Get Complete Project Material File(s) Now! »
Transatlantic Employment Outcomes
Over the past four decades, the French economy has been characterized by lower employment per-formances than in the U.S. Economists argue that these diﬀerences in employment highlight possible malfunctioning of the French labour market. In that respect, Piketty (1998) proposes a long-run structural analysis to identify which sectors lack employment in France with respect to the U.S. Despite the fact that both countries experienced a decline in agricultural and manufacturing em-ployment, he argues that France has not developed its service sector to the same extent as the U.S. The high labor cost appears to have obstructed the reallocation of labor across sectors.
Despite those diﬀerences, both countries experienced similar changes in their occupational structure. They both underwent job polarization as documented by Autor, Katz, and Kearney (2006) and Goos, Manning, and Salomons (2009). The occupational employment structure shifted from middle-paid jobs that perform routine tasks towards high- and low-paid jobs that perform abstract and manual tasks, respectively. This process reflects first and foremost the eﬀects of technological change and globalization. Nevertheless, the employment gains and losses arising from this reallocation process appear to occur at diﬀerent times for each country. As a consequence, omitting the occupational aspect of labor reallocation provides only a partial understanding of the transatlantic employment gap.
This chapter reassesses the long-run structural analysis initiated by Piketty (1998) by investigating the extent to which occupational and socio-demographic changes shape the transatlantic employment gap. I show that the process of job polarization is at the core of the employment dynamics in both countries and that specific socio-demographic groups account for the relative scarcity of French employment. Hence, the transatlantic employment gap does not solely reflect a disfunctioning labor market. It also reflects the occupational reallocation of labor that has been ongoing over the last four decades and which aﬀects the employment prospects and participation decisions of specific socio-demographic groups.
I first proceed by constructing long period time series for both France and the U.S. between 1982 and 2017 based on the French Labor Force and the U.S. Current Population Surveys. Those time se-ries include information on aggregate employment, occupational employment, the socio-demographic composition and propensities, as well as worker transitions across labor market states for both coun-tries. Two main challenges encountered are the treatment of cross-country and time inconsistencies inherent to survey data. First, surveys are subject to important redesigns over the 1982-2017 period bringing significant discontinuities in the resulting time series. Second, countries collect survey data following diﬀerent rules and methodologies making cross-country comparisons even more challenging. I tackle data discontinuities by relying on correction procedures, and cross-country inconsistencies by relying on crosswalks and finding compatibilities in variable definitions. The resulting time series are used to produce three sets of exercises describing thoroughly transat-lantic employment outcomes in light of occupational and socio-demographic trends. The first set of results describes the transatlantic employment gap. I identify whether it is accounted for by cross-country discrepancies in employment propensities or in socio-demographic compositions, which then allows to determine the occupational and socio-demographic content of the French employment deficit. In that sense, this comparative analysis builds on Piketty (1998), Cahuc and Debonneuil (2004), Passet (2015) and Catherine, Landier, and Thesmar (2015) who all have conducted sim-ilar counterfactual analyses to document the relative scarcity of French employment. This study distinguishes itself by disaggregating the transatlantic employment gap into occupational and socio-demographic groups shedding new implications in terms of economic policies. It also closely relates to Cohen, Lefranc, and Saint-paul (1997) who compare the French and U.S. labor markets. Never-theless, the authors focus on unemployment as they consider that it is a relevant indicator to capture labor market ineﬃciencies. In contrast, I focus on occupational employment since the transatlantic employment gap is accounted for by both unemployment and non-participation discrepancies. Furthermore, while employment dynamics reflect unemployment changes in France, they appear to reflect non-participation in the U.S. Thus, unemployment provides only a fragmented picture of the labor market that fails to capture the deterioration of U.S. labor market outcomes. Nevertheless, the finding that a large part of the transatlantic employment gap originates from diﬀerences in the tim-ing of employment gain and losses resonates with the authors’ point that unemployment diﬀerences do not originate from the behavior of the unemployed or the institutions underlying their decisions.
