” 2014. University of Michigan Working Paper. Aguiar, Mark and Erik Hurst, “Deconstructing Life Cycle Expenditure,” Journal of Political Economy, 2013, 121 (3), 437–492. , , and Loukas Karabarbounis, “The Life-Cycle Profile of Time Spent on Job Search,” American Economic Review Papers and Proceedings, 2013, 103 (3), 111–116. Altonji, Joseph G., Anthony A. Smith Jr., and Ivan Vidangos, “Modeling Earnings Dynamics,” Econometrica, 2013, 81 (4), 1395–1454. Azariadis, Costas, “Human Capital and Self-Enforcing Contracts,” Scandinavian Journal of Economics, 1988, 90 (4), 507–28. Bagger, Jesper, Francois Fontaine, Fabien Postel-Vinay, and Jean-Marc Robin, “Tenure, Experience, Human Capital, and Wages: A Tractable Equilibrium Search Model of Wage Dynamics,” American Economic Review, June 2014, 104 (6), 1551–96. Becker, Gary S., Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education, University of Chicago Press, 1964. Ben-Porath, Yoram, “The Production of Human Capital and the Life Cycle of Earnings,” Journal of Political Economy, 1967, 75 (4), 352–365. Bernhardt, Dan and Gerald C Timmis, “Multiperiod Wage Contracts and Productivity Profiles,” Journal of Labor Economics, October 1990, 8 (4), 529–63. Bhattacharya, Dhritman, Nezih Guner, and Gustavo Ventura, “Distortions, Endogenous Managerial Skills and Productivity Differences,” Review of Economic Dynamics, January 2013, 16 (1), 11–25. Bils, Mark and Peter J. Klenow, “Does Schooling Cause Growth?,” American Economic Review, December 2000, 90 (5), 1160–83. Bourguignon, Francois and Thierry Magnac, “Labor Supply and Taxation in France,” Journal of Human Resources, 1990, 25 (3), 358–389. Bowlus, Audra J. and Chris Robinson, “Human Capital Prices, Productivity, and Growth,” American Economic Review, 2012, 102 (7), 3483–3515. and Huju Liu, “The contributions of search and human capital to earnings growth over the life cycle,” European Economic Review, 2013, 64, 305–331. Burdett, Kenneth, “A Theory of Employee Job Search and Quit Rates,” The American Economic Review, 1978, 68 (1), 212–220. and Dale T Mortensen, “Wage Differentials, Employer Size, and Unemployment,” International Economic Review, May 1998, 39 (2), 257–73. , Carlos Carrillo-Tudela, and Melvyn G. Coles, “Human Capital Accumulation and Labor Market Equilibrium,” International Economic Review, August 2011, 52 (3), 657–677. Bureau of Labor Statistics, “Employee Tenure Summary,” 2012. Cagan, Phillip, “Measuring Quality Changes and the Purchasing Power of Money: an Exploratory Study of Automobiles,” The National Banking Review: a Journal of Policy and Practice, 1965. Carroll, Christopher D. and Lawrence H. Summers, “Consumption Growth Parallels Income Growth: Some New Evidence,” in B. Douglas Bernheim and John B. Shoven, eds., National Saving and Economic Performance, Chicago: University of Chicago Press, 1991, pp. 305–43. 35

Caselli, Francesco, “Accounting for Cross-Country Income Differences,” in P. Aghion and S. Durlauf., eds., Handbook of Economic Growth, 679-741, Elsevier, 2005. Christensen, Bent Jesper, Rasmus Lentz, Dale T. Mortensen, George R. Neumann, and Axel Werwatz, “On-the-Job Search and the Wage Distribution,” Journal of Labor Economics, 2005, 23 (1), 31–58. de la Croix, David, Matthias Doepke, and Joel Mokyr, “Clans, Guilds, and Markets: Apprenticeship Institutions and Growth in the Pre-Industrial Economy,” 2016. Unpublished Working Paper, Northwestern University. Deaton, Angus, The Analysis of Household Surveys: A Microeconometric Approach to Development Policy, The World Bank, 1997. Duflo, Esther, “Schooling and Labor Market Consequences of School Construction in Indonesia: Evidence from an Unusual Policy Experiment,” American Economic Review, 2001, 91 (4), 795–813. Gollin, Douglas, “Getting Income Shares Right,” Journal of Political Economy, April 2002, 110 (2), 458– 474. , David Lagakos, and Michael E. Waugh, “The Agricultural Productivity Gap,” Quarterly Journal of Economics, 2014, 129 (2), 939–993. Gould, William and Finis Welch, “Imputing Labor Market Experience,” 1976. mimeograph, Rand Corporation. Gunther, Isabel and Andrey Launov, “Informal employment in developing countries: Opportunity or last resort?,” Journal of Development Economics, 2012, 97 (1), 88 – 98. Guvenen, Fatih, “Learning Your Earning: Are Labor Income Shocks Really Very Persistent?,” American Economic Review, June 2007, 97 (3), 687–712. , Burhanettin Kuruscu, and Serdar Ozkan, “Taxation of Human Capital and Wage Inequality: A Cross-Country Analysis,” Review of Economic Studies, 2014, 81 (2), 818–850. , Fatih Karahan, Serdar Ozkan, and Jae Song, “What Do Data on Millions of U.S. Workers Reveal about Life-Cycle Earnings Risk?,” 2014. mimeo, Federal Reserve Bank of Minneapolis. Hall, Robert E., “Technical Change and Capital from the Point of View of the Dual,” The Review of Economic Studies, 1968, 35 (1), pp. 35–46. Hanoch, Giora, “An Economic Analysis of Earnings and Schooling,” Journal of Human Resources, 1967, 2 (3), 310–329. Heckman, James and Richard Robb, “Using Longitudinal Data to Estimate Age, Period and Cohort Effects in Earnings Equations,” in William M. Mason and Stephen E. Fienberg, eds., Cohort Analysis in Social Research, Springer New York, 1985, pp. 137–150. , Lance Lochner, and Christopher Taber, “Explaining Rising Wage Inequality: Explanations With A Dynamic General Equilibrium Model of Labor Earnings With Heterogeneous Agents,” Review of Economic Dynamics, January 1998, 1 (1), 1–58. Heise, Sebastian and Tommaso Porzio, “Unemployment Risk and Flat Job Ladders: Lessons from the German Reunification,” Yale University Working Paper 2015. Herrendorf, Berthold and Todd Schoellman, “Why is Measured Productivity so Low in Agriculture?,” Review of Economic Dynamics, 2015, 18 (4), 1003–1022. Hornstein, Andreas, Per Krusell, and Giovanni L. Violante, “Frictional Wage Dispersion in Search Models: A Quantitative Assessment,” American Economic Review, December 2011, 101, 2873–2898.

36

Hsieh, Chang-Tai, Erik Hurst, Charles I. Jones, and Peter J. Klenow, “The Allocation of Talent and U.S. Economic Growth,” NBER Working Papers 18693 January 2013. Huggett, Mark, Gustavo Ventura, and Amir Yaron, “Sources of Lifetime Inequality,” American Economic Review, 2011, 101 (7), 2923–54. Hurst, Erik, Geng Li, and Ben Pugsley, “Are Household Surveys Like Tax Forms: Evidence from Income Underreporting of the Self Employed,” Review of Economics and Statistics, 2014, 96 (1), 19–33. Interamerican Development Bank, “Soci’ometro-BID,” 2016. Jovanovic, Boyan, “Matching, Turnover, and Unemployment,” Journal of Political Economy, 1984, 92 (1), 972–990. Kambourov, Gueorgui and Iourii Manovskii, “Accounting for the Changing Life-Cycle Profile of Earnings,” Working Paper 2009. King, Miriam, Steven Ruggles, J. Trent Alexander, Sarah Flood, Katie Genadek, Matthew B. Schroeder, Brandon Trampe, and Rebecca Vick, Integrated Public Use Microdata Series, Current Population Survey: Version 3.0. [Machine-readable database], University of Minnesota, 2010. Klenow, Pete and Andres Rodriguez-Clare, “The Neoclassical Revival in Growth Economics: Has It Gone Too Far?,” in “NBER Macroeconomics Annual 1997, Volume 12,” National Bureau of Economic Research, 1997, pp. 73–114. La Porta, Rafael and Andrei Shleifer, “The Unofficial Economy and Economic Development,” Brookings Papers on Economic Activity, 2008, 39 (2 (Fall)), 275–363. Lagakos, David, Benjamin Moll, Tommaso Porzio, Nancy Qian, and Todd Schoellman, “Experience Matters: Human Capital and Development Accounting,” NBER Working Papers 18602 December 2012. , , , , and , “Lifecycle Human Capital Accumulation Across Countries: Lessons From U.S. Immigrants,” NBER Working Papers 21914 January 2016. Lazear, Edward P, “Why Is There Mandatory Retirement?,” Journal of Political Economy, December 1979, 87 (6), 1261–84. Lehmann, Erich Leo and Joseph P. Romano, Testing Statistical Hypotheses Springer Texts in Statistics, New York, NY: Springer Science+Business Media, 2005. Lemieux, Thomas, “The Mincer Equation Thirty Years After Schooling, Experience, and Earnings,” in S. Grossbard-Schechtman, ed., Jacob Mincer, A Pioneer of Modern Labor Economics, Springer, 2006, chapter 11, pp. 127–145. Lerner, Josh and Antoinette Schoar, eds, International Differences in Entrepreneurship, National Bureau of Economic Research, 2010. Lise, Jeremy and Jean-Marc Robin, “The macro-dynamics of sorting between workers and firms,” IFS Working Papers W13/22, Institute for Fiscal Studies August 2013. Lucas, Robert E., “Ideas and Growth,” Economica, 2009, 76 (301), 1–19. and Benjamin Moll, “Knowledge Growth and the Allocation of Time,” Journal of Political Economy, 2014, 122 (1), 1 – 51. Manning, Alan, “Movin’ on up: Interpreting the Earnings-Experience Profile,” Bulletin of Economic Research, October 2000, 52 (4), 261–95. Manuelli, Rodolfo E. and Ananth Seshadri, “Human Capital and the Wealth of Nations,” American Economic Review, 2014, 104 (9), 2736–62. 37

