Returns to experience and seniority: Evidence from Spanish Social Security Data Joaquin Garcia-Cabo⇤ November 30, 2017

Very Preliminary–Do Not Cite Abstract In this paper, I use the framework described in Buchinsky et al. (2010) to investigate the returns to seniority and experience in Spain. I model the participation/employment and interfirm mobility decision, together with a wage equation and estimate the three-equation system simultaneously using data from the Spanish Social Security Administration on workers’ labor market histories. I find that returns to experience for workers with 10 years of experience are 30%-60% lower than in the US for workers in each of three different education levels studied. I next extend the model to allow for experience to be accumulated under contracts with different firing costs and find that marginal returns to experience under fixed-term contracts are negligible after three contracts of this type.

JEL Classification: J08, J24, J30, J41 Keywords: labor economics, labor contracts, human capital

⇤

Department of Economics, University of Minnesota. E-mail: [email protected]

1

1

Introduction

There has been a debate in the literature in the sources of wage growth. While Topel (1991) found large returns to firm-specific human capital, other estimates by Abraham and Farber (1987) and Altonji and Williams (1997) suggested specific human capital accumulation was lower. Buchinsky et al. (2010) pointed out the endogeneity present in the previous studies from ignoring mobility decisions of workers, and estimate a wage equation together with participation and mobility decisions of workers with data from the Panel Study of Income Dynamics. Their results suggest that returns to specific skills are larger than those previously found. Lagakos et al. (2017) document that wages grow susbtantially more in rich countries than poorer countries, which can be explained by both human capital and search theories. Using Spanish Social Security data on worker’s wages and labor histories, I analyze the role of returns to general (experience) and specific (tenure) human capital in Spain using the same approach presented in Buchinsky et al. (2010). I find that the cumulative returns to 5 years of experience in the labor market are more than twice as large for college graduates compared to high school drop-outs, and 60% larger compared to workers with some college. However, marginal returns to experience basically disappear after 10 years of labor market experience for the most educated group of workers, while high school drop-outs still accumulate returns and decrease the differences with respect the other two groups. Next, I compare the role of general and specific human capital in wage growth in Spain and the US. The long-term returns to both experience and tenure in Spain are smaller than those for the US for all education groups. Returns to experience are between 30%-62% higher in the US for workers with 10 years of labor market experience, depending on the education group. The returns to tenure are also lower in Spain, with returns after 10 years in the same job being 30-54% of the returns for workers in the same educational group in the US. The Spanish labor market exhibits dual or two-tier labor market: highly-protected, long2

term employment relationships or permanent (open-end) workers; and workers with insecure, short-term employment relationships, or temporary (fixed term) workers. Finally, I extend the baseline model to distinguish between labor market experience and tenure accumulated under each type of contract. I estimate the model and find that the market penalizes workers with more than 7 years of labor market experience under fixed-term contracts, or in other words, three contracts of this type (the maximum duration is 2 years).

2

Econometric specification

In this section I describe the empirical model used to estimate the returns to experience and tenure. I follow Buchinsky et al. (2010) and define a system of three equations: a wage equation, participation equation and mobility equation. The first equation builds on a specification of the wage function that has been adopted in the literature of wage determination. The observed log wage equation for individual i in job j at time t is given by:

⇤ wijt = wijt 1 (yit = 1)

0

⇤ wijt = xwijt

0

(1)

W + Jijt + ↵wi + ⇠ijt

0

where xwjit is a vector of observed characteristics of the invididual in his/her current job. It includes sex, education, labor market experience and tenure among others. The subscript w on this vector refers to the observed characteristics in the wage equation. 1(·) is an indicator function and it takes value one when the individual participates in the labor market at time t, this is, when yit = 1, and 0 otherwise. Therefore, the wage offer is observed only if the individual chooses to work. The remaining elements in the wage equation are ↵wi which denotes a person specific fixed-effect, ⇠ijt is a contemporaneous idiosyncratic error

3

W term, and finally Jijt is a piecewise linear function that provides a summary statistic for the W individual labor market history. In particular, Jijt captures the timing and magnitude of all

the discontinuous jumps in the individual’s wages resulting from job all changes until time t. This summary statistic is approximated by a piece-wise linear function of experience and seniority:

