Project topics and materials
Get Complete Project Material File(s) Now! »
Specialization during Education : Impact of Specialization for New Comers in the Labor Market
This paper studies and quanti es the returns to specialization de ned as the schooling choice of a speci c eld. We consider two di erent di-mensions of education : a quantitative one, the number of accumulated years of schooling, and a qualitative one, the eld of study. In addition to the estimation of returns to specialization, the timing of specialization during education is also taken into account. To understand the impact of early and late specialization, we build a dynamic discrete choice model allowing for heterogeneity in returns to human capital : individuals are supposed to be forward looking agents, making sequential schooling decisions and then facing a labor market speci c to her eld. Returns to schooling, job destruction probability and o er arrival rates are measured through the labor market structure and di er by specialization.
We use French panel data Generation 98 , with more than 6500 men over 7 years. All these individuals exit school at the same date and then face the same macroeconomic context.
Our results underline that the choice of the specialization have a large impact on individual labor market trajectories with a particular role of utility of schooling. Our speci cations allow us to capture heterogeneity in both dimensions of schooling choices, especially distinct returns of schooling.
Introduction
Education is usually de ned as the number of years of schooling. Returns to edu-cation are then measured as the increase in earnings obtained after spending an additional year at school instead of entering the labor market.2 This simple way of measuring education may be too restrictive. The large diversity of schools and curri-cula may bring di erent skills that are likely to be rewarded di erently by the labor market. As a consequence, returns to education will di er : two individuals with the same number of years of schooling may actually have distinct labor income.
Furthermore, a recurrent debate about vocational education raises the question of specialization during education. Some degrees provide skills that are speci c to particular occupations on the labor market. And then, specialized individuals po-tentially face di erent job opportunities. Although speci c skills may respond to the needs of some industries or rms, policy makers might be interested in knowing the impact of early or late schooling specialization on future trajectories.
The aim of this paper is to describe the di erences in labor market trajectories of people acquiring di erent skills during their school path. Thus, we argue that education has at least two dimensions : a quantitative one – the number of years of schooling – and a qualitative one that characterizes the professional knowledge or skills acquired at school that is called specialization.
In the early years of the career path, specialization is often characterized by vocational education. Focusing on this aspect of education Adda, Dustmann, Meghir, and Robin (2013) show its importance on future trajectories.3 Skilled and non skilled workers in Germany have di erent wage pro les and job opportunities : although skilled workers experience a fast wage increase at the beginning of their career, they face lower destruction rates and receive less job opportunities which may prevent them from nding new jobs after a negative employment shock. In post secondary education, specialization mainly corresponds to the choice of a major. For higher level of schooling, Arcidiacono (2004) and Arcidiacono, Hotz, and Kang (2012) underline that major choice may have a large impact on future individual earnings.
Finally, Hanushek, Woessmann, and Zhang (2011) question the timing of the specialization. Considering the trade-o between immediate productivity and future adaptability, they compare di erent countries where the age of specialization di ers. They show that later specialization is associated to worse employment outcomes but that this initial penalty lowers with time. Focusing on this same question, Malamud (2010) compares England and Scotland who have di erent educational systems in terms of specialization. He shows that the timing in accumulation of eld speci c skills has important implications for career paths : switching to occupations unrela-ted to the eld of studies lowers wages and the cost of switching is much higher for English workers who specialize earlier than Scottish ones.
Altogether, these results con rm the intuition that individuals face di erent labor market conditions according to the skills acquired during education and the timing of the acquisition. In particular, returns to education, o er arrival rates and probability of job destruction may di er according to the educational path of individuals.
In this paper, we propose a model of individual labor market trajectories and schooling choices with a specialization dimension. The key issue in order to measure the di erences in job opportunities and returns to specialization is the endogeneity of schooling decisions. We build a dynamic discrete choice model of schooling and career choices a la Keane and Wolpin (1997) : individuals rst make sequential schooling decisions choosing a level of schooling and a specialization ; then they leave school once and for all and enter the labor market. At this second stage, people face job o ers, decide to accept or reject it and may lose their job while working.
This dynamic discrete choice structure allows us to take into account the ac-tualized values of alternatives in individual decisions and is relevant to identify the importance of the timing for optimal individual decisions. Moreover, unobserved he-terogeneity is added to the model in order to solve for endogeneity issues linked to the fact that individuals base their choice on future expected earnings.4 Following Belzil and Hansen (2002), each discrete type of unobserved heterogeneity has a schooling ability and a labor market ability and individuals belonging to this type base his decisions on these two elements. This allows us to identify returns to specializa-tion by accounting for potential correlation between both dimensions of unobserved heterogeneity.