McKenzie, David J, “Disentangling age, cohort and time effects in the additive model,” Oxford Bulletin of Economics and Statistics, 2006, 68 (4), 473–495. Mincer, Jacob A., Schooling, Experience, and Earnings, National Bureau of Economic Research, 1974. Minnesota Population Center, Integrated Public Use Microdata Series, International: Version 6.1 [Machine-readable database], University of Minnesota, 2011. Murphy, Kevin M and Finis Welch, “Empirical Age-Earnings Profiles,” Journal of Labor Economics, April 1990, 8 (2), 202–29. Paulson, Anna and Robert Townsend, “Financial Constraints and Entrepreneurship: Evidence from the Thai Financial Crisis,” Economic Perspectives, 2005, 29, pp. 34–48. Perla, Jesse and Christopher Tonetti, “Equilibrium Imitation and Growth,” Journal of Political Economy, 2014, 122 (1), 52 – 76. Poschke, Markus, “Entrepreneurs out of necessity: a snapshot,” Applied Economics Letters, 2013, 20 (7), 658–663. Postel-Vinay, Fabien and Hélène Turon, “The Impact of Firing Restrictions on Labour Market Equilibrium in the Presence of On-the-job Search,” Economic Journal, 2014, 124 (575), 31–61. Psacharopoulos, George, “Returns to Investment in Education: A Global Update,” World Development, 1994, 22 (9), 1325–43. Restuccia, Diego, Dennis Tao Yang, and Xiaodong Zhu, “Agriculture and Aggregate Productivity: A Quantitative Cross-Country Analysis,” Journal of Monetary Economics, 2008, 55, 234–250. Rubinstein, Yona and Yoram Weiss, “Post Schooling Wage Growth: Investment, Search and Learning,” in Erik Hanushek and F. Welch, eds., Erik Hanushek and F. Welch, eds., Vol. 1 of Handbook of the Economics of Education, Elsevier, 2006, chapter 1, pp. 1–67. Salthouse, Timothy A., “Interrelations of Aging, Knowledge, and Cognitive Performance,” in Ursula M. Staudinger and Ulman Lindenberger, eds., Understanding Human Development: Dialogues with Lifespan Psychology, Springer, 2003, chapter 12, pp. 265 – 287. , “Within-Cohort Age-Related Differences in Cognitive Functioning,” Psychological Science, 2013, 24 (2), 123 – 130. Schoar, Antoinette, “The Divide between Subsistence and Transformational Entrepreneurship,” in Josh Lerner and Scott Stern, eds., Innovation Policy and the Economy, Vol. 10, National Bureau of Economic Research, 2010, pp. 57–81. Schoellman, Todd, “Education Quality and Development Accounting,” The Review of Economic Studies, March 2012, 3 (1), 133–175. Schulhofer-Wohl, Sam, “The Age-Time-Cohort Problem and the Identification of Structural Parameters in Life-Cycle Models,” 2013. mimeo, Federal Reserve Bank of Minneapolis. Smith, James P. and Finis Welch, “Race Differences in Earnings: A Survey and New Evidence,” Technical Report R-2295-NSF, Rand Corporation 1978. Tanaka, Hirofumi and Douglas R. Seals, “Endurance exercise performance in Masters athletes: ageassociated changes and underlying physiological mechanisms,” The Journal of Physiology, 2008, 586 (1), 55–63. Todaro, Michael P. and Stephen C. Smith, Economic Development, 11th edition ed., Addison Wesley, 2012.

38

Topel, Robert H and Michael P Ward, “Job Mobility and the Careers of Young Men,” The Quarterly Journal of Economics, May 1992, 107 (2), 439–79. Welch, Finis, “Effects of Cohort Size on Earnings: The Baby Boom Babies’ Financial Bust,” Journal of Political Economy, 1979, 87 (5), S65–S97. World Masters Athletics Current Records World Masters Athletics Current Records, 2010. Accessed 4/4/2016 at http: // www. world-masters-athletics. org/ records/ current-records .

39

150

Percent Wage Increase Relative to Experience <5 Years 50 100

150

Figure 1: Cross-Sectional Experience-Wage Profiles, Core Countries

100

Germany

United States

Canada Brazil United Kingdom

50

Chile

Mexico

0

0

Jamaica

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Note: Experience-wage profiles are for full-time males working in the private sector, and are calculated using all available years of data for each country. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. The wage is defined to be earnings divided by hours worked. For each country and year, we compute the ratio of average wages for workers in each 5-year experience bin relative to the average wages of workers with less than five years of experience. The experience-wage profiles in the figure are the unweighted average wage ratios by experience across all years. The left-hand panel are the rich countries, defined as those with greater than twenty thousand dollars GDP per capita in 2011 at PPP, and the right-hand panel are the poor countries, defined as those with less than twenty thousand dollars GDP per capita in 2011 at PPP.

40

150

Germany

Australia

United States

100

Percent Wage Increase 50 100

150

Figure 2: Cross-Sectional Experience-Wage Profiles, All Countries

France

Uruguay

50

Canada

S. Korea

United Kingdom

0

0

Chile

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

100

150

5

Percent Wage Increase 50 100 150

0

Indonesia

Brazil India

50

Peru Mexico

0

0

Guatemala 0

5

10

15 20 25 30 Potential Experience

Bangladesh

Jamaica

35

40

0

5

10

Vietnam

15 20 25 30 Potential Experience

35

40

Note: Experience-wage profiles are for full-time males working in the private sector, and are calculated using all available years of data for each country. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. The wage is defined to be earnings divided by hours worked. For each country and year, we compute the ratio of average wages for workers in each 5-year experience bin relative to the average wages of workers with less than five years of experience. The experience-wage profiles in the figure are the unweighted average wage ratios by experience across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel

41

Figure 3: Deaton-Hall Experience-Wage Profiles

250 200

Percent Wage Increase Relative to Experience <5 Years 50 100 150 200

250

(a) All Growth Driven by Cohort Effects

150

Brazil

United Kingdom

Germany

100

Jamaica

Chile Canada Mexico

0

0

50

United States

0

5

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

150

150

(b) All Growth Driven by Time Effects

Percent Wage Increase Relative to Experience <5 Years 50 100

Germany

100

Canada

Brazil United Kingdom United States

Chile

50

Jamaica

0

0

Mexico

0

5

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

Note: Experience-wage profiles are for full-time males working in the private sector. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. The wage is defined to be earnings divided by hours worked. The top panel shows the experience-wage profiles estimated using equation (2), with time controls as well as cohort controls, assuming that all growth is driven by cohort effects. The bottom panel shows the experience-wage profiles estimated using equation (2), with time controls as well as cohort controls, assuming that all growth is driven by time effects. See Section 4.2 and Appendix A.3 for a detailed description of our methodology.

42

150

Percent Wage Increase Relative to Experience <5 Years 50 100

150

Figure 4: Heckman-Lochner-Taber (HLT) Experience-Wage Profiles

100

Germany

Canada

United States Brazil

50

United Kingdom

Mexico Chile

0

0

Jamaica

0

5

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

Note: Experience-wage profiles are for full-time males working in the private sector. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. The wage is defined to be earnings divided by hours worked. The experience-wage profiles are estimated using equation (2) with time controls as well as cohort controls, assuming that the last ten years of potential experience have zero percent growth. See Section 4.3 for a detailed description of our methodology.

43

Figure 5: Experience-Wage Profiles using Panel Data

0

Percentage Wage Increase 50 100

150

(a) Panel (U.S. PSID and Mexico FLS)

0

10

20 Potential Experience United States

30

40

Mexico

0

Percentage Wage Increase 50 100

150

(b) Repeated Cross-Sections for Comparison (reproduced from Figure 1)

0

10

20 Potential Experience United States

30

40

Mexico

Notes: In Figure 5a, a cohort group comprises fives years of experience levels (e.g., the youngest cohort in Mexico had 0-4 years of experience in 2002). Each line segment in the figure is the normalized mean wage of a cohort over time. Figure 5b shows the analogous profiles from the repeated cross-sectional data (these are identical to those in Figure 1). The data are from the U.S. PSID (bi-annually, 2003 – 2013) and the Mexican FLS (2002, 2005 and 2009).

44

100 150

Germany

Canada

0

50

United States

30

40

50

60

United Kingdom

20

30

40

50

60

Chile

20

30

40

50

60

30

40

50

60

Brazil

0

50

100 150

20

30

40

50

60

Mexico

20

30

40

50

60

20

Jamaica

College

50

100 150

20

High School Less than H.S.

0

Percent Wage Increase Relative to Age 21-25

Figure 6: Age-Wage Profiles by Education Group

20

30

40

50

60

20

30

40

50

60

Note: Age-wage profiles are for full-time males working in the private sector, and are calculated for each educational attainment group. The wage is defined to be earnings divided by hours worked. College, High School and Less than H.S. mean that the individual (i) attended some college or graduated from college, (ii) attended some high school or graduated high school, and (ii) did not attend high school. For each country, year and education group, we compute the ratio of average wages for workers in each 5-year age bin relative to the average wages of workers in the same education group that are 20-24 years of age. The age-wage profiles in the figure are the unweighted average wage ratios by age group, for each education group, across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

45

50 100 150

Germany

Canada

0

United States

10

20

30

40

United Kingdom

0

10

20

30

40

Chile

0

10

20

30

40

10

20

30

40

Brazil

0

50 100 150

0

50 100 150

0

10

20

30

40

Mexico

0

10

20

30

40

0

Jamaica College High School Less than H.S.

0

Percent Wage Increase Relative to Experience <5 Years

Figure 7: Experience-Wage Profiles by Education Group

0

10

20

30

40

0

10

20

30

40

Note: Experience-wage profiles are for full-time males working in the private sector, and are calculated for each educational attainment group. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. The wage is defined to be earnings divided by hours worked. College, High School and Less than High School mean that the individual (i) attended some college or graduated from college, (ii) attended some high school or graduated high school, and (ii) did not attend high school. For each country, year and education group, we compute the ratio of average wages for workers in each 5-year experience bin relative to the average wages of workers in the same education group with less than five years of experience. The experience-wage profiles in the figure are the unweighted average wage ratios by experience, for each education group, across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

46

100

Figure 8: Contribution of Education to Cross-Country Differences in Experience-Wage Profiles

Counterfactual Average Return 50

KOR BRA

100%

FRAURY USA CAN IDN GBR AUS CHL GTM BGD

PER

MEX

50%

25%

DEU

JAM

VNM

0

0% 0

50 Actual Average Return

100

Note: Each point on the graph represents the actual and counterfactual average height of the experience-wage profile for one country. The average height of the experience-wage profile is the height of the profile for experience bins other than the smallest, relative to the smallest experience bin. The counterfactual average height is the same statistic calculated under the assumption that the fraction of workers in each education bin – College, High School, and Less than High School – is the same as in the United States. See Section 6 for a more detailed description of our methodology.