W Jijt =(

s 0

+

e 0 ei0 ) di1

+

" 4 Mit X X l=1

(

k=1

k0

+

s k si,tl 1

+

e k ei,tl 1 ) dkitl

#

(2)

where di1 is a dummy that equals 1 if the individual i changed jobs in the first sample year, 0 otherwise. d1itl equals 1 if the l-th job of individual i lasted less than a year and 0 otherwise. Similarly, d2itl equals 1 if the l-th job of individual i lasted between 2 and 5 years, and 0 otherwise, d3itl equals 1 if the l-th job of individual i lasted between 6 and 10 years and 0 otherwise, and finally d4itl equals 1 if the l-th job of individual i lasted more than 10 years and equals 0 otherwise. Mit denotes the number of job changes the individual i has experienced throughout his career up to time t, without taking into account changes that occur in the individual’s first sample year. The quantities eitl and sitl denote the individual’s experience and seniority in year tl , respectively, when the individual i leaves job l. Even though the

’s are fixed parameters (13 identifiable parameters in total), the size of the

jumps will depend on the different values that experience and seniority can take when the invididual changes a job. This function captures the initial conditions specific to the individual at the start of a new job. It provides a measure of the opportunity wage of the worker if he/she were to change jobs. It captures differences between workers displaced from the firm and those who move directly to another firm from their current job. It also allows to control for the quality of the match by including tenure at every job. The model is completed by adding the participation decision of the worker and the mobility decision. 4

First, a participation (employment ) equation at any t > 1 is given by: yit = 1 (yit⇤

0

w yit⇤ = a0 Jijt + xyit

0

+

(3)

0)

y yi,t 1

+

m mi,t 1

+ ↵yi + uit

where yit⇤ is the latent variable for participation. The second equation is an inter-firm mobility equation, at any date t > 1 is given by: mit = 1 (m⇤it

0) 1 (yit

0

w m⇤it = a1 Jijt + xmit

0

+

1

(4)

= 1, yit = 1)

m yi,t 1

+ ↵mi + ⌫it

where m⇤it is the latent variable for mobility. Given the equation for participation, we can only observe a worker moving if he was participating both in t

1 and t.

Individuals in the sample are at different points in their labor market histories. It is crucial to control for initial conditions, and this is done by following Heckman (1981) and approximate the initial condition using a probit specification for participation and mobility: ⇤ yi1 = 1 (yi1

mi1 = 1 (m⇤i1

⇤ 0) , where yi1 = a0 xyi1 +

y ↵yi

(5)

+ ui1

0) 1 (yi1 = 1) , where m⇤i1 = b0 xmi1 +

m ↵mi

and where ↵yi and ↵mi are the individual fixed effects defined in (2) and (3). This is, m

(6)

+ ⌫i1 y

and

are allowed to differ from 1. The authors impose orthogonality conditions on the idiosyncratic shocks. In particular,

5

the fixed effects are stochastically independent of the idiosyncratic shocks

↵i ? (uit , ⌫it , ⇠ijt ) and ↵i follows the distribution given by:

↵i ⇠ N (f (xi1 , ..., xiT ) , i

= Di

{

⇢ Di ,

⇢ }j,l

Di = diag (

(7)

i ) , where

yi ,

mi ,

wi ) ,

and

= ⇢↵ji ↵li , f or j, l = y, m, w.

where the function f depends on all exogenous variables in all of the periods. ji 2 li

is allowed to be heteroskedastic and depend on xyit , xmit and xwit . In particular,

= exp (hl (xi1 , ..., xiT )) ,for l = y, m, w and h (·) is a real-valued function. 0

Finally, the idiosyncratic error term ⌧ij = (uit , ⌫it , ⇠ijt ) from the participation, mobility and wage equation respectively are assumed to be contemporaneously correlated white noises with ⌧it ⇠ N (0, ⌃) where

and

2 u

=

2 v

0

⇢uv ⇢u⇠ ⇠ B 1 B ⌃=B 1 ⇢v⇠ ⇠ B ⇢uv @ 2 ⇢u⇠ ⇠ ⇢v⇠ ⇠ ⇠

1 C C C C A

(8)

= 1 for identification purposes. Additional exclusion restrictions and sources

of identification are described in the appendix.