The second set of results describes the employment dynamics of each country over time. I determine the extent to which the transatlantic employment gap is determined by the improvement or the dete-rioration of employment outcomes in each country. I quantify the extent to which those changes are due to compositional shifts or to changes in employment propensities. I also investigate the aggregate relevance of socio-demographic groups in order to identify those in the midst of this occupational reallocation process. In that respect, I rely on a decomposition method as in Cortes, Jaimovich, and Siu (2017). The latter paper focuses on routine employment changes in the U.S. The authors provide a precise description as well as explanations for those trends through a standard neoclassical model of the labor market. In that spirit, Albertini, Hairault, Langot, and Sopraseuth (2017) provide a search and matching model to quantify the employment gains and losses arising from task-biased technological change, labor market institutions, and the rising educational attainment in France, Germany and the U.S. In contrast, I do not determine the causes of those employment gains and losses. I provide an empirical assessment of transatlantic employment performances by determining the occupational and socio-demographic content of transatlantic employment discrepancies.
The final set of results analyses transatlantic employment dynamics from a transitional perspective. It relies on annual transition rate data and a series of counterfactual experiments to identify the key transitions accounting for the occupational employment dynamics. I determine whether job polarization arises from occupational mobility or non-employment adjustments in the long run. Cortes, Jaimovich, Nekarda, and Siu (2014) also study worker flows and occupational employment but they mainly focus on the disappearance of routine jobs in the U.S. They also quantify the extent to which demographic factors explain changes in key transition rates accounting for the decline in routine employment. Charlot, Fontaine, and Sopraseuth (2019) provide a more thorough comparative study on worker flows in France and the U.S. However, they focus on quantifying the ontribution of labor market transitions to unemployment fluctuations, as well as the role played by job polarization in aﬀecting labor market dualism. In contrast, I focus on long-run occupational employment trends. I determine whether workers adjust to those changes through occupational mobility or non-employment transitions.
Those findings have several implications in terms of labor market policies. First, low-skilled em-ployment should be further supported as France started since the mid 1990s to implement of labor cost reduction policies targeted on low-paid workers. Nevertheless, the sustainability of the em-ployment gains arising from those policies remains partly threatened by further technological and trade developments. Second, labor market policies should also provide incentives to labor market participation as most of the transatlantic employment gap is accounted for by socio-demographic groups that face important participation decisions and deteriorating employment prospects. Finally, policies should promote occupational mobility as they could dampen aggregate employment losses by allowing displaced workers to transit towards in-demand occupations.
The chapter is organized as followed. Section 1.2 provides a description of the data and correction procedures used to produce time-consistent time series comparable across countries between 1982 and 2017. Section 1.3 documents employment trends as well as job polarization for both France and the U.S. Section 1.4 analyses the transatlantic employment gap and targets the employment deficit. Section 1.5 decomposes employment dynamics in order to determine whether those dynamics are due to socio-demographic factors or changes in employment propensities. Section 1.6 determines the most important flows explaining occupational employment dynamics. Finally, section 1.7 concludes.
In this section, I describe the data used and the construction of cross-country and time-consistent time series on labor market outcomes for France and U.S. between 1982 and 2017.
Labor force surveys and samples
France. This paper uses data from the 1982 to 2017 French Labor Force Survey (LFS) collected by the French National Institute for Statistics and Economic Studies. The French LFS is a repre-sentative sample of the French labor force. This database oﬀers information on individuals’ socio- demographic characteristics as well as their labor market and occupational status over a long period of time. The French LFS is a rotative panel. From 1982 to 2002, households were followed for at most three consecutive years. Since 2003, the French LFS became a quarterly survey in which house-holds are followed for at most six consecutive quarters. I use this configuration to match respondents across two subsequent interviews. This allows to monitor individual transitions across labor market states and across occupations between two successive years prior to 2003 and two successive quarters since then.
