The Career Costs of Children

Abstract

We estimate a dynamic life-cycle model of labor supply, fertility and savings, incorporating occupational choices, with specific wage paths and skill atrophy that vary over the career. This allows us to understand the trade-off between occupational choice and desired fertility, as well as the sorting both into the labor market and across occupations. We quantify the life-cycle career costs associated with children, how they decompose into loss of skills during interruptions, lost earnings opportunities and selection into more child-friendly occupations. We analyze the long-run effects of policies that encourage fertility and show that they are considerably smaller than short-run effects.

Full text

Forthcoming Journal of Political Economy
The Career Costs of Children∗
J´erˆome Adda
Bocconi University and IGIER
Christian Dustmann
University College London and Center for Research and Analysis of Migration
Katrien Stevens
University of Sydney
Abstract
We estimate a dynamic life-cycle model of labor supply, fertility and savings, in-
corporating occupational choices, with specific wage paths and skill atrophy that vary
over the career. This allows us to understand the trade-off between occupational choice
and desired fertility, as well as the sorting both into the labor market and across oc-
cupations. We quantify the life-cycle career costs associated with children, how they
decompose into loss of skills during interruptions, lost earnings opportunities and selec-
tion into more child-friendly occupations. We analyze the long-run effects of policies that
encourage fertility and show that they are considerably smaller than short-run effects.
∗Funding through the ESRC grant RES-000-22-0620 is gratefully acknowledge.
1

1 Introduction
In almost all developed countries, despite significant improvements over the last decades,
women still earn less than men (see Blau and Kahn (1996), and Weichselbaumer and Winter-
Ebmer (2005) for evidence), are often underrepresented in leading positions, and their careers
develop at a slower pace.1Having children may be one important reason for these disadvan-
tages, and the costs of children for women’s careers and lifetime earnings may be substantial.
One key question for investigation, therefore, is the magnitude of these costs and how they
decompose into loss of skills during interruptions, lost earnings opportunities, and lower ac-
cumulation of experience. Another important question is how intended fertility, even before
children are born, affects the type of career a woman chooses. Addressing these issues requires
an understanding of the dynamics of women’s choices, how unobserved fertility preferences
and ability affect the sorting into different career paths, and how intermittency patterns, work
decisions, savings decisions and fertility choices interact with each other.
This paper addresses these questions, by estimating a dynamic model which describes the
labor supply, occupational choices, assets, marital status and fertility decisions of women over
the life-cycle. Our model builds on the early work by Polachek (1981), Weiss and Gronau
(1981) and Gronau (1988), which emphasizes the important connection between expected
intermittency and occupational choice. Like Polachek, we allow different occupations to have
different entry wages and different rates of atrophy (skill depreciation) and wage growth.
In addition, we allow atrophy rates to vary over the career cycle, and occupations to vary
according to their amenity value with regards to children. We cast this in a dynamic setting
which endogenizes occupational choice, human capital, wages, savings decisions and fertility,
and which allows for unobserved heterogeneity in ability and the taste for children. Hence,
our model integrates occupational and fertility choices into a woman’s life cycle plan, where
women with different fertility plans opt for different occupations so as to balance a potentially
higher wage path with higher atrophy rates during work interruptions. Further, it explicitly
1See for instance Catalyst (2009).
2

implements risk aversion and savings, thus taking account of the trade-off between building
up assets early in the career and maternity during a woman’s most fertile period.
While many papers have addressed the issues of female labor supply and fertility, most
have dealt with them in isolation.2Early papers that analyze female labor supply and fertility
jointly use reduced-form models, such as Moffitt (1984) or Hotz and Miller (1988). More recent
papers that use dynamic life-cycle models to study these as joint decisions include Francesconi
(2002), Gayle and Miller (2006), Sheran (2007) and Keane and Wolpin (2010). We extend
this work in three significant ways. First, we incorporate occupational choices to better
understand the interplay between job characteristics, such as skill atrophy or differential wage
growth, and the planning of fertility, as well as the sorting that takes place both into the
labor market and across occupations. Second, we allow for skill atrophy, which can differ not
just between occupations, but also over the career cycle. This is important to capture the
trade-off between occupational choice and desired fertility, with possibly high atrophy rates at
career stages where fertility is most desirable. Third, we allow for an inter-temporal budget
constraint and risk aversion, which adds to our understanding of the relationship between
savings and fertility, and is important when investigating the dynamic aspects of policies that
incentivise fertility.
We study this for Germany, where individuals who choose to attend lower track schools at
age 10 (about 65% of each cohort) enroll after graduation (and at the age of 15-16) in a 2-3
2Early papers by Becker (1960), Willis (1973) and Becker and Lewis (1973) study fertility decisions and
their dependence on household background variables in a static context. Several authors, including Heckman
and Willis (1976), Ward and Butz (1980), Rosenzweig and Schultz (1983), Wolpin (1984), Rosenzweig and
Schultz (1985), Cigno and Ermisch (1989), Heckman and Walker (1990), Blackburn, Bloom, and Neumark
(1990), Hotz and Miller (1993), Leung (1994), Arroyo and Zhang (1997) and Altug and Miller (1998), propose
dynamic models of fertility, but assume labor supply decisions as exogenous. On the other hand, a related
literature on women’s labor supply behavior takes fertility decisions as exogenous; see, for example, Heckman
and Macurdy (1980), Blau and Robins (1988), Eckstein and Wolpin (1989), van der Klaauw (1996), Hyslop
(1999), Attanasio, Low, and Sanchez-Marcos (2008), Keane and Sauer (2009), Blundell et al. (2013); see
Blundell and Macurdy (1999) for a survey.
3

year vocational training program in one of 360 occupations within the German apprenticeship
system.3This unique setting enables us to observe initial occupational choices for these indi-
viduals before fertility decisions are made, but conditional on individual preferences for future
fertility. Our primary dataset is of administrative nature, and allows precise measurement of
wages, career interruptions, labor supply, and occupations, including the initial occupational
choice, for many cohorts across different regions over several decades. We combine this data
with survey data to measure fertility, household formation and savings decisions.
Our model and estimated parameters produce valuable insights into the different compo-
nents of the career costs of children, the contribution of fertility to explaining the male-female
wage differential, and the short-run and long-run impact of transfer policies on fertility. We
estimate that about three quarters of the career costs of children stem from lost earnings due
to intermittency or reduced labor supply, while the remainder is due to wage responses, as a
result of lost investments in skills and depreciation. More specifically, we show that skill de-
preciation rates are higher in mid-career and differ across occupations, as do the opportunity
costs of raising children and the child raising value, so that different occupational choices lead
to different costs of raising children and affect the timing of their birth. Our results highlight
that the selection into different careers is based not only on ability, but also on desired fertility,
so that some costs of fertility incur well before children are born.
Both atrophy and prior selection into child friendly occupations based on expected fertility
therefore contribute to the career costs of children. We also provide evidence on dynamic
selection, where fertility leads to changes in the ability composition of working women over the
life-cycle. Using a sample of comparable male cohorts who made similar educational choices,
we run simulations to understand better the wage differences between women and men over
the life course, and how these are affected by fertility decisions. We find that fertility explain
an important part of the gender wage-gap, especially for women in their mid-thirties.
3These occupations range from hairdresser to medical assistant to bank clerk, and two in three individuals
of each birth cohort follow an apprenticeship-based career route. See Fitzenberger and Kunze (2005) for details
on the occupational choices of males and females.
4

