Youth labour market histories, neighborhood of origin and diploma: a dynamic modeling

Transcription

1 Youth labour market histories, neighborhood of origin and diploma: a dynamic modeling Thierry Kamionka Xavier Vu Ngoc February 12, 2014 Abstract We use the survey Generation 98 to study the trajectories of youth between five states of the labour market: permanent employment contracts, temporary employment contracts, subsidized employments, unemployment, non-participation. These youths are leaving education system in We use a dynamic multinomial logit model with random effects in order to study labour market transitions of young workers. Initial conditions are treated using the method proposed by Heckman (1981). Initial conditions consist in the neighborhood of origin at the time they leave the education system, the level of the diploma and the initial position on the market. Censoring on the location at the time they leave education is formally taken into account in modeling. The model we use allow to disentangle true from spurious state dependence. We assess the impact of neighborhood of origin on youth labour market histories. We evaluate the impact of subsidized employment on labour market transitions on a recent period. JEL Classification : C33 ; C34; J60 ; C53; R23. Keywords : labour market histories, state dependence, subsidized employments, mobility, panel data, neighborhood effect, random effects, simulated maximum likelihood We would like to thank Cereq for providing the data. CNRS and CREST, Timbre J390, 15 Bd Gabriel Péri, Malakoff Cedex, France (kamionka@ensae.fr). Tel : Ecole Polytechnique, France (vungoc@poly.polytechnique.fr). 1

2 Introduction In recent years, young people seem to have more difficulties in entering the labor market. These growing difficulties they face can be characterized using several indicators. There is a structurally higher unemployment rate among young workers compared to other age. 22.8% of people aged are unemployed in late 2010, a rate much higher than for older individuals (8.3% for years and 6.3 % for those more than 50 years according to Insee s employment survey in the fourth quarter of 2010) 1. The rate of youth unemployment hides geographical disparities and according to the characteristics of individuals. Thus, in 2008 the unemployment rate in deprived urban areas (ZUS) was equal to 16.9% against 7.7% outside deprived urban areas in the neighborhoods of urban units comprising a deprived urban area 2. These geographical differences are accentuated for certain groups of the population. For instance, in 2008 and for young men (15-24 years), the unemployment rate was equal to 41.7 % in deprived urban area against 19.1 % off deprived urban areas, in neighborhoods of urban units comprising at least a deprived urban area (Zus). To cope with the development of youth unemployment, since the mid-70s, have been put in places government sponsored employment policies (cf. Dares, 1997). The government has developed the use of subsidized employment and has implemented policies to reduce payroll taxes. However, these policies have failed to limit the progression of youth unemployment. The impact of these programs is dependent on their content and of characteristics of individuals. Thus, Fougère, Kramarz et Magnac [2000] show, that in France, employment programs that include significant training are most effective. Thierry Magnac (2000) noted that the training programs do not improve the chances of youth to find a stable job. Nathalie Havet (2006) highlights the impact of those measures differs by gender. Nathalie Havet et Guy Lacroix (2012) show that only structured training have some effectiveness. Some also note that the integration problems encountered by unskilled youth could be amplified due to the protection of senior employment (insiders) by firms or because of public policies targeted on older workers. Rising of youth unemployment increases the period going from the end of the initial training to the first stable job for low skilled young. Some authors also note the existence of an intensification of the process of "downgrading" of graduates (Emmanuelle Nauze- Fichet et Magda Tomasini, 2002). The objective of this paper is to determine the impact of the location of the individual upon entry into the workforce on the dynamics of its insertion on the labor market. Such a spatial characterization has recently been studied for the Paris region by Gobillon, Magnac, Selod (2009, 2010), who evaluate the impact of residential segregation on the access to employment. Duguet, L Horty, du Parquet, Petit, Sari (2012), they use a testing technique to cross the impact of gender, of geographical origin, and spatial segregation, studying the recall rate on fictitious applications in three municipalities more or 1 The unemployment rate for year olds was equal to % in 1982 according to Insee s employment survey. 2 Source : Insee s Employment Survey and Observatoire des inégaliés. 2

3 less favored of the department of Val d Oise (an administrative division similar to an US county). They show that women, people of Moroccan origin and residents of disadvantaged municipalities are less likely to access employment. Selod and Zenou (2006) show that social networks are less developed in deprived urban areas, making it more difficult access to employment. In our article, we study the impact of initial geographical location - i.e. at the exit of education system - on transitions between states of the labor market during a sufficiently long period to account for the process of integration of young people into working life. The advantage of our approach is that we we are interested not only in access to employment but we are trying also to appreciate the quality of insertion. We do not seek to limit our analysis to the instantaneous impact of the localization but we wonder rather if a localization in a deprived urban area 3 during the formative years does not have a lasting impact on trajectories of young people from these areas. Because we analyze the individual paths between different states of the labor market, we can also study the impact of public programs of insertion even if we can not distinguish the various programs. We focused on the location of individuals according to whether they live in a deprived area (ZUS) or not. According to the definition of Insee, the deprived urban areas are infra-urban areas defined by the government to be the primary target of the urban policy depending on local considerations related to difficulties experienced by the inhabitants of these territories. The law of 14 November 1996 for implementation of the stimulus policy of the city policy it provides for derogations introduced for three categories of geographic units: deprived urban areas (Zus, 751 areas identified), areas of urban revitalization (Zru, 416 areas identified), urban free zones (Zfu, 100 areas identified) 4. Several measures, both fiscal and social are planned for compensate for the economic difficulties faced by the inhabitants of these territories. These measures are more important for the free urban areas (ZFU) than for areas of urban revitalization (Zru) and more for areas of urban revitalization (Zru) than for devprived urban areas (Zus). For instance, firms located in deprived urban areas (Zus) receive exemptions on business tax for creations or extensions of firms with fewer than 150 employees. Roland Rathelot et Patrick Sillard (2008) show that the replacement of some areas (Zru) by deprived urban areas (ZfU) generally has a significant and positive impact on job creation and establishment installation, while having a low impact on surrounding areas. However, according to the report of ONZUS of the year 2010 (National Observatory of Zus), the unemployment rate in Deprived Urban Areas (ZUS) 18.6 % in 2009 against 9.8 % for other areas. In Deprived urban (ZUS), unemployment is particularly large for youth: according to the same report, in 2009, 43 % of young men active and 37% of young women active were unemployed. The localization in a deprived urban area (ZUS) and educational level are not exoge- 3 Deprived urban areas (ZUS) are characterized by the presence of large housing complexes or degraded habitat areas and by an accentuated imbalance between housing and employment, Act No of 4 February According to the General Secretariat of the CIV: update of October 3,