United-States. For the U.S., I rely on micro data from the IPUMS Current Population Survey (CPS).1 As for France, this survey is a representative sample of the U.S. labor force. Raw data are collected by the United-States Census Bureau allowing the Bureau of Labor Statistics to monitor U.S. labor market outcomes. The survey covers the 1982 to 2017 period as for France. The survey is a monthly rotative panel in which households are followed for at most a total of eight months. Respondents are included in the CPS for four subsequent months, they are not interviewed for the next eight months. They are then comprised again for the next four months. As for France, the longitudinal aspect of the survey enables me to identify workers’ transitions across labor market states and occupations.
Samples. This chapter relies on both the French LFS and the U.S. CPS to study the evolution of aggregate employment and its occupational structure from 1982 to 2017. The study starts in 1982 because the French LFS lacks in precision about some variables prior to 1982 especially about occupational variables. It uses two samples for each country. Both of them focus on the 15 to 64 year-old working age population; military and farming occupations are excluded. The first sample contains all the observations and is used to compute aggregate labor market stocks per capita, propensities to be in a given labor market state as well as the composition of the working-age population. The second sample only keeps observations for which individuals are matched across two subsequent surveys. It is used to study individuals’ transitions across labor market states and occupations.
Variables and measurement
Occupational groups. In order to study transatlantic employment outcomes in light of job po-larization, I aggregate occupations into three task groups: manual, routine and abstract jobs. For the U.S., I rely on the classification proposed by Cortes, Jaimovich, Nekarda, and Siu (2014). In this chapter, one of the main challenges is to provide cross-country comparable time series. Since occupational codes are not consistent across time and countries, I produce a handmade crosswalk for occupational codes for France. It is close to the one proposed by Albertini, Hairault, Lan-got, and Sopraseuth (2017) and Charlot, Fontaine, and Sopraseuth (2019). Table B.3 displays the crosswalk between occupational codes and task groups. In this study, manual occupations refer to non-routine manual occupations. Those jobs require mostly social interactions and manual dexter-ity. This definition of manual jobs captures the bulk of the employment growth at the bottom of the occupational mean wage distribution. Thus, manual occupations include mostly personal ser-vice workers (CSE56), some specific public service civil servants (CSE52) as well as some protective services (CES53). Routine jobs are located in the middle of the occupational mean wage distri-bution. They include occupations such as intermediate health and social work personnel (CSE45), intermediate business administration and commerce personnel (CSE46), foremen (CSE48), business administrative personnel (CSE54), salespeople (CSE55), drivers (CSE64), maintenance, storage and transportation workers (CSE65), skilled industry and artisan laborers (CSE62 and 63), and unskilled industry and in construction finishing laborers (CSE67 and 68). A substantial portion of those jobs has been subject to either automation or computerization since the last three decades, which ex-plains why the middle class has been shrinking ever since. Abstract jobs include occupations that often require a relatively high diploma because of the complexity of cognitive tasks accomplished. They include occupations such as wholesalers (CSE22), heads of company (CSE23), liberal profes-sions (CSE31), public service professionals (CSE33), professors and scientific professions (CSE34), business administration and commerce jobs (CSE37), business engineers and technicians (CSE38), intermediate health and social work personnel (CSE43), technicians (CSE47).
Stocks and socio-demographic factors. In order to study the role played by socio-demographic factors in grasping diﬀerences in transatlantic labor market outcomes, I rely on time series on aggre-gate employment, unemployment and non-participation per capita, as well as population shares and propensities to be in abstract, routine, manual employment, unemployment and non-participation for 12 socio-demographic groups depending on
• Gender: men, women;
• Age: 15-25 (young), 26-54 (prime-aged) and 55-64 (old);
• Education: at most a high school degree and more than a high school degree.
Since surveys have been subject to significant modifications in both countries, I build those variables by proceeding as followed. First, I construct aggregate time series on employment, unemployment and non-participation per capita as the fraction of individuals in the considered state over the 15-64 year-old working age population. Then, I construct employment shares for 36 gender age education task groups, as well as unemployment and non-participation shares for 12 gender age education groups. I correct them for breaks as described in subsection 1.2.3. I use the resulting time series to obtain cross-country and time-consistent time series for aggregate stocks, population shares and propensities to be in a given labor market state for each of the 12 gender age education groups between 1982 and 2017.