Finally, we use our model to simulate the impact of pronatalist transfer policies. Most
previous studies that investigate the effect of these policies on fertility are based on difference-
in-difference (DiD) designs and focus on short-term impacts. 4In contrast, our model allows us
to evaluate both short term and long term effects and to distinguish between responses through
the timing of fertility in reaction to an announced policy, vs. a change in overall fertility.
In doing so, we show not only that the long-run effect of a subsidy policy is considerably
lower than the short-run effects estimated in the literature but that such policies may also
have a long-run impact on skill accumulation, labor supply and occupational choice. More
importantly, we demonstrate that these policies are likely to have a far larger impact on
younger women, as they can adjust many life course decisions older women have already
made. These younger cohorts however are typically not considered in DiD type studies as
their fertility does not respond in a narrow window around the policy.
2 Background, Data, and Descriptive Evidence
2.1 Institutional Background and Data
Following fourth grade (at about age 10), the German education system tracks individuals into
three different school types: low and intermediate track schools, which end after grade 9 and 10
(age 15/16), or high track schools, which end after grade 13. About one third of the cohorts
studied here attend each of the three school types. Traditionally, only high track schools
qualify individuals for university entrance, while low and intermediate track schools prepare
for highly structured 2 to 3-year apprenticeship training schemes that combine occupation-
specific on-site training 3-4 days a week with academic training at state schools 1-2 days
a week. 5These programs, which train for both blue- and white-collar professions, cover
4See, for instance, Milligan (2005), Laroque and Salanie (2014), Cohen, Dehejia, and Romanov (2013), and
Lalive and Zweim¨uller (2009).
5For instance, training is only provided in recognized occupations, skilled training personnel must be present
at the training site, and trainees must pass monitored exit examinations.
5

many occupations that in the U.S. require college attendance (e.g., nurse, medical assistant,
accountant). At the end of the training period, apprentices are examined based on centrally
monitored standards, and successful candidates are certified as skilled workers in the chosen
profession.
In our analysis, we concentrate on women born in West Germany between 1955 and 1975
who attend lower and intermediate track schools and then enroll in an apprenticeship training
scheme after school completion.6We follow these women throughout their careers for up to
26 years in the labor market. We draw on three main datasets (described in more detail
in the online appendix): register-based data from the German social security records (IABS
data) and survey data from the German Socio-Economic Panel (GSOEP) and the Income- and
Expenditure Survey (EVS). The IABS data covers a 2% sample of all employees in Germany
that contributed to the social security system between 1975 and 2001 and provides detailed
information on wage profiles, transitions in and out of work, occupational choice, education,
and age, and periods of apprenticeship training. The sample we construct contains about 2.7
million observations on wages and work spells. We use the GSOEP data to measure, for a
sample from the same birth cohorts as in the register-based dataset, women’s fertility behavior
over their careers, as well as family background and spousal information. Finally, we use the
EVS data to compute savings rates.
All analyzes concentrate on the German population. Because the register data exclude
the self-employed and civil servants, we exclude these groups from our analysis, as well as all
individuals who have ever worked in East Germany. We provide more detail about the sample
construction in Appendix A.7
6Women born in East Germany experienced different conditions while growing up behind the Iron Curtain
and we do not observe them in administrative data until after German reunification.
7Earnings in all data sets we use have been deflated using Consumer Price Index data for private households
(German Statistical Office) and converted into Euros, with the base year being 1995.
6

2.2 Occupation Groups
We allocate occupations to groups that reflect the tradeoff between careers that offer a higher
wage but punish interruptions, and careers that imply lower profiles but also lower atrophy
rates. To achieve that, we use information on the task content of occupations, drawing on
the task-based framework introduced by Autor, Levy, and Murnane (2003). This results in
an aggregation of the many occupations into three larger groups according to characteristics
that are meaningful in the context we study, distinguishing between occupations where tasks
performed are mostly routine, occupations where tasks are mostly analytic or interactive,
and occupations where tasks are mostly manual, but not routine. We refer to these three
occupational groups as routine, abstract and manual occupations.8
Requirements in jobs with mainly abstract tasks are likely to change at a faster pace
than those in routine dominated occupations, while those in manual occupations may take
an intermediate position. For instance, shop assistants and sewers are classified as routine
occupations, and require a set of skills that is acquired in the early stages of the career (like
product knowledge and relational skills), but unlikely to change much over time. On the
other hand, bank clerks and medical assistants (classified as abstract) are likely to require
constant updating of their skills because of rapidly changing information technologies or new
financial products, while nurses and stewards (classified as manual occupations) may take an
intermediate position. We show below that wage profiles, but also atrophy rates are indeed
higher in abstract occupations than in routine or manual occupations.
8Autor, Levy, and Murnane (2003) distinguish between (i) non-routine analytic, (ii) non-routine interactive,
(iii) routine cognitive, (iv) routine manual, and (v) non-routine manual jobs. These are often combined into
abstract (i, ii), routine (iii, iv) and manual (v) jobs. We follow Gathmann and Schonberg (2010), Black and
Spitz-Oener (2010) and Dustmann, Ludsteck and Schoenberg (2009) who allocate two digit occupations to
these three groups, using data from the German Qualification and Career Survey 1985/86, and which includes
survey information on tasks performed on the job. The construction of the task indicators and the classification
of occupations across the three groups are detailed in the online appendix.
7

2.3 Occupational Choice, Labor Supply, Fertility and Savings
In Table 1, we present descriptive statistics for the whole sample and by current occupation.
About 45% of all women in our sample choose an initial (training) occupation with more
abstract tasks, while 25% and 30%, respectively, choose routine or manual occupations. The
second row of the table illustrates that current occupational proportions are similar to those
for initial occupations, indicating that few women switch occupations over their careers. This
is confirmed by the transition rates across groups in the second panel of the Table, illustrating
that 98.6% of individuals remain in the same occupational group in two consecutive years.
In the third panel, we report initial wages at age 20 and real wage growth in each of the
three occupation categories, after 5, 10, and 15 years of potential experience. Women in more
abstract occupations not only earn higher wages than those in the two other groups at the
start of their careers, but also have a higher wage growth at each level of experience. The next
panel reports the accumulation of total labor market experience, and broken down by part-
time and full-time work, by occupational category and evaluated after 15 years of potential
labor market experience. Women who have chosen an occupation with predominantly abstract
tasks accumulate 1.2 years (or 10 %) more total work experience and about 1.6 more full-time
work experience over this period than women in routine task-dominated occupations. Finally,
the next panel reports changes in daily wages after an interruption of 1 or 3 years, where
changes in work hours, firm size, and occupation are conditioned out. The overall log wage
loss for a one (three) year interruption is about 0.12 (0.21) log points. Wage losses are highest
in abstract, and lowest in routine occupations.
Thus, the costs of interruptions both in terms of forgone earnings as well as atrophy differ
between occupational groups, and are highest in occupations dominated by abstract tasks.
This is reflected by different fertility patterns, as shown in the penultimate panel, where
women in abstract jobs are more likely to remain childless or to have only one child, while
being less likely to have two or more children, and being older at the birth of their first child.
While these figures suggest therefore sizeable differences between women who choose different
8

