Long-Run Changes and Short-Term Fluctuations in the US Wage Structure∗ Gonzalo Castex and Evgenia Dechter† August 2016

Abstract We study how technological change and business cycle shape the wage structure. Controlling for institutional and demographic changes, we examine patterns of skill prices and real wages between 1961-2012. Return to schooling is weakly countercyclical and positively correlated with technological change. Return to experience is countercyclical and negatively correlated with technology. Wages of unskilled inexperienced workers are positively correlated with technological change and business cycle. To explain these findings, we extend the capital-skill complementarity framework to incorporate schooling, experience, vintage capital equipment and new capital equipment. Our model generates the observed profiles of skill prices if vintage-capital-skill complementarity assumptions hold.

JEL Classification: E24, E32, J24, J31 Keywords: returns to education; returns to experience; business cycle; capital-skill complementarity; technological change.

∗ We would like to thank Mark Bils, Yongsung Chang, Michael Keane, Per Krusell, James Morley, Valentyn Pancheko, Victor Rios-Rull, as well as seminar participants at Atlanta Fed, Philadelphia Fed, UNSW, University of Wollongong, University of Santiago, University of Chile, Central Bank of Chile, Ghent University, KU Leuven, Catholic University of Chile, University of Hong Kong for comments and suggestions. † Gonzalo Castex, Central Bank of Chile. Email: [email protected]. Evgenia Dechter, School of Economics, University of New South Wales. Email: [email protected].

1

1

Introduction

The US economy have experienced significant changes in the distribution of earnings throughout the twentieth century. Developments in labor market returns to skills have played an important role in shaping the earnings distribution. Schooling premium in the US (and other developed countries) has been steadily increasing starting the early 1980s. This increase coincides with a substantial rise in the average education level of the workforce. A large number of studies investigate the role of skill-biased technological change in explaining the increasing demand for college graduates and the increasing schooling premium. See for example Bound and Johnson (1992), Katz and Murphy (1992), Levy and Murnane (1992), Juhn, Murphy, and Pierce (1993), Autor, Katz, and Krueger (1998), Acemoglu (2002), Autor, Katz, and Kearney (2005) and Autor, Katz, and Kearney (2008). Experience premium had also been increasing between the early 1970s and late 1980s, but starting the 1990s it shows a continuous decline till around 2008. Earlier studies explain the developments in schooling and experience premia using one mechanism. For example, Katz and Revenga (1989), Katz and Murphy (1992), Murphy and Welch (1992), Bound and Johnson (1992) and Levy and Murnane (1992) argue that the increasing returns to schooling and experience between the 1970s and early 1990s is a result of the skillbiased technological change and the associated shifts in the demand for skilled labor. Other explanations include globalization pressures that led to shrinking relative demand for less skilled and to the increase in skill premia, declining unionization and real minimum wage. Skill-biased technological change, globalization pressures and declining unionization cannot explain both the increase in experience premium in the 1970s - 1980s and its subsequent decline starting the late 1980s. Jeong, Kim, and Manovskii (2015) document a similar trend in returns to experience, and suggest that relative supply effects driven by the progression of the baby boom cohorts through the labor market and by the increase in female labor force participation, have contributed to the decline in experience premium starting the 1990s. Controlling for institutional and demographic changes, we examine how technological change and business cycle shape the developments in skill premia and real wages in the 1961-2012 period. We show that different types of labor respond differently to business cycles and technological change. Younger educated workers gain the most from improvements in technology, followed by unskilled workers; whereas experienced (or older workers) find themselves in a disadvantage as a result of technological progress. Unskilled workers gain relatively more from economic expansions and lose more in recessions, compared to high-educated workers (the relative disadvantage (advantage) at times of economic expansions (contractions) is more pronounced for high-educated workers with high experience). Real wage adjustments to aggregate fluctuations is a central piece of macroeconomic models. More recent studies, starting with Stockman (1983), Bils (1985), Keane, Moffitt, and Runkle (1988), Blank (1990) and Solon, Barsky, and Parker (1994), Carneiro, Guimar˜aes, and Portugal (2012) show that, controlling for the changing composition of the workforce over the business cycle, real wages are (mildly) procyclical. Failing to control for composition effects may generate a countercyclical bias in wages since the distribution of

2

unobservable workers’ characteristics is changing over the business cycle.1 Most previous studies find little heterogeneity in the cyclicality of real wages by education level, see for example Bils (1985), Keane and Prasad (1993), Solon, Barsky, and Parker (1994), Young (2003), Lindquist (2004), Castro and Coen-Pirani (2008). Ziliak, Wilson, and Stone (1999) find that return to education is procyclical especially for workers in white-collar occupations. Very few studies examine the cyclicality of returns to experience. Reder (1955) finds the experience premium to be countercyclical in the 1930s and 1940s; Raisian (1983) and Keane and Prasad (1993), document a procyclical or weakly procyclical return to experience. Our empirical analysis utilizes 1962 - 2013 March Current Population Surveys (CPS). We estimate the marginal return to experience, marginal return to schooling and common wage component for each year in the sample, controlling for individual and geographic characteristics. We examine how the estimated returns to skills vary with the technological change, institutional changes, demographic changes and the business cycle. To measure the business cycle we use HP-filtered real output per capita and unemployment rate. Technological change is measured by a relative price index of capital equipment and software, constructed by dividing the price index of nondurable consumption goods and nonhousing services by the price index of capital equipment.2 To account for compositional changes of the workforce over the business cycle we use three approaches. To assess the magnitude of compositional effects we compare returns to skills estimated at nearby quantiles (45th, 50th and 55th), assuming that if there was a shift in the unobserved skills distribution due to the changing economy conditions, an individual would not move too far from the initial location. We do not find significant differences in the estimates of nearby quantiles but take further precautions in the empirical analysis. First, we employ the predictions of positive assortative matching theory (see for example Lam 1988) and estimate the determinants of returns to skills controlling for spousal schooling and spousal ranking in the schooling distribution to proxy for unobserved skills. Second, using quantile regression, we estimate skill prices at the 50th and 75th percentiles of the wage distribution, assuming that employment is less sensitive to business cycle fluctuations for workers at the high end of wage distribution. Additionally to controlling for compositional changes associated with business cycles, we also take into account the changing distributions of demographic characteristics over time. Changing demographics have a direct effect on the estimated marginal returns to skills. To control for the changing characteristics we reweight each wave of the CPS to match age and schooling distributions of the 2000 CPS. We find that technological change is positively correlated with the return to schooling and negatively with the return to experience. Return to experience is strongly countercyclical in all specifications. Return to schooling is mildly countercyclical when controlling for composition effects and technological change. The common across skill types wage component, measured by the intercept in wage function, is procyclical in 1 Some of these studies also note that the countercyclical bias in wage pattern due to compositional changes is not sufficiently strong to produce a countercyclcal pattern in real wages, see for example Bils (1985). Bils (1985) also argues that countercyclical bias might result from failing to account for fluctuations in overtime earnings over the business cycle and also suggests that different time periods may have different patterns of real wage cyclicality. 2 Declining relative price index implies technological innovation. We follow Greenwood, Hercowitz, and Krusell (1997), Acemoglu (2002), Cummins and Violante (2002), and Fisher (2006) among others and interpret the relative price index of non-durable consumption goods divided by the price index of capital equipment as a measure of technological change. To construct the relative price index we follow Cummins and Violante (2002). See Data Section for details on construction of the index.

3

specifications that control for the changing workforce composition over the business cycle and it is positively correlated with the technological change. Thus, the overall procyclicality of real wages (reported in a number of existing studies) is driven by the common component. Most of the literature on wage structure study the earlier 1970s-1990s period and reach different conclusions. More recent studies focus on the increasing return to schooling and do not examine changes in return to experience. Jeong et al. (2015), who analyze the shift in experience premium over a similar time period as in our study, focus on demand and supply effects but do not examine the effects of business cycle fluctuations or technological change. Our findings suggest that to understand the wage dynamics one should consider heterogeneity in both schooling and experience levels and take into account a range of changing macroeconomic conditions. There is a large variety of theoretical models that aim to explain adjustments and developments in real wages and wage structure. For example, frameworks that incorporate implicit contracts, job match quality and/or on-the-job human capital accumulation can generate countercyclical returns to skills. However, these frameworks do not incorporate technological change. The link between technology on the demand skilled (educated) workers has been extensively studied in a range of frameworks. Griliches (1969), Tinbergen (1975), and Berman, Bound, and Griliches (1994) employ a model with technology-skill complementarity and show that the demand for educated workers rises with technology. Nelson and Phelps (1966), Welch (1970), Schultz (1975), Bartel and Lichtenberg (1987), Greenwood and Yorukoglu (1997) and Chun (2003) argue that the process of adopting new technologies raises the demand for educated workers. Krusell, Ohanian, R´ıos-Rull, and Violante (2000) and Lindquist (2004) incorporate technological change and neutral shocks into capital-skill complementarity in production framework. Lindquist (2004) in a business cycle model with neutral and investment-specific technology shocks and shows that schooling premium is increasing with technology but negatively correlated with neutral productivity shocks. However, these existing frameworks do not incorporate labor experience and therefore cannot explain the developments in return to experience. To help interpret our findings we outline a model that extends the standard capital-skill complementarity framework by incorporating vintage human capital and vintage physical capital. We distinguish between two types of skills, education and experience, and two types of capital equipment, new investment and vintage. Correlations between skill prices, technological change and business cycle are driven by responses of capital-skill ratios and labor supply effects. We show empirically that in times of economic expansions or high technological progress the ratio of new capital equipment and young high-educated labor is increasing. We argue that that vintage capital equipment does not respond to investment or productivity shocks, and therefore the ratio of vintage capital and more-experienced high-educated labor is declining with any type of shock. Our model generates relationships between skill prices, neutral and investment-specific technology shocks we observe in the data if vintage-capital-skill complementarity assumptions hold. The model generates countercyclical experience premium negatively correlated with technological change and countercyclical education premium positively correlated with technological change if five assumptions hold. (i) More-experienced high-educated labor is more complementary with vintage capital equipment than with new capital equipment. (ii) Less-experienced high-educated labor is more complementary with new capital equipment than with vintage capital. (iii) The complementarity between high-educated labor (at

4

any experience level) and capital equipment (of any vintage) is higher than it is between low-educated labor and high-educated labor or capital equipment. (iv) The magnitude of response of the ratio between new capital equipment and less-experienced high-educated labor to an investment-specific shock is larger than the magnitude of response of the ratio of vintage capital and more-experienced high-educated labor. (v) The magnitude of response of the ratio between new capital equipment and less-experienced high-educated labor to a neutral shock is smaller than the magnitude of response of the ratio between vintage capital and more-experienced high-educated labor. Assumptions (i) and (ii) require higher complementarity within vintages (of human capital and physical capital) than between vintages. If these two assumptions hold, the model produces a countercyclical experience premium negatively correlated with technological change.3 Assumptions (ii)-(v) lead to countercyclical education premium positively correlated with technological change.4 Assumption (iii) is a benchmark outcome in the capital-skill complementarity framework (see Krusell et al. 2000). The remainder of the paper is organized as follows. Section 2 describes and summarizes the data. Section 3 presents the empirical methods, reports results and discusses the practical relevance of our findings. Section 4 outlines the theoretical framework to explain the variation in skill prices. Section 5 concludes the paper.

2

Data

To examine the developments in wage structure we use annual data from the 1962-2013 March Current Population Surveys (CPS). The raw CPS sample contains approximately 5.5 million observations, 50,000 to 160,000 observations per year. We use a subsample of 18 to 61 years old, who work 9 hours or more per week, not self-employed, not in the armed forces and not enrolled in school. Our sample contains 962,653 observations, 7,700 to 27,000 observations per year, detailed information on sample size for each year is reported in column (2) of Table 1. The variables of interest are real hourly wage, education and work experience. Some questions in the March CPS refer to usual last year activity and some to labor market activity during the previous week. Weeks worked and income from wages and salaries refer to last year. Before 1976 weeks worked last year are reported in 6 categories (the intervals are 1-13, 14-26, 27-39, 40-47, 48-49 and 50+). Information on usual hours worked per week last year is available only for 1976-2013; hours worked last week is available for the entire period. For the 1962 - 1975 period, we follow Krusell, Ohanian, R´ıos-Rull, and Violante (2000) methodology to obtain measures of weeks and hours worked last year.5 We construct a measure of hours worked per week last year for the 1962 - 1975 period based on how they project on hours worked last week 3 Satisfying these two assumptions is a sufficient condition. In practice, even if only one of these two assumptions holds, the model can generate patterns in experience premium observed in the data. 4 These four assumptions are sufficient conditions to generate patterns in education premium observed in the data. For a given parameters set, assumption (ii) could be reduced if assumptions (iii)-(v) are satisfied. 5 We divide the raw March CPS data into 264 groups, consisting of 11 age groups (5 years/group), 3 race groups (white, black and other), 2 gender groups and 4 education groups (less than high school, high school, some college and college or more). For each group we calculate the weighted average of weeks worked using data after 1976, ignoring individuals with missing or zero weeks worked. Some individuals report working zero hours last week and strictly positive weeks las year. For these individuals we calculate weighted average of weeks worked last week within each age-race-gender-education group.