Labor market transitions. I use information on the labor market status and occupational cate-gories of individuals to construct transition rates across five labor market states: abstract, routine, manual employment, unemployment and non-participation for both France and the U.S. In that respect, I closely relate to Charlot, Fontaine, and Sopraseuth (2019). The evolution of stocks is described by the following law of motion St = Pt 1St 1 (1.1) with St = [At Rt Mt Ut Nt]0 the fraction of the working age population in each state. The transition matrix Pt 1 gives transition rates Pij;t 1 from state i in t 1 to state j in t. I choose to construct annual transition probabilities to produce consistent time series covering the 1982 to 2017 period.2 Low-frequency transition rates omit within-year transitions but they nevertheless capture long-run labor market trends which is the aim of this paper.
I build those transition rates by relying on the longitudinal aspect of U.S. and French labor force surveys. For the U.S., I first match monthly CPS files at annual intervals. I obtain yearly transition rates at a monthly frequency.3 Then, I average over each year to obtain annual rates observed at a yearly frequency. The resulting time series are corrected for breaks as described in subsection 1.2.3 and then for margin errors as in Elsby, Hobijn, and ẞahin (2015).4 In the case of France, I match annual surveys before 2003 which gives annual transition rates at an annual frequency.5 Since 2003, the survey became quarterly. I match individuals across quarters at annual intervals which yields yearly transition rates observed at a quarterly frequency. Similarly to the U.S., I then average the resulting transition rates over each year to obtain annual transition rates observed at a yearly frequency. I also correct the resulting time series for breaks and margin errors. Transition rates are displayed in Figures A.2 to A.11.
French and U.S. long period time series are both subject to sharp discontinuities. I first describe why such breaks occur for each country. I then present subsequently the two models used to correct those breaks.6
Data discontinuities. Long period time series built from the French LFS are subject to two significant breaks because of two major redesigns and changes in classifications. The first redesign was implemented in 1990 when the sample was renewed. The questionnaire was modified notably concerning the collection of information on education. Moreover, interviews switched to computer-assisted classification of occupations. The second redesign occurred in 2003. The survey shifted from an annual rotating panel to a quarterly one. Additionally, the occupational classification changed from the PCS1982 to the PCS2003.
Long period time series for the U.S. are subject to three important breaks due to survey redesigns and changes in classifications. For instance, the occupational classification changed in 1983 introducing some discontinuity in aggregate time series.7 In 1993, a redefinition of education variables introduced breaks in educational-based times series (Jaeger, 1997). In 1994, the survey experienced the most drastic redesign over the time period considered. The aim was to improve the quality of the data by introducing a new questionnaire and modernized data collection methods (R Cohany, Polivka, and M Rothgeb, 1994). Such modifications pose a challenge for constructing long period time series that I tackle by using correction factors.
In this section, I first broadly document aggregate employment, unemployment and non-participation dynamics to capture transatlantic employment performances. Second, I decompose aggregate em-ployment by occupational groups to show that both countries underwent job polarization.
Aggregate employment outcomes
Transatlantic labor market performances are computed as the fraction of the working age population in employment, unemployment and non-participation. In that respect, Figure 1.1 displays the dynamics of aggregate employment, unemployment and non-participation per capita between 1982 and 2017.
On average, employment performances in France are lower than in the U.S. This is reflected by higher unemployment and also lower labor market participation. Indeed, average employment per capita in France is of 61.89% against 69.29% in the U.S. while average unemployment and non-participation per capita are of 6.63% and 31.49% in France against 4.57% and 26.14% in the U.S.
The two countries exhibit opposite employment dynamics. Their employment performances first diverge from 1982 to 1998 and then converge. In France, employment per capita first declines from 62.85% to 60.33% between 1982 and 1998. It then increases by 4.06 pp to reach 64.39% in 2017. Most of this increase occurs in the late 1990s. On the contrary, U.S. employment per capita first increases from 64.97% to 72.56% between 1982 and 1998. It then decreases by 3.74 pp to reach 68.82% in 2017. It should be noted however that most of the decline occurs during economic recessions and that aggregate employment recently increases.
