Life-Cycle Wage Growth Across Countries David Lagakos UCSD Tommaso Porzio Yale Benjamin Moll Princeton Nancy Qian Yale Todd Schoellman ASU Northwestern, 18 April 2016 1
Life-Cycle Human Capital Accumulation Across Countries Lessons from U.S. Immigrants David Lagakos UCSD Tommaso Porzio Yale Benjamin Moll Princeton Nancy Qian Yale Todd Schoellman ASU Northwestern, 18 April 2016 2
What We Do Document new fact: experience-wage profiles in rich countries are steeper than in poor countries twice as steep wages double in rich countries, increase by 50% in poor countries based on representative large-sample micro data from 17 countries better data than previous studies 3
Why Care? How life-cycle wage growth differs across countries may help us understand cross-country income differences Key for evaluating importance of cross-country differences in human capital accumulation Manuelli-Seshadri, Klenow-RodriguezClare, Bils-Klenow, Caselli,... labor market frictions (job ladder) Burdett, Burdett-Mortensen, Jovanovic,... Hope: use profiles to discipline theories, available from my website Illustration of finding s quantitative bite: development accounting how much of income differences due to K and H? current consensus: K&H account for 40%, TFP for 60% same exercise but assuming profiles reflect life-cycle H : increases contribution of K&H from 40% to 60% 4
So what s the mechanism? Why are profiles flatter in poor countries? human capital accumulation labor market frictions (job ladder)... Provide two pieces of (tentative) evidence: 1. from same data: additional moments (variance profiles etc) 2. from alternative data: wage profiles of U.S. immigrants These point to theories of human capital accumulation 5
Data 6
Data Nationally representative surveys with detailed wage and hours data: Australia, Bangladesh, Brazil, Canada, Chile, France, Germany, Guatemala, India, Indonesia, Jamaica, Mexico, Peru, South Korea, United Kingdom, United States, Uruguay, Vietnam Focus on core set of 8 countries with repeated cross-sections spanning 15+ years Limitation: very poorest countries not in sample. 7
Sample Focus on full time male wage earners Income of self-employed is payment to labor income and capital income (Gollin, 2002); host of other measurement issues (Deaton, 1997); potential experience harder to interpret for female and part-time workers Wage = labor earnings hours Majority of countries: earnings last month & hours last week Later look at females, part time, self employed 8
Potential Experience Measure lifecycle using potential experience Definition age schooling 6, if schooling 12 experience := age 18, if schooling < 12 That is, years since turned 18 or finished school Keep individuals with 0 experience 40 9
Lifecycle Wage Growth 10
Simplest Measure of Lifecycle Wage Growth Group workers into 5-year experience bins (0-4, 5-9, etc) Compute average wages by bin relative to 0-4 bin Report simple averages across years of data 11
Core Countries Percent Wage Increase Relative to Experience <5 Years 0 50 100 150 Germany United States Canada United Kingdom 0 50 100 150 Brazil Chile Mexico Jamaica Potential Experience Potential Experience data available from http://www.princeton.edu/~moll/research.htm 12
All Countries Percent Wage Increase 0 50 100 150 Australia Germany United States France Canada 0 50 100 150 United Kingdom Uruguay Chile S. Korea Percent Wage Increase 0 50 100 150 Indonesia Brazil Peru Mexico 0 50 100 150 India Jamaica Bangladesh Guatemala Vietnam Potential Experience Potential Experience data available from http://www.princeton.edu/~moll/research.htm 13
Challenges with Simplest Measure No controls for schooling Age-cohort-time identification problem 14
Mincerian Measure of Lifecycle Wage Growth Consider individual i in cohort c at time t Estimate equations of the form: log w ict = α + g(s ict ) + f (x ict ) + γ t + ψ c + ε ict w ict : wages s ict : schooling; x ict : experience. γ t : time effect, ψ c : cohort effect Goal: estimate f ( ) and assess how it varies across countries 15