5

for 1976 - 2013. Hourly wage rates are calculated using last year annual income, divided by weeks worked and weekly hours worked last year. Income, weeks and hours data refer to last year therefore the actual sample period is 1961 - 2012. Wage rates are in 1999 dollars (deflated using the CPI). To construct the schooling variable, we convert information on the highest grade completed into years of schooling. Schooling information is not available in 1962 and we exclude this year from the analysis. We use potential experience (calculated as age minus years of education minus 6) and age to proxy for the actual work experience. To measure demographic transitions, we calculate the mean and standard deviation of experience and education using the entire sample of 18 to 61 years old in the labor force. Other aggregate demographic characteristics include unionization rate and female labor force participation. Union membership data for 1960-1983 is from Mayer (2004) and from the CPS for the remaining years. Female labor force participation is constructed using the CPS. To measure technological change we use a relative price index. We construct the relative price index using nondurable consumption goods and nonhousing services price index divided by a quality-adjusted price of equipment and software. We follow Cummins and Violante (2002) methodology to extend Gordon (2007) series for 24 categories of price indexes and aggregate them using the Tornqvist procedure. Price indexes are from the National Income and Product Accounts (NIPA).6 We follow Greenwood et al. (1997), Acemoglu (2002), Cummins and Violante (2002), and Fisher (2006) among others, to interpret the relative price index of capital equipment as an inverse measure of technological change. A declining index indicates an increasing state of technology. We use two measures of business cycle, HP-filtered real GDP per capita and unemployment rate. Real GDP per capita is in 2011 dollars. Unemployment rate is for 25 to 54 years old individuals, obtained from the U.S. Bureau of Labor Statistics. In estimations we use lagged business cycle measures and the relative price index of capital equipment. Figure 1 depicts the developments in HP-filtered real GDP per capita and unemployment rate, relative price of capital equipment, mean age and potential experience, average years of education, proportion of males in the labor force and unionization rate. Significant drop in the relative price of equipment and software indicates an increase in the state of technology over the last 50 years. The trend in mean work experience and age demonstrates the importance of the baby boom cohorts’ entry into the labor market. The average level of schooling is increasing throughout the 1961-2012 period, female labor participation is increasing and unionization rate is falling. 6 We retrieve data from Tables 2.3.4 and 2.3.5 of the NIPA series to obtain personal consumption price indexes and expenditures, respectively. We retrieve data from Table 7.8 of the NIPA series to obtain price indexes for private fixed investment in equipment and software (Table 7.8 contains information up to 2001. After 2001 we retrieve this information from Table 5.5.4 ).

6

21 19

.6

38

.8

20

40

Age and experience

39

Relative price of equipment

18 17

37 36

16

35 .25

LF male ratio

1960 1970 1980 1990 2000 2010

Union rate

.2

.64

1960 1970 1980 1990 2000 2010

.1

.58

11

.6

12

.15

.62

13

.66

Education

1960 1970 1980 1990 2000 2010

.68

1960 1970 1980 1990 2000 2010

age exp

0

U

14

.4

−1

−.04

GDP

.2

−.5

−.02

0

0

.5

1

.02

1.5

.04

Business cycle

1

Figure 1: Aggregate trends

1960 1970 1980 1990 2000 2010

1960 1970 1980 1990 2000 2010

Note: HP-filtered real GDP per capita and unemployment rate series are from the BLS. Relative price of capital equipment is quality adjusted price of capital equipment relative to the price of nondurable consumption goods. Average experience, age, female labor participation, and years of education are constructed for 18-61 years old in the labor force, using the CPS. Union membership data for 1960-1983 is from Mayer (2004) and from the CPS for 1984-2012.

3

Empirical Analysis

To examine how labor market rewards productive attributes we use the traditional Mincer (1974) earning function. The terms returns to schooling and returns to experience refer to the effects of education and experience on log wage rate. We perform the analysis in two steps. First, we estimate the average marginal returns to schooling and experience for each year in the sample. Second, we evaluate what drives the changes in wage structure.

3.1

Estimating returns to skills

For the first stage estimations, consider a year t cross-sectional regression of log real wage rates of individual i on labor market experience, education and other personal characteristics, 7

log wit = β0t + β1t Experienceit + β2t Experience2it + β3t Schoolingit + Xit γt + εt ,

(1)

where Xit is a set of variable that includes marital status, metropolitan status and state fixed effects; εt summarizes the measurement error in the data. Some specifications use a subsample of married men and include spousal schooling level and spousal ranking in the schooling distribution to proxy for unobserved skills. We estimate equation 1 for each year in the 1961-2012 period. We define returns to education and experience as PSCHOOLt = βb3t and PEXPt = βb1t + 2βb2t Experiencet , where Experiencet is the mean experience level of the workers in the sample in year t.7 We assume that the estimate of β0t , the common component, also measures variations in wage for individuals with low education and no work experience; thus we denote PU = βb0t . t

We estimate equation 1 using several approaches. A simple OLS specification using weights provided by the CPS delivers the mean marginal returns to skills. The interpretation of these results should be cautious since the average individual is changing over time. In particular, there are substantial changes in the distributions of education and experience. To keep these distributions constant over time, we reweight each year of the CPS survey to match distributions of experience and education as they are in CPS 2000. To construct the weights, we follow the methodology developed in DiNardo, Fortin, and Lemieux (1996). For each wave t of CPS, we pool data from wave t and from CPS 2000 and use a Probit model to estimate the probability that an observation is in the CPS 2000, conditional on S=(age, age2 , age3 , schooling, schooling 2 ).8 The estimated probabilities are used to construct the weights: ψ(S) =

P (d2000|S) 1−P (2000|S) ,

where

d2000 ∈ {0, 1} equals 1 when an observation is taken from the CPS 2000 and 0 otherwise, and P (d2000|S) is the conditional probability of appearing in CPS 2000. Obtaining ψ(S) we construct the product of weights ψ0 ψ(S), where ψ0 are the original weights provided by the BLS. In estimations we use the product of weights function, ψ0 ψ(S), to reweight observations in each wave of the CPS; this reweighting produces nearly equal age and schooling distributions across time. Using these adjusted weights the returns to schooling and experience are estimated at the same point of the age and education distribution (which also locks mean education and experience levels at the year 2000 level). The estimated returns to skills (from OLS, using the BLS weights and constructed weights), PEXPt and PSCHOOLt , are depicted in Figure 2. Trends in return to education are widely documented. It increases rapidly between the 1980s and 1990s, and more moderately in the later periods. The less documented trend is in return to experience. Return to experience is increasing substantially between the 1970s and 1990s. It declines starting the 1990s till around 2008. Using constructed weights mitigates the changes in return to experience, which emphasizes the importance of controlling for the changing distributions of age and education. Figure 2 also includes outcomes of equation (1) that uses age instead of experience. Patterns of skill prices are very similar. 7 To examine the robustness of our results we also use mean experience that does not vary over time. See next paragraph for details. 8 These probability estimations use sampling weights provided by the BLS to achieve population representative samples.

8

Figure 2: Estimated skill prices

Return to schooling

.04

.005

.06

.01

.08

.015

.1

.02

.12

Return to experience

1960

3.2

1970

1980

1990

2000

2010

1960

1970

1980

1990

2000

BLS weights, using potential exp

constructed weights, using potential exp

BLS weights, using age

constructed weights, using age

2010

Returns to skills

In this section we examine the determinants of returns to skills. The 1961-2012 is a period of important demographic and economic changes, some of which are documented in Figure 1. Autor et al. (1998), Bound and Johnson (1992), Mincer (1991) and others argue that these developments have played an important role in shaping the wage structure. Previous studies show that different skill groups are imperfect substitutes in production; therefore, shifts in the supply of and demand for labor skills can affect returns to these skills. Relative supply of skills could be affected by changing cohort size, female labor force participation, changes in access and cost of education and immigration. Skill-biased technological change and institutional changes could shift the demand for skills. We control for these factors when estimating the effects of business cycle and technological change on returns to skills. We include the mean and standard deviation of experience (or age) and education and the proportion of women in labor force to control for relative supply and demand shifts. We include the unionization rate to control for institutional changes. We examine the formation of skill premia using the estimated PEXPt and PSCHOOLt . We also estimate a similar specification for the estimate of β0t in equation (1), assuming that this coefficient measures variations

9

in wage for unskilled individuals, PUt . We use the following specifications.

PEXPt = η1 Yt + η2 pt + η3 Zt + η4 t + υ1t ,

(2)

PSCHOOLt = µ1 Yt + µ2 pt + µ3 Zt + µ4 t + υ2t ,

(3)

PUt = κ1 Yt + κ2 pt + κ3 Zt + κ4 t + υ3t ,

(4)

where Yt is a business cycle measure in year t, pt is the relative price index of capital equipment to measure (inverse of) technological change, Zt is a set of aggregate control variables and υ1t , υ2t and υ3t are uncorrelated measurement errors. Aggregate variables included in Zt are proportion of women in the labor force, mean and standard deviation of education and experience, and unionization rate. We report Newey-West robust standard errors with two lags to adjust for serial correlation in residuals. Our specifications of skill prices, PSCHOOLt and PEXPt , measure percentage return to an additional year of schooling or an additional year of experience. In equations (2)-(4) we estimate how wage structure responds to business cycle fluctuations, technological change, supply and demand shifts, demographical changes and institutional changes. The estimations include HP-filtered real output per capita (or unemployment rate) to measure the business cycle and levels of relative price of capital equipment to measure the technological change. In our specification, the level of return to schooling or experience are on the left hand side of the equation; explanatory variables of interest are the measure of business cycle and level of technological change, controlling for time trend, demographic change and institutional change. Controlling for time trend, changing demographics and institutional structure works to eliminate the effects of important trends not driven by business cycles or technology innovations. The choice between using levels of returns to skills and relative prices rather than their fluctuations around a given trend, links to an ongoing debate. Krusell et al. (2000) reports a positive correlation between the relative price index of capital equipment and schooling premium, which they interpret as an evidence of complementarily between capital equipment and skill. Balleer and Van Rens (2013) derive short-term fluctuations in schooling premium and relative price of capital and find a negative correlation between the two series. They argue that the production function exhibits substitutability between capital equipment and skill. According to Balleer and Van Rens (2013), what explains the difference between the two studies is that Krusell et al. (2000) base their argument on a correlation in the long run trends in the schooling premium and the relative price of investment goods. Our empirical approach takes into account considerations raised in Balleer and Van Rens (2013) by controlling for a range of changing conditions without imposing a strict structure on fluctuations frequency. Relationships between the residual skill prices, GDP and relative price index are depicted in Figure 3. Using the Augmented Dickey-Fuller test we reject the unit-root null hypothesis for all series in Figure 3.

10

Figure 3: Returns to skills, GDP and relative price index, residuals

Panel A: Skill prices and GDP

1970

1980

1990

2000

2010

.04 .02 0 −.04 −.02

0 −.01 −.005

.005

0 −.04 −.02

.02

.001 .002 0 −.002 −.001

1960

.01

return to education

.04

return to experience

1960

1970

1980

1990

2000

2010

2000

2010

Panel B: Skill prices and relative price index

.1 .05

.005

1970

1980

1990

2000

2010

−.05

−.01 −.005

0 −.05 1960

0

0

.05

.001 .002 0 −.002 −.001

.01

return to education

.1

return to experience

1960

return to skill

1970

1980

1990

GDP or relative price index

Note: Time series in this figure are obtained by computing residuals of projections of skill prices, real HPfiltered GDP per capita and relative price index of capital equipment on proportion of women in the labor force, mean and standard deviation of education and experience, unionization rate and time trend.