occupational careers at an early point in their life cycle, they cannot be interpreted causally
due to selection of women into occupations, fertility behavior, and labor supply patterns based
on fertility preferences and labor market abilities.
In Figure 1 we plot the average household savings rates as a function of the age of the
woman. Savings rates have a hump-shaped profile, at least until age 50, starting at about 7%
at age 20 and reaching a peak at age 28, after which they decrease. The figure suggests that
savings are build up before the arrival of children, indicating that savings are an important
element in the fertility decision, and that parents are smoothing consumption over the life
cycle and in response to the added expenditures linked to children. Building up a sufficient
stock of assets could therefore be an important reason to delay pregnancy, in addition to
career considerations.
3 A Life-Cycle Model of Fertility and Career Choice
Our objective is to develop an estimable life-cycle model to assess the career costs of children.
To achieve that, our model has to be able to evaluate the costs of fertility by considering
all associated decisions. There are at least three elements that determine the career costs of
children. First, children may require intermittency periods of unearned wages during which
women cannot work. Second, during intermittency, there will be no skill accumulation, and
existing skills may depreciate. Third, depending on ability and expected fertility, women
may sort into occupations that minimize the expected career costs of children. In particular,
occupations may differ in terms of opportunity costs of raising children and in how skills
depreciate. To understand how these different determinants of the costs of fertility operate,
we need to understand how fertility is planned. This requires, in addition to the above
components, modeling of the evolution of assets over a woman’s work career, which in turn
will interact with both fertility- and career decisions. Thus, consideration of savings decisions
is an important building block in our model. In the next section, we describe the main
components of our model. We provide a more detailed description in Appendix A and in the
9

online appendix.
3.1 The Set-Up
In each period, individuals choose consumption (and savings), whether to have an additional
child, labor supply, and the type of occupation they work in. Frictions in the labor market
imply that individuals have to wait for offers to adjust their labor supply. In the first period,
around the age of 15, they decide on a particular training occupation and enroll into a 2-3
year apprenticeship training scheme. Time is discrete, a period lasts for 6 months, and we
consider women in the age range between 15 to 80, thus starting at the age when occupational
decisions are made. We first present the building blocks of our model and then show how
decisions are taken.
Ex ante heterogeneity. We allow for ex-ante heterogeneity, which we model in terms of
discrete mass points, along four dimensions: labor market productivity - or ability - (fP
i),
taste for leisure (fL
i), taste for children (fC
i), and potential infertility (fF
i). We collect these
characteristics in the vector fi. As four dimensional heterogeneity is very demanding in terms
of identification and computation, we place some restrictions on how these characteristics
vary across individuals. We group together ability and the taste for leisure. While individuals
with high ability can have a different taste for leisure than low ability individuals, we do not
allow for heterogeneity in the taste for leisure, conditional on ability. We allow for unobserved
heterogeneity in the ”taste for children” to be correlated with ability and the taste for leisure,
and we estimate this correlation. Further, we assume that potential infertility is orthogonal to
the first three characteristics, meaning that, while women know the first three characteristics,
they do not anticipate infertility, and they do not learn from unsuccessful conception attempts.
Based on medical evidence, we fix the proportion of infertile women at 5%. 9
9Data from the U.S. indicate that about 8% of women aged 15 to 29 have impaired fecundity (see Centers
for Disease Control and Prevention (2002)), although some may give birth after treatment for infertility.
10

Occupation and labor supply. In our model, several features describe an ”occupation”
oit (which takes three values denoting whether an occupation is ”routine”, ”abstract” or
”manual”). First, each occupation has a particular wage path, characterized by different
log wage intercepts and different returns to work skills (denoted xit). Second, occupations
are characterized by the pace with which skills depreciate through intermittency (atrophy).
Third, arrival rates of offers when out of work differ across occupations. Finally, occupations
differ in their amenity value with regards to children, as in some occupations, women can
better vary their work hours to care for their children. By allowing for occupational choices,
we build into our model an important aspect of women’s career decisions, which has first been
emphasized by Polachek (1981). We extend Polachek’s formulation, by allowing these choices
to be taken in conjunction with fertility choices, and by considering the “child raising value”
of occupations.10
In any occupation, individuals can work either full time (F T ) or part time (P T ). They can
also choose to be unemployed (U), or out of the labor force (OLF ), and we record the choice in
the variable lit. We assume that offers for alternative occupations and working hours arrive at
random, but that arrival rates differ according to current occupation and labor supply status.
We refer the reader to Appendix A for further detail on functional forms. Furthermore, women
who are working face an exogenous and constant probability of layoff δ.
Budget constraint. The budget constraint of the household is given by:
Ait+1 = (1 + r)Ait +net(GIit ;hit, nit)−cHH
it −κ(ageK
it , nit)Ilit =F T ,P T,nit >0,(1)
where Ait is the stock of assets and rthe interest rate (which we assume as fixed and set at 4
percent). In our model, assets are accumulated for precautionary motives as individuals are
risk-averse and face shocks to wages, labor market participation and household size. Assets
are used to finance periods out of the labor force, fluctuations in household earnings, and costs
associated with children and retirement. Households cannot borrow against future income to
10See also Goldin (2014) who stresses this point as an important aspect of occupational choice.
11

finance the costs induced by having children and need to delay fertility to accumulate sufficient
assets (see also Heckman and Mosso (2014) for a discussion of imperfect borrowing in a model
of parenting). Total household consumption is denoted as cHH
it , which is equal to the woman’s
own consumption, cit, scaled by the number of adults and children in the household.11 Further,
we denote by GIit the gross income of the household, which consists of the labor earnings of
the husband, the labor earnings of the wife if she works, unemployment benefits or maternity
leave benefits if eligible, and government transfers according to the number of children.12
If children are present, but the father has left the household, the father contributes to the
household budget through child support.13 During retirement, women receive retirement
benefits, which are a fraction of their last earnings. Net income net(GIit ;hit, nit ) is derived
from gross income, using institutional features of the German tax code, and is a function of
the number of children and the presence of a husband (where hit = 1 if a husband is present),
as tax rates vary between singles and couples. Finally κis a cost incurred if children are
present and the mother decides to work and includes the cost of child care. We assume it
depends on the age of the youngest child (denoted ageK
it ) and the number of children, nit. We
estimate this cost along with the parameters of the model.
11We use a scale similar to the ”OECD modified scale”, where the ”number of adult equivalent” is equal
to one plus 0.5 for a second adult and 0.3 for each child, see Hagenaars, de Vos, and Zaidi (1994). As these
consumption weights are derived for the average household in OECD countries, they may not be pertinent for
the population we study. We therefore estimate the weight of children, while holding constant the weight of
adults.
12Unemployment benefits depend on past earnings, which in turn are functions of the previous occupation,
accumulated skills and unobserved productivity. Individuals are eligible for benefits if they had been working
prior to becoming unemployed. Maternity benefits consist of two components, a fixed one, and a variable one,
that depends on former labor market status.
13The father’s compulsory contribution is 15% of his income for each child, see Duesseldorfer Tabelle (2005)
for more detail.
12