Mincerian Measure of Lifecycle Wage Growth Assume g(s) = θs, but fully flexible f ( ) log w ict = α + θs ict + x X ϕ x D x ict + γ t + ψ c + ε ict where Dict x is a dummy for experience group x X = {5-9,10-14,...} Pointwise identification of f (x) via the {ϕ x } Cannot estimate as is, due to well-known collinearity problem 16
Mincerian Measure of Lifecycle Wage Growth 1. Time/cohort controls a la Hall (1968), Deaton (1997) Focus on core countries, which have repeated cross sections spanning 15+ years Assume that all growth is due either to time or cohort effects 2. New approach based on Heckman, Lochner and Taber (1998) Assume no wage gains due to experience in final working years Consistent with models of lifecycle H accumulation or search 17
Deaton-Hall Profiles: All Growth Due to Time Percent Wage Increase Relative to Experience <5 Years 0 50 100 150 Germany Canada United Kingdom United States 0 50 100 150 Brazil Chile Jamaica Mexico 18
Deaton-Hall Profiles: All Growth Due to Cohort Percent Wage Increase Relative to Experience <5 Years 0 50 100 150 200 250 United Kingdom Germany Canada United States 0 50 100 150 200 250 Brazil Jamaica Chile Mexico 19
Limitations of Deaton-Hall Approach Just guessing about relative roles of cohort and time Same roles of cohort and time in all countries? Hard to imagine world without strong time effects 20
Heckman-Lochner-Taber (HLT) Approach Assume no wage gains due to experience in last working years (e.g. 35-40 or 30-40 years of potential experience) With this assumption, and using repeated cross sections, can identify experience effects from cohort and time Intuition: follow different cohorts over time; wage growth from years 1999 to 2000 identified from oldest cohort s wage growth 21
Heckman-Lochner-Taber (HLT) Profiles Percent Wage Increase Relative to Experience <5 Years 0 50 100 150 Germany Canada United States United Kingdom 0 50 100 150 Brazil Mexico Jamaica Chile 22
HLT Profiles: Robustness to Age Heaping Percentage Wage Increase 0 50 100 150 Germany Canada United States United Kingdom 0 50 100 150 Brazil Mexico Chile Jamaica Potential Experience Potential Experience 23
HLT Profiles: Robustness to Education Measurement Percentage Wage Increase 0 50 100 150 Germany Canada United States United Kingdom 0 50 100 150 Brazil Mexico Jamaica Chile Potential Experience Potential Experience 24
Selection? Concern: in rich countries, less productive workers select out of wage employment as they age and/or...... in poor countries, less productive workers select into wage-employment as they age Examine using panel data from Mexico and U.S. (FLS and PSID) (a) panel data (b) cross section (from Fig 1) Percentage Wage Increase 0 50 100 150 Percentage Wage Increase 0 50 100 150 0 10 20 30 40 Potential Experience United States Mexico 0 10 20 30 40 Potential Experience United States Mexico 25
Alternative Sample Restrictions & Experience Def. Table 5: Robustness Height at 20-24 Years Experience, HLT Profiles Rich Poor Rich - Poor (1) Baseline 79.3 39.2 40.1** (2) Experience at 16 82.1 45.8 36.2** (3) Constructed experience 90 43.5 46.6** (4) Measurement error: age 76.5 39.2 37.3** (5) Measurement error: education 71.7 39.2 32.5** (6) Measurement error: age and education 71.2 39.2 32.0** (7) Include Self-Employed 80.3 36.6 43.6** (8) Include Public-Sector Employees 80.4 42.2 38.2** (9) Include women 70 29.1 41** (10) Constructed experience, men and women 76.6 25.5 51.1** (11) Include Part-Time (20+ hours) 83 38.2 44.8** (12) Include Part-Time (> 0 hours) 84.8 36.7 48.1** (13) Constructed experience, incl. Part-Time 100 42 58.0** 26
Lifecycle Wage Growth Across Countries Punchline: less lifecycle wage growth in poor countries Results present multiple assumptions about role of cohort and time, numerous alternative sample restrictions Some modest role for interactions between schooling and experience 27
Interactions Between Schooling and Experience 28
Experience-Wage Profiles by Education Level Percent Wage Increase Relative to Experience <5 Years 0 50 100 150 0 50 100 150 0 50 100 150 United States Germany Canada 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 United Kingdom Chile Brazil 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 Mexico Jamaica College High School Less than H.S. 0 10 20 30 40 0 10 20 30 40 29