6

3

Data and estimation

3.1

Muestra Continua de Vidas Laborales (MCVL)

This paper intends to provide a comparison between the US and Spain in terms of returns to experience and tenure. Buchinsky et al. (2010) use PSID data from 1975 to 1992. I replicate their results using the data and perform the same exercise for Spain. Next I outline the data sources and variables used in the estimation. I use Social Security data from Spanish taxpayers, waves 2006-2015. The dataset is known as “Muestra Continua de Vidas Laborales”. The dataset has three key characteristics: (i) large sample size; (ii) longitudinal study; and (iii) the administrative nature of this data. This is a sample of 4 percent of Spanish taxpayers for a given year (approximately 1.2 million individuals), reducing the sample size limitations that surveys with smaller samples have. It is a longitudinal dataset, which allows us to follow the working histories of all individuals, starting from 1980. This is, once the worker is in a wave, we can observe all the labor history associated with that worker. Finally, the data is provided by the Social Security Administration from administrative records, reducing substantially the measurement error from survey data.1 Regarding the population and content of the data, the reference population of taxpayers includes individuals that worked at least a day during the reference year, including selfemployment. It excludes individuals with provided health insurance or non-contributory subsidies, as well as individuals without any connection to the SSSA. The dataset contains monthly wage data back to 1980. There is an entry for each contract under which the worker has been employed and the type of contract. Associated with the contract, the dataset also reports the start date of the contract, the date where the contract was finalized, and the cause of dismissal, among other relevant variables that I will describe later in detail.2 . This 1

Although there is information prior to 1980, the further back in time the more subject to measurement error. Most of the studies that have used this dataset acknowledge the limitations of the data as one goes back to 1980. 2 There is information regarding the location, size and sector of the firm, particular characteristics about

7

is an improvement with respect to other datasets used in the literature, since I am able to exactly identify those workers displaced apart from those who quit or move from job to job, as well as calculate exact experience and tenure variables at each point in time.

3.2

Sample and variable construction

In order to be as close as possible to the exercise performed by Buchinsky et al. (2010) with PSID, I use a sample of 25-54 year old men during 1995-2014.3 . I have excluded workers that have been self-employed or that haved worked in Agriculture or in the Fishing industry at some point in their careers (this is, I only take workers in Régimen General). I also restrict the sample to workers that I observe for at least three consecutive years, and those with no missing information. This leaves the sample to 84,955 individuals. I perform a quarterly analysis4 , where I observe an entry for each contract under which the worker has been employed, the type of contract, and the wages obtained. It also reports the start date of the contract, the date where the contract was finalized, and the cause of dismissal. In the event of an unemployment spell, the duration and benefits collected are available. I also observe the anonymized identifier of the employer and the database contains information regarding age, tenure, sex, and education level, among other variables. Constructing tenure and experience variables is key in this setting. Papers that use survey data often have missing or contradicting information regarding the tenure of the worker at the firm. The typical solution is to proceed as suggested by Topel (1991): for jobs that begin in the panel, tenure starts at zero and incremented by 1 for each additional year in which the person worked for the same employer. If the job started before the sample sarted, tenure the worker on the contract (full or part-time, if the worker has a disability), and professional category of the worker as described in the contract. 3 Although it is true that the time period is different and their time of the analysis ends where mine starts, it is hard to extend it in the US to the present since PSID stopped being an annual survey in 1997, when it became bienial, making it harder to identify participation and mobility, as well as construct tenure and experience correctly. I try to construct the exercise the same way and the goal of this paper is to provide the literature with some estimates for a country different than the US where administrative data is available. 4 The dataset is monthly but I collapse the panel into a quarterly an annual one. I am currently repeating the exercise by aggregating the data at the annual level and compare the results.

8

is imputted from the longest sequence of consistent observations. However, given the characteristics of my data, I am able to construct at each point in time the exact tenure of the worker at the firm, since I observe the start day, month ,and year date (as well as end date) of the contract . Tenure starts at 0 the day the contract starts and I add every month the days worked while the contract is active. Then I can transform this variable into a quarterly or yearly measure of tenure. I construct experience in a similar fashion. For each worker in the sample, I observe the entry year into the labor market. I construct two measures of experience. The first and preferred one is to construct labor market experience as the difference between the year of entry and the current year. This is closer to how experience is usually computed in other studies, by defining potential experience as the difference between the age minus years of schooling, and also the total number of days worked since the worker joined the labor market. The second measure, or actual experience adds the actual days worked since joining the labor force, without taking into account episodes of non-participation. I divide the sample into three education groups: those with less than high school or equivalent (12 years of schooling), those with high-school and less than a college degree (12-16 years of schooling), and those with college degree or higher (16 or more years of schooling). A worker will be defined as participating if he is at least working for 25% of the days of a quarter, this is 23 days. A worker mobility decision will take value 1 if he was participating in t

1 and in t but the employer in t

1 is different from the employer in t. The dataset

provides nominal monthly earnings so I deflate them using the Spanish CPI with base year 2006 provided by the National Institute of Statistics (INE), and the reference currency is Euros. I define wages as monthly earnings divided by the number of days worked.