3.3

Composition effects

Stockman (1983), Bils (1985) and others show that cyclical changes in the work force composition may induce a countercyclical bias in the aggregate wage. Aggregate measures of real wages tend to give more weight to low-skill workers during expansions than during recessions because these workers are more vulnerable to layoffs. Less is known about how returns to skills respond to the changing composition of the work force. We use three approaches to examine the effects of changing unobserved characteristics. First, we compare estimation outcomes obtained for nearby quantiles (45th, 50th and 55th), assuming that if there was a shift in the unobserved skills distribution due to changing economy conditions, an individual would not move too far from the original location in the wage distribution. We estimate equation 1 for the 45th, 50th and 55th quantiles for t = 1961, ..., 2012 and test whether coefficients of interest are significantly different across q45 q50 q55 q45 q50 q55 q45 q50 q55 these percentiles. Our H0 ’s are β1t = β1t = β1t , β2t = β2t = β2t and β3t = β3t = β3t . Table 1

summarizes estimates for the 50th percentile and reports p-values for the three t-tests. For most years t-tests 11

fail to reject the null at the 5% significance level. In our estimations, β3t ’s are not statistically different across the three percentiles for all years. There are 2 years for which β1t ’s are statistically different and 13 years for β2t ’s (9 years if we set the significance level at 10%). We find no correlation between the business cycle and the probability to reject the null in β2t test. The results suggest that the effects of composition changes on the estimated returns to skills are small and should not drive the main findings. Two other approaches alternate the baseline specification of equation (1) to control for composition effects. First, we employ the predictions of positive assortative matching theory (see for example Lam 1988) and estimate equation (1) controlling for spousal schooling and spousal ranking in the schooling distribution to proxy for unobserved skills. These estimation use the subsample of married men. Second, we use quantile regression to estimate skill prices at the 50th and 75th percentiles of the wage distribution, assuming that employment is less sensitive to business cycle fluctuations for workers at the mid to high end of wage distribution.9

3.4

Results

Tables 2 and 3 report the main selective estimates of equations (2)-(4). Full results are reported in Appendix Tables A.1-A.6. Table 2 reports results for OLS, OLS with spousal controls and quantile regression specifications, using BLS weights. Table 3 reports similar estimates using constructed weights to match age and education distributions across years. Columns (1)-(4) show results obtained using the output per capita as the business cycle measure; columns (5)-(8) report results using the unemployment rate. Columns (1) and (5) in each table report the baseline OLS results. Columns (2) and (6) report results obtained from estimations of equation (1) that include spousal education and education ranking as controls to address the composition effects. Columns (3) and (7) report results for the 50th quantile estimates. Columns (4) and (8) report results for the 75th quantile.10 The upper panel in Tables 2 and 3 reports results for return to experience. Return to experience is strongly countercyclical in all specifications. There is a positive correlation between the return to experience and relative price index, implying that return to experience is declining when the state of technology is increasing. Results are robust across estimation techniques, using BLS or constructed weights. Full estimation results are reported in Appendix Tables A.1 and A.2. Other controls behave in line with the existing theory. Similarly to Jeong et al. (2015), we find a strong negative relationship between the price of experience and the supply of experience. In Appendix Table A.1 this relationship is to some extent induced by using the average experience to construct the return to experience. Estimations in Appendix Table A.2 address this issue using the adjusted weights which lock average individual attributes to be constant over time.11 The middle panel in Tables 2 and 3 reports results for return to schooling. We find the return to 9 Lindquist (2004) follows a similar approach to clean composition effects and estimates the return to education for the 50th percentile. 10 Appendix Tables B.1-B.6 report estimation results using age to proxy for work experience. Results are very similar. 11 In Appendix Table A.2, experience and schooling means and standard deviations used as controls in the estimations are calculated using the original BLS weights.

12

schooling to exhibit weak countercyclicality, which is more pronounced in estimations that use constructed weights (see Table 3). Returns to schooling are negatively correlated with the relative price index, i.e. positively correlated with the state of technology. This result is also more pronounced in estimations that use constructed weights (see Table 3). Appendix Tables A.3 and A.4 report full results for the return to schooling. Return to schooling is positively correlated with the average level of experience and it increases with the declining in unionization rate, (similar relationship between unionization rate and return to schooling is documented in DiNardo et al. 1996, Freeman and Katz 2007, Lee 1999). The bottom panel in Tables 2 and 3 reports results for the common component, the estimated intercept in equation (1). We use the common component to analyze movements in wages of unskilled workers. Unskilled wages are procyclical in estimations that control for composition effects, this procyclical pattern is more pronounced in estimations that use the constructed weights. This result is in line with the previous literature that finds wages to be procyclical when composition effects are accounted for. Unskilled wages are negatively correlated with the relative price of equipment, i.e. positively with the state of technology. Full outputs in Appendix Tables A.5 and A.6 also show that unskilled wages are positively correlated with unionization rate and decline over time. The positive correlation between return to schooling and technological change is widely documented. For early work see Bound and Johnson (1992), Katz and Murphy (1992), Levy and Murnane (1992), Juhn et al. (1993). Acemoglu (2002) and Hornstein et al. (2005), Autor et al. (2008) and Acemoglu and Autor (2011) provide extensive surveys of this literature. This literature focuses on the substantial increase in the college premium (documented in Figure 2) despite growing supply of college graduates (documented in Figure 1); the leading explanation for this is that skill biased technological change shifts demand towards college graduates. On the other hand, the relationship between technological change and experience premium is not well documented. Majority of previous studies that explore the relationship between schooling premium and business cycle find it to be acyclical. However, it should be noted that most studies base this observation on unconditional correlations, i.e. not controlling for changing state of technology, demographics or institutional trends. The evidence on cyclicality of experience premium is mixed and does not take into account the recent trends. The existing estimates also do not control for changing technology, demographics or institutional trends.

4

A Stylized Model

We show empirically that both return to schooling and return experience play an important role in real wage adjustment to business cycle fluctuations and technological change. Returns to skills are countercyclical; return to schooling is positively correlated with technology and return to experience has a negative correlation with technology. To explain our findings, we propose a model that, under certain assumptions, generates the relationships we observe in the data. Our model extends the capital-skill complementarity framework by incorporating vintages of physical and human capital. In particular, we identify two types of labor skills, education and experience and by differentiating vintages in capital equipment.

13

Capital-skill complementarity hypothesis states that capital is more complementary to skilled labor than to unskilled labor, see for example Krusell et al. (2000). In the standard capital-skill complementarity framework there are two types of capital, structures and equipment, and two types of labor, skilled and unskilled (where skill is determined by level of schooling). In our extended model the production process incorporates four types of labor inputs and three types of physical capital. Four types of workers are low-educated inexperienced (Ut ), low-educated experienced (U Xt ), high-educated inexperienced (Et ), and high-educated experienced (EXt ). Capital inputs are capital structures (kst ), new capital equipment (kent ), and vintage capital equipment (keot ). All factor inputs are in efficiency units. Flexible framework allows for different substitution possibilities between education-experience combinations of labor and different vintages of capital equipment. Similarly to Krusell et al. (2000) we assume an economy with two production sectors and three types of final goods, consumption (ct ), investment in capital structures (Ist ) and investment in capital equipment (Iet ). The output in each sector is given by:

ct + Ist = At G(kst , keot , kent , Ut , U Xt , Et , EXt ),

(5)

Iet = At qt G(kst , keot , kent , Ut , U Xt , Et , EXt ),

(6)

where Ist and Iet are investments in capital structures and new capital equipment in year t, and ct is consumption. Inputs of the same factor in each sector are different, superscript indicators are omitted to simplify notation. There are two types of productivity shocks, aggregate neutral shock is denoted by At and investment-specific technology shock is denoted by qt . The function G(.) is common to both sectors and homogeneous of degree one, which allows to define total output in consumption units and aggregate both sectors.12 We assume the following laws of motion for the three types of capital. kst+1 = (1 − δst )kst + Ist ,

(7)

kent+1 = (1 − δent )kent + Iet , keot+1 = (1 − δeot )keot + δent kent ,

(8)

where the parameters δst , δent and δeot denote capital specific depreciation rates. The first two equations are standard. The stocks of capital structures and capital equipment are increasing with investments in these factors. The law of motion for keot is a new addition to the standard framework; the stock of vintage capital equipment is increasing when new capital equipment becomes obsolete. We propose a production function that is tractable in the context of capital-skill complementarity framework. Our extensions to the standard framework incorporate introduction of the two dimensional skill of 12 See

Krusell et al. (2000) for details.

14

labor and vintage of capital. Our framework reduces to the Krusell et al. (2000) specification if the distinction between the skills of labor and vintages of capital is not important. The production function is a Cobb-Douglas in capital structures and a combination of CES functions of the remaining factor inputs: h i 1−α γ γ , σ α βFUγ + (1 − β) (τ FEσ + (1 − τ )FEX )σ G(Ωt ) = kst

(9)

where G(Ωt ) is the aggregate output, Ωt ≡ {kst , keot , kent , Ut , U Xt , Et , EXt } are factor inputs. Composite production factors FU , FE and FEX are defined as follows.  1 FU = λ1 Utθ + (1 − λ1 )U Xtθ θ , 1

η FE = {λ2 kent + (1 − λ2 )Etη } η ,

FEX =

µ {λ3 koet

+ (1 −

1 λ3 )EXtµ } µ

(10) .

The parameters α, β, τ , λ1 , λ2 , λ3 ∈ (0, 1) govern the income shares and θ, γ, σ, η, µ ∈ (−∞, 1) govern the elasticities of substitution. Equation (9) simplifies to the Krusell et al. (2000) production function if there is no difference in the degrees of substitution within each production factor FU , FE and FEX , i.e., if θ = 1, and σ = η = µ . On the other hand, σ > η and σ > µ imply that the complementarity within FE and FEX is higher than between the components of FE and FEX . For simplicity of the analysis we assume one type of unskilled workers by setting θ = 1. One interpretation of σ > η and σ > µ is that skilled workers get attached to capital equipment technology which is adopted when their generation enters the labor market. Younger educated workers adopt the new technologies while older workers continue operating the older technologies. This production structure is in line with predictions of the human capital theory. Workers decide whether to use the established technology or to adopt a new and better one. The new technology is more productive but costly to adopt and older workers would lose the specific human capital they have accumulated using the old capital equipment. Thus, workers with prior experience with a given technology are less likely to be adopting new technologies if the older technologies are still in place. Chari and Hopenhayn (1991) develop a vintage capital framework and show that under such assumptions the older workers become experts in the specific vintage technology they have adopted when young. Assuming perfectly competitive factor markets, we derive wage rates for each type of labor (WE , WEX and WU ) by taking derivatives of equation  respect to labor inputs (U , E and EX). Using the wage  (9) with WEX E rates, we define the experience premium WE and education premium W WU , which are given as follows, (time subscripts are omitted to simplify presentation),

15

h λ3 WEX = κ1 h WE λ2

i σ−µ µ  σ−1 + (1 − λ3 ) EX i σ−η
E η ken η + (1 − λ ) 2 E


keo µ EX

(11)

"    ση  σµ  η   µ σ # γ−σ σ WE ken keo EX + (1 − λ2 ) + (1 − λ3 ) = κ2 τ λ2 + (1 − τ ) λ3 WU E EX E σ−η  η  γ−1   η ken E + (1 − λ2 ) × λ2 E U where κ1 =

(1−τ )(1−λ3 ) τ (1−λ2 )

and κ2 =

(1−β)τ (1−λ2 ) β

(12)

are positive constants.

Schooling premium and experience premium in equations (11) and (12) are functions of capital-labor ratios

keo ken EX , E

and relative supplies of skills,

EX E E , U.

Coefficients σ, η, µ and γ determine the relationships

between the capital-skill ratios and skill premia. The ratios rise in

EX E

reduces the experience premium and a rise in

E U

EX E

and

E U

drive the relative supply effects. A

reduces the education premium.