Skills and wages. While working, individuals accumulate skills. Skills are increased by
one unit for each year of full time work and by 0.5 units for each year of part time work.14
When out of work, skills depreciate, and the rate of atrophy ρ(xit, oit)<1 depends on the
occupation and the previous level of skills:
xit+1 =xit ρ(xit, oit ),(2)
ρ(xit, oit ) = ρ1(oit)Ixit ∈[0,5[ +ρ2(oit )Ixit∈[5,7] +ρ3(oit )Ixit ∈[8,∞[,
where ρ1, ρ2, ρ3are vectors of parameters, specific to each occupation, and Ijis an indicator
variable taking the value of one if jis true (i.e. if the level of skills xit falls in the interval in
brackets). We thus allow for career interruptions being more detrimental at career stages where
learning is intense or where individuals compete for key workplace positions, a potentially
important factor to understand the timing of fertility.15
Female full-time daily wages depend on skills, xit , occupation, oit , and individual produc-
tivity fP
i:
ln wit =fP
i+αO(oit) + αX(oit )xit +αX X (oit)x2
it +ηit ,(3)
where ηit is an iid shock to log wages. The wage profile is specific to a given occupation, with
different intercepts and different returns to skills.16
Marriage, divorce and husbands’ earnings. Women’s probability of getting married in
each period depends on their age, skills, and taste for children (fC
i). Conditional on being
married, women face a probability of divorce which depends on their age and the presence of
14We do not consider occupation-specific skills as in the data we observe very few individuals switching
occupation during the life-cycle.
15This extends the empirical literature that assumes constant depreciation rates across occupations or career
stages, see e.g. Kim and Polachek (1994) and Albrecht et al. (1999). We chose the nodes of 5, 7 and 8 based
on results from reduced form regressions.
16As wage profiles are flat after 15 years of accumulated work experience, we assume that there exists a
threshold, ¯x, beyond which the marginal effect of skills on wages is zero. We estimate this threshold along
with the other parameters.
13

children. The functional forms for these probabilities are shown in Appendix A.1. Our model
therefore allows for the age of marriage to vary with unobserved characteristics, where women
with a higher taste for fertility may marry at younger ages. To the extent that unobserved
heterogeneity such as differences in productivity, taste for children or leisure affect labor
market attachment, these characteristics will also influence marital status through the effect
on skills.
We model the earnings of the husband, earnh
it, which capture both their wages and labor
supply. We assume earnings depend on observed characteristics of the woman, as in van der
Klaauw (1996) or Sheran (2007), which in our case include age and occupation. We extend
these papers by allowing earnings to depend also on her unobserved ability, fp
i:
earnh
it =αh
0+αh
a1ageM
it +αh
a2ageM2
it +X
j
αh
jIoit=j+αh
PfP
i+ηh
it ,(4)
where Ioit is an indicator variable equal to one if the wife is working in occupation j, and
ηh
it is a shock assumed to be iid and normally distributed with mean zero.17 As we allow
for a rich set of characteristics, both observed and unobserved, to influence marital status
and husbands’ earnings, our model captures essential ingredients of a marriage market with
assortative matching and differential marriage rates across women, while at the same time re-
maining tractable.18 Husbands contribute to the income of the household, providing resources
and insurance against income or labor market shocks.
Conception. If a woman decides to conceive a child, a child is born in the next period with
a probability π(ageM
it , f F
i). This probability takes into account potential infertility, although
17We estimate the variance of ηh
it using data on earnings for the spouses, including non-employment spells,
so that ηh
it takes into account unemployment shocks as well. We assume that the shock to the husband’s
earning is orthogonal - conditional on observables and a fixed effect specified in equations (3) and (4)- to the
shock to the woman’s wage. In the data, the wage/earnings residuals within a household are weakly correlated
with a coefficient of -0.04 and a standard deviation of 0.001.
18Dynamic models of marriage markets and schooling or labor supply have been derived by e.g. Chiappori,
Iyigun, and Weiss (2009) and Eckstein and Lifshitz (2014).
14

women do not know or learn about it. Drawing on medical evidence we allow the probability
of conception to decline with age. 19 Note that a child can be conceived out of wedlock,
although this is uncommon in Germany during the period we consider.20
Dynamic choice. At the start of each period, individuals take as given the variables that
form their state space Ωit:
Ωit =nlit−1, oit−1, Ait−1, hit−1, ageM
it , xit, nit , ageK
it ,Υit, fio.
The state space is composed of variables set at the end of the previous period: labor supply
lit−1, occupation oit−1, assets Ait−1and marital status (presence of a husband) hit−1. It also
comprises variables updated at the start of the period: age (ageM
it ), skills xit, number of
children (including any newborn child) nit, and the age of the youngest child ageK
it . The state
space includes a vector of iid shocks to preferences affecting labor market status, occupation
and conception as well as income or earning shocks (which we collect in a vector Υit), and the
different dimensions of ex-ante heterogeneity, collected in the vector fi.
The value function for individual iin period tis given by:
Vt(Ωit) = max
{bit,cit ,oit,lit }u(cit , oit, lit ;nit , hit, ageK
it ,Υit, fi) + βEtVt+1(Ωit+1),(5)
where βis a discount factor, and Etis the expectation operator conditional on information in
period t. The expectation of the individual is over the vector of future preference and income
shocks Υit+1, and future shocks to marital status.
In each period the individual chooses whether to conceive or not (denoted by the indicator
variable bit), her consumption (or equivalently household consumption), occupation oit and
19Khatamee and Rosenthal (2002) estimate that a woman has a 90% chance of conceiving within a year at
age 20, a 70% chance at age 30, and a 6% chance at age 45. After age 50, the probability of conception is
almost zero.
20We do not allow for conception errors, as in Sheran (2007), as we do not have such information, but our
model allows for shocks to preferences regarding conception, so that two seemingly similar women may differ
in their decision to conceive.
15

labor market status lit. The choice of occupation and labor market status become effective at
the end of the period.
Utility is derived from her own consumption cit , labor market status lit (which reflects the
amount of leisure time available), the amenity value of an occupation oit, and the number of
children nit (we abstract from modelling the quality of children). We write the utility function
as:
uit =u1(cit, lit ;nit, f L
i) + u2(nit;fC
i, ageK
it , lit, oit , hit,) + u3(bit ,Υit ) (6)
The utility function has three parts. The first sub-function is the utility derived from con-
sumption and leisure. We allow for curvature in the utility function over consumption to allow
for risk aversion, by specifying a constant relative risk averse function. The utility of con-
sumption is interacted with leisure (labor supply), as in Attanasio, Low, and Sanchez-Marcos
(2008), the taste for leisure as well as the number of children.
The second sub-utility is the utility of children. The utility a woman derives from children
depends on her taste for children and on four further factors, the age of the youngest child,
ageK
it , labor supply, lit, occupation, oit, and her marital status, hit. The age of the youngest
child reflects that leisure may be particularly valuable when children are young. The spec-
ification also allows for complementarity between children and leisure. These features help
explaining why many mothers take time off from the labor market.21 Occupation may affect
the marginal utility of children as some occupations may be more demanding and impose
constraints for working mothers. Finally, marital status is part of the utility function to allow
for the possibility that raising children imposes less of a utility cost if a partner is present.
The third sub-utility collects preference shocks pertaining to the choice of conception (bit).
We describe in Appendix A.3 in detail the functional form for the subutility functions.
21We do not model child quality, as the goal of our analysis is the study of the careers of women and
not the production of child quality per se. See Del Boca, Flinn, and Wiswall (2014) for a model of child
quality and parental inputs, which however does not consider fertility choices, savings, occupational choices
and depreciation of skills. With data on labor supply and fertility, our formulation is observationally equivalent
to models where mothers derive utility over child quality, which is produced with parental time inputs.
16