Accounting for Aggregate Experience-Wage Profiles Counterfactual Average Return 0 50 100 100% 50% 25% 0% GTM BGD VNM JAM MEX CHL PER KOR BRA FRAURY USA CAN GBR IDN AUS DEU 0 50 100 Actual Average Return 30
Distinguishing Between Mechanisms (new!) 31
Potential Mechanisms 1. human capital accumulation 2. search and matching/job ladder 3. long-term contracts with w MP L 4. what else? Large literature studies rel. importance of 1 to 3 in U.S./rich countries Topel-Ward, Rubinstein-Weiss, Altonji-Smith-Vidangos, Bagger-Fontaine-PostelVinay-Robin,... ASV 32
Moments we would like to look at search and matching/job ladder data on job-to-job transitions long-term contracts tenure profiles problem: both require panel data (or matched employer-employee data) which we don t have 33
Moments we can look at profiles for particular groups of workers workers with short-term contracts long-term contracts?... hours and earnings profiles human capital, long-term contracts variance profiles human capital 34
Workers with Short-Term Contracts Long-term contracts flatter profiles in poor countries if w MP L and wages front-loaded in poor countries w MP L and wages back-loaded in rich countries a priori reason to be skeptical: median tenure in U.S. = 4.6 years (BLS) Nevertheless went through survey codebooks to identify workers for which long-term contracts, tenure concerns seem unlikely 35
Workers with Short-Term Contracts (a) India (b) Mexico Percentage Wage Increase 0 50 100 150 Percentage Wage Increase 0 50 100 150 Potential Experience Potential Experience Daily Workers Long Term Workers Daily Workers Long Term Workers (c) United States Percentage Wage Increase 0 50 100 150 Potential Experience Marginally Attached Baseline 36
Hours, Earnings and Variance Profiles Two predictions of simple human capital theories (Ben-Porath,...): 1. time investment into H declines over life-cycle if hours worked reflect time not investing steep hours profiles in rich countries flat hours profiles in poor countries 2. V ar(log ear ni ngs) are U-shaped Mincer, Polachek, Rubinstein-Weiss individuals differ in learning ability steep profiles start below flat ones and cross ( overtaking age ) 37
Lifecyle Hours Profiles Percent Hours Increase 0 25 50 75 100 Canada Australia United States Germany France 0 25 50 75 100 United Kingdom Uruguay S. Korea Chile Percent Hours Increase 0 25 50 75 100 Mexico Indonesia Brazil Peru 0 25 50 75 100 Vietnam India Jamaica Guatemala Bangladesh Potential Experience Potential Experience 38
Lifecyle Earnings Profiles Percent Earnings Increase 0 50 100 150 200 250 Germany United States Canada Australia France 0 50 100 150 200 250 S. Korea Uruguay United Kingdom Chile Percent Earnings Increase 0 50 100 150 200 250 Brazil Indonesia Peru Mexico 0 50 100 150 200 250 Guatemala India Jamaica Bangladesh Vietnam Potential Experience Potential Experience 39
Lifecyle Variance Profiles (within education groups) Variance of Log Earnings.25.5.75 1 1.25 Australia United States Canada Germany France.25.5.75 1 1.25 Chile Uruguay United Kingdom S. Korea Variance of Log Earnings.25.5.75 1 1.25 Peru Brazil Indonesia Mexico.25.5.75 1 1.25 Bangladesh Jamaica India Vietnam Guatemala Potential Experience Potential Experience 40
Summary Additional moments from our data not supportive of long-term contracts consistent with human capital theories, not definitive inconclusive about search and matching/job ladder Next: bring another dataset to the table 41
Lessons from U.S. Immigrants 42
Returns to Experience for Immigrants Study returns to experience for immigrants in the U.S. foreign experience, but also U.S.-acquired experience Advantages: common labor market, institutions, data set Challenges: immigrants may be selected, suffer skill loss 43
Three Main Findings 1. Return to foreign experience is much lower for poor country immigrants, similar to that for non-migrants 2. Return to U.S. experience is modestly lower 3. Return to U.S. experience for U.S.-educated immigrants is independent of birth country 44