We analyze the model in the context of our empirical findings. We find that technological change is positively correlated with the return to schooling and negatively with the return to experience. Return to experience is strongly countercyclical in all specifications. Return to schooling is mildly countercyclical when controlling for composition effects. To be consistent with the model specification, we also construct wage ratios to derive skill premia and use those in our empirical analysis. To obtain the alternative measures of education premium and experience premium we define what constitutes uneducated inexperienced, (Ut ), inexperienced educated, (Et ), and experienced educated, (EXt ). We use two definitions for low/high experience, more than 10 years of potential experience (10 years roughly corresponds to the lowest quartile of experience over the 50 years of CPS data - defined as model 1) and more than 20 years of potential experience (20 years roughly corresponds to the mean experience level over the 50 years of CPS data - defined as model 2). We consider an individual to be highly educated if he has 16 years of schooling or more (as in Krusell et al. 2000). We project wages for each skill group using the 50th percentile estimates of equation (1) (with additional spousal variables to control for composition effects). To control for supply effects we calculate the ratios

EX E

and

E U

using total hours worked for each skill group, including all employed men and women above

18 years old. We estimate equations (2) and (3) using the alternative skill premium definitions and including measures of supply effects,

EX E

in equation (2) and

E U

in equation (3). We also include the proportion of

men in the labor force, mean and standard deviation of education and experience, and unionization rate to control for changes in efficiency units of different types of labor. Key outcomes are reported in Table 4. Findings are very similar to those reported in Section 3.4. Schooling premium and experience premium are countercyclical; schooling premium is negatively correlated with the relative price index whereas experience premium is positively correlated with the relative price index. Coefficients of supply effects are negative and consistent with model assumptions. It follows from equation (11), keeping

EX E

constant, a rise in 16

keo EX

or decline in

ken E

increase the experience

premium if σ > µ and σ > η, respectively. It follows from equation (12), keeping in

ken E

keo EX

E U

constant, an increase

has a positive effect on schooling premium if γ > σ and σ > η. On the other hand, a decline in

will negatively affect education premium if γ > σ. In our vintage-capital-skill complementarity model,

γ > σ implies a higher complementarity between the composites of FE and FEX than between unskilled labor and composites of FE or FEX . If γ > σ, then there is a positive relationship between

keo EX

(and

EX E )

and education premium. Results in Table 4 show that controlling for supply effects, experience premium, defined in equation (11), and education premium, defined in equation (12) are countercyclical. We also find that, controlling for supply effects, experience premium is positively correlated with the relative price of capital equipment, whereas education premium is negatively correlated with the relative price index. We follow Fisher (2006) and Balleer and Van Rens (2013) and use the relative price of capital equipment to identify investmentspecific shock and to define investment-neutral technology shocks as all remaining shocks that drive labor productivity. In our theoretical framework, neutral shock, At , drives the business cycle; investment specific shock, qt , drives technological innovations. Assuming that both investment specific shock and neutral shock affect the size of investment but not the size of the exiting vintage capital, ken is increasing when these shocks are positive whereas keo is not affected by contemporaneous productivity or technology shocks. We test empirically how capital-labor ratio,

ken E ,

and relative supply effects,

E U

and

EX E ,

move with the neutral

shock and investment specific shock. A measure of ken is real capital equipment investment series (deflated using the GDP deflator) reported by the U.S. Bureau of Economic Analysis (BEA), (see Table 5.5.5., Private Fixed Investment in Equipment by Type). Tables 5 and 6 report the estimation results for capital-labor ratios and labor ratios, respectively. We use two definitions for high experience, more than 10 years of potential experience (model 1) and more than 20 years of potential experience (model 2). We consider an individual to be highly educated if he has 16 years of schooling or more. Similarly to the previous estimations, we use two measures of neutral shocks, the lagged HP-filtered real GDP per capita and unemployment rate. The ratio types of shocks. Labor ratio

E U

ken E

is positively correlated with both

is negatively correlated with both types of shocks. Labor ratio

EX E

is not

correlated with either type of shock when EX is defined as 10 or more years of experience and exhibits weak positive correlations with both types of shocks when when using 20 or more years of experience to define EX. Assuming that keo is not affected by contemporaneous productivity or technology shocks, we derive that

keo EX

is countercyclical and negatively correlated with investment-specific shock (due to the increase in

EX in response to both types of shocks). Theoretical analysis of the model shows that it can generate countercyclical experience premium negatively correlated with technological change and countercyclical education premium positively correlated with technological change if five assumptions hold. (i) More-experienced high-educated labor, EX, is more complementary with vintage capital equipment, keo , than with new capital equipment, ken ; i.e., σ > µ. (ii) Less-experienced high-educated labor, E, is more complementary with new capital equipment, ken , than with vintage capital, keo ; i.e., σ > η. (iii) The complementarity between high-educated labor (at any experience level), EX or E, and capital equipment (of any vintage), ken or keo , is higher than it is between low-educated labor, U , and high-educated labor or capital equipment, i.e., γ > σ. (iv) The magnitude of

17

response of ken /E to an investment-specific shock is larger than the magnitude of response of keo /EX. (v) The magnitude of response of ken /E to a neutral shock is smaller than the magnitude of response of keo /EX. It is evident from equation (11) that only assumptions (i) and (ii) are required to produce a countercyclical experience premium negatively correlated with technological change. It is also evident that satisfying these two assumptions is a sufficient condition. The model can generate patterns in experience premium observed in the data if only one of these two assumptions holds. From equation (11), given that WWEX is countercyclical E and negatively correlated with investment-specific shock, procyclicality of

ken E

and countercyclicality of

keo EX

(and their positive and negative correlations with the investment-specific shock) suggest that assumptions (i) and/or (ii) hold, i.e., σ > µ and/or σ > η. It follows from equation (12) that assumptions (ii)-(v) lead to countercyclical education premium positively correlated with technological change. These four assumptions are sufficient conditions to generate patterns in education premium observed in the data. For a given parameters set, assumption (ii) could be reduced if assumptions (iii)-(v) are satisfied. From equation (12), given that

WE WU

is countercyclical and positively correlated with investment-specific shock, considering the

Krusell et al. (2000) outcome of γ > σ, implies that the countercyclical response of procyclical response of

ken E ,

keo EX

dominates the

leading to a countercyclical education premium. On the other hand, it follows

that the positive response of

ken E

to investment-specific shock dominates the negative response of

keo EX .

The

countercyclical education premium positively correlated with investment-specific shock is also in line with the vintage-capital-skill complementarity assumptions of σ > µ and σ > η.

5

Conclusion

We study how technological change and business cycle shape the wage structure. Controlling for institutional and demographic changes, we examine patterns of skill prices and real wages using data from 1962 - 2013 CPS. We estimate the marginal return to experience, marginal return to schooling and common wage component (i.e., mean wage of unskilled inexperienced labor) for each year in the sample, controlling for individual and geographic characteristics. We examine how the estimated returns to skills vary with the technological change, institutional changes, demographic changes and the business cycle. To measure the business cycle we use HP-filtered real output per capita and unemployment rate. Technological change is measured by a relative price index of capital equipment and software, constructed by dividing the price index of nondurable consumption goods and nonhousing services by the price index of capital equipment. We find that technological change is positively correlated with the return to schooling and negatively with the return to experience. Return to experience is strongly countercyclical in all specifications. Return to schooling is mildly countercyclical when controlling for composition effects and technological change. The common across skill types wage component, measured by the intercept in wage function, is procyclical in specifications that control for the changing workforce composition over the business cycle and it is positively correlated with the technological change. Thus, the overall procyclicality of real wages (reported in a number of existing studies) is driven by the common component. Previous studies on wage structure focus on the earlier, 1970s-1990s, period and show different results.

18

Both education and experience premia had been increasing between the early 1970s and late 1980s. Earlier studies argue that similar channels lead to these developments in wage structure. The dominant explanations are the skill-biased technological change and the associated shifts in the demand for skilled labor; globalization pressures that led to shrinking relative demand for less skilled and to the increase in skill premia; declining unionization and real minimum wage. Our conclusions are different; we point out the different response of experience and education premia to technological change. To interpret the findings we outline a model that extends the standard capital-skill complementarity framework by incorporating vintage human capital and vintage physical capital. We distinguish between two types of skills, education and experience, and two types of capital equipment, new investment and vintage. Our model generates relationships between skill prices, neutral and investment-specific technology shocks we observe in the data if vintage-capital-skill complementarity assumptions hold. These vintagecapital-skill complementarity assumptions suggest that skilled workers get attached to capital equipment technology which is adopted when their generation enters the labor market. Younger educated workers adopt the new technologies while older workers continue operating the older technologies.

19

References Acemoglu, D. (2002). Technical change, inequality, and the labor market. Journal of economic literature 40 (1), 7–72. Acemoglu, D. and D. Autor (2011). Skills, tasks and technologies: Implications for employment and earnings. Handbook of labor economics 4, 1043–1171. Autor, D. H., L. F. Katz, and M. Kearney (2005). Trends in us wage inequality: Responding to the revisionists. NBER Working Paper 11628. Autor, D. H., L. F. Katz, and M. S. Kearney (2008). Trends in us wage inequality: Revising the revisionists. The Review of economics and statistics 90 (2), 300–323. Autor, David, H., L. F. Katz, and A. B. Krueger (1998). Computing inequality: have computers changed the labor market? Quarterly Journal of Economics 113 (4). Balleer, A. and T. Van Rens (2013). Skill-biased technological change and the business cycle. Review of Economics and Statistics 95 (4), 1222–1237. Bartel, A. P. and F. R. Lichtenberg (1987). The comparative advantage of educated workers in implementing new technology. The Review of Economics and statistics, 1–11. Berman, E., J. Bound, and Z. Griliches (1994). Changes in the demand for skilled labor within us manufacturing: Evidence from the annual survey of manufacturers. The Quarterly Journal of Economics, 367–397. Bils, M. J. (1985). Real wages over the business cycle: evidence from panel data. Journal of Political economy 93 (4), 666–689. Blank, R. M. (1990). Why are wages cyclical in the 1970s? Journal of Labor Economics 8 (1, Part 1), 16–47. Bound, J. and G. Johnson (1992). Changes in the structure of wages in the 1980’s: An evaluation of alternative explanations. American Economic Review 82 (3), 371–92. Carneiro, A., P. Guimar˜ aes, and P. Portugal (2012). Real wages and the business cycle: Accounting for worker, firm, and job title heterogeneity. American Economic Journal: Macroeconomics 4 (2), 133–152. Castro, R. and D. Coen-Pirani (2008). Why have aggregate skilled hours become so cyclical since the mid-1980s? International Economic Review 49 (1), 135–185. Chari, V. V. and H. Hopenhayn (1991). Vintage human capital, growth, and the diffusion of new technology. Journal of political Economy, 1142–1165. Chun, H. (2003). Information technology and the demand for educated workers: Disentangling the impacts of adoption versus use. The Review of Economics and Statistics 85 (1), 1–8. Cummins, J. G. and G. L. Violante (2002). Investment-specific technical change in the united states (1947– 2000): Measurement and macroeconomic consequences. Review of Economic dynamics 5 (2), 243–284. 20

DiNardo, J., N. M. Fortin, and T. Lemieux (1996). Labor market institutions and the distribution of wages, 1973-1992: A semiparametric approach. Econometrica 64 (5), 1001–1044. Fisher, J. D. (2006). The dynamic effects of neutral and investment-specific technology shocks. Journal of political Economy 114 (3), 413–451. Freeman, R. B. and L. F. Katz (2007). Differences and changes in wage structures. University of Chicago Press. Gordon, R. J. (2007). The measurement of durable goods prices. University of Chicago Press. Greenwood, J., Z. Hercowitz, and P. Krusell (1997). Long-run implications of investment-specific technological change. The American Economic Review , 342–362. Greenwood, J. and M. Yorukoglu (1997). 1974. In Carnegie-Rochester Conference Series on Public Policy, Volume 46, pp. 49–95. Elsevier. Griliches, Z. (1969). Capital-skill complementarity. The Review of Economics and Statistics 51 (4), 465–468. Hornstein, A., P. Krusell, and G. L. Violante (2005). The replacement problem in frictional economies: A near-equivalence result. Journal of the European Economic Association 3 (5), 1007–1057. Jeong, H., Y. Kim, and I. Manovskii (2015). The price of experience. The American Economic Review 105 (2), 784–815. Juhn, C., K. M. Murphy, and B. Pierce (1993). Wage inequality and the rise in returns to skill. Journal of political Economy, 410–442. Katz, L. F. and K. M. Murphy (1992). Changes in relative wages, 1963-1987: Supply and demand factors. The Quarterly Journal of Economics 107 (1), 35–78. Katz, L. F. and A. L. Revenga (1989). Changes in the structure of wages: The united states vs japan. Journal of the Japanese and International Economies 3 (4), 522–553. Keane, M., R. Moffitt, and D. Runkle (1988). Real wages over the business cycle: Estimating the impact of heterogeneity with micro data. The Journal of Political Economy, 1232–1266. Keane, M. P. and E. Prasad (1993). The relation between skill levels and the cyclical variability of employment, hours and wages. Technical report. Krusell, P., L. E. Ohanian, J.-V. R´ıos-Rull, and G. L. Violante (2000). Capital-skill complementarity and inequality: A macroeconomic analysis. Econometrica 68 (5), 1029–1053. Lam, D. (1988). Marriage markets and assortative mating with household public goods: Theoretical results and empirical implications. Journal of Human resources, 462–487. Lee, D. S. (1999). Wage inequality in the united states during the 1980s: Rising dispersion or falling minimum wage? Quarterly Journal of Economics, 977–1023.