Labor market choices are taken until 60, at the age at which women retire and live an addi-
tional 20 years, deriving utility from consumption, leisure, and children. During that period,
they finance consumption from retirement benefits and by de-cumulating assets. Choices are
made under the constraints detailed above, as well as some additional institutional constraints,
which we describe in Appendix A.4. For instance, women who are out of the labor force can-
not apply for unemployment benefits, and pregnant women in employment have the option
to return to their previous occupation after their maternity leave (although not necessarily at
the same wage, as skills depreciate when out of work).
Initial choice of occupation. At time t= 0, women enter apprenticeship training, typi-
cally around age 15 or 16, and decide on a specific training occupation oi0by comparing the
expected flow of utility for each occupational choice with the current cost, which depends on
the region of residence Riand the year of labor market entry (Y eari), as well as a preference
shock, ωio, drawn from an extreme value distribution and specific to each possible occupa-
tion o. These costs arise from temporary or local shortages of training positions in particular
occupations and we use those as instruments to identify the choice of occupation. The ap-
prenticeship training lasts three years, so the payoff is received six periods later (as a period
in our model lasts for half a year):
oi0= arg max
oβ6E0V6(Ωi6)−cost(o, Ri, Y eari)−ωio .(7)
We do not model choices during the training periods. Training regulations in Germany commit
firms to fulfilling the entire period of the apprenticeship contract, so individuals cannot be
fired. We assume that women begin making choices about fertility once they have completed
their training.22
22Teenage pregnancy rates are very low in Germany. For instance, in 1998, only 1.3% of women between 15
and 19 gave birth, compared to 5.2% in the U.S. (see UNICEF (2001)), and 4.7% between 15 and 18 in the
UK ( see http://www.fpa.org.uk/professionals/factsheets/teenagepregnancy).
17

3.2 Estimation
3.3 Method and Moments
The model is estimated using the method of simulated moments (MSM) (see Pakes and
Pollard (1989) and Duffie and Singleton (1993)), which allows us to combine information from
different data sources on career choices, wages, savings and fertility decisions over the life
cycle. The method also allows us to address time aggregation, through simulations, as the
sample frames of our data sets vary, from semi-annual (in the IAB data) to annual frequencies
(in the GSOEP).
In this approach, the model is solved by backward induction (value function iterations)
based on an initial set of parameters, and then simulated for individuals over their life cycles.23
The simulated data provide a panel dataset used to construct moments that can be matched
to moments obtained from the observed data. Using a quadratic loss function, the parameters
of the model are then chosen such that the simulated moments are as close as possible to
the observed moments. Because the focus of our model is on describing life-cycle choices, we
remove regional means and an aggregate time trend from all our moments.24
The method of simulated moments yields consistent estimates. However, as shown by
Eisenhauer, Heckman, and Mosso (2014), its finite distance properties depend on the choice of
moments, the number of simulations, and the weighting matrix, and we follow their suggestion
and choose both static and dynamic moments. To obtain a smoother criterion function, we
weight the moments with a diagonal matrix which contains the variances of the observed
moments.
23We refer the reader to the online appendix for a discussion on the numerical solution of the model.
24Removing regional and aggregate effects when calculating moments implies that we are relying on
difference-in-difference variations to identify our model. In other words, the model is not identified from
pure cross-sectional variations or time-series variations that could introduce spurious correlation. An alterna-
tive choice would be to model regional differences together with a choice of residence, which would be infeasible
within our framework. Kennan and Walker (2011) model location choices In a simpler setting.
18

The full list of moments used to identify the model is displayed in Table 2, grouped by
the choices they identify, i.e. labor supply and occupational choices, wages, savings, fertility
and marital status. In each of these categories, we rely on simple statistics that ensure that
the model reproduces the basic trends and levels in the real data. These moments include
variables such as the proportion of women in each occupation, average wages, hours of work,
number of children and savings rates, all computed at different ages ranging from 15 to 55. 25
We further describe fertility choices by ages at first and second birth and their heterogeneity
by including centiles of the age at first and second birth in our list of moments.
In addition, we add conditional moments, which relate the main outcome variables to other
endogenous variables, either for the same period or adjacent periods. Eisenhauer, Heckman,
and Mosso (2014) argue that such moments are crucial to identify the parameters of dynamic
models such as ours.26
Information contained in regressions of log wages on career interruptions contribute to the
identification of the atrophy rate parameters in equation (2), as in Polachek (1981). More
specifically, we use information from regressions of the change in log wages for individuals who
interrupt their career on the duration of the interruption, dummies for experience levels, occu-
pational groups, and the interaction of duration and experience. This information alone is not
sufficient to identify the atrophy parameters, due to the non-random selection into maternity
and more generally into non-working spells. However, by matching the simulated moments to
those obtained from the observational data, our model, which specifies the process through
which these choices are made, allows us to identify the underlying structural parameters. We
follow here similar identification schemes that have been used in previous literature (see for
25We follow the cohorts in our main data from age 15 to 40. To completely characterize their life cycle,
however, we also use supplemental data from slightly older cohorts to construct moments that describe wages
and labor supply at ages 45, 50 and 55. We verify that at age 40, the labor supply and wages of these older
cohorts are very similar to those of the younger ones.
26For instance, dynamic moments we use link current wages to past and future wages, or past labor supply
or fertility decisions, current labor supply to previous labor supply, and current savings with past fertility
decisions.
19

instance Del Boca, Flinn, and Wiswall (2014)).
Our model also allows for unobserved heterogeneity in the level of wages (ability), as in
equation (3), and in the utility derived from children.27 We model the heterogeneity as a
mass point distribution and allow for a correlation between both dimensions. We use discrete
mass points, which are estimated together with the relative proportion in the sample in a
similar way as in Heckman and Singer (1984). To identify the proportion of individuals in
each ability “type”, we proceed in several steps. We first regress log wages on experience and
occupation and compute wage residuals for each individual. This residual contains information
on unobserved ability, fP
i. We then use the cross-sectional variance of these wage residuals
as a moment. We also regress the number of children on age to compute a fertility residual,
which contains information about the unobserved taste for children, fC
i, and we correlate it
with the wage residual to provide information on the correlation between ability and desired
fertility.
We construct these conditional moments from different data sets and use for each moment
the data set that contains the most precise information. For instance, moments pertaining
to wages and labor supply are computed from the administrative IAB data. Information on
fertility is gathered from the GSOEP, and information on savings rates from the EVS. In
total, we rely on 763 moments to identify our model. The online appendix provides further
evidence on the identification of the model.
3.4 Model Fit
Overall, the model fits the sample moments well in an economic sense. However, due to our
very large sample (we observe 2.7 million work spells and earnings), our moments are estimated
with very high precision, which leads, not surprisingly, to statistical rejection of global equality
of estimated and simulated moments. For instance, the proportion of women working full time
at age 30 is 37.5% in the data, while the model prediction is 37.1%. Statistically the equality
27As explained earlier, we groups together heterogeneity in ability and taste for leisure.
20