Interpretation of Findings Evidence leads us to a human capital interpretation: Less human capital formation through experience in poor countries Part of this effect is explained by the work environment Part of this effect stems from school type/quality 45
Data Data: 1980 2000 Census, 2005 12 ACS Immigrant: born outside the fifty states Restrictions: employed wage worker, 0 45 years experience Positive income, valid responses to other key variables Nice feature: extremely large sample 1.6 million immigrants, 120 birth countries 102 countries with 1000+; 29 with 10, 000+ Wide variety of controls 46
Fact 1: Returns Similar for Immigrants, Non-Migrants Height at 20-24 Years of Experience 0 50 100 BGD BGD VNM VNM IDN IDN BRA PER BRA PER JAM GTM JAM GTM CHL MEX MEX CHL GER GER AUS FRA GBR GBR KOR AUS CAN CAN FRA KOR 1000 10000 100000 PPP GDP p.c., 2010 Non-Migrants Immigrants 47
Fact 1: Returns Similar for Immigrants, Non-Migrants Immigrants 0 50 100 IDN KOR PER CHL MEX JAM VNM BGD GTM BRA CAN GBR AUS FRA GER 0 50 100 Non-Migrants 48
Implication of Fact 1 Simplest explanation: Less lifecycle human capital accumulated in poor countries. Alternative explanation: Non-migrant returns are biased Labor market frictions, implicit contracts, measurement error Returns for immigrants biased Selection, skill transferability These biases affect only poor countries, negatively, by same magnitude 49
Fact 2: No Relation Between Income, Skill Transfer US Immigrants 0 20 40 60 80 100 Percent of Workers in High-Skill Occupations College Graduates TZA KHM ZAF IND CAN FRA IRL ESP MYS CMR IRN GRC JAMITA NPL SLE TUR UGA VEN NLD CHL ROM CHE SEN PRI GHA BRA EGY JOR IRQ MAR URY PRT FJI AUT PHL PAN CRI HUN PER HTI THA BOL MEX ECU NIC CUB SLV GIN 0 20 40 60 80 100 Non-Migrants 50
Fact 2: No Relation Between Income, Skill Transfer Percent of Workers in High-Skill Occupations Ratio of Immigrants to Non-Migrants, College Grads Only Ratio of Immigrants to Non-Migrants 0.5 1 1.5 2 GIN TZA SLENPL CMR KHM UGA SEN HTI GHA NIC IND IRQ FRACAN PHL ZAF TUR IRN MYS GRC ESP IRL MAR PER BRA JAM VEN PRI FJIEGY PAN URY CHL ITA NLD JOR THAROM BOL ECU CRI MEX PRT CHE HUN AUT SLV CUB 500 5000 50000 Ln(GDP per Capita) 51
Fact 3: Schooling Selection Declines in Income Years of Schooling, Immigrants 4 6 8 10 12 14 YEM IND EGY CMR KEN KWT SGP TWN UGA TZA ZWE ZAF IDN SAU BGR DZA CHE AUS FIN FRA LKA IRN BEL BGD GBR HKG JPN NPL VEN SWE NZL PAK DNK GHA ISRCZE LBR MAR PHL MYS LVA BRA BOL ARGAUT GER KOR NLD SDN SLE JOR CAN CHL LTU MMR PAN NOR THA TUR MDA UKR ESPCYP ARM IRL HUN AFG PRY PER SVK SEN TTO JAM BRB SYR URY COL CHN GUY BLZ POL ALB CRIHRV HTI TON NIC FJI VNM IRQ ECU GRC CUB ITA DOM HND GTM KHM PRT SLV LAO MEX USA 4 6 8 10 12 14 Years of Schooling, Non-Migrants 52
Development Accounting 53
Development Accounting So far, new fact: experience-wage profiles flatter in poor countries than rich countries Now: development accounting exercise same as previous literature... except returns to experience vary across countries Conclusion: Importance of H now 60%, rather than 40% 54
Development Accounting Use same accounting method as Caselli (2005). Real GDP in a country Assume α = 1/3. Y = K α (AH) 1 α Re-construct Caselli s success 1 measure: Y KH = K α H 1 α success 1 = var(ln Y KH) var(ln Y ) 55
Development Accounting Human Capital Measure Schooling Experience Schooling + Experience Success 1 0.40 0.40 0.63 Slope(log(Y KH ),log(gdp)) 0.53 0.56 0.65 Cohort & Time Effects 56
Conclusion Less lifecycle wage growth in poor countries Some evidence in favor of human capital explanation Through lens of development accounting framework: H and K account for 60% of income differences, not 40% Priority for future work: panel data for poor countries 57
Altonji-Smith-Vidangos using PSID back FIGURE 1. Decomposing the experience profile of wages. Baseline model, full SRC sample. The figure displays the model s decomposition of wage growth over a career (or the experience profile of log wages) into the contributions of job shopping (the mean value of the job-specific wage component ν), the accumulation of tenure (the contribution of the mean value of tenure on the wage experience profile), and the accumulation of general human capital. 58