9

Table 1: Summary Statistics for the MCVL extract for selected years and variables Variable

1995

Year 2005

2014

Age

42.05

38.29

36.84

(7.47)

(9.79)

(7.43)

Education Experience Tenure

Temporary (%) Monthly Earnings

No. of observations

3.3

8.33

9.94

10.50

(3.94)

(3.79)

(3.91)

11.56

11.44

6.99

(4.10)

(9.00)

(6.58)

6.84

6.59

3.21

(5.58)

(7.73)

(4.90)

1.02%

22.11%

47.79%

1260.7

1786.6

1400.51

(460.9)

(793.4)

(808.59)

11098

12022

16797

Descriptive statistics

Table 1 provides a summary statistics of some relevant variables of the sample. The average age decreases over time, since our restriction of age will imply that more individuals join the dataset in later years, and old ones will progressively leave the labor force. Average education raises by more than two years, since newly entrants tend to have a higher level of education. Average experience decreases from 11.56 years in 1995 to 7 years in 2014, which is expected since average age decreases. Tenure also goes down, and this fact is attributable to two factors: the increase in temporary forms of employment experienced durin the 2000’s and the impact of the Great Recession on the destruction of jobs after 2007 (when unemployment rose to more than 20%). An interesting fact is that monthly real earnings increase between 1995 and 2014, but so does the dispersion, which is consistent to what has happened in the US and evidence for Spain from Bonhomme and Hospido (2013).

10

3.4

Model estimation and identification

The model is estimated using a MCMC algorithm to summarize the joint posterior distribution of the parameters of the model. In particular, I denote the set of parameters to be estimated as ✓, and let ⇡ (✓) be the prior density of the model’s parameters. Given observed data z, the posterior distribution of parameters is given by ⇡ (✓ | z)

Pr (z | ✓) ⇡ (✓). I follow

the approach presented in Chib and Greenberg (1998) and augment the parameter space to include the latent variables zit⇤ = (yit⇤ , m⇤it , wit⇤ ) and be able to implement the Gibbs sampler5 more easily. This is done because of the practical intractability of Pr (z | ✓) and the difficulty to simulate the posterior distribution of parameters. I use 20,000 MCMC repetitions and discard the first 5,000 to report the results. I estimate the model separately for three education groups: high school dropouts (or less than 12 years of education), some college or professional degree (less than 16 years of education) and college graduates (16 years of education or more). The participation equation includes the following variables: a constant, years of schooling of the individual, a fourth-order polynomial in lagged labor market experience, a regional dummy, a dummy variable for living in a municipality with more than 30,000 individuals, a dummy for non-spanish individuals (foreigners), the province quarterly unemployment rate, the size of the household, a dummy for holding a fixed-term contract, a cohort group dummy, and year fixed effects. I also include the J W piece-wise function that contains the labor history of the worker as argued earlier. The mobility equation contains the same variables as the participation equation, and additionally a fourth-order polynomial for lagged tenure at the job and industry fixed effects. Finally, the wage equation, where the dependent variable is the deflated quarterly wage, includes the same regressors as the mobility equation, with the difference that tenure and experience are current rather than lagged and the exclusion of the household size from the 5

The sampler iterates through the set of conditional distributions z ⇤ conditional on ✓ and on ✓ conditional on z ⇤ .

11

wage equation. Identification comes from both cross-section and panel dimension of the data. The crosssection provides enough variation to identify returns to tenure and experience. However, without including the panel dimension, we would be incurring into a bias from ignoring heterogeneity across workers. The panel dimension allows me to include fixed effects to control for unobserved heterogeneity and reduce the bias in the estimates, since many individuals change jobs during the time period covered. This allows me to observe the same individual at different tenure levels and correctly identify the returns to tenure.