of these two moments is nonetheless rejected, with a t-statistic of 3.8.28 Locally, however, we
cannot reject the equality of the observed and simulated moments at the 95% confidence level
in 53% of the cases, despite our large sample. Furthermore, the model matches trends in labor
supply and work hours well, as well as the number of children and spacing of births by age. It
is also able to match wage profiles by age and initial occupation, the savings rate by age, and
the coefficients of a regression of log wage on work experience by occupation. The model also
replicates closely the dynamic moments. For instance, a regression of our simulated data for
log wages on the lead and lagged log wages give very similar results to the ones in the data
(respectively 0.51 versus 0.53 and 0.48 versus 0.46, see Table A8 in the online appendix. To
economize on space in the main text, we refer the reader to the online Appendix for a detailed
presentation of the model fit and an extended set of tables.
4 Results
4.1 Estimated Parameters
To describe wages, hours of work, occupational choices, the number of children, the spacing
of births, and savings decisions over the life cycle, we estimate a structure that is defined by
a total of 73 parameters.29 We now discuss subsets of these parameters.
28The J-statistic for the overall fit of the model is equal to 123,352, which implies the rejection of the equality
of predicted and observed moments at any confidence level. The chi-square critical value at the 5 percent level
and 670 degrees of freedom is equal to 731. The equality of the predicted and observed moment would not
be rejected were the sample size about 100 times lower, which would still be larger than the samples sizes
typically used in structural models.
29In addition, the initial choice of occupation, allowing for a fully interacted model with regional and time
effects in the cost function (see equation (7)) is defined by 88 parameters.
21

4.2 Atrophy rates, wages and amenity values by occupation
We display in the first panel of Table 3 the atrophy rates, measured as the value of skill loss
resulting from a one-year work interruption, which we allow to vary by level of skills and by
occupation (see equation (2) ). As skills accumulate in the same way as work experience, but
depreciate when out of work, a skill level of xis equivalent to xyears of uninterrupted work
experience, and we report the atrophy rates at 3, 5, and 10 years.
In routine occupations, atrophy rates are low and vary between 0.06% and 0.6%. In con-
trast, in abstract occupations, atrophy rates are far higher, and vary substantially over the
career cycle (between 0.1% and 6.9% per year), while manual jobs take an intermediate posi-
tion. In both types of occupations, atrophy rates are highest at about 6 years of uninterrupted
work experience, suggesting that intermittency is far more costly at intermediate career stages,
possibly due to differing learning intensity over the career cycle, and important career steps
being decided at those career points.30 An uninterrupted work experience of 6 years corre-
sponds on average to an age of 26 years, which is when many women find it desirable to have
children. Fertility decisions are therefore likely to be affected far more by career concerns in
abstract (and to some extent in manual) jobs than in routine occupations. This is in line
with the evidence in Table 1 on the age at first birth across occupational groups. In addition,
the non-linear evolution of atrophy rates in these occupations, being highest at mid-career
stages, adds a further important (and so far largely ignored) consideration when considering
occupational choices of women, and how these interact with desired fertility.31
30 Kim and Polachek (1994) estimate atrophy rates of about 2% to 5%, depending on the sample and the
estimation techniques. Albrecht et al. (1999) find atrophy rates of about 2% per year. Both papers do not
allow for differences by occupation, or level of skills.
31Contrasting these estimates with those from simple (fixed effects) regressions as in Polachek (1981) and
Kim and Polachek (1994), obtained by first simulating life-cycle careers using our model and then regress-
ing changes in log wages following interruptions on the time out of work by occupation, and allowing for
non-linearities, leads to estimates that understate the role of atrophy, but re-produce the ranking across occu-
pations. This is mainly due to such regressions ignoring that those who return to work are positively selected,
as they are more likely to have drawn a positive wage shock, something that is built into our model and
22

In Panel B of Table 3 we display the parameters of the (log) wage as a function of un-
interrupted work experience (defined as in equation (3)), which represent average treatment
effects of working full time in a given occupation.32 The estimates suggest that wage returns
to human capital (the constant terms) are highest in abstract occupations. Furthermore,
while wage increases are similar across occupational groups in the early years, wage profiles
are less concave in abstract occupations, and continue to grow at a faster pace at higher to-
tal work experience.33 Thus, interruptions at mid-career in abstract jobs are not only more
costly because of higher atrophy rates, but also due to considerably higher opportunity costs
as individuals forgo earnings while not in work.
An additional dimension when balancing occupational choice with labor supply and fertility
decisions, besides atrophy rates and opportunity costs, is the amenity value of an occupation
with regards to children (which can be interpreted as the ease with which women in these
occupations can combine work with childraising).34 We present estimates of these amenity
values, normalized to be zero for routine occupations, in Panel C of Table 3. The figures
show that - in comparison to routine jobs - abstract jobs are least desirable when children are
present. Our estimates imply that if abstract and manual occupations had the same amenity
value as routine ones, the proportion of women opting for abstract or manual occupations
would increase by 5%. The amenity of part-time work - an option chosen by many mothers
in our data -is likewise lower in abstract jobs, as the second row of this panel shows. Our
estimates imply that if women in abstract jobs had the same amenity value for part-time jobs
than in routine ones, the proportion of part-time work in abstract jobs would be 7% higher
estimation.
32Compared to the OLS coefficients shown in Table A8 in the online appendix, these structural parameters
are ”causal”, taking account of selection; further, they refer to a measure of skills which, unlike real experience,
depreciates when the individual is out of work.
33For instance, after 10 years of uninterrupted work experience, wages in abstract jobs increase by about
2% more per additional year than in the other two occupations.
34These parameters are identified through variations in labor supply of women with and without children in
various occupations, which cannot be explained by differences in atrophy rates, opportunity costs or selection.
23

by the age of 30. These estimates point at a complex interaction between career- and fertility
decisions. Women in abstract occupations face higher atrophy rates, higher opportunity costs
of leaving the labor market, and have a higher utility cost of handling children and work.
This will induce women with a higher desired fertility to chose more often careers in routine
occupations and to have children earlier. On the other hand, as children are costly, higher
wages in abstract jobs - and the prospect of marrying a better husband - makes this career
also desirable, as assets can be built up faster to smooth consumption when children arrive.
We illustrate these tradeoffs below.
4.3 Utility of Consumption
Table 4 presents estimated parameters that characterize consumption decisions. The estimate
of the discount factor (first row) is 0.96 annually, similar to values in the previous literature
(e.g. Cooley and Prescott (1995)). Our estimate for the curvature of the utility function
(or the relative risk aversion, row 2) is close to 2, again a common value in the literature.
Row 3 displays the estimated weights children have in the consumption equivalence scale.
This parameter is important as it drives not only fertility choices, but also savings choices
to smooth consumption over the life cycle. Our estimate is close to 0.4, slightly higher than
the one in the “modified OECD scale”, which is equal to 0.3. The last two rows present the
estimates of the cost of childcare, distinguishing between the age of the youngest child. These
are estimated to be 31 euros per day for infants and 12 euros per day for older children.35
The consumption costs of children and the cost of childcare suggest that households may
gain from smoothing consumption, by anticipating births. We illustrate this in Figure 2,
which shows that savings rates start to increase at least four years prior to birth, and decline
afterwards. Hence, savings are likely an important factor to understand fertility decisions and
35The average cost of day care in Germany is estimated by the OECD to represent about 11% of net family
income (http://www.oecd.org/els/soc/PF3.4%20Childcare%20support%20-%20290713.pdf), which amounts
to about 15 euros per day. Our estimated parameter takes also into account other expenses linked to work
and children such as transportation.
24