4 nous variables when we consider the situation on the labor market. The knowledge of the localization at the exit of the education system authorizes, however, to obtain a unique characterization of the spatial and economic segregation, allowing to study its impact on youth employment Several studies find some constants inherent to deprived urban areas (Zus). Aeberhardt, Fougère, Pouget et Rathelot (2010) show, for instance, that outside paris region, wage differentials between workers living in deprived urban areas (Zus) and those living off deprived urban areas are about equal to 4%. Moreover, the probability difference of being employed between the two groups is of the order of 26 %. Aeberhardt, Coudin et Rathelot (2010) emphasize moreover a high correlation between geographical origin and localization in a deprived urban area (Zus): 24% of French whose parents come from du Maghreb 5 live in deprived urban areas (Zus), against 6 % of French with a French origin. If such a descriptive study provides useful information on the local labor market, it does not allow to assess the dynamic effects of the localization on the insertion of young people on this market. Surveys Génération of CEREQ allow to track the professional integration of young leaving the education system in From these data, we can realize that the process of insertion of young people on the labor market is heterogeneous depending on the localization, gender and educational level. Thus, male graduates living outside of deprived urban areas (Zus) are inserted faster on the labor market and more durably than the rest of the population. To account for the potential endogeneity of the localization in deprived urban areas (ZUS), we perform a joint dynamic modeling of transitions on the labor market, of localization at the exit of the education system and of the level of education. Due to the longitudinal nature of the data, we use a conditional multinomial logit model with unobserved heterogeneity (Card and Sullivan [1988], Magnac [2000], Jones and Riddell [1999], Havet [2006]). Such a model allows modeling the conditional probability for the individual to be in a state of the labor market given its individual characteristics, but also taking into account its its past trajectory. We use a specification of unobservable heterogeneity via individual correlated random effects (Havet [2006], Gilbert, Kamionka, Lacroix, 2011). In order to take into account that we observe the individuals from their entry into the workforce, that localization and the education level are fixed at the exit of the education system, we model the initial conditions using the method proposed by Heckman (1981). Our study has several new innovations compared to previous research on the effects of geographic localization. Unlike previous studies, we have, by the nature of our database, the opportunity to study a representative sample of young French: this sample has no geographical limitation or of level of education and does not concern only a particular population in terms of the situation on the labor market. We consider formally nonresponse to the question of the localization in deprived urban areas. This database opens up a broader field of possibilities compared to other existing French databases: FHS database for instance which is limited to job seekers, sample obtained using testing by Duguet, L Horty, du Parquet, Petit et Sari (2012) and limited to three municipalities. 5 States of North Africa. 4

5 Our model specification has many advantages. It allows to take into account several explanatory components (unobserved individual characteristics, observed explanatory variables and previous situation on the labor market) and to compare their respective impacts on the situation of the individual at a given time. Our main results show that subsidized jobs not significantly improve the insertion in stable employment. The neighborhood of origin of these young has a significant and negative impact on their insertion on the labor market. This effect of the initial neighborhood lasts for the entire observation period and it concerns, particularly, the access to insecure jobs. This phenomenon can be explained by a spatial mismatch between jobs and labor supply from these young. It also raises the issue of geographical mobility, which is probably insufficient and could be encouraged by policies that favor the removals or, by an aid to setting up boarding facilities for colleges and high schools. These policies could nevertheless have limitations as these results may also reflect the effects of discrimination or may be due to inadequate social networks available to these young. Also, the introduction pathways for grands lycées (prestigious high schools) and great schools reserved for these young may be beneficial. The paper plan is the following one: section 1 includes a presentation of data and Descriptive statistics performed on the sample. In section 2, we present the econometric model. Section 3 includes a presentation of estimation results. In the last section we conclude. 1 Descriptive Statistics and Data 1.1 Data Data are from the Survey "Generation 98" and allow to follow the history on the labor market of young people leaving the education system in In this survey are defined as "young" individual under 30 yo at the time of survey or people who have more than 30 years and in 1998 have exited the education system since less than 7 years. Four successive interrogations were conducted by Céreq on a representative sample of leavers of the education system in 1998, allowing to follow up on 10 years. The first survey was conducted in 2001 and covers 55,345 young; The second was conducted in 2003 and covers 22,000 youth interviewed again; The third was conducted in 2005 and and covers youth who are re-interviewed; The most recent was conducted in 2008 and covers 10,641 youth interviewed again. These questions are based on a set of questions asked to the individual on personal characteristics, its situation on the labor market, working conditions, his job search, among others. 6 For a presentation of the survey 98 Generation, see Couppié et alii (2007). 5

6 These questions have resulted in a database on the individuals that can be divided into three portions: An individual data base, with individual characteristics and occupational status on a monthly basis : identifier, diploma, social and geographical origin, family status, professional situation month by month; A firm data base, it lists the characteristics of employment spells of individuals: wage, contract type and duration, characteristics of the job and the company, responsibilities and experiences. A non-employment data base, lists the characteristics of non-employment episodes: reason of non-employment, duration, characterization of the research process, lost resources, aspirations, recovery or not of studies. 1.2 Data Processing Variables selected for analysis We pooled data from different surveys in order to have a comprehensive database of career of these young, knowing that for each new survey, less and less young provide information. We retain the variables that characterize the trajectory of these young people on the labor market: The trajectory of the labor market among 5 states: employment with an unlimited duration contract, employment with a limited duration, subsidized employment, unemployment and inactivity (including the resumption of studies). We have wanted to distinguish between regular jobs, from precarious jobs and those subsidized to reflect the duality of contract types on the job market. We also have the evolution of wage. The variables representing social and geographical origins of the individual (variables considered as fixed): gender, place of birth, nationality, occupation of parents and spouse, localization in a deprived urban area, highest degree obtained. Variables representing the social trajectory of the individual likely to influence his professional life: marital status, number of children, household type Data structure We chose, for practical reasons, to consider situation of individuals in the months of December and June of each year, from December This choice allows us to reduce our calculations while maintaining frequent observations of the trajectory of the individual on the labor market. We chose to begin the study in December 1998 as, at that date, only a minority of young people in the sample has not yet entered the labor market (1913 individuals on a total of 55345), while this number is much higher in previous months. 6