To secure identi cation, we use local capacity constraints of schools as exclu-sion variables that only in uence schooling decisions but are supposed to have no direct impact on future earnings. These variables are considered as exogenous from the individual point of view : in the presence of moving costs, individual decisions are guided by local schooling opportunities, measured by the number of available positions for each specialization and degree.
The model is estimated on the Generation 98 survey, a French panel data of young people leaving school in 1998. We measure specialization from the observed eld of studies or the domain of the highest diploma. We nd large heterogeneity in labor market outputs according to the degree of individuals. In particular, we nd that the choice of the specialization have a large impact on individual labor market trajectories. This heterogeneity comes from direct returns of education but also from di erences in subsequent returns to labor market experience and costs of schooling faced by individuals.
The next section details the model and identi cation strategy. Section 1.2 pre-sents the data. Then, section 1.3 analyzes the empirical results. The last section concludes.
Model
Our goal is to measure di erences in labor market trajectories taking into account the endogeneity of education decisions. Therefore, we build a two-stage model. The rst part explains education decisions and the second one describes labor market decisions. The two stages are sequential : once they entered the labor market, indivi-duals cannot change their education level or specialization by going back to school. This feature of the model signi cantly simpli es the resolution of the model and is empirically supported by the fact that in our data less than 1% of individuals actually go back to school after entering the labor market.
During their schooling path, individuals are supposed to make sequential discrete decisions. At each level of education, they choose to enter the labor market or to continue to the next level of schooling. When continuing in the schooling system, individuals have to choose to specialize toward a particular set of occupations or not. We consider three types of schooling opportunities : no specialization, specialization toward production occupations and specialization toward service occupations.
Thus, during the schooling stage, an individual state is characterized by a position in the path in terms of level and specialization and he faces four alternatives : three specialization alternatives at the next level of the schooling system and one labor market alternative.
During the second stage, individuals face a labor market that is speci c to the nal level and chosen specialization. At each period, they receive job o ers and decide to accept or reject it. When employed, positions can also be destroyed at a rate that is specialization speci c.
Individuals are supposed to be forward looking in the sense that they take into account the future impact of their present decisions. This allows us to link the two stages of the model : the value of an educational degree will depend on its labor market value.
The two stages are also related through individual heterogeneity linking unob-served cost of schooling to unobserved ability on the labor market. Letting these two dimensions of unobserved heterogeneity to be correlated allows us to explicitly model the correlation between education and labor market and then to solve for the endogeneity problem.
In order to formalize this model, let t be the state space at time t, including state variables and random draws, and djt an indicator variable for choice j at time t. The Bellman equations are :
Vt t max V j t Vtj( t) = ujt + E[Vt+1( t+1)Sdjt = 1; t]
where is the discount factor and ujt the direct utility of choice j. To avoid notation burden, we make the conditioning in the E max term implicit and we omit individual subscripts. Given initial characteristics 0, the set t can be written in three parts : state variables at time t, !t, random draws at time t, !t, and past state space : t = t−1 ∪ !t ∪ !t .
We value functions for schooling decision V S j ( t ) from value distinguish the functions for the second stage of the model V Ltj( t).5
Schooling decisions
In the schooling stage, individuals choose a level of schooling and a specialization maximizing their utility. At the end of her schooling path, an individual is charac-terized by a level of schooling d and a specialization k. We distinguish six di erent levels of schooling :
• d = 1 : Junior high school : most of children have age 16 which corresponds to 5A summary of notations and details on the derivation of value functions and the likelihood are given in Appendix 1.A. the maximum of compulsory age of schooling in France. These individuals are high school drop outs.
• d = 2 : First year of High school or Short professional track that ended with a quali cation (BEP/CAP).
• d = 3 : High school diploma (Baccalaureat)
• d = 4 : Early higher education : Community college (BTS, DUT) or University rst year
• d = 5 : Bachelor degree
• d = 6 : Higher degrees (Master/PhD degree and French Grandes Ecoles)
And as stated before, three specializations are available at each level of schooling : no specialization, specialization toward production occupation and specialization toward service occupation.
At each state (d, k), individuals have the possibility to leave school and face the labor market corresponding to the pair (d, k). Alternatively, they have the possibility continue at school to the next schooling level and choose a new specialization. We denote by Cd;k, the set of possible choices at state (d; k). Cd;k = {(d + 1; 1); :::; (d + 1; K); LM} where LM is the labor market alternative with level d and specialization k. At the last level of schooling, d = 6 individuals necessarily enter the labor market.