4

Results: A cross-country comparison between Spain and the US

Table 2 presents the results of the estimation for the wage equation. As we would expect, the returns to schooling are higher the higher the achieved education level. The estimated penalization of being employed under a fixed-term contract is increasing in the level of education, with college graduate workers under fixed-term contract 5 times the losses compared to high school drop-outs under the same type of contracts. In Table 3 I report the estimates for the returns to experience. The results indicate important difference across education groups. The cumulative returns to 5 years of experience in the labor market are more than twice as large for college graduates compared to high school drop-outs, and 60% larger compared to workers with some college. However, marginal returns to experience basically disappear after 10 years of labor market experience for the most educated group of workers, while high school drop-outs still accumulate returns and decrease the differences with respect the other two groups. Table 4 presents cumulative and marginal returns to tenure at the job. Cumulative returns to tenure are smaller for individuals with some college and college graduates, and there are not significant increases in returns to staying at the firm after 7 years. On the 12

Table 2: Wage equation by education group High school drop-outs

Some college

College graduates

Variable

Mean

Std

Mean

Std

Mean

Std

Constant

7.888

0.012

7.939

0.049

7.326

0.218

Education

0.025

0.001

0.031

0.004

0.059

0.013

Fixed-term

0.057

0.005

0.026

0.002

0.013

0.001

Experience

0.041

0.001

0.082

0.002

0.145

0.004

Exp.2 /100

0.243

0.015

0.765

0.017

1.597

0.043

Exp.3 /100

0.049

0.007

0.262

0.009

0.653

0.023

Exp.4 /100

0.003

0.001

0.033

0.001

0.096

0.004

Seniority

0.057

0.001

0.055

0.001

0.058

0.002

Sen.2 /100

0.592

0.011

0.663

0.016

0.782

0.043

Sen.3 /100

0.283

0.007

0.329

0.009

0.412

0.027

Sen.4 /100

0.054

0.002

0.070

0.005

0.045

0.001

s 0

0.079

0.015

0.258

0.030

0.066

0.089

e 0

0.013

0.002

0.018

0.003

0.008

0.010

10

0.005

0.000

0.003

0.000

0.013

0.001

20

0.008

0.001

0.014

0.001

0.018

0.003

30

0.019

0.004

0.028

0.004

0.045

0.009

40

0.010

0.006

0.005

0.008

0.041

0.026

s 2

0.025

0.001

0.026

0.001

0.004

0.002

s 3

0.008

0.001

0.006

0.001

0.008

0.003

s 4

0.007

0.001

0.000

0.001

0.005

0.003

e 1

0.002

0.000

0.002

0.000

0.006

0.001

e 2

0.004

0.000

0.006

0.000

0.002

0.001

e 3

0.000

0.000

0.002

0.000

0.003

0.001

e 4

0.003

0.000

0.003

0.001

0.001

0.003

Notes: The regression also includes: year dummy variables, four cohort group dummies, three regional location dummies, a dummy for living in a municipality with more than 30,000 inhabitants, province of residence unemployment rate, a dummy for foreign workers, and industry dummies.

Table 3: Estimated cumulative and marginal returns to experience Cumulative returns

Marginal returns (in %)

Years of experience

Years of experience

Group

2

5

10

2

5

10

HS drop-outs

0.072

0.150

0.213

3.18

2.02

0.60

Some college

0.136

0.252

0.288

5.50

2.39

0.50

College graduates

0.230

0.399

0.407

8.83

2.91

1.70

13

Table 4: Estimated cumulative and marginal returns to tenure Cumulative returns

Marginal returns (in %)

Years of tenure

Years of tenure

Group

2

5

10

2

5

10

HS drop-outs

0.093

0.171

0.219

3.69

1.70

0.58

Some college

0.088

0.151

0.167

3.32

1.10

0.00

College graduates

0.088

0.148

0.145

3.18

0.76

0.20

other hand, returns to tenure are of similar magnitude compared to returns to experience for high school drop-outs, and surprisingly larger compared to the other two groups. In Table 5 I compare with the findings for the US of Buchinsky et al. (2010). The long-term returns to both experience and tenure in Spain are smaller than those for the US for all education groups. In particular, returns to experience after 2 years are very similar in both countries, but because of rapidly decreasing marginal returns in Spain, substantial differences start appearing shortly after, creating a divergence in returns over time. Returns to experience are between 30%, 39%, and 62% higher for high school drop-outs, workers with some college, and college graduates respectively, with 10 years of labor market experience. This generates important differences in wage growth across countries, and also has policy implications, since the market does not value the experience of senior workers differently after 10 years in the market.