Long-run employment dynamics appear to be foremost reflected by unemployment dynamics in France and by non-participation dynamics in the U.S. In France, unemployment mirrors the non-monotonous employment dynamics. It first increases from 4.63% to 7.99% between 1982 and 1998 while it falls by 1.14 pp to reach 6.85% in 2017. Most of this decline occurred during the late 1990s and early 2000s. On the contrary, non-participation follows a steady downward trend. It falls steadily from 32.52% to 28.76% over the entire time period. In the U.S., unemployment absorbs the cyclical variations in employment while non-participation mostly mirrors its non-monotonous long-run pattern. For instance, unemployment per capita increases by 3.59 pp between 2007 and 2010 reflecting the eﬀect of the Great recession. On the contrary, non-participation dynamics are smoother. It first decreases from 28.29% to 23.96% between 1982 and 2000 and then increases by 4.03 pp to reach 27.99% in 2017. Thus, employment is the relevant indicator when comparing French and U.S. labor markets. Unemployment does not fully capture the ongoing changes aﬀecting them.
Table of contents :
1 Transatlantic Employment Outcomes
1.2.1 Labor force surveys and samples
1.2.2 Variables and measurement
1.2.3 Break correction
1.3 Aggregate employment and job polarization
1.3.1 Aggregate employment outcomes
1.3.2 Job polarization
1.4 The transatlantic employment gap
1.4.1 Socio-demographic composition and propensities
1.4.2 The transatlantic employment gap
1.4.3 Decomposition by socio-demographic groups
1.5 Employment dynamics
1.5.1 Employment change decomposition
1.5.2 Aggregate relevance of socio-demographic groups
1.6 Labor market transitions
1.6.1 Average transition rates
1.6.2 Job polarization and occupational mobility
2 Job Polarization and Unskilled Employment Losses in France
2.2 Stylized facts
2.2.1 The deterioration of unskilled employment outcomes
2.2.2 Job polarization
2.2.3 Occupational wage dynamics
2.2.4 Labor taxation policies
2.2.5 An incomplete reallocation of unskilled labor
2.3 A general equilibrium model
2.3.1 The environment
2.3.3 Occupational choice
2.3.4 The representative consumer
2.3.5 Market clearing conditions
2.5.1 The obstructed reallocation of unskilled labor
2.5.2 Labor taxation and technological change
2.5.3 Accounting for the decline in unskilled employment
3 Routine-Biased Technological Change and HoursWorked over the Business Cycle
3.2 A general equilibrium model
3.2.1 The model
3.2.3 Comparative statics
3.3.1 Data construction
3.3.2 Stylized facts
3.4 A VAR model
3.4.1 Bayesian estimation
3.5.1 Specification I – Is Gali’s technological shock neutral?
3.5.2 Specification II – RBTC, neutral and task-supply shocks
3.5.3 Technological shocks and aggregate fluctuations
A Appendix of chapter 1
A.1 Alternative measure of the employment deficit
A.2 Margin-error adjustment
A.3 Additional Tables and Figures
B Appendix of chapter 2
B.1 Additional Figures and Tables
B.2.1 French Labor Force Survey and samples
B.2.2 Building variables
B.3 Re-weigthing methods
B.3.1 Counterfactual employment structure
B.3.2 Wage change decompositions
B.4 Labor taxation policies
B.4.1 Labor taxation time series
B.4.2 A brief history of labor taxation policies
B.5 Asymptotic equilibrium
B.5.1 Preliminary computations
B.5.2 Asymptotic wages
B.5.3 Asymptotic allocation of labor
B.5.4 Asymptotic wage inequality
C Appendix of chapter 3
C.1 Comparative statics analysis
C.2 Univariate time series analysis
C.2.1 Unit root tests
C.2.2 Robustness of business cycle moments
C.3 VAR algorithm
C.4 Additional results
C.4.1 Impulse responses
C.4.2 Forecast error variance decompositions
C.4.3 Historical decompositions
C.5 Empirical robustness
List of Figures
List of Tables