how these are affected by policy interventions, something that we investigate in Section 4.7.
4.4 Unobserved heterogeneity
Panel A in Table 5 lists the parameters that characterize individual types. As explained above,
we allow for unobserved heterogeneity, with two different levels of ability/taste for leisure, and
two types of preferences towards fertility. Columns LA/HC, LA/LC, HA/LC, HA/HC refer
to combinations of Low Ability (LA) - High Ability (HA) and Low taste for children (LC) -
High taste for children (HC) types. Note that as we explain in Section 3, we allow groups with
different ability to have different taste for leisure. The first row reports the proportions for
the four type combinations in our data. The next two rows show the differences in log wages
across ability types, and the utility of leisure, where we have normalized LA type women to
zero. High ability women earn wages that are 0.14 log points higher than low ability women
and have a lower utility of leisure (by about 26%). Rows 4 and 5 display the utility of children
for the different categories, showing that women with a high taste for children (HC) obtain
positive utility for both the first and second child, while LC types only derive a positive utility
for the first child.36 The correlation between taste for children and ability is close to zero,
suggesting that it is not the combination of high ability and low taste for children that leads
women in better paid careers to have fewer children; rather, the choice of a steeper career
paths for these women induces considerable costs through the sacrifice of fertility.
Panel B, which reports estimated type-specific fertility rates and proportions in the dif-
ferent occupations, shows that women with a low taste for children have on average about
one child, while women with a high taste have 1.9 children. Interestingly, we do not find
much difference in terms of total fertility with respect to ability. The last three rows in the
table show the proportion of each type in the three different occupational groups, providing
evidence of sorting on ability and desired fertility: close to 50% of women with a low taste for
36 As the model also allows for idiosyncratic preference shocks towards conception, women with low perma-
nent taste for children (fC
i) may nonetheless have more than one child.
25

children opt for an abstract occupation, while routine occupations are relatively more frequent
for women with a high taste for fertility. 37
4.5 Career Costs of Fertility
We now use our model to assess the career costs of children, which is how much a woman
would gain in monetary terms if she decided not to have children. We evaluate these costs
by simulating life-cycle outcomes under two scenarios. First, we simply match the model
to the fertility pattern in our data, which serves as our baseline scenario. Second, we set
the conception probability to zero, so that a woman knows ex ante that no children will be
conceived, and will therefore base all her decisions on that knowledge. This includes the initial
choice of occupation, as well as labor market decisions and savings over the entire life-cycle.38
We first present the differences in career paths for the two scenarios along various dimensions.
We then compute the cost of children as the net present value of the difference in life cycle
earnings at age 15 between the two scenarios.39
Occupational choice and labor supply. Figure 3a displays the differences in occupational
choices at age 15 between the two scenarios. It shows that the expectation about future fertility
affects the choice of occupation even before fertility decisions are taken: a woman who knows
that she will remain childless is less likely to work in routine and manual occupations (by about
3% and 2% respectively), and more likely to work in occupations involving mainly abstract
tasks (by about 5%). This is an important insight, suggesting that key career decisions are
affected by the expectation about future fertility, possibly long before fertility decisions are
37We report other parameter estimates as well as the arrival rates of offers in different states in the online
appendix.
38Note that, as the probability of marriage and the type of husband depend on endogenous choices such as
occupation and skills, a change in fertility will also affect women along this margin. We find however that
these indirect effects are small.
39As we are interested in the career costs for a single individual, we compute partial equilibrium results.
The results might differ if all women chose not to have children.
26

taken, and implies that some of the career costs of children are determined even before a child
is born. Below we will assess the magnitude of these costs.
Figure 3b plots the difference in labor supply over the life cycle between the two scenarios.
In the no-fertility scenario, a woman is more likely to work at any age: the difference increases
from about 10% in her early twenties, to 30% in her mid thirties. It then declines to about a
10%, as women who had children gradually return to the labor market. Hence, the difference
in labor supply over the life cycle is an important component of the overall costs of children,
as we demonstrate below. Fertility affects labor supply also at the intensive margin: Figure 3c
shows that children increase the probability to work part-time (conditional on working), and
the difference increases with age to reach about 25 percentage points between age 35 and 45.
Interestingly, and comparing Figures 3b and 3c, women who return to the labor market in
their late 30’s and early 40’s tend to remain in part time jobs, compared to the non-fertility
scenario. This feature comes from the fact that women derive a higher utility of leisure when
children are present, but need to work to finance higher consumption needs. The effect of
fertility on women’s labor supply over the life cycle has also a stark impact on work experience:
our simulations show that by the time they retire at age 60, mothers have on average 22%
less work experience per child.
Wages and selection over the life-cycle. Figure 3d plots the deviation of wages in the
no-fertility scenario from the baseline scenario. Here, and in the simulations below, we report
average daily (rather than hourly) wages conditional on working. Hence, differences across
scenarios result from differences in skills, the number of hours worked per day, occupational
choices, and differences in the ability composition of women who choose to work. The figure
shows that, while at age 20, the daily wage in the no fertility scenario is only about 0.03 log
points higher than in the baseline scenario, this difference rises to 0.22 log points by age 40,
and then slightly declines when women return to the labor market. Hours of work contribute
only partly to these differences in daily wages. We find that by age 40, the full time wage in
the non fertility scenario is on average 10% higher than in the baseline scenario.
27