21

Levy, F. and R. J. Murnane (1992). Us earnings levels and earnings inequality: A review of recent trends and proposed explanations. Journal of economic literature 30 (3), 1333–1381. Lindquist, M. J. (2004). Capital–skill complementarity and inequality over the business cycle. Review of Economic Dynamics 7 (3), 519–540. Mayer, G. (2004). Union membership trends in the united states. Mincer, J. (1974). Schooling, experience, and earnings. New York: NBER Press. Mincer, J. (1991). Human capital, technology, and the wage structure: what do time series show? Technical report, National Bureau of Economic Research. Murphy, K. M. and F. Welch (1992). The structure of wages. The Quarterly Journal of Economics, 285–326. Nelson, R. R. and E. S. Phelps (1966). Investment in humans, technological diffusion, and economic growth. The American economic review 56 (1/2), 69–75. Raisian, J. (1983). Contracts, job experience, and cyclical labor market adjustments. Journal of Labor Economics, 152–170. Reder, M. W. (1955). The theory of occupational wage differentials. The American Economic Review 45 (5), 833–852. Schultz, T. W. (1975). The value of the ability to deal with disequilibria. Journal of economic literature 13 (3), 827–846. Solon, G., R. Barsky, and J. A. Parker (1994). Measuring the cyclicality of real wages: How important is composition bias? The Quarterly Journal of Economics 109 (1), 1–25. Stockman, A. C. (1983). Aggregation bias and the cyclical behavior of real wages. Unpublished manuscript. Tinbergen, J. (1975). Income differences: recent research. Welch, F. (1970). Education in production. Journal of political economy 78 (1), 35–59. Young, E. R. (2003). The wage premium: A puzzle. Florida State University 33. Ziliak, J. P., B. A. Wilson, and J. A. Stone (1999, May). Spatial Dynamics And Heterogeneity In The Cyclicality Of Real Wages. The Review of Economics and Statistics 81 (2), 227–236.

22

Table 1: 50th quantile estimates of returns to schooling and experience; t-test outcomes for null hypothesis that there are no difference in estimates for nearby percentiles, p45, p50 and p55. H0: b(p45)=b(p50)=b(p55) experience experience2 schooling return to return to school exp year N (p50) (p50) F-stat p-value F-stat p-value F-stat p-value (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

1961 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990

7705 8158 8089 17272 10878 17386 17617 17030 17104 16437 16429 16193 15866 15450 18172 17927 17960 20971 21252 19047 18686 18330 19032 18885 18786 18807 17706 19326 19269

0.0178 0.0190 0.0206 0.0219 0.0213 0.0216 0.0207 0.0239 0.0279 0.0329 0.0363 0.0446 0.0504 0.0624 0.0704 0.0752 0.0884 0.1000 0.1201 0.1312 0.1464 0.1561 0.1667 0.1715 0.1878 0.1815 0.1793 0.1902 0.1866

0.1216 0.1353 0.1472 0.1576 0.1590 0.1726 0.1710 0.2021 0.2182 0.2339 0.2540 0.2533 0.2828 0.3404 0.3343 0.3508 0.3819 0.4122 0.4869 0.5665 0.6563 0.6936 0.7714 0.8449 0.8516 0.8945 0.8892 1.0258 1.0334

1.95 8.58 1.01 16.90 5.51 10.93 8.26 17.69 14.34 12.37 13.47 15.61 26.86 16.51 10.44 18.21 14.54 16.81 14.98 9.39 12.52 14.85 13.69 17.32 12.00 10.92 15.99 16.34 9.19

0.14 0.00 0.37 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.90 3.25 0.16 8.18 3.13 4.49 4.57 8.24 5.80 5.39 4.96 5.36 11.40 7.57 3.73 6.37 4.44 8.85 7.22 3.12 4.88 5.23 4.73 5.58 4.66 1.81 4.76 4.37 2.38

0.41 0.04 0.85 0.00 0.04 0.01 0.01 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.02 0.00 0.01 0.00 0.00 0.04 0.01 0.01 0.01 0.00 0.01 0.16 0.01 0.01 0.09

23.70 19.27 14.34 25.81 18.82 39.25 34.56 44.74 33.55 46.96 26.15 27.36 36.25 35.88 30.52 30.83 38.00 26.33 23.77 34.71 34.43 43.35 36.85 51.59 45.77 61.86 73.96 69.11 75.80

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

See Table notes on next page. Continued 23

Table 1 (continued): 50th quantile estimates of returns to schooling and experience; t-test outcomes for null hypothesis that there are no difference in estimates for nearby percentiles, p45, p50 and p55. H0: b(p45)=b(p50)=b(p55) experience experience2 schooling return to return to school year N (p50) exp (p50) F-stat p-value F-stat p-value F-stat p-value (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

19034 18658 17917 18037 16073 16607 16518 17200 17394 26901 26899 26352 25315 24640 24757 24787 24418 24285 23149 22420 22184 22338

0.1863 0.2004 0.2050 0.2044 0.1827 0.1897 0.1848 0.1832 0.1825 0.1702 0.1701 0.1615 0.1620 0.1686 0.1790 0.1664 0.1754 0.1713 0.1787 0.2089 0.1997 0.2016

1.0744 1.1369 1.1916 1.2400 1.1969 1.2617 1.2615 1.3744 1.4051 1.5040 1.5912 1.6843 1.6352 1.6894 1.7703 1.8025 1.8547 2.0293 2.0565 2.0799 2.2191 2.1661

19.59 6.87 10.36 9.77 13.85 8.23 9.02 10.10 7.82 7.85 19.58 14.73 10.69 8.02 8.99 13.98 15.05 14.31 18.19 10.42 13.38 14.90

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

6.21 1.84 2.58 2.47 3.86 2.06 1.27 3.18 1.48 2.16 8.43 7.09 2.20 2.83 3.67 3.80 5.45 5.43 10.34 3.08 6.08 7.95

0.00 0.16 0.08 0.08 0.02 0.13 0.28 0.04 0.23 0.12 0.00 0.00 0.11 0.06 0.03 0.02 0.00 0.00 0.00 0.05 0.00 0.00

75.68 62.58 73.73 58.83 55.61 59.21 71.50 53.93 66.12 88.34 128.54 93.68 105.37 95.86 103.21 111.52 138.17 98.29 89.61 90.16 78.38 96.24

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Note: This table reports quantile regressoin estimates of equation (1) for each year in the 19612013 CPS using OLS. Columnss (3) and (4) report the p50 estimates. Columns (5)-(10) report t-test results for H0: b45=50=b55 (for three estimates, coefficients of schooling, experience and experience squared). The variables and data sources are defined in the text.

24

Table 2: Determinants of returns to experience, returns to schooling and the common component, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) Return to experience b.cycle

-0.0376*** -0.0325*** -0.0410*** -0.0298*** 0.0007*** 0.0007*** 0.0008*** 0.0007*** (0.008) (0.007) (0.009) (0.008) (0.000) (0.000) (0.000) (0.000) rel price 0.0237*** 0.0232*** 0.0346*** 0.0217*** 0.0220*** 0.0219*** 0.0326*** 0.0206*** (0.006) (0.007) (0.006) (0.007) (0.006) (0.007) (0.006) (0.006)

Return to schooling b.cycle rel price

-0.0462 (0.029) -0.0269 (0.021)

-0.0236 (0.024) -0.0198 (0.018)

-0.0622* (0.034) 0.0078 (0.022)

-0.0463 (0.031) -0.0467** (0.022)

0.0008 (0.001) -0.0293 (0.022)

0.0003 (0.001) -0.0215 (0.018)

0.0012 (0.001) 0.0049 (0.022)

0.0010* (0.001) -0.0485** (0.022)

Common component b.cycle

1.3406 1.2493 2.8358*** 1.8530*** -0.0199 -0.0047 -0.0475** -0.0432*** (1.013) (1.094) (0.955) (0.652) (0.022) (0.024) (0.022) (0.013) rel price -1.4255*** -2.2048*** -1.9686*** -0.6096 -1.3415*** -2.0772*** -1.8097*** -0.5491 (0.424) (0.631) (0.528) (0.477) (0.437) (0.673) (0.539) (0.411)

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. Selected estimates, full estimation results are in Appendix Tables A.1, A.3 and A.5. Return to experience, return to schooling and common component are estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

25

Table 3: Determinants of returns to experience, returns to schooling and the common component, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8)

b.cycle

Return to experience -0.0380***-0.0352***-0.0408***-0.0279*** 0.0008*** 0.0007*** 0.0008*** 0.0006*** (0.007)

rel price

(0.008)

(0.007)

-0.0515*

-0.0416

-0.0697*

(0.029)

(0.028)

(0.036)

rel price -0.0774*** -0.0410* (0.021)

b.cycle

(0.008)

(0.009)

(0.000)

(0.000)

(0.000)

(0.000)

0.0161** 0.0206** 0.0292*** 0.0185** 0.0145** 0.0191** 0.0273*** 0.0176** (0.007)

b.cycle

(0.008)

1.1637 (1.545)

(0.022)

(0.007)

(0.007)

(0.007)

(0.007)

(0.007)

Return to schooling -0.0417 0.001

0.0006

0.0014*

0.0011*

(0.030)

(0.001)

(0.001)

(0.001)

(0.001)

-0.0325 -0.0922***-0.0799*** -0.0437*

-0.0357 -0.0932***

(0.024)

(0.025)

(0.024)

(0.021)

(0.023)

(0.023)

Common component 2.3336*** 2.8879*** 1.8740*** -0.0182 -0.0400** -0.0485** -0.0392*** (0.829)

(0.981)

(0.693)

(0.027)

(0.016)

(0.023)

(0.013)

rel price -1.7419***-3.1449***-2.4860*** -1.1739** -1.6721***-3.0174***-2.3247***-1.0968*** (0.591)

(0.655)

(0.497)

(0.437)

(0.601)

(0.663)

(0.492)

(0.392)

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. Selected estimates, full estimation results are in Appendix Tables A.2, A.4 and A.6. Return to experience, return to schooling and common component are estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables , constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

26

Table 4: Determinants of wage ratios output unemployment rate model 1 model 2 model 1 model 2 We/Wu Wex/We We/Wu Wex/We We/Wu Wex/We We/Wu Wex/We (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

-0.2573 (0.206) rel price -0.4830** (0.204) mean exp 0.013 (0.013) mean educ -0.1350** (0.067) sd exp -0.0396** (0.018) sd educ 0.2536 (0.176) union % -0.2725 (0.418) male % 0.3517 (1.143) trend 0.0006 (0.006) supply effect 0.3314 (0.241) const 1.5376 (12.082)

N R2 adj.