47

100 150

Germany

Canada

0

50

United States

10

20

30

40

United Kingdom

0

10

20

30

40

Chile

0

10

20

30

40

20

30

40

Mexico

0

50

100 150

0

10

20

30

40

Brazil

0

10

20

30

40

0

10

Jamaica Cognitive Manual

50

100 150

0

0

Percent Wage Increase Relative to Experience <5 Years

Figure 9: Experience-Wage Profiles by Occupation Group

0

10

20

30

40

0

10

20

30

40

Note: Experience-wage profiles are for full-time males working in the private sector, and are calculated separately for each occupation group. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. The wage is defined to be earnings divided by hours worked. Cognitive occupations comprise categories 1-5, and 7 of the ISCO-08 (International Standard Classification of Occupations). They are: (1) Managers, (2) Professionals, (3) Technicians and associate professionals, (4) Clerical support workers, (5) Service and sales workers, and (7) Craft and related trades worker. Manual occupations comprise categories 6, 8 and 9 of the ISCO-08. They are: (6) Skilled agricultural, forestry and fishery workers; (8) Plant and machine operators, and assemblers, and (9) Elementary occupations. For each country, year and occupation group, we compute the ratio of average wages for workers in each 5-year experience bin relative to the average wages of workers in the same education group with less than five years of experience. The experience-wage profiles in the figure are the unweighted average wage ratios by experience, for each occupation group, across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

48

1 .75

Variance of Log Wages .5 .75

1

Figure 10: Wage Variance Profiles, All Countries

Chile

United States

Uruguay S. Korea .5

Canada Australia Germany

United Kingdom .25

.25

France 10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

35

40

1

5

1

0

Variance of Log Wages .5 .75

Brazil Jamaica

Indonesia

India

Guatemala

.5

Peru

.75

Mexico

.25

.25

Bangladesh Vietnam

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

Note: The wage variances profiles are for males working in the private sector and are calculated using all available years of data for each country. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. For each country, year and education group, we compute the variance of log wages. The wage-variance profiles in the figure are the weighted averages of the log wage variances across years and education groups, weighted by the shares of workers in each education group. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

49

Table 1: Summary of Data

United States* Germany* Australia Canada* France United Kingdom* South Korea Chile* Uruguay Mexico* Brazil* Peru Indonesia Jamaica* Guatemala Vietnam India Bangladesh

(1) GDPpc (2011) 49,781 42,143 41,763 41,567 37,325 36,590 31,327 20,266 17,905 15,730 14,831 10,379 8,870 8,481 6,799 4,717 4,686 2,579

(2) Data Source Census of Population, American Community Survey German Socioeconomic Panel (SOEP) Household Income & Labour Dynamics Census of Canada Survey of Employment British Household Panel Survey (BHPS) Korea Labor and Income Panel Study National Socioeconomic Survey (CASEN) Extended National Survey of Households General Population and Housing Census General Census of Brazil National Household Survey National Labor Force Survey Population Census National Living Standards Survey Living Standards Survey Human Development Survey Household Income and Expenditure Survey

(3) Years Covered 1960-2013 1991-2009 2001-2009 1971-2001 1993-2001 1994-2008 1999-2008 1990-2011 2006 1990-2010 1991-2010 2004, 2010 2001-2010 1982-2001 2000, 2006 1998, 2002 2012 2005, 2010

Note: * indicates core countries. GDP data are from the World Bank’s World Development Indicators (2016), and the measure used is 2011 GDP per capita (PPP) in constant 2011 international dollars. For exact years of each survey, see Appendix A.1.

50

Table 2: Experience-Wage Profiles

United States Germany Canada United Kingdom Chile Brazil Mexico Jamaica

Rich Mean Poor Mean Rich - Poor

Panel A: Summary Statistics by Country (1) (2) (3) Height at Height at Average Experience 20-24 Experience 35-39 Height 88.7 88.1 67.9 105.3 108.0 84.2 78.1 68.7 59.3 85.1 61.5 61.6 45.0 37.6 35.7 71.8 60.0 53.2 41.1 23.5 30.0 32.5 26.4 25.1

(4) Discounted Avg Height 30.5 38.8 27.3 29.3 16.7 24.3 14.5 11.8

Panel B: Test of Differences in Means, Rich and Poor Groups (1) (2) (3) (4) Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height 89.3 81.6 68.3 31.5 47.6 36.9 36.0 16.8 41.7∗∗ 44.7∗∗ 32.3∗∗ 14.7∗∗ (.012) (.014) (.014) (.015)

Note: The first column of Panel A is the average height of the experience-wage profile at potential experience of 20-24 years, defined as the ratio of average wages for workers with 20-24 years of potential experience to average wages for workers with 0-4 years of potential experience. The second column is the average height of the experience-wage profile at experience 35-39 years, defined as the ratio of average wages for workers with 35-39 years of potential experience to average wages for workers with 0-4 years of potential experience. The third column is the average height of the profile relative to workers with 0-4 years of potential experience. The fourth column is the discounted average height of the profile relative to workers with 0-4 years of potential experience, where wages are discounted at a rate of four percent per year. The sample is restricted to full-time males in the private sector. Panel B shows the results of permutation tests of the null hypothesis that the experience-wage profiles are the same in rich and poor countries. *** denotes p-value less than 0.01; ** denotes p-value less than 0.05; * denotes p-value less than 0.10.

51

Table 3: Deaton-Hall Experience-Wage Profiles

Rich Mean Poor Mean Rich - Poor

Panel A: All Growth Explained by Cohort Effects (1) (2) (3) (4) Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height 90.9 94.1 70.9 32.0 80.1 132.7 70.5 28.9 10.8 −38.7 0.4 3.1 (.352) (.839) (.489) (.365)

Panel B: Growth Explained Equally by Cohort and Time Effects (1) (2) (3) (4) Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height Rich Mean 93.2 96.7 72.7 32.7 Poor Mean 65.8 99.5 56.6 23.7 Rich - Poor 27.4∗ −2.8 16.1 9.0∗ (.082) (.562) (.106) (.089)

Rich Mean Poor Mean Rich - Poor

Panel C: All Growth Explained by Time Effects (1) (2) (3) Height at Height at Average Experience 20-24 Experience 35-39 Height 95.8 101.1 75.0 51.4 69.9 43.4 44.4∗∗ 31.1 31.6∗∗ (.029) (.117) (.042)

(4) Discounted Avg Height 33.6 18.6 15.0∗∗ (.024)

Note: This table reports summary statistics of experience-wage profiles estimated using the Deaton-Hall method under the assumptions that all growth is driven by cohort effects (Panel A), growth is equally explained by cohort and time effects (Panel B), and that all growth is driven by time effects (Panel C). The rows present the average of the rich countries, the average of the poor countries and the difference between the rich and poor means, plus the results of permutation tests of the null hypothesis that the experience-wage profiles for rich and poor are the same. *** denotes p-value less than 0.01; ** denotes p-value less than 0.05; * denotes p-value less than 0.10. The first column is the average height of the experience-wage profile at potential experience of 20-24 years, defined as the ratio of average wages for workers with 20-24 years of potential experience to average wages for workers with 0-4 years of potential experience. The second column is the average height of the experience-wage profile at experience 35-39 years, defined as the ratio of average wages for workers with 35-39 years of potential experience to average wages for workers with 0-4 years of potential experience. The third column is the average height of the profile relative to workers with 0-4 years of potential experience. The fourth column is the discounted average height of the profile relative to workers with 0-4 years of potential experience, where wages are discounted at a rate of four percent per year.

52

Table 4: Heckman-Lochner-Taber (HLT) Experience-Wage Profiles Panel A: No Experience Effects in last 10 Years, 0% Depreciation Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height Rich Mean 79.3 80.8 62.5 28.5 Poor Mean 39.2 43.3 31.3 14.0 Rich - Poor 40.1∗∗ 37.5∗∗ 31.2∗∗ 14.5∗∗ (.013) (.013) (.013) (.013) Panel B: No Experience Effects in last 5 Years, 0% Depreciation Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height Rich Mean 90.3 100.7 72.1 32.3 Poor Mean 33.2 33.1 26.2 12.0 ∗∗ ∗∗ ∗∗ Rich - Poor 57.0 67.6 45.9 20.3∗∗ (.013) (.013) (.016) (.015) Panel C: No Experience Effects in last 10 Years, 1% Depreciation Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height Rich Mean 47.1 27.5 35.2 17.7 Poor Mean 14.3 1.3 10.0 5.6 Rich - Poor 32.7∗∗ 26.2∗∗ 25.3∗∗ 12.2∗∗ (.015) (.017) (.015) (.014) Panel D: No Experience Effects in last 5 Years, 1% Depreciation Height at Height at Average Discounted Avg Experience 20-24 Experience 35-39 Height Height Rich Mean 55.9 41.3 42.6 20.7 Poor Mean 9.4 -6.0 5.9 3.9 Rich - Poor 46.5∗∗ 47.3∗∗ 36.7∗∗ 16.8∗∗ (.014) (.013) (.014) (.015) Note: This table reports summary statistics of the estimated experience-wage profiles estimated under the assumption that there are no experience effects in the last ten years of potential experience and no depreciation (Panel A), no experience effects in the last five years and no depreciation (Panel B), no experience effects in the ten years and one percent depreciation (Panel C), and no experience effects in the last five years and one percent depreciation (Panel D). The rows present the average of the rich countries, the average of the poor countries and the difference between the rich and poor mean, plus the results of permutation tests of the null hypothesis that the experience-wage profiles for rich and poor are the same. *** denotes p-value less than 0.01; ** denotes p-value less than 0.05; * denotes p-value less than 0.10. The first column is the average height of the experience-wage profile at potential experience of 20-24 years, defined as the ratio of average wages for workers with 20-24 years of potential experience to average wages for workers with 0-4 years of potential experience. The second column is the average height of the experience-wage profile at experience 35-39 years, defined as the ratio of average wages for workers with 35-39 years of potential experience to average wages for workers with 0-4 years of potential experience. The third column is the average height of the profile relative to workers with 0-4 years of potential experience. The fourth column is the discounted average height of the profile relative to workers with 0-4 years of potential experience, where wages are discounted at a rate of four percent per year.