5

Extension: allowing for differences in returns by contract

I next allow for experience and tenure to have different returns depending on the type of contract the worker holds. In particular, I use the institutional background from the Spanish labor market to distinguish between two types of contracts under which the worker can be employed. Spain is characterized by a dual or two-tier labor market: highly-protected, long14

Table 5: Returns to experience (top) and tenure (bottom): US (BFKT) vs Spain Group

HS drop-outs

Some College

Years of experience

College graduates

Years of experience

Years of experience

2

5

10

2

5

10

2

5

10

US (BFKT)

0.114

0.241

0.362

0.134

0.278

0.402

0.202

0.430

0.661

Spain

0.072

0.150

0.213

0.136

0.252

0.288

0.230

0.399

0.407

Group

HS drop-outs

Some College

College graduates

Years of tenure

Years of tenure

Years of tenure

2

5

10

2

5

10

2

5

10

US (BFKT)

0.133

0.297

0.507

0.127

0.283

0.294

0.136

0.294

0.477

Spain

0.093

0.171

0.219

0.088

0.151

0.167

0.088

0.148

0.145

Notes: The first row is taken from BFKT, Table 5 and Table 6.

term employment relationships or permanent (open-end) workers; and workers with insecure, short-term employment relationships, or temporary (fixed term) workers. Temporary contracts were introduced in Spain during 1984 in order to increase flexibility of the labor market. These contracts dramatically reduced the cost of firing workers compared to permanent ones. In particular, permanent contract firing costs depend on seniority (2045 days’ wages per year of service), while temporary contracts have maximum duration of 3 years and firing costs betwenn 0 and 8-12 days’ wages per year of service. As a result of this regulation, the proportion of temporary contracts in dependent employment has been above 30% since the early 1990s. There were mild attemps during the first half of the 1990s to correct the excessive use of temporary contracts. In particular, in 1994 the government introduced restrictions for the use of temporary contracts, but this reform only lasted until 1997. The long term effects of two tier temporary contracts are stark now: there exist two groups of workers in the economy, with different evolutions over the business cycle. In particular, permanent workers enjoy protection to the cycle and longer tenures. Temporary workers on the other hand often transition to unemployment or concatenate several temporary contracts. The number of temporary workers upgraded to a permanent contract is less than 5%. This high turnover is likely to have effects on the composition of the two tiers of the labor force. In particular, in temporary relationships both firms and 15

workers have little incentives to invest in general and specific human capital, reducing the aggregate productivity of this tier, and of the total economy. I explore the implications of two-tier contracts on returns to experience and tenure. In particular, I use information on the type of contract present in Spanish Social Security records to distinguish between labor market experience accumulated under permanent contract and under fixed-term contracts. I follow the same approach to estimate returns to tenure under each contracts, with the limitation that fixed-term contracts are restricted by law to last less than 24 months. I now extend the model to allow for this possibility.

5.1

Incorporating two-tier contracts into the model

I extend the baseline model in several directions: I adapt the piecewise linear function of w experience and seniority Jijt to take into account the type of the previous contracts and I

extend the participation, mobility and wage equation to include experience and tenure under both contracts. Next I go over these extensions in detail. w I adapt the function Jijt described in equation (2), and now takes into account if the

previous contracts were temporary or permanent: w Jijt

+

" 4 Mit X h⇣ X l=1

k=1

P k0

+

=

⇣

s,P 0

e,P P k eitl 1

+

+

e,P 0 ei0

⌘

s,P P k sitl 1

dPi1

⌘

+

⇣

dPkitl +

s,F T 0

⇣

FT k0

+

+

e,F T ei0 0

⌘

(9)

dFi1T

e,F T F T eitl 1 k

+

s,F T F T sitl 1 k

⌘

dFkitTl

i

#

The super-index P refers to previous contracts that were permanent, while F T refers to fixed-term (or temporary) contracts. I group previous jobs by duration into four bins and type of contract and represent them with dummy variables d. In particular, dP1itl = 1 if the l th job of the i-th individual was permanent and lasted less than a year, 0 otherwise. In 16

a similar fashion, dP2itl = 1 if the l th job of the i-th was permanent and lasted between 2 and 5 years, and will be 0 otherwise, dP3itl = 1 if the l th job of the i-th was permanent and lasted between 6 and 10 years, 0 otherwise and finally, dP4itl = 1 if the l th job of the i-th was permanent and lasted more than 10 years, 0 otherwise. A similar classification applies to previous temporary jobs, although dF3itTl = dF4itTl = 0 since by law the duration of a temporary contract cannot exceed 24 months6 . The total number of jobs held by the i-th invidual up to time t (excluding the first job observed in t = 1) is summarized by Mit . Finally, if the i-th individual changed jobs in the first sample year, dP1 = 1 if the job was permanent, dF1 T = 1 if if was temporary, and 0 otherwise. ePitl , eFitlT , sPitl , sFitlT represent experience and tenure for the i-th individual at time tl , when individual i left job l. It is important to realize that even if the parameters ’s are fixed, the size of jumps, represented by the four bins of seniority and the type of contract, may differ depending on level of market experience, tenure and type of contract.