50 0.963

-0.7380*** -0.5491*** -0.6981*** (0.172) (0.137) (0.251) 0.5445*** -0.0956 0.4191** (0.106) (0.178) (0.178) -0.0087 0.0091 -0.0013 (0.012) (0.009) (0.021) 0.0363 -0.0239 -0.0078 (0.054) (0.061) (0.079) 0.0241** -0.0323** -0.0475* (0.012) (0.013) (0.028) -0.3324*** 0.3668** -0.5336** (0.120) (0.161) (0.261) -0.5038 -0.423 -0.4831 (0.303) (0.379) (0.598) -0.3319 -0.7206 1.5956 (0.575) (0.716) (0.959) 0.0100** 0.0015 0.0128* (0.004) (0.004) (0.007) -0.3864*** -0.0593 -0.1131 (0.098) (0.157) (0.070) -18.0812** -1.3811 -23.0141* (7.330) (7.333) (12.733) 50 0.963

50 0.966

50 0.915

0.003 (0.005) -0.5363** (0.217) 0.0139 (0.013) -0.1345* (0.069) -0.0391** (0.018) 0.2366 (0.177) -0.3149 (0.425) 0.5127 (1.164) -0.0007 (0.007) 0.3877 (0.259) 3.9102 (12.792)

0.0144*** (0.003) 0.4983*** (0.098) -0.0002 (0.012) 0.0407 (0.058) 0.0306** (0.013) -0.3776** (0.150) -0.6369* (0.360) -0.3582 (0.671) 0.0091** (0.004) -0.4410*** (0.089) -16.3371** (7.474)

0.0075** (0.003) -0.1909 (0.193) 0.0108 (0.008) -0.0309 (0.068) -0.0305** (0.015) 0.3470* (0.177) -0.5223 (0.431) -0.4714 (0.776) -0.0002 (0.004) 0.0217 (0.172) 2.0061 (8.047)

0.0150*** (0.005) 0.3887** (0.154) 0.0056 (0.022) -0.0124 (0.085) -0.0491* (0.028) -0.5632* (0.285) -0.6063 (0.588) 1.5425 (0.994) 0.0123* (0.007) -0.1278* (0.074) -21.7926* (12.341)

50 0.962

50 0.962

50 0.963

50 0.917

Note: This table reports estimates with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. We project wages for each skill group using the 50th percentile estimates (with spousal controls) of equation (1). We is wage of low-experienced employed with 16 or more years of schooling. Wex is wage of high-experienced employed with 16 or more years of schooling. Wu is wage of low-experienced employed with less than 16 years of schooling. Model 1 defines high experience if individual's potential experience is 10 years or more. Model 2 defines high experience if individual's potential experience is 20 years or more. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

27

Table 5: Investment in capital equipment (Ken), capital labor ratios (Ken/E) and shocks output unemployment rate model 1 model 2 model 1 model 2 Ken Ken/E Ken/E Ken Ken/E Ken/E (1) (2) (3) (4) (5) (6) b.cycle rel price mean exp mean educ sd exp sd educ union % male % trend const

N R2 adj.

8.9048*** (2.850) -8.0644*** (2.696) -0.1849 (0.232) -1.1826 (1.230) 0.0327 (0.227) 6.9984* (3.762) -1.4282 (8.229) 1.6391 (13.345) 0.0115 (0.103) -16.1038 (192.276)

0.6893* (0.357) -0.534 (0.418) 0.0401 (0.030) -0.3551* (0.210) -0.0932*** (0.034) 0.7595 (0.489) 2.2386* (1.241) -0.8659 (2.718) 0.0053 (0.014) -6.5898 (24.635)

0.9284** (0.432) -0.7947* (0.455) 0.0248 (0.033) -0.2406 (0.207) 0.0273 (0.034) 0.442 (0.542) 2.2641 (1.431) -0.2915 (2.819) -0.0014 (0.014) 4.6534 (25.935)

-0.1492* (0.079) -7.5662*** (2.671) -0.2333 (0.233) -1.2825 (1.258) -0.0061 (0.220) 7.5878* (3.803) 0.7153 (8.599) 0.6856 (13.782) 0.0289 (0.106) -49.5023 (196.855)

-0.0064 (0.008) -0.4771 (0.451) 0.0365 (0.031) -0.3715* (0.210) -0.0957** (0.036) 0.8015 (0.496) 2.4115* (1.306) -1.1464 (2.939) 0.007 (0.014) -9.695 (25.818)

-0.0106 (0.010) -0.725 (0.489) 0.0199 (0.034) -0.2594 (0.213) 0.0238 (0.038) 0.4999 (0.562) 2.4944 (1.532) -0.5912 (3.108) 0.0008 (0.015) 0.6677 (27.968)

50 0.938

50 0.921

50 0.915

50 0.935

50 0.918

50 0.911

Note: This table reports estimates with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Ken is (real) capital equipment investment from BEA Table 5.5.5. E measures total annual hours of low-experienced employed with 16 or more years of schooling, obtained from the CPS. Model 1 defines high experience if individual's potential experience is 10 years or more. Model 2 defines high experience if individual's potential experience is 20 years or more. High education is defined as 16 years of schooling or more. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

28

Table 6: Determinants of labor ratios output unemployment rate model 1 model 2 model 1 model 2 E/U EX/E E/U EX/E E/U EX/E E/U EX/E (1) (2) (3) (4) (5) (6) (7) (8) b.cycle rel price mean exp mean educ sd exp sd educ union % male % trend const

N R2 adj.

-0.4338*** 0.1539 -0.4048*** 0.2509* (0.135) (0.353) (0.113) (0.138) 0.7716*** -0.2271 0.6566*** -0.2522** (0.168) (0.329) (0.136) (0.122) 0.013 0.2066*** 0.0047 0.0837*** (0.014) (0.039) (0.012) (0.017) 0.1800* -0.4605** 0.0657 0.0023 (0.092) (0.217) (0.068) (0.054) 0.0041 -0.3401*** 0.0078 0.0563*** (0.016) (0.049) (0.013) (0.019) -0.1649 1.2740*** 0.0376 0.0802 (0.120) (0.382) (0.119) (0.159) -0.3207 2.6944*** -0.2716 0.7000** (0.336) (0.911) (0.357) (0.330) -1.8406 0.4486 -1.7510* -0.2335 (1.110) (3.478) (0.896) (1.215) 0.0079 0.0378*** 0.0126*** 0.006 (0.005) (0.014) (0.004) (0.005) -16.6243* -71.2006***-24.8829*** -13.6325 (9.687) (25.935) (7.961) (9.238) (50.000) 0.989

(50.000) 0.99

(50.000) 0.992

(50.000) 0.987

0.0068** 0.0022 0.0064** -0.0028 (0.003) (0.007) (0.003) (0.003) 0.7457*** -0.2015 0.6325*** -0.2332* (0.171) (0.342) (0.139) (0.120) 0.0154 0.2059*** 0.0069 0.0824*** (0.015) (0.039) (0.012) (0.017) 0.1857* -0.4702** 0.0709 -0.0029 (0.095) (0.213) (0.073) (0.057) 0.006 -0.3403*** 0.0095 0.0553*** (0.017) (0.048) (0.015) (0.020) -0.1933 1.2807*** 0.0111 0.0958 (0.128) (0.368) (0.132) (0.156) -0.4258 2.7379*** -0.3696 0.7623** (0.364) (0.877) (0.379) (0.322) -1.7751 0.24 -1.691 -0.3168 (1.223) (3.563) (1.015) (1.303) 0.007 0.0384*** 0.0117** 0.0065 (0.006) (0.014) (0.005) (0.005) -14.9495 -72.2607***-23.3227*** -14.7152 (10.105) (25.290) (8.568) (9.491) (50.000) 0.988

(50.000) 0.99

(50.000) 0.991

(50.000) 0.987

Note: This table reports estimates with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. All data are from 1962-2012 CPS. E measures total annual hours of low-experienced employed with 16 or more years of schooling. U measures total annual hours of low-experienced employed with less than 16 years of schooling. EX measures total annual hours of high-experienced employed with 16 or more years of schooling. Model 1 defines high experience if individual's potential experience is 10 years or more. Model 2 defines high experience if individual's potential experience is 20 years or more. High education is defined as 16 years of schooling or more. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

29

Table A.1: Determinants of return to experience, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% (1) (2) (3) (4) (5) (6) (7) b.cycle

-0.0376*** (0.008) rel price 0.0237*** (0.006) mean exp -0.0035*** (0.000) mean educ -0.0025 (0.002) sd exp -0.0022*** (0.001) sd educ -0.0057 (0.006) union % -0.0052 (0.022) male % 0.0581 (0.035) trend 0.0006*** (0.000) const -1.0542*** (0.329)

R2 adj.

0.969

q 75% (8)

-0.0325*** (0.007) 0.0232*** (0.007) -0.0032*** (0.000) -0.0019 (0.003) -0.0020*** (0.001) -0.0088 (0.007) -0.0242 (0.023) 0.0572 (0.035) 0.0005** (0.000) -0.9036** (0.381)

-0.0410*** (0.009) 0.0346*** (0.006) -0.0041*** (0.000) -0.0025 (0.002) -0.0026*** (0.001) -0.0057 (0.006) 0 (0.019) 0.0298 (0.038) 0.0007*** (0.000) -1.3131*** (0.375)

-0.0298*** (0.008) 0.0217*** (0.007) -0.0030*** (0.000) -0.0006 (0.003) -0.0027*** (0.001) -0.0129* (0.007) -0.0137 (0.023) 0.0607* (0.031) 0.0004** (0.000) -0.7507* (0.376)

0.0007*** (0.000) 0.0220*** (0.006) -0.0033*** (0.000) -0.0023 (0.002) -0.0020*** (0.001) -0.0083 (0.007) -0.0141 (0.021) 0.0577 (0.044) 0.0005*** (0.000) -0.9244** (0.358)

0.0007*** (0.000) 0.0219*** (0.007) -0.0030*** (0.000) -0.0018 (0.003) -0.0019*** (0.001) -0.011 (0.008) -0.0318 (0.022) 0.055 (0.040) 0.0005** (0.000) -0.7960** (0.373)

0.0008*** (0.000) 0.0326*** (0.006) -0.0039*** (0.001) -0.0022 (0.003) -0.0024*** (0.001) -0.0085 (0.008) -0.0098 (0.019) 0.0307 (0.048) 0.0007*** (0.000) -1.1682*** (0.415)

0.0007*** (0.000) 0.0206*** (0.006) -0.0028*** (0.000) -0.0006 (0.003) -0.0026*** (0.001) -0.0150* (0.008) -0.0207 (0.022) 0.0576 (0.035) 0.0004* (0.000) -0.6548* (0.359)

0.961

0.974

0.967

0.968

0.962

0.971

0.969

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to experience is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

30

Table A.2: Determinants of return to experience, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% (1) (2) (3) (4) (5) (6) (7) b.cycle

-0.0380*** (0.007) rel price 0.0161** (0.007) mean exp -0.0021*** (0.000) mean educ -0.0021 (0.002) sd exp -0.0017** (0.001) sd educ -0.0102* (0.006) union % -0.0064 (0.021) male % 0.0668** (0.032) trend 0.0005*** (0.000) const -0.8940** (0.343)

R2 adj.

0.923

q 75% (8)

-0.0352*** (0.008) 0.0206** (0.008) -0.0018*** (0.000) -0.0013 (0.003) -0.0016** (0.001) -0.0108 (0.007) -0.0244 (0.023) 0.0663* (0.036) 0.0005** (0.000) -0.8692** (0.397)

-0.0408*** (0.008) 0.0292*** (0.007) -0.0026*** (0.000) -0.0024 (0.002) -0.0021*** (0.001) -0.0110* (0.006) -0.0086 (0.017) 0.0411 (0.035) 0.0007*** (0.000) -1.1995*** (0.379)

-0.0279*** (0.009) 0.0185** (0.007) -0.0014*** (0.000) 0.0005 (0.003) -0.0024*** (0.001) -0.0164** (0.007) -0.0148 (0.022) 0.0730** (0.033) 0.0003 (0.000) -0.6298 (0.378)

0.0008*** (0.000) 0.0145** (0.007) -0.0019*** (0.000) -0.0019 (0.002) -0.0015** (0.001) -0.0128* (0.007) -0.0154 (0.020) 0.0649 (0.040) 0.0004** (0.000) -0.7665** (0.352)

0.0007*** (0.000) 0.0191** (0.007) -0.0016*** (0.000) -0.0012 (0.003) -0.0015** (0.001) -0.0132* (0.008) -0.0327 (0.023) 0.064 (0.042) 0.0004* (0.000) -0.7527* (0.392)

0.0008*** (0.000) 0.0273*** (0.007) -0.0023*** (0.001) -0.0021 (0.003) -0.0019*** (0.001) -0.0138* (0.007) -0.0183 (0.017) 0.0411 (0.046) 0.0006** (0.000) -1.0572** (0.410)

0.0006*** (0.000) 0.0176** (0.007) -0.0012*** (0.000) 0.0005 (0.003) -0.0022*** (0.001) -0.0183** (0.007) -0.0213 (0.021) 0.0690* (0.037) 0.0003 (0.000) -0.5429 (0.359)

0.891

0.941

0.896

0.923

0.894

0.937

0.904

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to experience is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables, constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

31

Table A.3: Determinants of return to schooling, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% (1) (2) (3) (4) (5) (6) (7) b.cycle

-0.0462 (0.029) rel price -0.0269 (0.021) mean exp 0.0039** (0.002) mean educ -0.0084 (0.009) sd exp -0.0047* (0.003) sd educ -0.0202 (0.027) union % -0.1388 (0.092) male % -0.1868 (0.126) trend 0.0003 (0.001) const -0.215 (1.562)

R2 adj.