7 Furthermore, begin our study on that date allows us to have the largest average number of observations on the first sample of youngs followed from 1998 to On the first period from June 1998 to June 2001, the trajectories of individuals are informed, this number decreases on the other periods of observation: trajectories between June 2001 to June 2005, trajectories between June 2005 and December Thus, our panel is not a balanced panel. That is why there is a right censoring mecanism from the first date there is no information on the situation of the individual. If some information constituting fixed variables are systematically indicated (age, diploma, gender), the localization variable (lives in a deprived area, outside a deprived area) is not filled for individuals out of So we chose to build a variable indicating the fact of having or not an information about the localization. 1.3 Descriptive statistics at the exit of the school system The table below (cf. Table 1) contains descriptive statistics by gender and area of origin and it is made using 1998 data. These statistics highlight the heterogeneity of individuals right out of school. The age of the individuals in the sample ranges from 13 to 35 years, the cumulative percentage of individuals under the age of 16 is very low (of about 0.1%). The average age at the exit of the school system of individuals in a deprived area is lower than elsewhere, for men and women: 21.5 yo outside a deprived area and 20.8 yo in a deprived urban area. This category of individuals made shorter studies than the average of the population, This is confirmed by the level of education obtained in a deprived urban area: 45% of the male sample living outside of a deprived urban area do not have the Baccalaureate (high school degree) against 60.5% for men living in deprived urban area. We obtain, respectively, 30 % and 50 % for the female sample. Individuals located in a deprived urban area come from more disadvantaged backgrounds. Thus, an individual localized in a deprived area is two times less likely to have an executive father and he is about twice as likely to have a worker father. However, the profession of the mother varies less by location. The proportion of mothers working as employed is nearly equal to 60 % in both areas. Individuals located in a deprived urban area are more likely to live with their parents and the proportion of women who are raising children is greater in a deprived urban area. This is due to a higher proportion of single mothers in deprived urban areas. Whatever its location, a woman is more likely to raise children than a man. From the beginning, men earn on average about 10 % more than women during the first employment contract whatever the area. Moreover, a disparity between wages in the first job can be observed between employed individuals located in deprived urban areas and those living outside deprived urban areas. This difference is on the order of 4% for individuals of the same gender. 7

8 Table 1: Descriptive statistics (1998) Outside Deprived Areas Deprived Area Men Women Men Women Mean age in Average starting wage (Euro, nov 1998) Family status in 1998 Lives with her parents 75.48% 62.14% 83.78% 72.83% Cohabiting 10.73% 23.55% 8.11% 17.64% Alone 13.78% 14.31% 8.11% 9.53% Level of education Without diploma 18.31% 10.84% 34.37% 24.46% Vocational-technical school, basic (BEP) 26.18% 19.29% 26.14% 25.12% Vocational-technical school (Bac Pro) 17.49% 17.19% 13.64% 16.61% High school (Baccalauréat) 3.77% 5.60% 3.05% 5.06% Post secondary education (Bac+2) 17.35% 22.55% 11.86% 13.57% Master 7.69% 15.63% 6.68% 9.82% Graduate School or Ph.D. 9.20% 8.89% 4.26% 5.36% Has at least one child 7.24% 18.43% 8.35% 21.37% Number of children (if any) % 81.93% 80.69% 80.61% % 15.20% 16.55% 17.45% % 2.65% 2.07% 1.66% % 0.22% 0.00% 0.28% Father s occupation 1=farmer 4.95% 5.11% 0.97% 1.28% 2=craftsman, trader, entrepreneur 12.05% 11.47% 7.73% 8.10% 3=executive, engineer, professional, teacher 17.22% 17.87% 8.44% 8.91% 4=technician, foreman, packman, intermediate occupation 9.46% 8.60% 5.39% 6.61% 5=employee 28.68% 29.62% 30.71% 29.62% 6=worker 24.99% 24.38% 42.08% 40.69% 7=not available 2.65% 2.94% 4.68% 4.79% Mother s occupation 1=farmer 3.50% 3.61% 0.56% 0.56% 2=craftswoman, trader, entrepreneur 4.81% 5.22% 3.08% 3.26% 3=executive, engineer, professional, teacher 10.90% 11.51% 6.62% 7.17% 4=technician, foreman, packman, intermediate occupation 5.53% 5.38% 4.57% 5.21% 5=employee 57.66% 58.98% 62.72% 64.06% 6=worker 12.87% 11.29% 15.47% 14.06% 7=not available 4.74% 4.00% 6.99% 5.68% 8

9 1.4 Insertion on the labor market General Evolution Some general trends are remarkable, regardless of gender, degree and location of individuals. The insertion of youth is more rapid in the first two years of professional life. During this period, the number of young employed on a contract of unlimited duration increases very rapidly while the percentage of young people out of employment declines. This trend is particularly pronounced for men, graduates and people living outside deprived urban areas. The insertion is slower for women, those living in deprived urban areas and less qualified. The rate of increase in the proportion of employees on unlimited duration contract among youth decreases and becomes constant between 2001 and From 2007, the progression of this proportion is low or even negative. This is particularly true for women, qualified individuals and the inhabitants of the deprived urban areas. Unlike employment on an unlimited duration contract, the proportion of fixed-term employments from 2000 remains relatively stable or decreases whatever the gender, diploma and localization of the individual Evolution of employment by gender and location in 98 Right out of school appears a gap between men and women, between young people living in a deprived area and those living off deprived urban areas (cf. figure 1). Inequalities in terms of integration into the labor market remain throughout the trajectory on this market. Between 1998 and 2008, the proportion of young unemployed is on average 111% more important for young men located in a deprived urban area than for those living outside a deprived urban areas. This difference is on average equal to 63% for women. Inactivity is more important for women located in a deprived urban area: 51% more on average than women living outside deprived areas. This compares with fertility differences (largest in deprived urban areas) and differences in diploma (on average higher outside deprived urban areas). The insertion on the labor market is generally faster off deprived urban areas: the proportion of employment on unlimited duration contract is 21.5 % for men off deprived urban areas in 98 against 16.6 % for men in deprived urban areas (respectively % against % for women). However, there is a convergence over time of employment in deprived urban areas and outside these areas. In 2008, 59.3 % of male off deprived urban areas are employed on an unlimited duration contract against 55.4 % only in deprived urban areas (respectively 46.8 % against 43.1 % for women). This convergence is also evident at the level of wages. This convergence is also patent at the level of wages. 9

10 1.4.3 Evolution of employment by diploma and localization in 1998 Unsurprisingly, the more the diploma is high, more integration on the labor market is easy (cf. figures 2 et 3). Individuals living outside a deprived urban area are inserted on the market more easily than individuals in deprived urban areas. All degrees combined, in deprived urban areas, the proportion of young unemployed is higher and the proportion of young employed on an unlimited duration contract is lower. In deprived urban areas, the level of unemployment in December 1998 does not seem correlated with educational level. On the contrary, outside deprived urban areas, more the diploma is high and less the proportion of young unemployed is important. As might be expected, more the diploma is high, more the proportion of young unemployed decreases rapidly and more the proportion of employed on an unlimited duration contract increases. Less the diploma is high, more inequalities tend to fade with time in terms of the proportion of contracts of long duration. Thus, we can notice that, for a low level of qualification, being in a deprived urban area or not is less important in order to find an employment on an unlimited duration contract. However, for non-graduates, the proportion of young unemployed and in inactivity is higher in deprived urban areas but but the proportion occupying an employment with a limited duration contract is much lower. The proportion of subsidized jobs and limited-term employments remains relatively constant over time for holders of a diploma below bachelor s degree plus two years. Beyond the Baccalaureate, the share of such employment contracts tends to decrease with time. Thus, in 1998, the proportion of subsidized contracts concerns a very small proportion of the population who holds a diploma above Bachelor (high school degree) and tends to disappear in 2008 among young people holding a Master diploma or a PhD (going from less than 3 % to 0 % of this population). More the diploma is high, the lower is the proportion of young in subsidized employment and on limited-term employment Evolution of wages by gender, diploma and localization in 1998 Wages follow the same kind of evolution (cf. figure 4): a man earns on average 12 % more than a woman in the same area of residence on the period In average, a person living outside a deprived urban area earns 4 % more than a person of the same sex localized in a deprived urban area. Key fact, the trajectories of wages in and outside deprived urban areas follow a parallel trend different according gender. Wage differentials between men and women increases with experience. These findings are similar to those listed on wage inequality in the economic literature (cf. Manning and Swaffield, 2008). The salary is positively correlated with education level. We distinguish three groups: on average holders of a diploma below bachelor (high school degree) earn roughly the same salary. A second group consisting of university graduates emerges: the average wage for these individuals is approximately 23 % higher than the average wage of the holders of a Bachelor. The third group consists of PhDs and graduates of Graduate Schools, who earn, outside deprived urban areas, an average 79 % more than holders of 10