53

Table 5: Robustness Height at 20-24 Years Experience, HLT Profiles

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17)

Baseline Experience at 16 Constructed experience Measurement error: age Measurement error: education Measurement error: age and education Include Self-Employed Include Public-Sector Employees Include women Constructed experience, men and women Include Part-Time (20+ hours) Include Part-Time (> 0 hours) Constructed experience, incl. Part-Time 1% depreciation for last 10 years in Poor 2% depreciation for last 10 years in Poor 1% depreciation for last 5 years in Poor 2% depreciation for last 5 years in Poor

Rich 79.3 82.1 90 81.1 71.7 74.4 80.3 80.4 70 76.6 83 84.8 100 79.3 79.3 90.3 90.3

Poor 39.2 45.8 43.5 39.2 39.2 39.2 36.6 42.2 29.1 25.5 38.2 36.7 42 14.3 -6.1 9.4 -10.2

Rich - Poor 40.1** 36.3** 46.5** 41.9** 32.5** 35.2** 43.6** 38.2** 41** 51.1** 44.8** 48.1** 58.0** 64.9** 85.4*** 80.9*** 100.5***

Note: Row (1) uses the baseline sample and measures. Row (2) expands the sample to include individuals who are age 16 and 17. Row (3) uses constructed experience instead of potential experience (see Section 5.1.1). Row (4) adds noise to the age variable in rich countries in order to match the same amount of age heaping observed in Chilean data (see Section 5.1.2). Row (5) adds noise to the years of education variable in rich countries by assuming that the distribution in the years of education for a given level of educational attainment is the same as in Chile (see Section 5.1.2). Row (6) adds noise to both the age and years of education variables in rich countries. Row (7) includes self-employed workers. Row (8) includes public-sector workers. Row (9) includes female workers. Row (10) includes female workers and uses constructed experience, constructed separately for male and female workers. Row (11) includes part-time workers who work at least 20 hours per week. Row (12) includes all part-time workers. Row (13) includes all part-time workers and use constructed experience, where it is constructed separately for full-time and part-time workers. Row (14) assumes 1% depreciation in the last 10 years in the Poor core countries, and 0% depreciation in the last 10 years in the Rich core countries. Row (15) assumes 2% depreciation in the last 10 years in the Poor core countries, and 0% depreciation in the last 10 years in the Rich core countries. Row (16) assumes 1% depreciation in the last 5 years in the Poor core countries, and 0% depreciation in the last 5 years in the Rich core countries. Row (17) assumes 2% depreciation in the last 5 years in the Poor core countries, and 0% depreciation in the last 5 years in the Rich core countries. *** denotes p-value less than 0.01; ** denotes p-value less than 0.05; ** denotes p-value less than 0.05; * denotes p-value less than 0.10.

54

A

Appendix – For Online Publication

A.1

Data Sources

The surveys we employ in our analysis are listed below for each country. All surveys are nationally representative. We attempted to obtain data for every country in the world with a population greater than one million people. We obtained a number of surveys through the Integrated Public Use Microdata Series (IPUMS) (Minnesota Population Center, 2011; King et al., 2010), which can be found here: www.ipums.org. The remaining surveys were made available to us by the statistical agencies of the countries in question or other sources, as listed below. • Australia: Household Income and Labour Dynamics in Australia, yearly from 2001 to 2009, from the Australian Department of Families, Housing, Community Services and Indigenous Affairs, available from the Cornell Department of Policy Analysis and Management. • Bangladesh: Household Income and Expenditure Survey, 2005 and 2010, from the Bangladesh Bureau of Statistics, available from the FAO RIGA database. • Brazil: Recenseamento Geral do Brasil, Censo Demográfico, 1991 (5.8% sample), and 2000 (6% sample), 2010 (5% sample) from the Instituto Brasileiro de Geografia e Estatística (IBGE), available from IPUMS. • Canada: Census of Canada, 1971 (1% Sample), 1981 (2% Sample), 1991 (3% Sample) and 2001 (2.7% Sample), available from IPUMS. • Chile: National Socioeconomic Characterization Survey (CASEN), 1990, 1992, 1994, 1996, 1998, 2000, 2003, 2006 and 2009, 2011 from the Chilean Ministry of Planning and Cooperation. • France: Enquete Emploi, yearly from 1993 to 2001, from the Ministre de l’Économie de l’Industrie et de l’Emploi. • Germany: German Socioeconomic Panel (SOEP), yearly from 1991 to 2009, from the German Institute for Economic Research (DIW Berlin). • Guatemala: Encuesta Nacional de Condiciones de Vida, 2000 and 2006, from the Instituto Nacional de Estadistica. • India: India Human Development Survey (IHDS), 2012, from the Indian National Council of Applied Economic Research, available from Inter-university Consortium for Political and Social Research (ICPSR) at the University of Michigan • Indonesia: National Labour Force Survey (SAKERNAS), 2001 to 2010 from the Indonesia Badan Pusat Statistik. • Jamaica: Population Census, 1982, 1991 and 2001, (10% samples) from the Statistical Institute of Jamaica, available from IPUMS. • Mexico: XI General Population and Housing Census, 1990 (10% sample); Population and Dwelling Count, 1995 (0.4% of sample); XII General Population and Housing Census, 2000 (10.6% of sample); Population and Housing Census, 2010 (10% of sample) available from IPUMS. • Peru: Encuesta Nacional de Hogares, 2004 and 2010, from the from the Instituto Nacional de Estadística y Informática. • South Korea: Korea Labor and Income Panel Study, yearly from 1999 to 2008, from the Korea Labor Institute, available from the Cornell Department of Policy Analysis and Management.

55

• United Kingdom: British Household Panel Survey, yearly from 1994 to 2008, from the Institute for Social & Economic Research at the University of Essex. • United States: Census of Population and Housing, 1960 (1% Sample), 1970 (1% Sample), 1980 (5% Sample), 1990 (5% Sample) , 2000 (5% Sample); American Community Survey, yearly from 2005 to 2013 (1% Sample); all available from IPUMS. • Uruguay: Extended National Survey of Households, 2006, from the Uruguay National Institute of Statistics, available from IPUMS. • Vietnam: Living Standards Survey, 1998 and Household Living Standards Survey, 2002, both from the General Statistics Office of Vietnam, available from the FAO RIGA. All calculations in our analysis are weighted using the applicable sample weights for each survey. We express all earnings and wage data in local currency units of the most recent year in the data using the consumer price index of the country in question, taken from the IMF’s International Financial Statistics database. In each survey, we drop the top and bottom one percent of earners to remove potential outliers, and to minimize the impact of potential cross-country differences in top-coding procedures. For most countries, we measure hours as the actual hours worked in the past week (or in some recent reference week.) For the United States and Brazil, we measure hours as the usual weekly hours worked (which is what is available). We define an individual to be a full-time worker if she works more than 30 hours per week. For most countries, labor earnings and hours worked are for both primary and secondary jobs. In Chile, France, South Korea and Uruguay, labor earnings and hours worked are for just the primary job. In Australia, Canada, Germany, Jamaica, South Korea, and the United States, earnings are measured at the annual frequency. In the remaining countries, earnings are measured at the monthly frequency. In all surveys, earnings are before taxes. The numbers for per capita GDP at PPP in 2000 that we use are taken from the World Bank’s World Development Indicators (WDI).

A.2

Details on Deaton-Hall Approach

As stated in the main text, we use an approach suggested by Deaton (1997) and Hall (1968) and impose one additional linear restriction to estimate (1) (or in practice (2)). We report results for three different versions of this approach. The purpose of this Appendix is to explain in detail the three different linear restrictions we impose and how we implement these in practice. To derive restrictions on these, consider total wages in year t Nct XX Wt = wict c∈Ct i=1

where Ct is the set of cohorts working at time t and Nct is the number of members of cohort c working at time t. Assuming that wages satisfy the Mincerian equation (1), it is not hard to show that total wages can be written as ¯ t F¯t Wt = Γt X

(3)

Γt = exp(γt ) X Fct ¯t = X exp(χc ) ¯ Ft c∈C

(4) (5)

t

F¯t =

X c∈Ct

Fct ,

Fct =

Nct X

exp(θsict + f (xict ) + εict )

(6)

i=1

Now consider the growth of total wages from one year to the next. Equation (3) says that there are three ¯ t , and growth due sources to total wage growth: growth due to time effects Γt , growth due to cohort effects X 56

to changes in the composition of schooling and experience as captured by F¯t . We will impose restrictions on the term ¯t, Ωt = Γt X (7) which can be thought of as a year-specific aggregate labor productivity term. This term can improve over time for two reasons: (i) directly, due to increases in cohort-neutral changes captured by Γt ; and (ii) indirectly, due to changes in the composition of cohorts active in the labor market captured by X¯t . For example, suppose that young cohorts are more productive than old ones (i.e. they have higher χc ). Then aggregate productivity will improve over time as old cohorts exit the labor force and young cohorts enter. Equation (7) is the key equation we use to derive meaningful restrictions on cohort and time effects. The basic idea is to decompose the time series for Ωt into a trend component and a cyclical component and to make assumptions on whether the trend component (“productivity growth”) is attributed to time or to cohort effects. ¯t. In practice, this is implemented as follows. First, it is convenient to work in logs:Pωt = log Ωt , χ ¯t = log X T Second, define time periods in deviations from the sample mean, i.e. such that T1 t=0 t = 0, and similarly PT PT renormalize γt and χ ¯t such that T1 t=0 γt = T1 t=0 χ ¯t = 0. This can be achieved by writing ωt = log Ωt as ωt = ω ¯ + γt + χ ¯t , (8) where ω ¯ is an appropriately chosen constant. Second, the series for γt and χ ¯t can be decomposed into a trend component and a cyclical component γt = gγ t + uγ,t ,

χ ¯t = gχ t + uχ,t ,

PT

PT

where gγ =

t=0 γt t PT 2 , t=0 t

gχ = Pt=0 T

χ ¯t t

t=0

t2

.

(9)

(10)

Intuitively, one simply runs a regression of γt and χt on time, thereby decomposing each time series into a trend component and a component that is orthogonal to the time trend. Finally, from (8) and (9), the logarithm of aggregate productivity ωt = log Ωt is ωt = ω ¯ + gM t + uM,t , where gM = gγ + gχ and uM,t = uγ,t + uχ,t . The different restrictions we use are then simply different ways of splitting gM between gγ and gχ . The three restrictions for which we present results in the main text (see in particular Figure 3 and Table 3) are as follows. Restriction 1 (All Growth Explained by Cohort Effects): the time trend in productivity growth is entirely due to cohort effects: gM = gχ , gγ = 0. PT From (10) this implies the linear restriction t=0 γt t = 0, i.e. that time effects are orthogonal to a time trend and capture only cyclical variation. This is the same restriction as equation (2.94) in Deaton (1997). Restriction 2 (All Growth Explained by Time Effects): the time trend in productivity growth is entirely due to year effects: gM = gγ , gχ = 0. PT ¯t t = 0 or From (10) this implies the linear restriction t=0 χ T X t=0

log

X c∈Ct

Fct exp(χc ) ¯ Ft

! t = 0.