7

The result of estimating equation (9) gives an answer to how much general and specific human capital is accumulated under each type of contract, this is, returns to experience and tenure by type of contract. These parameter estimates allow me to quantify how being employed under a temporary contract affects the slope of wages, as well as compute predicted wages and quantify wage losses for unemployment spells under each contract. Next, I adapt the participation equation to include quadratic lagged labor market experience under permanent contracts and quadratic lagged labor market experience under fixed-term contracts8 . The mobility equation is additionally extended with a quadratic polynomial for lagged tenure under each type of contract–as well as lagged experience as in 6

I allow for durations longer than 24 months but lower than 40 between 2011 and 2012 given that labor reform that suspended the application of article 15.5 of the Workers’ Estatute on duration of temporary contracts during this period. 7 In total there are 13 identifiable parameters from permanent contracts and 7 from temporary contracts, corresponding to the four bins of tenure and the first year. 8 The choice of quadratic rather than quartic as in the baseline is done to reduce the set of parameters that are estimated in the extended specification. I have also estimated the model with quartic experience and tenure for the three groups, and although the parameter estimates are very similar, the significance is decreased because of the larger parameter space.

17

the participation equation, and lastly the wage equation includes the same regressors as the mobility equation, with the difference that both types of tenure and experience are current rather than lagged.

5.2

Preliminary results

Preliminary results for the specification are presented in Table 6. These results suggest that the returns to experience from working under fixed-term contracts is higher initially than under permanent contracts, but these returns are short-lived. In particular, after concatenating three fixed-term contracts, the marginal value of an extra year of experience rapidly decreases, and because of marginal returns from the quadratic specification, they can even penalize the worker after accumulating more than 10 years under these contracts. In particular the cumulative returns of 7 years of experience under fixed-term contracts for a college graduate are 0.52, but they start decreasing and are down to 0.42 after 10 years of experience and basically 0 after 13 years. On the other hand, returns to experience under permanent contracts increase more slowly but do not hit marginal returns until 15 years of experience. From these results we can infer than workers that are able to start under a fixed-term contract and obtain later in their career a permanent contract achieve the highest returns to experience, compared to obtaining a permanent contract since the beginning of their labor market career9 .

6

Conclusions and future research

This paper has analyzed the role of returns to general (experience) and specific (tenure) human capital in Spain using the same approach presented in Buchinsky et al. (2010). I find that the cumulative returns to 5 years of experience in the labor market are more than twice 9 Workers that fit this description would be graduate students with temporary contracts, or interim goverment workers. Tenure positions for these workers require some experience that can be achieved through these contracts.

18

Table 6: Wage equation by education group: two-tier labor makets High school drop-outs