11 a bachelor. This difference is less significant in deprived urban areas (61 %). Remarkably, the wage differential between urban areas increases significantly when the diploma is high. For comparison, in deprived urban area, a young out of school without a diploma earns on average 4 % more than a young person living outside deprived urban areas, while someone who holds a doctorate or is graduated from a Graduate School, earns 12.3 % less in deprived urban areas compared to what he can earn on average outside these areas. In conclusion, men living outside deprived areas at the exit of the education system are inserted on the labor market more rapidly and durably than the rest of the population. However, these differences may reflect differences in individual characteristics between the two locations we must now consider. 2 Econometrics Modeling 2.1 The Model In order to model the trajectory of young people on the labor market, we use a conditional polytomic logit model with individual effects (cf. Magnac (2000), Havet (2006)). The propensity yitl of individual i to occupy state l of the labour market is described as a function of socio-demographic variables x itl and the state in which he stayed in t 1, denoted y i(t 1). We are going to consider a set of n individuals all leaving the education system in Let i denote the index of the individual, i {1, 2,..., n}. Let t denote the index of the period, t {1, 2,..., T }. The history of a given individual on the labour market is described by a set K+1 states. Let l denote the index of the state of the labour market occupied by the individual, l {0, 1,..., K}. Let z 2 denote a binary variable indicating the location of the individual in a sensitive urban area at the exit of the education system. The diploma obtained at the end of studies is denoted z 3. It is 1 if the individual did not receive any degree or obtained a diploma below bachelor level (high school degree or baccalauréat ), 2 if he got a bachelor level and 3 if he has a higher diploma than a bachelor degree. Let z 32 = 1 {z3 =2} denote the binary variable representing the obtention of a of a high school diploma and z 33 = 1 {z3 =3} the binary variables indicating the obtention a diploma higher than a high school degree. The attachment to the state l of the labour market is described by the variable y itl = η l z 2 + µ l z 32 + λ l z 33 + x itβ l + K δ kl y i(t 1)k + ε itl where η l, µ l, λ l, δ kl IR and β l IR p. The residual term ε itl incorporates both an individual effect and a term specific to the period. Thus, one has ε itl = α il + u itl k=0 11

12 where α il represents an individual effect specific to state l of the labour market and u itl is an idiosyncratic error term. The vector x it IR p includes individual exogenous explanatory variables (gender, family status, citizenship). The vector α il includes K composnents of unobservable heterogeneity. We assume that ( αi1, α i2,, α ik ) N(0, Vα ) and that these vectors are independent across individuals. Being able or unable to locate the individual at the exit of the education system, in other words, to determine if he lives in a deprived urban area (Zus) or not, is the result of the dichotomization of the latent variable z1. Thus, z 1 = 1 {z 1 >0}. The localization in a deprived urban area (Zus) at the exit of studies is the result of the dichotomization the latent variable z2. Wa have z 2 = 1 {z 2 >0}. The level of education z 3 can take three possible values. We get that z 3 = 1 {z 3 0} {0<z3 ξ} {z3 >ξ}. A we consider a dynamic model, the initial conditional on the labour market, i.e. y 1 = (y 10,..., y 1K ), must be treated carefully. The solution we use is that proposed by Heckman (1981). Finally, one has to consider the following initial equations: zi1 = w i1 γ 1 + v i1, zi2 = w i2 γ 2 + v i2, zi3 = w i3 γ 3 + v i3, (1) where vectors of error terms are independent normal variables: ( ) v 1, v 2, v 3 N(0, Vv ). γ j IR p j, j {1, 2, 3}, with p j IN. V v is a variance-covariance matrix with the following structure: 1 ρ 0 12 ρ 0 13 V v = ρ ρ 0 23 ρ 0 13 ρ where ρ 0 jk [ 1; 1] is a parameter to be estimated (j, k {1, 2, 3}, j < k). We assume that u itl u i t l, if t t or i i or l l. Under the assumption that the error terms u itl are distributed according a type 1 extreme value distribution, we obtain that: P [Y it =l y i1, y i(t 1), x it, z i1, z i2, z i3, α i ; θ] = exp(η l z i2 +µ l z i32 +λ l z i33 +x it β l+ K δ kl y i(t 1)k +α il ) k=0 K K exp(η j z i2 +µ j z i32 +λ j z i33 +x itβ j + δ kj y i(t 1)k +α ij ) j=0 k=0 (2) 12

13 where η 0 = µ 0 = λ 0 = 0, β 0 = 0 and δ k0 = 0, for k {0, 1,..., K}. These constraints are established in order to obtain the identification of the parameters of the model. We assume, moreover, that α i0 = 0. For the initial distribution in the states of the labour market, we assume P [Y i1 =l x i1, z i, α i ; θ] = exp(ηl 1 z i2+µ 1 l z i32+λ 1 l z i33+x i1 β1 l +ν1 l α il) K exp(ηj 1 z i2 +µ 1 j z i32 +λ 1 j z i33 +x i1βj 1 +νj 1 α ij ) j=0 (3) where ν 1 l IR for all l {0, 1,..., K} and ν 1 0 = η1 0 = µ1 0 = λ1 0 = 0, β1 0 = Structure of the variance-covariance matrix The variance-covariance matrix of the individual effects (α i1, α i2,..., α ik ) can be written: σ 2 1 ρ 12 σ 1 σ ρ 1K σ 1 σ K ρ 12 σ 1 σ 2 σ 2 2. V α =..... σk 1 2 ρ K 1K σ K 1 σ K ρ 1K σ 1 σ K ρ K 1K σ K 1 σ K σk 2 Let us consider the following vector r i = ( v 1, v 2, v 3, α i1,..., α ik ) IR K+3. We are going to assume that r i N(0, Ω) where Ω = V v ρ 21 σ 1... ρ 2K σ K ρ 11 σ 1... ρ 1K σ K ρ 31 σ 1... ρ 3K σ K ρ 11 σ 1 ρ 21 σ 1 ρ 31 σ 1... V α ρ 1K σ K ρ 2K σ K ρ 3K σ K with r i r i, for i i and u i r i ( ρ j,k [ 1; 1]). 2.3 Estimation Likelihood function and initial conditions The initial conditions are treated in the spirit of the method proposed by Heckman (1981). The method consists in correlating freely the error terms of the intial conditions with the individual effects of next dates. We will omit in what follows, the index of the individual. Let us consider the following integration limits: 13