0.984

-0.0236 -0.0622* -0.0463 0.0008 (0.024) (0.034) (0.031) (0.001) -0.0198 0.0078 -0.0467** -0.0293 (0.018) (0.022) (0.022) (0.022) 0.0022* 0.0021 0.0047*** 0.0041** (0.001) (0.002) (0.002) (0.002) -0.0013 -0.0035 -0.0043 -0.008 (0.010) (0.009) (0.011) (0.010) -0.0044** -0.0080*** -0.0056*** -0.0045* (0.002) (0.002) (0.002) (0.003) 0.0141 -0.025 -0.0063 -0.0233 (0.024) (0.025) (0.026) (0.028) -0.1210* -0.0834 -0.1331* -0.1499 (0.064) (0.092) (0.077) (0.094) -0.0483 -0.1355 -0.0923 -0.1839 (0.103) (0.128) (0.126) (0.135) 0.0002 0.0008 0 0.0002 (0.001) (0.001) (0.001) (0.001) -0.3731 -1.3472 0.2042 -0.0468 (1.178) (1.607) (1.433) (1.631) 0.981

0.983

0.987

0.983

q 75% (8)

0.0003 0.0012 0.0010* (0.001) (0.001) (0.001) -0.0215 0.0049 -0.0485** (0.018) (0.022) (0.022) 0.0023* 0.0025 0.0049*** (0.001) (0.002) (0.002) -0.0008 -0.0031 -0.0041 (0.010) (0.010) (0.011) -0.0043** -0.0078*** -0.0054*** (0.002) (0.003) (0.002) 0.0126 -0.0292 -0.0095 (0.025) (0.026) (0.027) -0.1268* -0.0981 -0.1440* (0.066) (0.093) (0.077) -0.0412 -0.1353 -0.0959 (0.110) (0.140) (0.127) 0.0002 0.0007 -0.0001 (0.001) (0.001) (0.001) -0.2729 -1.1301 0.3561 (1.210) (1.660) (1.461) 0.981

0.983

0.987

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to schooling is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

32

Table A.4: Determinants of return to schooling, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% (1) (2) (3) (4) (5) (6) (7)

q 75% (8)

b.cycle

-0.0515* -0.0416 -0.0697* -0.0417 0.001 0.0006 0.0014* 0.0011* (0.029) (0.028) (0.036) (0.030) (0.001) (0.001) (0.001) (0.001) rel price -0.0774*** -0.0410* -0.0325 -0.0922*** -0.0799*** -0.0437* -0.0357 -0.0932*** (0.021) (0.022) (0.024) (0.024) (0.021) (0.023) (0.025) (0.023) mean exp 0.0064*** 0.0050*** 0.0053*** 0.0078*** 0.0067*** 0.0052*** 0.0057*** 0.0081*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) mean educ -0.0056 0.0003 -0.0032 -0.0048 -0.0053 0.0009 -0.0027 -0.0049 (0.009) (0.010) (0.010) (0.011) (0.009) (0.011) (0.011) (0.011) sd exp -0.001 -0.0004 -0.0060** -0.002 -0.0008 -0.0003 -0.0057** -0.0018 (0.003) (0.002) (0.003) (0.002) (0.003) (0.002) (0.003) (0.002) sd educ -0.0201 0.009 -0.031 -0.0077 -0.0236 0.0063 -0.0358 -0.0107 (0.026) (0.028) (0.028) (0.029) (0.027) (0.030) (0.029) (0.029) union % -0.1678* -0.1859** -0.1214 -0.1434* -0.1800* -0.1960** -0.1379 -0.1530* (0.097) (0.078) (0.102) (0.078) (0.098) (0.082) (0.104) (0.077) male % -0.2148* -0.162 -0.1253 -0.1309 -0.2142* -0.1536 -0.1255 -0.141 (0.120) (0.122) (0.139) (0.140) (0.124) (0.131) (0.147) (0.133) trend -0.0007 -0.0005 0.0001 -0.0009 -0.0008 -0.0006 -0.0001 -0.0009 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) const 1.6184 1.1605 0.1713 1.9186 1.799 1.3266 0.4133 2.0383 (1.563) (1.380) (1.763) (1.524) (1.600) (1.422) (1.816) (1.501)

R2 adj.

0.984

0.972

0.979

0.982

0.984

0.972

0.978

0.982

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to schooling is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables, constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

33

Table A.5: Determinants of the common component, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% (1) (2) (3) (4) (5) (6) (7) b.cycle

1.3406 (1.013) rel price -1.4255*** (0.424) mean exp -0.0023 (0.044) mean educ -0.0559 (0.205) sd exp 0.2155*** (0.062) sd educ 0.9163 (0.619) union % 4.3214** (1.842) male % -4.6608 (4.314) trend -0.0279 (0.018) const 55.2981 (34.855)

R2 adj.

0.958

q 75% (8)

1.2493 2.8358*** 1.8530*** -0.0199 -0.0047 -0.0475** -0.0432*** (1.094) (0.955) (0.652) (0.022) (0.024) (0.022) (0.013) -2.2048*** -1.9686*** -0.6096 -1.3415*** -2.0772*** -1.8097*** -0.5491 (0.631) (0.528) (0.477) (0.437) (0.673) (0.539) (0.411) -0.1106 0.0987** 0.0179 -0.0095 -0.1170* 0.0833* 0.0075 (0.072) (0.047) (0.036) (0.046) (0.068) (0.049) (0.036) -0.6483 0.4186 0.1766 -0.0752 -0.6894* 0.3867 0.1761 (0.392) (0.274) (0.211) (0.212) (0.402) (0.306) (0.216) 0.1490* 0.4016*** 0.2995*** 0.2099*** 0.1451 0.3892*** 0.2903*** (0.084) (0.069) (0.045) (0.070) (0.087) (0.082) (0.047) 1.1802 1.1724 0.1221 1.0032 1.2512 1.3601* 0.2536 (0.924) (0.731) (0.573) (0.644) (0.984) (0.793) (0.590) 6.7468*** -0.1067 0.5341 4.6475** 7.0696*** 0.576 0.9637 (2.153) (1.891) (1.245) (1.966) (2.153) (1.997) (1.160) 4.9726 -8.3909** -5.5288** -4.9054 4.1863 -8.6963* -5.2387* (5.268) (3.888) (2.697) (4.665) (5.410) (4.789) (2.910) 0.015 -0.0804*** -0.0388** -0.0251 0.0187 -0.0749*** -0.0361** (0.029) (0.023) (0.017) (0.020) (0.029) (0.025) (0.017) -26.1908 152.2484*** 75.7800** 50.016 -32.5172 141.6081***70.0584** (53.839) (42.923) (31.784) (38.302) (53.894) (47.245) (30.794) 0.932

0.96

0.972

0.957

0.93

0.957

0.974

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. The common component is the estimated coefficient in equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

34

Table A.6: Determinants of the common component, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% (1) (2) (3) (4) (5) (6) (7) b.cycle

1.1637 (1.545) rel price -1.7419*** (0.591) mean exp -0.0062 (0.046) mean educ -0.0223 (0.354) sd exp 0.2296*** (0.078) sd educ 0.9907 (0.675) union % 3.526 (2.524) male % -1.4203 (4.147) trend -0.0298 (0.026) const 56.768 (47.135)

R2 adj.

0.915

2.3336*** 2.8879*** 1.8740*** (0.829) (0.981) (0.693) -3.1449*** -2.4860*** -1.1739** (0.655) (0.497) (0.437) -0.0086 0.1459*** 0.0589 (0.054) (0.048) (0.038) -0.1284 0.5123* 0.232 (0.317) (0.254) (0.216) 0.3038*** 0.4304*** 0.3312*** (0.071) (0.075) (0.048) 0.723 1.2538* 0.3811 (0.898) (0.665) (0.493) 1.388 0.1706 1.0406 (2.034) (1.975) (1.323) 6.3654* -8.7590** -5.8395** (3.513) (4.053) (2.826) -0.0361 -0.0939*** -0.0517*** (0.023) (0.022) (0.017) 66.667 176.8621***99.3405*** (42.163) (41.763) (31.547) 0.944

0.967

0.98

-0.0182 (0.027) -1.6721*** (0.601) -0.0125 (0.046) -0.0375 (0.344) 0.2246** (0.084) 1.0668 (0.713) 3.8079 (2.593) -1.5986 (4.392) -0.0275 (0.026) 52.2685 (47.563) 0.914

q 75% (8)

-0.0400** -0.0485** -0.0392*** (0.016) (0.023) (0.013) -3.0174*** -2.3247*** -1.0968*** (0.663) (0.492) (0.392) -0.0213 0.1302** 0.0485 (0.053) (0.050) (0.038) -0.1531 0.48 0.224 (0.332) (0.291) (0.223) 0.2935*** 0.4178*** 0.3223*** (0.078) (0.087) (0.053) 0.8782 1.4451* 0.5108 (0.964) (0.727) (0.521) 1.9485 0.8656 1.481 (2.067) (2.051) (1.291) 6.1512 -9.0646* -5.7261* (3.752) (5.014) (3.182) -0.0316 -0.0883*** -0.0487*** (0.024) (0.025) (0.017) 58.0042 166.0399***93.1017*** (44.175) (46.509) (32.023) 0.941

0.963

0.98

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. The common component is the estimated coefficient in equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables, constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

35

Table B.1: Determinants of return to experience, experience=age, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

-0.0361*** (0.008) rel price 0.0226*** (0.006) mean exp -0.0036*** (0.000) mean educ 0.0006 (0.002) sd exp -0.0027*** (0.001) sd educ -0.0058 (0.006) union % -0.0038 (0.020) male % 0.0699** (0.034) trend 0.0006*** (0.000) const -1.1398*** (0.317)

R2 adj.

0.969

-0.0304*** (0.007) 0.0223*** (0.007) -0.0033*** (0.000) 0.001 (0.002) -0.0028*** (0.001) -0.0100* (0.006) -0.0225 (0.020) 0.0762** (0.035) 0.0006*** (0.000) -1.0146*** (0.368)

-0.0385*** (0.009) 0.0315*** (0.007) -0.0042*** (0.000) 0.0002 (0.002) -0.0032*** (0.001) -0.0066 (0.006) 0.0051 (0.019) 0.0456 (0.036) 0.0008*** (0.000) -1.4534*** (0.365)

-0.0327*** (0.010) 0.0215*** (0.007) -0.0033*** (0.000) 0.0007 (0.002) -0.0035*** (0.001) -0.0140** (0.006) -0.0129 (0.023) 0.0744** (0.030) 0.0006*** (0.000) -1.0018** (0.379)

0.0007*** (0.000) 0.0212*** (0.006) -0.0035*** (0.000) 0.0006 (0.002) -0.0025*** (0.001) -0.0084 (0.007) -0.0117 (0.020) 0.0703* (0.041) 0.0006*** (0.000) -1.0331*** (0.343)

0.0007*** (0.000) 0.0213*** (0.007) -0.0031*** (0.000) 0.0009 (0.002) -0.0027*** (0.001) -0.0122* (0.006) -0.0292 (0.020) 0.0747* (0.039) 0.0005*** (0.000) -0.9255** (0.360)

0.0007*** (0.000) 0.0298*** (0.006) -0.0041*** (0.001) 0.0003 (0.003) -0.0031*** (0.001) -0.0093 (0.007) -0.0033 (0.020) 0.0482 (0.046) 0.0008*** (0.000) -1.3385*** (0.409)

0.0007*** (0.000) 0.0205*** (0.007) -0.0031*** (0.000) 0.0006 (0.003) -0.0034*** (0.001) -0.0164** (0.007) -0.0202 (0.023) 0.0718** (0.034) 0.0005** (0.000) -0.9063** (0.364)

0.964

0.972

0.967

0.968

0.965

0.97

0.969

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to experience=age is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

36

Table B.2: Determinants of return to experience, experience=age, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

-0.0353*** (0.007) rel price 0.0145** (0.007) mean exp -0.0021*** (0.000) mean educ -0.0003 (0.002) sd exp -0.0021*** (0.001) sd educ -0.0098* (0.005) union % -0.0041 (0.017) male % 0.0840*** (0.028) trend 0.0005*** (0.000) const -0.9712*** (0.299)

R2 adj.