Some college

College graduates

Variable

Mean

Std

Mean

Std

Mean

Std

Constant

7.829

0.074

7.915

0.072

6.424

0.458

Education

0.043

0.009

0.045

0.006

0.102

0.032

Experience, permanent

0.036

0.001

0.036

0.001

0.036

0.005

Exp.2 /100,

permanent

0.053

0.002

0.092

0.002

0.117

0.014

Experience, fixed-term

0.063

0.003

0.153

0.003

0.150

0.008

Exp.2 /100, fixed-term

0.237

0.030

1.283

0.036

1.077

0.059

Seniority, permanent

0.006

0.000

0.005

0.001

0.010

0.042

Sen.2 /100, permanent

0.020

0.002

0.019

0.002

0.042

0.018

Seniority, fixed-term

0.174

0.007

0.154

0.007

0.099

0.013

Sen.2 /100,

fixed-term

4.482

0.203

3.380

0.182

2.896

0.311

p 10

0.000

0.007

0.015

0.006

0.034

0.010

P 20

0.071

0.014

0.095

0.029

0.063

0.070

P 30

0.337

0.063

0.068

0.112

0.266

0.296

P 40 s,P 2 s,P 3 s,P 4 e,P 1 e,P 2 e,P 3 e,P 4

1.217

0.084

0.222

0.082

1.216

0.848

0.026

0.008

0.023

0.013

0.020

0.027

0.022

0.011

0.007

0.017

0.040

0.048

0.006

0.006

0.011

0.017

0.079

0.058

0.001

0.001

0.000

0.000

0.007

0.013

0.002

0.003

0.004

0.003

0.024

0.018

0.026

0.003

0.005

0.004

0.019

0.036

0.062

0.006

0.016

0.006

0.024

0.065

FT 10

0.020

0.005

0.007

0.005

0.003

0.008

FT 20 s,F T 2 e,F T 1 e,F T 2

0.032

0.087

0.062

0.047

0.062

0.059

0.008

0.037

0.025

0.028

0.033

0.001

0.017

0.003

0.016

0.005

0.027

0.011

0.014

0.006

0.009

0.011

0.010

0.012

Notes: The regression also includes: year dummy variables, four cohort group dummies, three regional location dummies, a dummy for living in a municipality with more than 30,000 inhabitants, province of residence unemployment rate, a dummy for foreign workers, and industry dummies.

19

as large for college graduates compared to high school drop-outs, and 60% larger compared to workers with some college. However, marginal returns to experience basically disappear after 10 years of labor market experience for the most educated group of workers, while high school drop-outs still accumulate returns and decrease the differences with respect the other two groups. The framework can be used to compare returns to general and specific human capital across Spain and the US, using PSID data. The long-term returns to both experience and tenure in Spain are smaller than those for the US for all education groups. Returns to experience are between 30%-62% higher in the US for workers with 10 years of labor market experience, depending on the education group. The returns to tenure are also lower in Spain, with returns after 10 years in the same job being 30-54% of the returns for workers in the same educational group in the US. The Spanish labor market exhibits dual or two-tier labor market: highly-protected, longterm employment relationships or permanent (open-end) workers; and workers with insecure, short-term employment relationships, or temporary (fixed term) workers. Finally, I extend the baseline model to distinguish between labor market experience and tenure accumulated under each type of contract. I estimate the model and find that the market penalizes workers with more than 7 years of labor market experience under fixed-term contracts, or in other words, three contracts of this type (the maximum duration is 2 years). Next steps include addressing extending the model to control for selection into both type of contracts, which is just done through the participation equation at this point. If there is selection on how workers are sorted into permanent and temporary contracts, then the estimates will be biased. Finally, I would like to compare the evolution of returns to experience and tenure over time and see if there has been a decline in recent years, that could explain wage stagnation.

20

References Abraham, K. G. and H. S. Farber, “Job Duration, Seniority and Earnings,” American Economic Review, 1987, 77 (3), 278–297. Altonji, J. G. and N. Williams, “Do Wages Rise with Job Seniority? A Reassessment,” NBER Working Paper, No. 6010, 1997. and R. A. Shakotko, “Do Wages Rise with Seniority?,” Review of Economic Studies, 1987, 54, 437–459. Bonhomme, Stéphane and Laura Hospido, “Earnings inequality in Spain: new evidence using tax data,” Applied Economics, 2013, 45 (30), 4212–4225. Buchinsky, Moshe, Denis Fougere, Francis Kramarz, and Rusty Tchernis, “Interfirm Mobility, Wages and the Returns to Seniority and Experience in the United States,” Review of Economic Studies, 2010, 77 (3), 972–1001. Chib, S. and E. Greenberg, “Analysis of Multivariate Probit Models,” Biometrika, 1998, 85, 347–361. Dustmann, Christian and Costas Meghir, “Wages, experience and seniority,” Review of Economic Studies, 2005, 72, 77–108. Heckman, James Joseph, “Selection Bias as a Specification Error,” Econometrica, 1979, 47, 153–162. Hyslop, Dean R., “State Dependence, Serial Correlation and Heterogeneity in Intertemporal Labor Force Participation of Married Women,” Econometrica, November 1999, 67 (6), 1255–1294. Lagakos, David, Benjamin Moll, Tommaso Porzio, Nancy Qian, and Todd Schoellman, “Life-Cycle Wage Growth Across Countries Life-Cycle Wage Growth Across 21

Countries Life-Cycle Wage Growth Across Countries,” Journal of Political Economy, 2017, forthcoming. Neal, Derek, “Industry-Specific Human Capital: Evidence from Displaced Workers,” Journal of Labor Economics, 1995, 13, 653–677. Topel, Robert, “Specific capital, mobility and wages: Wage rise with job seniority,” Journal of Political Economy, 1991, 99 (1), 145–176. and M. Ward, “Job Mobility and the Careers of Young Men,” Quarterly Journal of Economics, 1992, 107 (439-479).

22