14 { a 1 = w 1 γ 1 if z 1 = 1 and a 1 = if z 1 = 0, b 1 = + if z 1 = 1 and b 1 = w 1 γ 1 if z 1 = 0, { a 2 = w 2 γ 2 if z 2 = 1 and a 2 = if z 2 = 0, b 2 = + if z 2 = 1 and b 2 = w 2 γ 2 if z 2 = 0, { a 3 = if z 3 = 1 and a 3 = w 3 γ 3 if z 3 = 2 and a 3 = ξ w 3 γ 2 if z 3 = 3, b 3 = w 3 γ 3 if z 3 = 1 and b 3 = ξ w 3 γ 3 if z 3 = 2 and b 3 = + if z 3 = 3, If k 4, a k = and b k = +. The domain of integration is denoted A z = K+3 k=1 [a k; b k ]. The contribution to the likelihood function for an individual with a known location in 1998 can be written (z 1 = 1): ˆ b1 ˆ bk+3 T L(θ)= P [Y 1 =y 1 x 1, z, α; θ] P [Y t = y t y t 1, x t, z, α; θ] φ K+3 (r, Ω) d r a 1 a K+3 t=2 ˆ T = P [Y 1 =y 1 x 1, z, α; θ] P [Y t = y t y t 1, x t, z, α; θ] φ K+3 (r, Ω) d r A z t=2 where r = (v, α) and z = (z 1, z 2, z 3 ). Let φ K+3 (., Ω) denote the p.d.f. of the N K+3 (0, Ω) distribution. The contribution to the likelihood function for an individual with an unknown location in 1998 can be written (z 1 = 0): L(θ)= z 2 =0,1 ˆ A z P [Y 1 =y 1 x 1, z, α; θ] (4) T P [Y t = y t y t 1, x t, z, α; θ] φ K+3 (r, Ω) d r (5) t=2 We can notice that the limits of integration depend, at the same time, of the parameters of the model, of the realizations of exogenous explanatory variables and of the values of endogenous variables. Let us assume that we have an i.i.d. sample. Thus, the likelihood function can be written n L(θ) = L i (θ) i=1 where L i (θ) is the expression of the contribution of individual i to the likelihood function. Note that when individual could not be localized in 1998, the contribution to the likelihood is the sum of two components even if it could have been written as an unique integral. We can construct, on such a basis a continuous simulator with respect to the parameters for each one of these probabilities, whereas this cannot be done when the contribution is written as an unique integral. 14

15 2.3.2 Estimation Method The vector θ of parameters can be estimated maximizing the logarithm of the simulated likelihood function: ˆl N,H (θ) = n ln((1 z i1 ) i=1 z 2 =0,1 1 H H h=1 p(r h,z 2 i ; θ) + z i1 1 H H h=1 p(r h,z i2 i ; θ)) (6) where H is the number of random draws used for each individual. r h,z 2 i A zi where z i = (z i1, z 2, z i3 ). We assume that the draws r h,k i = (v h,k i1, vh,k i2, vh,k i3, αh,k i1,..., αh,k ik ) are i.i.d. and specific to the individual. Each drawing r h,k i is obtained using an iterative algorithm described in appendix 1. We use a G.H.K. simulator in order to obtain a given drawing (cf. Gouriéroux et Monfort 1997, Kamionka 1998, Edon et Kamionka 2007 et 2011). The main advantage of this technique is that it provides a no-rejection algorithm. In order to take into account the characteristics that r h,k i A zi for all h {1,..., H}, ; θ))where we use an unbiased simulator of P [r A zi ]. This simulator is denoted p(r h,k i the expression of p(r h,k i ; θ)) is given in the appendix 1. In order to obtain the value of the simulated maximum likelihood estimator of the vector of parameters θ, we maximize the logarithm of the simulated likelihood function (6). This estimator of the parameters of the model is consistent and efficient if H/n 0 when n, H + (cf. Gouriéroux and Monfort, 1997). In practice, many authors have underlined the quality of the behavior of this estimator for a reasonable and fixed number drawings (for instance, H = 30). The value of the estimator does not change much if we increase then the number of drawings we obtain for each individual (cf. Kamionka and Lacroix, 2008, Edon and Kamionka, 2011, Gilbert, Kamionka and Lacroix, 2011). 3 Estimation results We focus, in this section, on the presentation of results obtained from the estimation of parameters of the model presented in the previous section 7. We chose to retain as explanatory variables a set of individual characteristics such as family status, gender, educational level, geographic origin, parental occupation, citizenship of parents. We introduce in the specificaiton, in addition, a dynamic component related to the history on the labour market. To do this, we consider a form of true state dependence through 7 For reasons related to technical limitations of the total physical memory (RAM) addressed by the version of Gauss that we used, we realized the estimates using a random sample of size N = 5000 representative of the initial sample and with a number of drawings specific to the individual H =

16 lagged variables corresponding to the past realizations of the position on the labour market. We will examine, first, the initial conditions. As the model is dynamic we need to consider the initial position on the labour market with special attention. The method we use is in the spirit of what has been proposed by Heckman (1981). However, we can not limit ourselves to describe or formalize the position on the labor market at the time of the first survey. Indeed, it can be expected that, jointly, the initial location, the educational level and the non response are not exogenous relatively to the situation the labour market. Then, we have considered all of these variables as components of these initial conditions and used a proper treatment of these quantities. We then comment the results associated to the conditional modeling of the individual histories on the labour market. We conclude, finally, by examining the structure of the correlations between the variables described by the model and the structure of the correlations between the unobservable individual characteristics. 3.1 Modeling of initial conditions We have modeled the initial conditions using a system of equations described in the previous section (cf. equations 1 and 3). Three variables were used: the initial location - at the end of the study - in a deprived area or not (ZUS), the response or non-response to the question on the original location and, finally, the education level. In the conditional model we introduce some individual characteristics that have an explanatory power on these three variables but not on the labour market history that follows conditionally to the initial position. First, we use an indicator of the geographical origin of the parents: it is the parents nationality that appears in the equation associated to the location at the exit of the education system. An indicator of mobility, built using the information indicating whether or not a person has left the town in which she lived in 6th grade 8, is used in the equation for the response or non-response to the question about the location in a deprived area (ZUS). The socio-occupational category of the parents is used in the equation describing the level of education. The degree level can take three values: without a degree, a high school degree (baccalauréat), more than a high school degree Diploma, initial location (Zus) and non response Young people living in a couple respond less frequently to the question relative to location in a deprived area (cf. table 4). There is no significant gender difference. We can hypothesize that this non response (or difficulty of coding) is related to the geographical mobility of individuals. This explanation is consistent with the sign and the importance of parameter of the variable lack of mobility since the 6th grade which is positive and highly significant. It is possible that cohabitation is frequently associated to geographic mobility. Young people born outside Europe and, in particular, those born in North Africa, live more often in deprived urban areas. This is a signal of the presence of a form of urban 8 Classe de sixième". 16