(11)

Note that the variable Fct enters this restriction. Since this requires estimating equation (1), it is necessary to use an iterative procedure. Restriction 3 (Growth Explained Equally by Cohort and Time Effects): a share β of the time

57

trend in productivity growth is due to year effects: gγ = βgM ,

gχ = (1 − β)gM .

From (10), this implies the linear restriction β

T X

χ ¯t t = (1 − β)

t=0

T X

γt t.

(12)

t=0

In the main text, we represent results for the case β = 1/2. Finally, it can be seen that restrictions 1 and 2 are the special cases β = 0 and β = 1.

A.3

Returns to Experience over the Life Cycle: Additional Statistics

Table A.4 presents additional results on the differences in returns to experience between rich and poor countries that complement the cross-sectional results in Table 2 as well as the HLT results in Table 4. In particular we examine in more detail at which experience levels the largest gap in returns to experience between rich and poor countries occurs. Panel A shows the relative wages of those with 5-9 years of experience relative to those with 0-4 years of experience, panel B for those with 10-14 years of experience, panel C for those with 20-24 years and panel D those for 30-34 years. Column (1) states the cross-sectional results. Column (2) shows the Heckman-Lochner-Taber results. Panel C simply reproduces the main results from Table 2 column (1), which we re-state here for comparison purposes. These results show that, for our preferred HLT specification, out of the 40.1 percent gap in returns to experience between rich and poor countries, 23.1 percent are due to the first 5 years and by 10 years, the gap is already 31.5 percent. In contrast, going from 20-24 to 30-34 years of experience hardly affects the gap. These findings confirm the point emphasized in the main text that the majority of the difference in returns to experience between rich and poor countries occurs early on in the life cycle.

A.4

Attrition in Panel Data

In the main text, we ignore attrition from the panels when we estimate the wage residuals for workers who remain in wage employment and those who move into self-employment, which is on average 23 percent and 9.3 percent across survey waves of the Mexican FLS and U.S. PSID. This can bias the estimates if attrition is correlated to wages and the propensity to move into self-employment. Since this propensity is unobservable, we address this concern by adding in all of the individuals that earned wages in year t − 1, but who exited the panel in year t and estimate two bounds for the wages gaps. The first assumes that all those who drop out of the panel would have moved to self-employment. The second assumes that they all move into wage employment. The estimated wage gaps using these samples are shown in Appendix Figures A.10a - A.10d. They are very similar to the main results in the text because, on average, those who exit the sample earn slightly lower wages than those who remain in wage-employment and slightly higher wages than those who move to self-employment.42

A.5

Selection into or out of the Public Sector and Part-Time Employment

This section presents the empirical results examining the importance of selection into or out of the public sector as well as part-time employment discussed in section 5.2.2. Figures A.8a and A.8b show that U.S. workers who remain in the private sector earn higher wages than those who move from the private to the 42 A similar exercise addresses how attrition can affect our examination of selection out of self-employment. The main exercise ignores entrants into the panel data – i.e., it excludes workers in year t who were not also in the panel in year t − 1, which is, on average, 24% and 8.5% across survey waves for the FLS and PSID. Thus, as before we can add these individuals into the sample and alternatively assume that all new entrants worked in wage or self-employment in year t − 1. The results are similar. We conduct similar exercises to investigate whether attrition from the panel can affect our examination of selection into and out of the private sector and full-time employment. In all cases, we find that including those who exit the panel does little to change the main results. These results are available upon request.

58

public sector. In Mexico, such workers earn slightly lower wages when they have 25-30 years of experience. Similarly, Figures A.8c and A.8d show that workers who are in the private sector for two periods in a row always earn higher wages than workers who move in from the public sector in the United States, and slightly lower wages for workers who have 15 to 20 years of experience in Mexico.43 The product of the wage gap between those who switch in and out of public-sector employment and the rate of moving across sectors is very small, which implies that selection into our out of public-sector employment is unlikely to drive our findings that experience-wage profiles are steeper in rich countries. Figures A.9a – A.9d present analogous results for selection into and out of part-time employment, and show that U.S. workers who move into or out of part-time employment across survey waves earn lower wages, while wages for the different types of workers are similar in Mexico. The product of the wage gap between those who switch in and out of full-time employment and the rate of moving is very small, which implies that selection into our out of full-time employment is unlikely to drive our findings that experience-wage profiles are steeper in rich countries.

A.6

Age-Wage Profiles

Figure A.11 presents age-wage profiles for our eight core countries without conditioning on educational attainment. As in the main text, wages are defined simply as earnings divided by the number of hours worked, and we restrict the sample to full-time males in the private sector. Figure A.11 plots age-wage profiles for our core countries – for rich countries on the left-hand panel and for poor countries on the righthand panel, analogous to Figure 1. We find that the profiles of richer countries are steeper. Panel A of Table A.3 reports the summary statistics for each country, analgous to Table 2. Panel B of Table A.3 presents permutation tests of the null hypothesis that age-wage profiles are the same in the rich and poor countries. The mean for richer countries is larger and statistically different from that for poor countries for all of the summary statistics. Thus, steeper age-wage profiles in rich countries show that our main result on life-cycle wage growth does not depend on our measure of years of education. One reason for preferring experience profiles like those in the main text is that age profiles inevitably mix some less-educated workers who have been working for several years with some other highly-educated workers who have only just entered the labor force.

A.7 A.7.1

Full- and Part-Time Workers Hours and Earnings Profiles for Full- and Part-Time Workers

Some theories of life-cycle wage growth have predictions for earnings and hours worked separately from wages. To address this, we examine how hours worked and earnings vary over the life cycle in our eighteen countries. Most theories of human capital predict that time investments into human capital accumulation are high early in the life cycle, and then decline. To the extent that hours worked reflect time not investing, hours increases over the life cycle may reflect the decrease in investment time predicted by the theory.44 Since earnings equal the product of wages and hours, human capital theories similarly have implications for earnings profiles. In order to do this, we relax the restriction to only full-time workers that we have made so far, as movements from part-time work to full-time work may be one part of the decline in investment. A.16 plots experience-wage profiles under this alternative sample restriction. These are exactly analogous to the experience-wage profiles in Figure 2, only here we include all workers, regardless of whether they work full time or not. The results are similar to our baseline results. Figure A.17 plots experience-hours profiles for all eighteen countries in our sample. In the United States, Canada, Australia and Germany, hours increase by around 35 percent to 50 percent in the first ten years of potential experience, remaining largely constant after that. France, in contrast has a very modest change 43 Note that in Figure A.8d for Mexico, there are too few workers with 0 to 5 years of experience who move into the private sector for us to estimate the differential wages for this bin. 44 The challenge with mapping human capital investment theory into the data is that hours spent accumulating human capital are rarely observed directly. Some fraction of the hours worked reported in household surveys like ours may actually correspond more with human capital accumulation than with producing output.

59

in hours over the life cycle, consistent with the conclusions of e.g. Bourguignon and Magnac (1990), that France’s labor market lacks flexibility in the number of hours worked conditional on employment. Among the poor countries, India and Peru have the largest increase, at around 15 percent, while the rest have increases of ten percent or less. Thus, while not all rich countries have large increases in hours over the life cycle, on average, the rich countries have substantially larger hours increases than do the poor countries. Figure A.18 shows the profiles for earnings over the life-cycle, again including both full-time and parttime workers. The richer countries generally have dramatic increases in earnings in the first five to ten years, and overall have much higher earnings growth over the life cycle than do the poor countries. Among the rich countries, Germany has the largest increase in earnings, at around 200 percent within ten years, with the United States, Canada and Australia having increases of around 150 percent, the United Kingdom at just under 100 percent, and France at 60 percent. Not surprisingly, the earnings profiles echo the hours profiles, with the highest earnings increases coming for those (rich) countries that have substantial hours increases.45 These cross-country patterns are consistent with theories of human capital accumulation, in particular the prediction that workers in rich countries initially invest a lot of time into human capital accumulation, and therefore work little, but then increase their time working over their life-cycle. The fact that poor countries generally do not see such hours increases is consistent with the prediction of less investment early in the life cycle. These patterns are also consistent with some search theories, for example that of Christensen et al. (2005), where workers search more early on in the life cycle and this is also reflected in lower initial hours worked.46 As discussed in the beginning of section 7.1, searching for a better match is conceptually similar to a form of human capital accumulation, theories based on human capital investment and search are quite similar, if not isomorphic. Finally, note that this pattern is inconsistent with theories in which long-term contracts in rich countries induce high effort for early periods in life-cycle. Our data show that, on average, hours in rich countries are substantially lower for workers with less than ten years of experience than for workers with more than ten years of experience. Even conditioning on full-time employment, hours change very little over the life cycle, and in no country do we observe hours decreasing in potential experience. A.7.2

Variance Profiles for Full- and Part-Time Workers

Figure A.19 plots profiles for the variance of the logarithm of earnings across countries. In the seven richest countries, i.e. those in the top-left quadrant plus the United Kingdom and South Korea, variance profiles are all declining in the beginning of the life cycle, with an increase in the variance at later years of experience, a pattern that can be interpreted as a U-shape. In contrast, in most poor countries, variance profiles are flat or weakly increasing. India, Bangladesh and Peru have modest decreases early in the life cycle, while the remaining eight countries are flat or increasing.

45 When restricting to full-time workers, as in our main analysis, earnings profiles are similar to wage profiles of Figure 2. Similarly, when computing the hours profiles but restricting to full-time workers, the profiles are flat in every single country, with no country having an hours increase of even ten percent over the life cycle. Thus, the hours increases in profiles in Figure A.17 almost entirely reflect movements from part-time work to full-time work. 46 Aguiar et al. (2013) look at job search among the unemployed over the life cycle using time-use surveys in a set of rich economies. They find a hump-shaped pattern in the United States, decreasing job search over the life cycle in the United Kingdom, Germany, France and Spain, and a flat (but imprecisely estimated) profile in Italy. To our knowledge, no study has looked at life-cycle patterns of job search in poorer countries, or life-cycle patterns of on-the-job search even in richer countries.

60

150

Percentage Wage Increase Relative to Experience <5 Years 50 100

150

Figure A.1: Experience-Wage Profiles (Cross-Sectional Estimates) with 95% Confidence Intervals

100

Germany United States Canada Brazil United Kingdom 50

Chile

Mexico

0

0

Jamaica

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Note: These profiles are the same as Figure 1, except that they also plot the 95% confidence intervals.