Transition cost from level d to level d′ is denoted by cdd′;k′ (Wd′;k′ ; dd′ ) where Wd′;k′ is a cost shifter of schooling and dd′ is a normally distributed random shock. It is important to acknowledge that French public schools are free and that there are no fee at all level of schooling. Thus, we do not include additional covariates is the cost speci cation which we assume to be linear. These costs are written : cdd′;k′ = d′;k′ + ’d′ ⋅ Wd′;k′ + dd′;k′
Prior the last period of schooling, the value function of choice (d′; k′) conditional on being at state d :
V Sd ′ = cd ′ ′ ( W ; d max V j ( + )] ∀ d 1 ≤ d < D ′ ;k d ;k d′;k′ d ′ ) + E [ ∈C ′ ′ d j d ;k t 1
where V j can be value functions for schooling V Sj or labor market values V Lj, de ned below ; the expectation is taken on future realizations of random draws. The ows cdd′;k′ can be interpreted as utility of schooling, usually reported as costs of schooling or psychic costs in the literature.
Labor Market Alternatives
At each state of the schooling path, individual can choose to enter the labor market.
The value of working depends on both his level of schooling and his specialization.
While in the schooling system, the labor market value is given by : V Lt(d; k) = 0(d; k) × V Et(d; k) + (1 − 0(d; k)) × V Ut(d; k)
wher V Et(d; k) is the intertemporal value of being employed, V Ut(d; k) the inter-temporal value of being unemployed at time t and 0(d; k) the probability to receive an o er while not working. Thus individuals do not receive automatically a job o er and then can be unemployed.
Reward functions
Intertemporal values of being employed and unemployed depend on the potential ows of utility characterized by the rewards associated to the decision to accept or not an o er.
The reward functions of employment for an individual with schooling level d and specialization k is given by : RtE(d; k; Xt) = g(d; k; Xt) × e »t(k) where « t(k) is the error term which distribution depends of the specialization and Xt denote the working experience. The g function has the following shape :
D g(d; k; Xt) = exp Œ 0k + 1kXt + 2kXt2 + Q 3kd1(s = d)‘ s=1
Xt is thus a state variable. Its evolution is deterministic given the decision to work (dEt = 1) : Xt+1 = Xt + dEt.
To allow for more exible on-the-job wage evolution we model error terms as autoregressive processes (see Adda, Dustmann, Meghir, and Robin (2013)). When entering a new job, unobserved heterogeneity of labor income is supposed to follow a normal distribution with variance parameter speci c to the chosen specialization : « t(k) ∼ N (0; k). Then, while staying in the same job, individuals face shocks that are added to the previous term. Thus we have : « mt(k) = « mt−1(k) + et(k)
Table of contents :
Introduction générale
1 Specialization during Education
1.1 Model
1.1.1 Schooling decisions
1.1.2 Labor Market Alternatives
1.1.3 Identication Issues
1.2 Data
1.2.1 The Generation 98 Panel Data
1.2.2 Computing Capacity Constraints of Schools
1.2.3 Descriptive Analysis of labor market trajectories
1.3 Results
1.3.1 Distinct Returns of Specializations
1.3.2 Costs of schooling
1.3.3 Labor market trajectories
1.3.4 Ex-ante returns to schooling
1.A Summary of Notations
1.B Model Likelihood
1.B.1 Labor market contributions
1.B.2 Schooling contributions
1.C Estimates
1.C.1 Schooling costs Intercepts
1.C.2 Oers arrival rate
2 Evidence on Risk Aversion and Psychological Traits
2.1 The GSOEP Data
2.1.1 The German Education System
2.1.2 Risk Aversion Measurement
2.1.3 Non Cognitive Traits
2.1.4 Cognitive Skills
2.1.5 Social Preferences : Trust and Reciprocity
2.1.6 Parental Background Variables
2.2 The Econometric Model
2.2.1 Dening Risk Aversion
2.2.2 Modeling Psychological Factors and Risk Aversion
2.2.3 Measurement Equations
2.2.4 Identication of the Factor Structure
2.2.5 Estimation Method and Likelihood Function
2.3 Empirical Results
2.3.1 The Distribution of Factors and Dierences between Tracks
2.3.2 The Correlation betweens Factors and Risk Aversion
2.4 Concluding Remarks
2.A Psychometric Questions in the GSOEP
2.A.1 Locus of control
2.A.2 Conscientiousness
2.A.3 Trust
2.A.4 Positive Reciprocity
2.B Descriptive Statistics
2.C Counterfactual Distributions
3 Risk Tolerance and Life Cycle Income Proles
3.1 Modeling Risk Tolerance
3.1.1 The Lottery and CRRA Assumption
3.1.2 Risk Tolerance Categories
3.2 A Model of Life-Cycle Income Proles
3.3 Results
3.3.1 Sample description
3.3.2 Model estimates
3.3.3 Eect of the Estimated Risk Tolerance
3.A Estimating Risk Tolerance From the PSID
3.B From Initial sample to proper data
3.C Full Model Estimates
Bibliographie