17 segregation. Women more often live in deprived urban areas. This may result from the fact that they are less frequently active and therefore less mobile for professional reasons. Young couple more frequently live off deprived urban areas. Young people who leave the home of their parents when they decide to live in couple frequently leave the neighborhood of origin and settle in a less deprived environment. We will see that this result is consistent with the observation that the most educated people and having, therefore, more revenue at the beginning of working life are more mobile. We can also verify that the young people living in couples are in average more graduated. When they stop their studies, youths born of French parents reside less frequently in deprived urban areas. We find once again this signal of urban segregation that can be explained by the fact that their parents have higher average incomes. As we have already pointed out, youth living in couple are more frequently graduated. Women of this generation often have a better level of education. It was noted that women are on average more graduated when leaving the education system (cf. Ministère de l éducation Nationale, 2009). This result can be compared to similar results obtained on the NLSY data for the U.S. and the same generation (cf. Reynolds et Pemberton, 2001). Young born in North Africa are on average less educated than those born in Europe. These young people often live in environments - neighborhoods - more deprived and families who have less knowledge of the French education system. Thus, on the impact of neighborhood effects between young people in the same neighborhood on educational outcomes, Goux and Maurin (2007) show, using data from the French employment surveys (Insee), that these interactions are important. On the contrary, young foreignborn outside Europe and North Africa are no less or more graduated than those born in Europe. A priori, parental occupation is informative of chances to leave the education system graduated or with a high diploma when this individual has obtained a diploma. For example, a child whose father is a worker have on average a lower level of education. In contrast, youth whose father or mother is executive or engineer or teacher or professional, in average, have a higher degree Initial position on the labor market In order to start, we can notice that young people leave school with at least the baccalauréat in hand, are, in December 1998, more frequently inactive (cf. tableau 3). It should be noted that the date of December was chosen to ensure that the vast majority of these young people, to this date, completed their studies. Entry into the active life is thus not automatically obtained after graduation but can be shifted by some months. Women are, at the output of the studies, more frequently active than men, and more particularly, more frequently employed. This reflects an integration of women into the labor market faster the first few months but does not continue thereafter. They also are more likely than young men to occupy a subsidized employment the first weeks of their working life. The youngs who are living in couple are more likely to occupy an employment immediately after leaving school. This may be the result of a more intensive job search the first few weeks that is associated, as we have already noted, to a greater geographical 17

18 mobility. Young people living in deprived urban areas at the output of studies occupy less frequently than other a job in the first weeks following their entry into the workforce. This demonstrates the difficulties that young people from these neighborhoods in entering the labor market. This may be the result of a lack of employment opportunities at the local level or a lower mobility of these young or, even, of a discrimination in hiring (ref. Duguet et alii, 2012). Difficulties encountered by young people in integrating the labor market have already been highlighted in the literature (cf. Fougère et Kamionka, 1992). However, it is notable that these young people are less likely to be employed on a temporary contract while one might expect that access to these jobs to be easier. 3.2 Trajectory on the labor market We now consider the parameters of the transition probabilities between the states of the labor market. These states are unemployment, subsidized employment, fixed term employment contract (Temporary Employment), unlimited duration employment (Unlimited duration contract). The time interval between two successful observations of the transition process is 6 months. We have two annual observations of the position on the labor market in December and June State dependence Let us remark, first, that the modeling approach allows us to distinguish the true state dependence from the spurious state dependence. Indeed, our model incorporates both the unobservable heterogeneity and lagged values of the indicators of the state of labor market. This specification is similar to those used, for example, by Hyslop (1999) or Edon and Kamionka (2007) but is more general in the sense that the model considered here is based on more than two states. We can observe the existence of a diagonal effect (see tableau 2): there is a strong persistence in the states of the labor market. This effect is particularly strong for unlimited duration contracts and subsidized jobs. This persistence is lower for unemployment. The persistence of stays in unlimited duration jobs is due to the legal stability of these contracts. The persistence of subsidized employments is more surprising and may result in both the average duration of these contracts, but also the existence of individuals who chain several episodes of subsidized jobs. Persistence of stays in the states of the labor market may result from the existence of search costs (cf. Hyslop, 1999). One can notice that transitions to employment (Unlimited duration contracts, Limited duration contracts and Subsidized employments) are common from unemployment. This phenomenon results from the search intensity that is higher for unemployed persons. Transitions between subsidized employments and unemployment are also frequent attesting the existence of unstable trajectories. (see Fougère et Kamionka, 2003). Interestingly, for these youth, exits from employment to unemployment are not significantly different according to whether the individual has a Limited duration contract or a Unlimited duration contract or occupies a subsidized employment. This reflects the time 18

19 required so that the relationship between these workers and employment stabilizes (cf. Fougère et Kamionka, 1992, Topel et Ward, 1992, Magnac, 2000). A passage through subsidized employment does not significantly improve integration on the labor market with respect to a period of unemployment or a sojourn in an employment with a limited duration contract (cf. table 3). The data from the Generation surveys do not allow to distinguish government sponsored employment programs, but from these data it is possible to obtain an evaluation of the overall effectiveness of employment programs for young people in effect during the period ( ) 9. Young people who come from subsidized employment do not enter significantly more into stable employment than those who arrive from unemployment. Subsidized jobs allow young people entering the workforce to avoid situations of unemployment without any benefits at ages where they had no right - on the period - to any welfare benefits (the RMI, which has since been replaced by the RSA). It may be noted, also, that young people who make a transition from a subsidized employment, enter significantly less frequently in an employment with a limited duration contract compared to those who are coming from unemployment. If subsidized jobs are a way to accumulate human capital, they do not allow to enhance this investment immediately Individual characteristics Young people living in couple are more often employed (on unsubsidized employment) relatively to an individual living alone (cf. table 2). This is in line with what we found for the initial conditions on the labour market (cf. table 3) and reflects a greater intensity of job search for people who live with a partner. This result is also consistent with the existence of a greater geographic mobility of young people. Women are less frequently employed than men. Thus, women lose during the insertion process on the labour market the initial advantage they had on young men. This may be due to the difficulties which they meet to conciliate family life and professional life (cf. Edon et Kamionka, 2007, 2011, Hyslop, 1999). This can also be the result of a form of discrimination against women (see also Havet, 2006). Graduates occupy more often a stable job (unlimited duration contract), but less frequently an unstable job (limited duration contract). Graduates - they have at least a bachelor s degree - are less frequently unemployed. In addition, graduates occupied more frequently than non-graduates an employment. This may be the result of self-selection of graduates - a better understanding of these programs - or, conversely, this can be due to a selection at the entry in these employment programs. Consequently, these employment programs could not succeed to target less skilled workers. Young born in North Africa are more frequently unemployed and, to a lesser extent, are more often in an employment with a limited duration contract. This can result in a form of discrimination against these youth. The youths who lived, at the time of exiting school, in a deprived urban area, occupy 9 Other accompanying programs for the unemployed individuals were also evaluated: cf. Fougère, Kamionka et Priéto (2010). 19