0.93

-0.0315*** (0.007) 0.0179** (0.007) -0.0018*** (0.000) 0.0006 (0.002) -0.0023*** (0.001) -0.0118** (0.006) -0.0229 (0.019) 0.0904*** (0.031) 0.0005** (0.000) -0.9193** (0.347)

-0.0379*** (0.008) 0.0272*** (0.005) -0.0025*** (0.000) 0.0019 (0.002) -0.0029*** (0.001) -0.009 (0.006) 0.0018 (0.015) 0.0593** (0.028) 0.0006*** (0.000) -1.1240*** (0.316)

-0.0302*** (0.009) 0.0169*** (0.006) -0.0017*** (0.000) 0.0019 (0.002) -0.0031*** (0.001) -0.0161** (0.006) -0.0146 (0.022) 0.0964*** (0.029) 0.0004** (0.000) -0.7850** (0.387)

0.0007*** (0.000) 0.0133** (0.006) -0.0019*** (0.000) -0.0003 (0.002) -0.0020*** (0.001) -0.0124** (0.006) -0.0119 (0.017) 0.0830** (0.034) 0.0005*** (0.000) -0.8673*** (0.309)

0.0007*** (0.000) 0.0168** (0.007) -0.0016*** (0.000) 0.0005 (0.002) -0.0022*** (0.001) -0.0141** (0.006) -0.0298 (0.019) 0.0890** (0.035) 0.0005** (0.000) -0.8268** (0.340)

0.0007*** (0.000) 0.0255*** (0.005) -0.0023*** (0.000) 0.002 (0.002) -0.0028*** (0.001) -0.0117 (0.007) -0.0064 (0.017) 0.0624* (0.037) 0.0006*** (0.000) -1.0108*** (0.357)

0.0006*** (0.000) 0.0159*** (0.006) -0.0016*** (0.000) 0.0018 (0.002) -0.0030*** (0.001) -0.0183** (0.007) -0.0213 (0.021) 0.0948*** (0.030) 0.0004* (0.000) -0.6964* (0.371)

0.915

0.949

0.911

0.932

0.918

0.945

0.919

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to experience=age is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables, constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

37

Table B.3: Determinants of return to schooling, experience=age, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

-0.0111 0.0023 -0.017 -0.0151 0.0001 -0.0003 0.0002 0.0003 (0.025) (0.021) (0.027) (0.025) (0.001) (0.000) (0.001) (0.001) rel price -0.0505*** -0.0385*** -0.019 -0.0545** -0.0513*** -0.0393*** -0.0201 -0.0552** (0.015) (0.013) (0.018) (0.021) (0.016) (0.013) (0.019) (0.022) mean exp 0.0064*** 0.0035*** 0.0055*** 0.0070*** 0.0065*** 0.0035*** 0.0055*** 0.0070*** (0.001) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.002) mean educ -0.0142* -0.0053 -0.0009 -0.0083 -0.0141* -0.0049 -0.0007 -0.0083 (0.007) (0.007) (0.009) (0.011) (0.008) (0.008) (0.009) (0.011) sd exp -0.0035* -0.0028* -0.0069*** -0.0028 -0.0035* -0.0029* -0.0068*** -0.0028 (0.002) (0.001) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) sd educ -0.0164 0.0231 -0.0159 -0.0024 -0.0171 0.0235 -0.017 -0.0035 (0.022) (0.021) (0.023) (0.026) (0.021) (0.022) (0.023) (0.026) union % -0.0997 -0.0609 -0.0612 -0.0903 -0.1019 -0.0599 -0.0647 -0.0936 (0.068) (0.053) (0.075) (0.076) (0.071) (0.056) (0.079) (0.077) male % -0.1714* -0.0369 -0.087 -0.1436 -0.1655* -0.0243 -0.0796 -0.1422 (0.091) (0.078) (0.096) (0.105) (0.094) (0.082) (0.101) (0.106) trend 0.0001 0.0003 0.0002 -0.0002 0.0001 0.0002 0.0001 -0.0002 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) const 0.0627 -0.4809 -0.2474 0.4701 0.098 -0.4826 -0.194 0.5152 (1.209) (0.898) (1.263) (1.321) (1.282) (0.943) (1.338) (1.344)

R2 adj.

0.988

0.985

0.988

0.989

0.988

0.985

0.987

0.989

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to schooling is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

38

Table B.4: Determinants of return to schooling, experience=age, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

-0.0202 -0.0169 -0.017 -0.0151 0.0003 0 0.0002 0.0003 (0.027) (0.026) (0.027) (0.025) (0.001) (0.001) (0.001) (0.001) rel price -0.0828*** -0.0520*** -0.019 -0.0545** -0.0839*** -0.0537*** -0.0201 -0.0552** (0.016) (0.016) (0.018) (0.021) (0.016) (0.017) (0.019) (0.022) mean exp 0.0077*** 0.0057*** 0.0055*** 0.0070*** 0.0078*** 0.0058*** 0.0055*** 0.0070*** (0.002) (0.002) (0.001) (0.002) (0.002) (0.002) (0.001) (0.002) mean educ -0.0132* -0.0053 -0.0009 -0.0083 -0.0130* -0.0049 -0.0007 -0.0083 (0.007) (0.008) (0.009) (0.011) (0.007) (0.008) (0.009) (0.011) sd exp -0.001 0.0003 -0.0069*** -0.0028 -0.001 0.0003 -0.0068*** -0.0028 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) sd educ -0.0137 0.0175 -0.0159 -0.0024 -0.0151 0.0165 -0.017 -0.0035 (0.022) (0.026) (0.023) (0.026) (0.022) (0.027) (0.023) (0.026) union % -0.1255 -0.1300* -0.0612 -0.0903 -0.1297 -0.1331* -0.0647 -0.0936 (0.077) (0.068) (0.075) (0.076) (0.080) (0.073) (0.079) (0.077) male % -0.2062** -0.1656 -0.087 -0.1436 -0.2014** -0.1507 -0.0796 -0.1422 (0.097) (0.102) (0.096) (0.105) (0.098) (0.110) (0.101) (0.106) trend -0.0007 -0.0006 0.0002 -0.0002 -0.0007 -0.0006 0.0001 -0.0002 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) const 1.5108 1.1655 -0.2474 0.4701 1.5723 1.2216 -0.194 0.5152 (1.304) (1.136) (1.263) (1.321) (1.365) (1.196) (1.338) (1.344)

R2 adj.

0.986

0.975

0.984

0.985

0.986

0.974

0.983

0.985

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. Return to schooling is estimated from equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables, constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

39

Table B.5: Determinants of the common component, experience=age, BLS weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

1.4015 (1.509) rel price -1.8780** (0.773) mean exp -0.1068* (0.063) mean educ -0.3883 (0.301) sd exp 0.1441 (0.088) sd educ 2.2013*** (0.726) union % 7.6853** (3.270) male % -10.6076** (4.225) trend 0.0272 (0.029) const -46.5574 (53.766)

R2 adj.

0.949

2.0310** (0.930) -2.7032** (1.003) -0.0957 (0.076) -0.3769 (0.362) 0.3099*** (0.108) 2.3537** (0.994) 4.3896 (3.453) -7.8776 (5.270) 0.0061 (0.034) -8.3922 (62.685) 0.905

3.3108*** 2.3485** -0.0317 (1.104) (1.053) (0.029) -1.8815** -0.5609 -1.8367** (0.738) (0.808) (0.755) 0.081 -0.0023 -0.1143* (0.065) (0.074) (0.062) 0.1964 0.1425 -0.3825 (0.245) (0.257) (0.294) 0.4058*** 0.2785*** 0.1375 (0.085) (0.091) (0.090) 2.1031*** 1.3894** 2.3046*** (0.615) (0.615) (0.764) 1.6884 3.7396 7.9993** (3.212) (3.538) (3.319) -18.5280***-14.2377*** -10.4578** (4.020) (4.248) (4.425) -0.0487 -0.008 0.0292 (0.029) (0.031) (0.028) 92.9154* 15.8558 -50.6359 (54.888) (58.041) (53.329) 0.932

0.918

0.949

-0.0484** -0.0621** -0.0568** (0.021) (0.026) (0.025) -2.6515*** -1.7395** -0.5044 (0.946) (0.733) (0.757) -0.1069 0.0647 -0.0154 (0.072) (0.064) (0.070) -0.365 0.1913 0.1576 (0.361) (0.274) (0.259) 0.2996*** 0.3940*** 0.2663*** (0.105) (0.094) (0.084) 2.5052** 2.3371*** 1.5653** (1.022) (0.673) (0.633) 4.8494 2.4044 4.2731 (3.275) (3.253) (3.420) -7.5474 -18.7908***-13.8114*** (5.042) (4.872) (4.210) 0.0089 -0.0436 -0.0048 (0.032) (0.031) (0.030) -14.2556 83.0266 9.0939 (59.412) (57.440) (56.154) 0.907

0.929

0.922

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. The common component is the estimated coefficient in equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

40

Table B.6: Determinants of the common component, experience=age, constructed weights output unemployment rate OLS with OLS with spouse spouse OLS controls q 50% q 75% OLS controls q 50% q 75% (1) (2) (3) (4) (5) (6) (7) (8) b.cycle

2.1347* 0.3523 3.3108*** 2.3485** -0.0425 -0.0059 -0.0621** -0.0568** (1.175) (1.428) (1.104) (1.053) (0.029) (0.029) (0.026) (0.025) rel price -0.4977 -1.4971 -1.8815** -0.5609 -0.4147 -1.4796 -1.7395** -0.5044 (0.857) (0.921) (0.738) (0.808) (0.850) (0.918) (0.733) (0.757) mean exp -0.1044* -0.2689*** 0.081 -0.0023 -0.1152* -0.2705*** 0.0647 -0.0154 (0.061) (0.093) (0.065) (0.074) (0.061) (0.092) (0.064) (0.070) mean educ -0.0921 -0.8633* 0.1964 0.1425 -0.0919 -0.8648* 0.1913 0.1576 (0.250) (0.456) (0.245) (0.257) (0.253) (0.455) (0.274) (0.259) sd exp 0.1540* 0.0524 0.4058*** 0.2785*** 0.1457* 0.0514 0.3940*** 0.2663*** (0.081) (0.119) (0.085) (0.091) (0.085) (0.119) (0.094) (0.084) sd educ 2.3885*** 2.8465** 2.1031*** 1.3894** 2.5413*** 2.8709** 2.3371*** 1.5653** (0.732) (1.128) (0.615) (0.615) (0.760) (1.172) (0.673) (0.633) union % 6.2745* 10.7822*** 1.6884 3.7396 6.7411** 10.8571*** 2.4044 4.2731 (3.222) (2.895) (3.212) (3.538) (3.302) (2.937) (3.253) (3.420) male % -13.7366*** -4.1732 -18.5280***-14.2377***-13.7883*** -4.234 -18.7908***-13.8114*** (4.393) (7.526) (4.020) (4.248) (4.789) (7.436) (4.872) (4.210) trend 0.0252 0.0870** -0.0487 -0.008 0.0284 0.0875** -0.0436 -0.0048 (0.027) (0.040) (0.029) (0.031) (0.027) (0.039) (0.031) (0.030) const -45.5602 -158.5757** 92.9154* 15.8558 -51.8875 -159.6415** 83.0266 9.0939 (50.226) (73.485) (54.888) (58.041) (51.033) (72.336) (57.440) (56.154)

R2 adj.

0.952

0.891

0.932

0.918

0.952

0.891

0.929

0.922

Note: This table reports estimates of equation (2) with Newey-West standard errors (lag 2) in parentheses. R2 adj. is obtained from similar OLS specifications. The common component is the estimated coefficient in equation (1), for each year in the 1961-2013 CPS using OLS, OLS with spousal controls and quantile regression. The variables, constructed weights and data sources are defined in the text. Statistical significance is denoted as ∗ 10%, ∗∗ 5%, and ∗∗∗ 1% levels.

41