150

Percentage Wage Increase Relative to Experience <5 Years 50 100

150

Figure A.2: Experience-Wage Profiles (HLT) with 95% Confidence Intervals

100

Germany Canada United States Brazil 50

United Kingdom

Mexico Chile

0

0

Jamaica

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Note: These profiles are the same as Figure 4, except that they also plot the 95% confidence intervals.

61

150

150

Figure A.3: HLT Experience-Wage Profiles – Robustness to Age Heaping

100

Percentage Wage Increase 50 100

Germany

Canada United States United Kingdom 50

Brazil

Mexico Chile

0

0

Jamaica

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Notes: These figures plot the experience-wage profiles estimated using the HLT method as in Figure 4. In both panels, the solid lines are estimated using actual data. In the left panel, the dashed line is estimated using data where we add noise to the age variable in order to match the same amount of age heaping observed in Chilean data.

62

150 100

Percentage Wage Increase 50 100

150

Figure A.4: HLT Experience-Wage Profiles – Robustness to Dispersion in the Years of Education

Germany Canada

United States

Brazil 50

United Kingdom Mexico Chile

0

0

Jamaica

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Notes: These figures plot the experience-wage profiles estimated using the HLT method as in Figure 4. In both panels, the solid lines are estimated using actual data. In the left panel, the dashed line is estimated using data where the years of education for workers are forced to have the same distribution for each reported level of educational attainment as in Chile (shown in Appendix Tables A.1).

63

150

150

Figure A.5: HLT Experience-Wage Profiles – Robustness to Age Heaping and Dispersion in the Years of Education

100

Percentage Wage Increase 50 100

Germany

Canada United States

Brazil 50

United Kingdom

Mexico Chile

0

0

Jamaica

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Notes: These figures plot the experience-wage profiles estimated using the HLT method as in Figure 4. In both panels, the solid lines are estimated using actual data. In the left panel, the dashed line is estimated using data where we add noise to the age variable in order to match the same amount of age heaping observed in Chilean data and the years of education for workers are forced to have the same distribution for each reported level of educational attainment as in Chile (shown in Appendix Tables A.1).

64

Figure A.6: Actual and Distorted Age Distributions of Core Countries (a) United States .04 Fraction of Individual at Age x .01 .02 .03 0

0

Fraction of Individual at Age x .01 .02 .03

.04

(b) Canada

18

25

30

35

40 age

45

50

55

60

18

25

30

Age Distribution Age Heaping as in Chile

35

40 age

45

50

55

60

50

55

60

45

50

55

60

45

50

55

60

Age Distribution Age Heaping as in Chile

Fraction of Individual at Age x .02 .03 .01

.01

Fraction of Individual at Age x .02 .03

.04

(d) United Kingdom

.04

(c) Germany

18

25

30

35

40 age

45

50

55

60

18

25

30

Age Distribution Age Heaping as in Chile

35

40 age

45

Age Distribution Age Heaping as in Chile

Fraction of Individual at Age x .02 .03

0

.01

Fraction of Individual at Age x .01 .02 .03

.04

(f) Chile

.04

(e) Mexico

18

25

30

35

40 age

45

50

55

60

18

25

30

40 age

(h) Jamaica

0

.01

Fraction of Individual at Age x .01 .02 .03

Fraction of Individual at Age x .02 .03

.04

.04

(g) Brazil

35

18

25

30

35

40 age

45

50

55

60

18

25

30

35

40 age

Notes: These figures are histograms of worker ages for each of the eight countries in the core sample. In Figures A.6a–A.6d, the light gray bars are the actual data while the white bars with black outlines are the artificially distorted data. Figures A.6e–A.6h only plot actual data. These data are used to estimate the profiles shown in Figure A.3.

65

Figure A.7: Selection into and out of Self-Employment – Panel Data Estimates

20 Potential Experience

30

5 -5 0 Percent -10 -15

-50 -75

Percent -25 0

-10

-5 0 Percent

10

-15

Percent -25 0 -50 -75

0

10

50 25

10

(b) Selection into self-employment – Mexico

5

25

50

(a) Selection into self-employment – U.S.

0

Wage Gap: Wage Workers - Self Employed (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

30

10

50

30

-5 0 Percent -10 -15

-50 -75

Percent -25 0

5

25

10 -10

-5 0 Percent

20 Potential Experience

-15

Percent -25 0 -50 -75

10

20 Potential Experience

(d) Selection out of self-employment – Mexico

5

25

50

(c) Selection out of self-employment –U.S.

0

10

Wage Gap: Wage Workers - Self Employed (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

0

Wage Gap: Wage Workers - Self Employed (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

10

20 Potential Experience

30

Wage Gap: Wage Workers - Self Employed (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

Notes: Figures A.7a and A.7b plot the estimated differences in wage residuals and their 95% confidence intervals for those who move from wage employment in year t − 1 to self-employment in year t and those who are in wage employment in both years. Wages are predicted from regressing log wagest−1 on dummy variables of five-year experience bins, the interactions of each experience-bin dummy variable with a dummy variable for whether the worker is self-employedt , controlling for the years of educational attainment and year fixed effects. The reference group is the 0-4 years of experience bin for workers who remain in wage work. t refers to the year of the survey. We transform the coefficients from logs to levels so that the y-axis reflects the level of percentage increase relative to the reference group. The figures also present the rate at which workers move from wage- to self-employment at each point on the life cycle and the product of this rate and the wage-gap. An analogous regression predicts the wages of those who move from self employment in year t − 1 to wage employment in year t and those who are in wage employment in both years shown in Figures A.7c and A.7d. The figures also present the rate at which workers move from self- to wage-employment at each point in the life cycle and the product of this rate and the wage-gap. The data are from the U.S. PSID (1975-1997 annually, 1999-2013 bi-annually) and the Mexican FLS (2002, 2005 and 2009).

66

Figure A.8: Selection into and out of Public Sector Employment – Panel Data Estimates

20 Potential Experience

30

5 -5 0 Percent -10 -15

-50 -75

Percent -25 0

-10

-5 0 Percent

10

-15

Percent -25 0 -50 -75

0

10

50 25

10

(b) Selection into the Public Sector – Mexico

5

25

50

(a) Selection into the Public Sector – U.S.

0

10

Wage Gap: Private Sector - Public Sector (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

20 Potential Experience

30

10 5 -5 0 Percent -10 -15

-50 -75

Percent -25 0

-10 -15

-5 0 Percent

5

25

50

(d) Selection out of the Public Sector – Mexico

10

50 25 Percent -25 0 -50 -75

10

30

Wage Gap: Private Sector - Public Sector (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

(c) Selection out of the Public Sector – U.S.

0

20 Potential Experience

10

Wage Gap: Private Sector - Public Sector (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

20 Potential Experience

30

Wage Gap: Private Sector - Public Sector (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

Notes: Figures A.8a and A.8b plot the estimated differences in wage residuals and their 95% confidence intervals for those who move from private-sector employment in year t − 1 to public-sector employment in year t and those who are in private-sector employment in both years. Wages are predicted from regressing log wagest−1 on dummy variables of five-year experience bins, the interactions of each experience-bin dummy variable with a dummy variable for whether the worker is in the public sectort , controlling for the years of educational attainment and year fixed effects. The reference group is the 0-4 years of experience bin for workers who remain in wage work. t refers to the year of the survey. We transform the coefficients from logs to levels so that the y-axis reflects the level of percentage increase relative to the reference group. The figures also present the rate at which workers move from private-sector to public-sector employment at each point on the life cycle and the product of this rate and the wage-gap. An analogous regression predicts the wages of those who move from public-sector employment in year t − 1 to private-sector employment in year t and those who are in private-sector employment in both years shown in Figures A.8d and A.8c. The figures also present the rate at which workers move from public-sector to private-sector employment at each point on the life cycle and the product of this rate and the wage-gap. The data are from the U.S. PSID (1975-1997 annually, 1999-2013 bi-annually) and the Mexican FLS (2002, 2005 and 2009).

67

Figure A.9: Selection into and out of Part-Time Employment – Panel Data Estimates

10

20 Potential Experience

30

5 -5 0 Percent -10 -15

-50 -75

Percent -25 0

-10 -15

-5 0 Percent

25

5

50

10

50 25 Percent -25 0 -50 -75

0

10

(b) Selection into Part-time Employment – Mexico

(a) Selection into Part-time Employment – U.S.

0

Wage Gap: Full-Time - Part-Time (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

10

50

30

-15

-10

-5 0 Percent

5

25 -50 -75

Percent -25 0

5 -10 -15

-5 0 Percent

20 Potential Experience

30

(d) Selection out of Part-time Employment – Mexico

10

50 25 Percent -25 0 -50 -75

10

20 Potential Experience

Wage Gap: Full-Time - Part-Time (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

(c) Selection out of Part-time Employment – U.S.

0

10

0

Wage Gap: Full-Time - Part-Time (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

10

20 Potential Experience

30

Wage Gap: Full-Time - Part-Time (Left Axis) Switch Rate (Right Axis) Wage Gap x Switch Rate (Right Axis)

Notes: Figures A.9b and A.9a plot the estimated differences in wage residuals and their 95% confidence intervals for those who move from full-time employment in year t − 1 to part-time employment in year t and those who are in full-time employment in both years. Wages are predicted from regressing log wagest−1 on dummy variables of five-year experience bins, the interactions of each experience-bin dummy variable with a dummy variable for whether the worker is part-timet , controlling for the years of educational attainment and year fixed effects. The reference group is the 0-4 years of experience bin for workers who remain in full-time employment. t refers to the year of the survey. We transform the coefficients from logs to levels so that the y-axis reflects the level of percentage increase relative to the reference group. The figures also present the rate at which workers move from full-time to part-time employment at each point on the life cycle and the product of this rate and the wage-gap. An analogous regression predicts the wages of those who move from part-time employment in year t − 1 to full-time employment in year t and those who are in full-time employment in both years shown in Figures A.9d and A.9c. The figures also present the rate at which workers move from part-time to full-time employment at each point on the life cycle and the product of this rate and the wage-gap. The data are from the U.S. PSID (1975-1997 annually, 1999-2013 bi-annually) and the Mexican FLS (2002, 2005 and 2009).

68

Figure A.10: Selection into and out of Self-Employment – Adding in individuals who exit the panel

Wage Gap, Percent -25 0 25

0

10

20 Potential Experience

-50

-50

Wage Gap, Percent -25 0 25

50

(b) Selection out of Self Emp./Attrition – Mexico

50

(a) Selection into Self Emp./Attrition – U.S.

30

0

10

Baseline (Balanced Panel) Exiters Remain in Wage Employment Exiters Move to Self-Employment

30

Baseline (Balanced Panel) Exiters Remain in Wage Employment Exiters Move to Self-Employment

Wage Gap, Percent -25 0 25

0

10

20 Potential Experience

-50

Wage Gap, Percent -25 0 25

50

(d) Selection out of Self Emp./Attrition – Mexico

50

(c) Selection into Self Emp./Attrition – U.S.