20 less frequently jobs with a limited duration contract. The effect is particularly significant and illustrates the difficulties faced by young people from deprived urban areas in entering the labor market. Let us note that the initial conditions of localization persist over time: the study period covers, indeed, the first ten years of working life. The phenomenon also affects jobs that can allow, that is what is generally considered, to enter more easily in stable employment through the accumulation of human capital achieved during these periods of employment with limited duration contracts. 3.3 Structure of correlations Correlations between the initial conditions Let us consider the correlations between diploma, location in sensitive urban areas at the exit of the education system and the response indicator to the question on the localization (see Table 6). The estimated correlation between the indicator of the response to the localization and the localization in a deprived urban area is positive. The estimated relationship between the indicator of the response to the localization and level of education is negative. One can hypothesize that the most geographically mobile people are also, on average, better educated, and are those for which the information relative to the localization at the end of the studies is less often available. The correlation between level of education and localization in deprived areas is negative and significantly different from zero. Those who benefit from an better educational environment reside less often in deprived urban areas Correlations between individual effects We found positive correlations between the unobservable characteristics corresponding to the level of education and those associated with temporary employment and unemployment (cf. table 7). Graduates have an important opportunity cost to remain inactive. There are positive correlations between the unobservable characteristics relative to localization in deprived areas and those associated with temporary employment and unemployment. It is interesting to note that the fact of living in a deprived urban area has a negative impact via true state dependence (cf. table 2) that probably reflect a form of discrimination or, at least, a difficulty in accessing employment. We see here that the localization has a positive effect via spurious state dependence that might be explained by a motivation on average greater of these young people to find a job. We found, finally, positive correlations between the unobservable characteristic associated with the response to the question about the geographic localization and those relative to temporary employment and subsidized employment: individuals who are less mobile geographically are also those who accept more frequently a temporary job Correlations and variances of individual effects The variances of the individual effects in employment with limited duration contract and unemployment are important, they reflect greater heterogeneity of individuals in these 20

21 states. (cf. table 5). There is a positive correlation between individual effects related to permanent employment and subsidized employment, firstly, and a positive correlation between the random effect associated to permanent employment and the random effects associated with unemployment and subsidized employment, secondly. This result can be explained by an heterogeneity of individuals passing through subsidized employment: those who are more qualified who pass through subsidized employment to avoid unemployment, on one hand, those who are less skilled who alternate some episodes of unemployment and various forms of unstable jobs, on the other hand Random effects coefficients in initial conditions The share of the total variance explained by individual effects decreases over time for stable employment and unemployment. It increases over time for precarious employment and subsidized employment (cf. table 8). For instance, we have less and less young skilled unemployed gradually as time passes. In stable employment, the share of the most qualified young people in total employment increases significantly over time (initially, we have probably has a larger share of low-skilled in employment because of shorter duration of some of vocational courses such those delivering a vocational technical degree as a "CAP"). Conclusion In this article we investigate the integration on the labor market of young people leaving the education system. The study allows us to study the impact of the initial localization in a deprived urban area on the transitions on the labor market. We also evaluate for a recent period to the impact of subsidized employments to determine whether or not these jobs facilitate the integration of these young people in stable employment (jobs with an unlimited duration contract). Dynamic modeling that we use allows to describe transitions between states of the labor market and to distinguish stable employment from precarious employment episodes (jobs with a limited duration contract). We also distinguish the episodes in unemployment from those in subsidized employment. By comparing the intensities of transitions between unemployment and stable employment, on one hand, and the intensities of transitions between subsidized employment and stable employment, on the other hand, we can make an assessment of the impact of subsidized employment over a recent period of time. Subsidized jobs do not improve the integration with respect to unemployment. We also study the probability of transition in subsidized employment based on individual characteristics and and we show that these subsidized jobs benefit more to graduated youth emphasizing, thus, the lack of targeting of these programs. This model allows us to distinguish between true state dependence and spurious state dependence resulting from the presence of omitted individual heterogeneity may summarize, for example, the individual preferences. We model the impact of localization in a deprived area at the exit of studies. This localization has a persistent negative effect 21

22 on transitions from unemployment to precarious employment. This result suggests that young people from these areas face particular difficulties in entering the labor market. The signs of the correlations between habitat in a deprived urban area, temporary employment and unemployment is consistent with the assumption of a greater motivation on average to find a job for young people from these neighborhoods and even if they are in average less qualified. An extension would be to compare the results with those from a discrete modeling of unobservable heterogeneity. One could also study the integration of youth into the labor market using a fixed effects approach (cf. Beck and Kamionka, 2012). This approach does not allow, however, to study the impact of fixed characteristics over time - as the initial localization - on the transition probabilities. A Modeling with fixed effects could nevertheless provide a complementary perspective on the integration process of young people. It would be interesting to implement programs to promote residential mobility on an experimental basis. Such programs have been evaluated in the USA (cf. Quigley et Raphael, 2008, Kling, Liebman et Katz, 2007) in a different context. 22

23 Bibliographie Aeberhardt R., Coudin E. and Rathelot R. (2010). Les écarts de taux d emploi selon l origine des parents: comment varient-ils avec l âge et le diplôme?, France, Portrait Social, édition 2010, Aeberhardt R., Fougère D., Pouget J. and Rathelot R. (2010). L emploi et les salaires des enfants d immigrés, Economie et Statistique, , Beck S. and Kamionka T. (2012). Mobilités, Inégalités et trajectoires professionnelles, to be published in Revue Economique. Card D. and Sullivan D. (1988). Measuring the Effect of Subsidized Training Programs on Movements in and out of Employment, Econometrica, vol. 56(3), Couppié T., Gasquet C. and Lopez A. (2007). Quand la carrière commence... Les sept premières années de vie active de la Génération 98, Cereq, Enquête Génération 98. DARES (1997). 40 ans de politiques de l emploi, La Documentation Française, Paris. Duguet E., L Horty Y., du Parquet L., Pascale P. and Sari F. (2012). Les effets du lieu de résidence sur l accès à l emploi : une expérience contrôlée sur des jeunes qualifiés en Ile-de-France". To be published in Economie et Statistique. Havet N. (2006). L insertion professionnelle des jeunes et mesures publiques : des trajectoires différenciées entre hommes et femmes, Annales d Economie et Statistique, vol.81, Havet N. and Lacroix G. (2013). La formation continue, un moyen de réduire les inégalités salariales entre hommes et femmes?, Revue Economique, vol.64, 2, Jones S. R.G. and Craig Ridell W. (1999). The Measurement of Unemployment: An Empirical Approach, Econometrica, vol. 67(1), Edon C. and Kamionka T. (2007). Modélisation dynamique de la participation au marché du travail des femmes en couple, Annales d Economie et de Statistique, , Edon C. and Kamionka T. (2011). Modélisation dynamique de la fécondité et de la participation au marché du travail des femmes en couple, Revue Economique, vol. 62, n 3, Fougère D. and Kamionka T. (1992). Un modèle markovien du marché du travail, Annales d Economie et de Statistique, vol. 27, Fougère D. and Kamionka T. (2003). Bayesian inference for the mover-stayer model in continuous time with an application to labour market transition data, Journal of Applied Econometrics, 18, Fougère D., Kamionka T. and Priéto A. (2010). L efficacité des mesures d accompagnement sur le retour en emploi, Revue économique, 61, 3,

24 Fougère D., Kramarz F. and Magnac T. (2000). Youth Employment Policies in France, European Economic Review, 44, Geweke J. (1991). Efficient simulation from the multivariate normal and student-t distributions subject to linear constraints, Computing Science and Statistics: Proceedings of the twenty-third symposium on the interface. Computing science and statistics, American statistical association, Alexandria, Gilbert L., Kamionka T. and G. Lacroix (2011). The Impact of Governement- Sponsored Training Programs On The Labor Market Transitions of Disadvantaged Men, in Advances In Econometrics, Theory and Applications, Edited by Miroslav Verbic, Gobillon L., Magnac T. and H. Selod (2011). The effect of location on finding a job in the Paris region. Journal of Applied Econometrics, Vol. 26, No. 7, 2011, Gobillon L., Magnac T. and H. Selod (2011). Do unemployed workers benefit from enterprise zones: the french experience, CEPR Working Papers Gouriéroux Ch. and Monfort A. (1997). Simulation-Based Econometric Methods, Core Lecture Series, Oxford: Oxford University Press. Goux D. and Maurin E. (2007). Close Neighbours Matter: Neighbourhood Effects on Early Performance at School, Economic Journal, 117, Havet N. (2006). L insertion professionnelle des jeunes et mesures publiques : des trajectoires différenciées entre hommes et femmes, Annals of Economics and Statistics, 81, Heckman J. J. (1981). The Incidental Parameters Problem and the Problem of Initial Conditions in Estimating a Discrete Time-Discrete Data Stochastic Process, in Structural Analysis of Discrete Data, eds. C. Manski et D. McFadden. Cambridge, MA, MIT Press. Hyslop D. R. (1999). State Dependence, Serial Correlation And Heterogeneity in Intertemporal Labor Force Participation of Married Women, Econometrica, 67, Kamionka T. (1998). SML Estimation in Transition Models, The Econometrics Journal, 1, C129 C153. Kamionka T. and Lacroix G. (2008). Assessing the External Validity of an Experimental Wage Subsidy, Annales d Economie et de Statistique, 91-92, Kling J.R., Liebman J.B. and Katz L.F. (2007). Experimental Analysis of Neighborhood Effects, Econometrica, 75, 1, Magnac T. (2000). Subsidised Training and Youth Employment: Distinguishing Unobserved Heterogeneity From State Dependence in Labour Market Histories, Economic Journal, 110,

25 Manning A. and Swaffield J. (2008). The Gender Gap in Early-Career Wage Growth, The Economic Journal, Vol. 118, 530, Ministère de l éducation Nationale (2009). Filles et Garçons sur le chemin de l égalité de l école à l enseignement supérieur, DGES, DGES, DEPP, ouvrage collectif. Nauze-Fichet E. and Tomasini M. (2002). Diplôme et insertion sur le marché du travail : approches socioprofessionnelle et salariale du déclassement, Economie et Statistique, 354, Quigley J.M. and Raphael S. (2008). Neighborhoods, Economic Self-Sufficiency, and the MTO program, Brooking-Wharton Papers on Urban Affairs, Rathelot R. (2010) Une autre utilisation des aires de l enquête Emploi : l étude des écarts d emploi et de salaire selon l origine et le quartier de résidence, Courrier des Statistiques, 129. Rathelot R. and Sillard P. (2009). Zones Franches Urbaines : quels effets sur l emploi salarié et les créations d établissements?, Economie et Statistique, , Reynolds J.R. and Pemberton J. (2001). Rising College Expectations Among Youth in the United States: A Comparison of the 1979 and 1997 NLSY, The Journal of Human Ressources, 36, 4, Selod H. and Zenou Y. (2006). City Structure, Job Search, and Labour Discrimination. Theory and Policy Implications, Economic Journal, 116, Topel R.H. and Ward M.P. (1992). Job Mobility And The Careers of Young Men, The Quartely Journal of Economics, 107, 2,

26 Appendixes Appendix 1: GHK algorithm (adaptation) We have to obtain drawings of the vector r = (v 1, v 2, v 3, α 1, α 2,..., α K ) such that r ɛ A = K+3 k=1 [a k; b k ], given r N(0, Ω). We use then a Cholesky decomposotion of the matrix Ω in order to obtain a variable u = (u 1,...u K+3 ) such that r = Lu. L is lower triangular matrix, with : l L = l 21 l l K+3,1 l K+3,2... l K+3,K+3 We must have : a 1 < v 1 < b 1 a 2 < v 2 < b 2 a 3 < v 3 < b 3 a 4 < α 1 < b 4. a K+3 < α K < b K+3 a 1 < u 1 < b 1 a 2 < u 2 + δ 21 u 1 < b 2. a K+3 < u K+3 + δ K+3,1 u δ K+3,K+2 u K+2 < b K+3 where δ k,j = l k,j /l j,j, a k = a k/l k,k, b k = b k/l k,k The algorithm incorporates a sequential structure. We generate successively the variables u k according to a standard normal distribution truncated on the interval [a k (δ k,1u δ k,k 1 u k 1 ); b k (δ k,1u δ k,k 1 u k 1 )], for k = 1,..., K+3. Let p(r h ; θ) = g(r h, θ) [Φ(b 1 ) Φ(a 1 )] K+3 k=2 [ Φ(b k (δ k,1 u h δ k,k 1u h k 1 )) Φ(a k (δ k,1u h δ k,k 1u h k 1 ))] (7) where g(r h, θ) = P [Y 1 =y 1 x 1, z, α h ; θ] T P [Y t = y t y 1,..., y t 1, x t, z, α h ; θ] t=2 26

27 Appendix 2: Descriptive Statistics fig 1. Insertion of young people by location and gender 27

28 fig 2. Insertion of young people by location and diploma 28

29 fig 2. Insertion of young people by location and diploma (continued) 29

30 fig 2. Insertion of young people by location and diploma (continued) 30