-50

20 Potential Experience

30

Baseline (Balanced Panel) Entrants Previously Wage Workers Entrants Previously Self-Employed

0

10

20 Potential Experience

30

Baseline (Balanced Panel) Entrants Previously Wage Workers Entrants Previously Self-Employed

Note: These figures are similar to those in Figure A.7a to A.7d of the paper. The solid lines plot the estimated wage gap between workers who remain in wage employment and those who move into/out of self-employment using the reported data. The thick dashed lines are the estimated wage gaps when individuals who exit the panel are added back as those who remain in wage employment. The thin dashed lines are the estimated wage gaps when individuals who exit the pannel are added back as those who switch into or out of self-employment.

69

150

Percent Wage Increase Relative to Age 21-25 50 100

150

Figure A.11: Cross-Sectional Age-Wage Profiles

100

United States

Germany Brazil

Canada

50

Chile United Kingdom

Mexico

0

0

Jamaica

20

30

40 Age

50

60

20

30

40 Age

50

60

Note: The age-wage profiles are for full-time males working in the private sector, and are calculated using all available years of data for each country. The wage is defined to be earnings divided by hours worked. For each country and year, we compute the ratio of average wages for workers in each 5-year age bin relative to the average wages of workers aged 21-25. The age-wage profiles in the figure are the unweighted average wage ratios by age across all years. The left-hand panel are the rich countries, defined as those with at least half the per-capita income level of the United States, and the right-hand panel are the poor countries, defined as those with less than half the per-capita income level of the United States.

70

0

Average Percentage Wage Increase 50 100

Figure A.12: Returns to Experience by Fine-Grained Schooling Category

0

5

10 Years of Education

High Income Countries

15

20

Low Income Countries

Note: Each observation in the figure represents a country, schooling-year pair (i.e. for each country there are multiple observations equalling the number of distinct schooling years in the data for that country). The x-axis measures the number of schooling years and the y-axis measures the average percentage wage increase (average profile height). Specifically, we estimate for each country and for each groups of full-time males with identical number of years of schooling an experience-wage profile. We then compute, for each group within each country, the average height of the experience-wage profile relative to the smallest experience bin and plot it in the figure as a function of the number of education years of the group. The size of the each dot is weighted by the share of individuals in each country that have that number of years of education. High income countries, defined as those with greater than twenty thousand dollars GDP per capita in 2011 at PPP, are colored in black. Low income countries, defined as those with less than twenty thousand dollars GDP per capita in 2011 at PPP, are colored in grey. The dotted lines are the fit lines of a regression of average percentage wage increase on year of education of the relevant group, computed separately for high and low income countries, and weighted by the share of individuals in each education groups within each country.

71

United States

Germany

Canada

0

10

20

30

40

United Kingdom

0

10

20

30

40

Chile

0

10

20

30

40

20

30

40

Mexico

0

50

100

0

10

20

30

40

Brazil

0

10

20

30

40

0

10

Jamaica Cognitive

50

100

0

Manual

0

Percentage of Workers in the Occupation

50

100

Figure A.13: Employment Shares in Cognitive and Manual Occupations

0

10

20

30

40

0

10

20

30

40

Note: Employment shares in cognitive and manual occupations are for full-time males working in the private sector. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. Cognitive occupations comprise categories 1-5, and 7 of the ISCO-08 (International Standard Classification of Occupations). They are: (1) Managers, (2) Professionals, (3) Technicians and associate professionals, (4) Clerical support workers, (5) Service and sales workers, and (7) Craft and related trades worker. Manual occupations comprise categories 6, 8 and 9 of the ISCO-08. They are: (6) Skilled agricultural, forestry and fishery workers; (8) Plant and machine operators, and assemblers, and (9) Elementary occupations. For each country and year, we compute the share of workers at each 5-year experience bin in either occupation group. The employment shares in the figure are the unweighted average employment shares by experience, for each occupation group, across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

72

200

Germany

Canada

0

100

United States

10

20

30

40

United Kingdom

0

10

20

30

40

Chile

0

10

20

30

40

20

30

40

Mexico

0

100

200

0

200

0

10

20

30

40

Brazil

0

10

20

30

40

Jamaica

0

10

Cognitive

100

Elementary

0

Percent Wage Increase Relative to Experience <5 Years

Figure A.14: Experience-Wage Profiles by Occupation Group for High-School Educated Workers

0

10

20

30

40

0

10

20

30

40

Note: These profiles are the same as Figure 9, except that they condition on high-school educated workers only.

73

Figure A.15: Experience-Wage Profiles of Workers with Short-Term Contracts (b) Mexico 150 Percentage Wage Increase 50 100 0

0

Percentage Wage Increase 50 100

150

(a) India

0

5

10

15 20 25 Potential Experience

Daily Workers

30

35

40

0

5

Long Term Workers

10

15 20 25 Potential Experience

Daily Workers

30

35

40

Long Term Workers

0

Percentage Wage Increase 50 100

150

(c) United States CPS

0

5

10

15 20 25 Potential Experience Marginally Attached

30

35

40

Baseline

Notes: These figures plot cross-sectional experience-wage profiles in India and Mexico separately for daily workers and those with longer-term work arrangements, and for the United States, workers that are marginally attached versus all workers.

74

150

Percent Wage Increase 50 100

150

Figure A.16: Cross-Sectional Experience-Wage Profiles, Expanded Sample with Part-Time Workers

100

Germany United States

United Kingdom S. Korea

Australia Uruguay

50

Canada

0

0

Chile

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

100

150

5

Percent Wage Increase 50 100 150

0

Indonesia

Brazil

India 50

Peru

Jamaica Guatemala

0

0

Mexico

0

5

10

15 20 25 30 Potential Experience

35

40

Bangladesh

0

5

10

Vietnam

15 20 25 30 Potential Experience

Note: These profiles are the same as Figure 2, except that they include part-time workers.

75

35

40

100 75

Percent Hours Increase 25 50 75

100

Figure A.17: Cross-Sectional Experience-Hours Profiles, Expanded Sample with Part-Time Workers

United States 50

Canada

Germany 25

Australia

Uruguay

United Kingdom

S. Korea Chile

0

0

France 10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

50

75

100

5

Percent Hours Increase 25 50 75 100

0

25

Vietnam

Brazil Peru

Jamaica

Guatemala

0

Mexico Indonesia

India

0

Bangladesh 0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Note: Experience-wage profiles are for males working in the private sector, including both part-time and full-time workers, and are calculated using all available years of data for each country. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. For each country and year, we compute the ratio of average hours worked for workers in each 5-year experience bin relative to the average hours worked of workers with less than five years of experience. The experience-hours profiles in the figure are the unweighted average hours ratios by experience across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

76

250

Germany United States

200

Percent Earnings Increase 50 100 150 200 250

Figure A.18: Cross-Sectional Experience-Earnings Profiles, Expanded Sample with Part-Time Workers

150

Canada Australia

S. Korea 100

United Kingdom Uruguay Chile

0

0

50

France

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

100

150

200

250

5

Percent Earnings Increase 50 100 150 200 250

0

Brazil

50

Indonesia Mexico

Vietnam

0

0

Peru

Guatemala Bangladesh Jamaica

India

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Note: Experience-earnings profiles are for males working in the private sector, including both part-time and full-time workers, and are calculated using all available years of data for each country. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. For each country and year, we compute the ratio of average earnings for workers in each 5-year experience bin relative to the average earnings of workers with less than five years of experience. The experience-earnings profiles in the figure are the unweighted average earnings ratios by experience across all years. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

77

1.25

Variance of Log Earnings .5 .75 1 1.25

Figure A.19: Earnings Variance Profiles, Expanded Sample with Part-Time Workers

1

United States

Uruguay

.75

Canada Australia

Chile

S. Korea

.5

Germany

United Kingdom .25

.25

France

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

40

1.25

5

1

Variance of Log Earnings .5 .75 1 1.25

0

Bangladesh Jamaica

Brazil

Peru

.75

Mexico

India

Guatemala

Indonesia

.25

.25

.5

Vietnam

0

5

10

15 20 25 30 Potential Experience

35

40

0

5

10

15 20 25 30 Potential Experience

35

40

Note: The earnings variances profiles are for males working in the private sector, including both part-time and full-time workers, and are calculated using all available years of data for each country. Potential experience is defined as the number of years elapsed since a worker finished schooling or turned 18, whichever is smaller. For each country, year and education group, we compute the variance of log earnings. The earnings-variance profiles in the figure are the weighted averages of the log earnings variances across years and education groups, weighted by the shares of workers in each education group. Countries are sorted in order of 2011 PPP GDP per capita from the top left to the bottom right panel.

78

79

Panel B: Chile 2009 in Percents Imputed Years Percentage Deviation from the Imputed Years of Schooling of Schooling 10th 25th 50th 75th 90th 0 0 0 0 0 0 3 -33 0 67 100 100 5 -40 0 40 60 60 8 0 0 25 50 50 12 -17 -8 0 0 0 13 0 0 8 15 23 14 0 0 7 14 14 16 0 6 6 6 13 20 -15 -15 -5 0 0

from our imputed years of schooling in percent. The sample is full-time males in the private sector.

Note: Panel A presents the distribution of reported years of schooling for each reported educational attainment level in 2009 in Chile, as well as our imputed number of schooling years for each educational attainment level. Panel B presents the differences

None Preparatory Primary School Secondary, Humanities/Special Secondary, Scientific/Commercial/Industrial Some Tertiary Education Two-year Tertiary Education Four-year College Education Post-graduate Studies

Highest Level of Education

None Preparatory Primary School Secondary, Humanities/Special Secondary, Scientific/Commercial/Industrial Some Tertiary Education Two-year Tertiary Education Four-year College Education Post-graduate Studies

Highest Level of Education

Panel A: Chile 2009 Imputed Years Directly Reported Schooling Years, Percentiles of Distribution of Schooling 10th 25th 50th 75th 90th 0 0 0 0 0 0 3 2 3 5 6 6 5 3 5 7 8 8 8 8 8 10 12 12 12 10 11 12 12 12 13 13 13 14 15 16 14 14 14 15 16 16 16 16 17 17 17 18 20 17 17 19 20 20

Table A.1: Distribution of the Years of Education by Level of Highest Attainment in Chile

Table A.2: Robustness of Cross-Sectional Estimates

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

Height at 20-24 Years Experience Panel A: Cross-Section, Core Countries Rich Baseline 89.3 Experience at 16 91.5 Constructed experience 112.4 Measurement error: age 110.6 Measurement error: education 76.7 Measurement error: age and education 93.5 Include Self-Employed 90.6 Include Public-Sector Employees 89 Include women 78.2 Constructed experience, men and women 102.9 Include Part-Time (20+ hours) 91.9 Include Part-Time (> 0 hours) 89.3 Constructed experience, incl. Part-Time 105.2

Poor 47.6 47.6 48.2 47.6 47.6 47.6 47.5 51.9 37.8 41.6 47.7 44.5 45.7

Rich - Poor 41.7** 43.9** 64.2** 63** 29.1** 45.9** 43.1** 37.1** 40.4** 61.3** 44.5** 44.8** 59.5**

(14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26)

Panel B: Cross-Section, All Countries Rich Baseline 83.5 Experience at 16 85.4 Constructed experience 104.9 Measurement error: age 95.5 Measurement error: education 71.4 Measurement error: age and education 80.2 Include Self-Employed 84.2 Include Public-Sector Employees 83.3 Include women 69.3 Constructed experience, men and women 87 Include Part-Time (20+ hours) 82 Include Part-Time (> 0 hours) 80.4 Constructed experience, incl. Part-Time 98

Poor 43.7 46.3 49.9 43.7 43.7 43.7 41.4 52.4 37.9 47.2 45.4 37.9 53.3

Rich - Poor 39.8*** 39.1*** 55.0*** 51.8*** 27.7*** 36.5*** 42.8*** 30.9*** 31.4*** 39.8** 36.6*** 42.5*** 44.7***

Note: Row (1) uses the baseline sample and measures. Row (2) expands the sample to include individuals who are age 16 and 17. Row (3) uses constructed experience instead of potential experience (see Section 5.1.1). Row (4) adds noise to the age variable in order to match the same amount of age heaping observed in Chilean data (see Section 5.1.2). Row (5) adds noise to the years of education variable in rich countries by assuming that the distribution in the years of education for a given level of educational attainment is the same as in Chile (see Section 5.1.2). Row (6) adds noise to both the age and years of education variables in rich countries. Row (7) includes self-employed workers. Row (8) includes public-sector workers. Row (9) includes female workers. Row (10) includes female workers and use measures of constructed experience, where it is constructed separately for male and female workers. Row (11) includes part-time workers who work at least 20 hours per week. Row (12) includes all part-time workers. Row (13) includes all part-time workers and use constructed experience, where it is constructed separately for full-time and part-time workers. *** denotes p-value less than 0.01; ** denotes p-value less than 0.05; * denotes p-value less than 0.10.

80

Table A.3: Age-Wage Profiles

United States Germany Canada United Kingdom Chile Brazil Mexico Jamaica

Panel A: Summary Statistics by Country (1) (2) (3) Height at Height at Average Age 41-45 Age 56-60 Height 92.6 98.7 71.6 64.6 69.4 50.6 62.7 53.1 46.7 67.6 42.8 47.6 51.3 40.7 39.4 63.1 47.7 45.1 39.5 17.1 26.7 28.4 15.7 19.1

(4) Discounted Avg Height 31.8 22.7 21.4 22.9 18.6 20.8 13.1 9.3

Panel B: Test of Differences in Means, Rich and Poor Groups (1) (2) (3) (4) Height at Height at Average Discounted Avg Age 41-45 Age 56-60 Height Height Rich Mean 71.9 66.0 54.2 24.7 Poor Mean 45.6 30.3 32.6 15.4 Rich - Poor 26.3∗∗ 35.7∗∗ 21.6∗∗ 9.3∗∗ (.027) (.027) (.016) (.015) Note: The first column of Panel A is the average height of the age-wage profile at ages 41-45, defined as the ratio of average wages for workers aged 41-45 to average wages for workers aged 21-25. The second column is the average height of the age-wage profile at ages 56-60, defined as the ratio of average wages for workers aged 56-60 to average wages for workers aged 21-25. The third column is the average height of the profile relative to age 21-25. The fourth column is the discounted average height of the profile relative to age 21-25, where wages are discounted at a rate of four percent per year. The sample is restricted to full-time males in the private sector. Panel B shows the results of permutation tests of the null hypothesis that the age-wage profiles are the same in rich and poor countries. *** denotes p-value less than 0.01; ** denotes p-value less than 0.05; * denotes p-value less than 0.10.

81

Table A.4: Returns to Experience for Different Experience Levels

(1) (2) Cross-section Heckman-Lochner-Taber A. Height of experience-wage profiles at 5-9 relative to 0-4 Rich mean 43.4 40.3 Poor mean 23.9 17.2 Rich-poor 19.5** 23.1** B. Height of experience-wage profiles at 10-14 relative to 0-4 Rich mean 69.4 61.5 Poor mean 37.5 30.0 Rich-poor 31.9** 31.5** C. Height of experience-wage profiles at 20-24 relative to 0-4 Rich mean 89.3 79.3 Poor mean 47.6 39.2 Rich-poor 41.7** 40.1** D. Height of experience-wage profiles at 30-34 relative to 0-4 Rich mean 88.7 83.4 Poor mean 44.7 41.6 Rich-poor 44.0** 41.8**

82

Life-Cycle Wage Growth Across Countries - Princeton University

May 22, 2016. Abstract ..... different time trends. Figure 1 plots experience-wage profiles ..... business-cycle effects that average to zero over the long run” (p.126).

1MB Sizes 6 Downloads 276 Views

Recommend Documents

Life-Cycle Wage Growth Across Countries - Princeton University
May 22, 2016 - business-cycle effects that average to zero over the long run” (p.126). ..... taking their reported income to be their wage and salary income. ..... that the output of a firm is Leontief in the firm's technology and the human capital

Sources of Wage Inequality - Princeton University
Jan 14, 2013 - strong empirical support. Helpman et al. ... facts that support the mechanism of firm$ ..... An International Comparison, Chicago: University of ...

The Lifecycle Wage Growth of Men and Women
register, which covers the entire population of Sweden aged 16-75. .... For this individual – a male business major with somewhat above average income .... gender wage gap at age 25 plus Ga, the difference in men's and women's total wage.

New Immigrants.qxd - Princeton University
Westminster Register Office. Standing beside a framed photo ..... satellite television, cheap phone calls and the internet, people in developing countries are more ...

Painting with Triangles - Princeton Graphics - Princeton University
By con- trast, programs like Adobe Illustrator and Inkscape let a user paint ... effect. These arbitrary polygons are costly to render however, and. “smooth” effects are only created via many .... Next the outer stroke polygons are rendered in 50

Attenuation of Adaptation - Princeton University
strategy, it cannot capture the rapid initial reduction in error or the overcompensatory drift. Therefore, we modeled the strategy as a setpoint/reference signal that ...

New Immigrants.qxd - Princeton University Press
2. Immigrants: Your Country Needs Them grey-haired man in a bright red, fur-trimmed robe ..... ing for a few years in the Valley and set up companies that trade.

net neutrality - cs.Princeton - Princeton University
Jul 6, 2006 - of traffic when your browser needs to fetch a new page from a server. If a network provider is using ... hand, applications like online gaming or Internet telephony (VoIP), which rely on steady streaming of interactive .... The VPN user

New Immigrants.qxd - Princeton University Press
months while learning English; a forty-six-year-old Romanian dental technician who described ..... remote locations to complete their degree courses online. And.

Agricultural Productivity Differences Across Countries
Jan 14, 2014 - Host of measurement concerns. • Exclude home production, which is very common in poor countries? • Agricultural output data otherwise badly ...

Public-private wage differentials in euro area countries ...
Feb 14, 2014 - the years 2004-2007 from the European Union Statistics on Income .... denote the group (s = {0, 1}) and ys the outcome of interest in group s. .... private schools). ..... against workers in agriculture, construction and retail trade.

Wage Inequality and Firm Growth
West Fourth Street, New York, NY 10012, NBER, CEPR, and ECGI. (e-mail: [email protected]); Ouimet: University of North. Carolina at Chapel Hill, Kenan-Flagler Business School, Campus Box .... provided by Income Data Services (IDS), an independen

Chapter 2 [PDF] - Princeton University Press
enables us to apply the tools used in the analysis of stationary models to study economies with sustained ...... of computer hardware and software. Thus we may ...

Career Choice and Wage Growth
an important extension because both in the data and the model, wage growth .... my baseline definition is the best choice for the empirical analysis of this paper.

1 Inter-industry wage differentials in EU countries
Oct 18, 2009 - differentials across a large number of industries for 8 EU countries at two ... The analysis uses the European Structure of Earnings Survey, an ... The data used are drawn from the first two waves (1995 and 2002) of the ..... “Compar

Prior Expectations Bias Sensory ... - Princeton University
In a separate analysis, we estimated BOLD amplitudes for each single trial, using the .... weights, we take our training data Bloc and regress those onto our hypo-.

Vision Based Self-driving Car - Princeton University
The world is very complicated. • We don't know the exact model/mechanism between input and output. • Find an approximate (usually simplified) model between input and output through learning. • Principles of learning are “universal”. – soc

Chapter 2 [PDF] - Princeton University Press
For more information send email to: ... These economists published two pathbreaking articles in the same year, 1956 (Solow, 1956;. Swan .... will study in Chapter 8) is that technology is free: it is publicly available as a nonexcludable, ... Definit

Prior Expectations Bias Sensory ... - Princeton University
segment in a 360° circle (Fig. 1A) using two buttons of an MR-compatible button box to rotate the line clockwise or anticlockwise. The initial di- rection of the line ...

Exporting and Organizational Change - Princeton University
Jul 18, 2017 - The computations in this paper were done at a secure data center .... of management (or L + 1 layers of employees, given that we call the ... of length z costs ¯wcz (c teachers per unit of knowledge at cost ¯w per teacher).

Exporting and Organizational Change - Princeton University
Jul 18, 2017 - We study the effect of exporting on the organization of production within firms. .... their technology (and so the marginal product of labor is higher) or .... Learning how to solve problems in an interval of knowledge .... We use conf

When Human Rights Pressure is ... - Princeton University
Sep 9, 2016 - have access to a good VPN to access foreign websites (Time 2013). On the other side, Chinese ... Times 2015), which was debated on social media. .... comparison to AU criticism 19, while AU criticism itself has no effect20.

How Do Laffer Curves Differ Across Countries?
do so, we calculate the implied peak of the Laffer curve and compute the maximum interest rate on outstanding ... has important effects for the labor income tax Laffer curve. Several countries are ...... Calculated as the residual of compensa-.