One reason for this difference in wages is composition. There is a long tradition in eco-
nomics - dating back to the seminal work of Heckman (1974) and including work by Blau
and Kahn (1996), Blundell et al. (2007), Mulligan and Rubinstein (2008) and Olivetti and
Petrongolo (2008) - of evaluating the selection of women into the labor market. Our model
allows us to assess the role of fertility decisions in shaping the ability composition of women in
the work force over the life cycle. In Figure 3e, we present the ratios of working women of low
versus high ability over the life cycle under the two scenarios. In the no-fertility scenario, this
ratio is close to 0.43 and relatively stable, while in the fertility scenario, the composition of
working women changes substantially over the career cycle. While at age 20, the ratio is equal
to 0.42, it decreases to 0.37 by age 35, when low ability women are less likely to work than
high ability women,40 and rises again toward the end of the working life as mothers return
to the labor market. Hence, selection into the labor market due to fertility is time-varying,
and depends both on fertility choices and the timing of births. Part of the rise in the wage
differential in Figure 3d is therefore due to this dynamic selection, with the difference in wages
due to differential ability (fP
i) at age 35 being equal to about 0.01 log points out of a total
difference of 0.22 log points.
Decomposing the net present value of fertility choices. The graphs presented above
show various aspects of the career costs of fertility in terms of occupational choice, labor
supply, and wages. We summarize these costs by calculating their net present value at the
start of the career (at age 15) taking account of all earnings, unemployment and maternity
benefits (ws
it,bs
U,it,bs
M,it), where the index s=F, N F stands for the baseline (F) and the no-
fertility scenario (N F ). Defining an indicator variable Ijs
it , which is equal to one if jis true
under scenario s, the net present value for individual iis given by:
N P V s
i=
T
X
t=0
βtwS
itIwor ks
it +bs
U,itIU nempls
it +bs
M,itIM at.Leaves
it .(8)
40As shown above, this is not because low ability women have more children, but rather because they are
less likely to come back to work, once their children are older.
28

We evaluate the relative costs of children by computing 1 −N P V N F /N P V F, using an annual
discount factor of β=0.95. These costs reflect the difference in earnings, labor supply, and
occupational choice induced by the presence of children. Based on this calculation, and
comparing the baseline scenario with the no-fertility scenario, the overall costs of children are
close to 35% of the net present value of income at age 15 (see Table 6).
To better understand the sources of these costs, we isolate two components: the contri-
bution of labor supply, and the contribution of wages (see the second panel in Table 6). The
first component is obtained by fixing wages at the scenario with children, while computing
the difference in terms of labor supply for women with and without children. The second
component fixes labor supply for the no children scenario, and computes the difference in
wages for women with and without children. 41 According to this decomposition, about three
quarters of the costs (i.e., 27% of the total 35.3% overall reduction in lifetime income) re-
sult from differences in the labor supply over the life cycle, while about one quarter results
from differences in wages. Thus, although wages are an important component of the cost of
children, more important are unearned wages of women who drop out of the labor force for
considerable periods over their career.
In the third and fourth blocks of Table 6, we provide two decompositions that break down
the contribution of wages into occupational choice and atrophy when out of work, and a
respective residual term (”other factors”). The figures in the Table show that the overall
contribution of atrophy to the lower wage in the fertility scenario is about 20%, or 5% of the
total life-cycle earnings difference in the fertility vs. the non-fertility scenario. Occupational
choices at the beginning of the career, and before any fertility decision is taken, represent
19% of the overall costs induced through wages, indicating that a substantial portion of the
wage-induced career costs of children is already determined before fertility decisions are made,
through occupational choices conditioned on expected fertility pattern.
Table 7 displays the costs of fertility of having one or two children (in terms of net present
41As in the standard Oaxaca-type decompositions, there are two alternative reference groups. In the table,
we present estimates based on the average of the two.
29

value at age 15), where we allow the spacing of births to differ. The figures in the Table
show that the cost of a second child is lower than the cost of the first child. For instance, a
first child at age 20 induces a total career costs of 31 percent compared to a scenario without
children. A second child conceived at age 22 increases these costs to 36 percent. The results
are qualitatively similar to those in Bertrand, Goldin, and Katz (2010), who also find that the
cost of a second child is much lower than the first. The cost of a second child is increasing in
the spacing of birth, as it prolongs the time the mother spends out of the labor market. The
costs of children are also decreasing in the age at birth, for two reasons. First, as we measure
the discounted costs, more distant costs are valued less. Second, older mothers have time
to establish themselves in the labor market and accumulate sufficient human capital, which
lowers the depreciation rate (see Table 3, panel A). The fact that children impose a lower cost
for older mothers does not imply that it is optimal to have children late, however, as women
also derive utility from their children. The optimal timing of births is therefore a trade-off
between the various costs of children and their utility.
4.6 Fertility and the Gender Gap
Having shown that fertility leads to a sizeable reduction in life-cycle earnings and affects
women’s wage profiles throughout their careers, we now examine the extent to which the
gender gap in earnings can be explained by fertility. To do so, we compare the women studied
here to men of similar qualifications.42 Again, we compute this difference for the average
daily (rather than hourly) wage, which we believe is the most appropriate measure because
it includes the change from full-time to part- time work as an important margin of fertility
adjustment (see Figure 3c).
In Figure 3f, we show the observed daily wages for working males (solid line) and females
42These are men belonging to the same birth cohorts, having the same education (lower or intermediate
secondary school), who enrolled in an apprenticeship training schemes before labor market entry, and whom
we observe from labor market entry onwards
30

(dashed line) by age, as well as the predicted profile for females from our model (dotted line).
The observed and predicted wages for females are very similar, illustrating that the model fits
the data well. Whereas men’s daily wages increase monotonically with age, women’s wages in
the baseline scenario increase up to age 27 but then decrease and only begin increasing again
after age 38. The overall gap increases as women reduce the number of hours worked between
ages 25 and 45 and then return to the labor market with lower labor market experience and
depreciated skills once their children are older.
To assess the contribution of fertility to the gender wage gap, we compute, as above,
the counterfactual wage profile (conditional on working) of a woman who remains childless
and conditions on that knowledge from the start of her career, which in Figure 3f is labeled
”Predicted Females, No Fertility”. The gender gap closes by about 0.2 log points when women
are in their thirties, which corresponds to about a third of the overall gap. 43
4.7 The Effect of Pro-Fertility Policies on Fertility and Women’s
Careers
In many countries, fertility is encouraged in the form of tax relief or transfers. A stream
of literature has evolved on the effects of such policies on fertility, and sometimes on labor
supply. 44 Some of this research, which typically identifies these effects based on policy changes,
43These findings are in line with Bertrand, Goldin, and Katz (2010) who, using different techniques, show
that fertility-induced differences in the labor supply of MBAs explains a large part of the male-female annual
earnings differential, although our population of women is on average less skilled. See also Rosenzweig and
Schultz (1985) and Goldin and Katz (2002) who illustrate the impact of fertility shocks on labor market
participation and wages.
44See, for instance, Cohen, Dehejia, and Romanov (2013) who investigate the effect of Israel’s child subsidy
program on fertility; Laroque and Salanie (2014) who study the impact of child subsidies in France on total
fertility and labor supply; Milligan (2005) who investigates the impact of a new lump sum transfer to families
that have a child in Quebec; Sinclair, Boymal, and De Silva (2012) who analyze the effect of a similar policy
on fertility in Australia; Haan and Wrohlich (2011) who estimate the effect of child care subsidies on fertility
and employment in Germany; Lalive and Zweim¨uller (2009) who investigate the effects of parental leave policy
31

nonlinearities in the tax and transfer system, regional variation, and/or changes or differences
in entitlements across family characteristics, reports considerable effects on total fertility. The
focus of this literature, however, tends to be limited to short-run responses of fertility because
of two important problems: first, it is difficult in many datasets to track women affected by
such policies over an extended period, and until the completion of their fertility cycle. Second,
and more importantly, because data become contaminated over time by other factors that
affect the fertility and careers of particular birth cohorts, making a causal statement about
the effect of a policy some years after its implementation requires restrictive assumptions.
Hence, extant studies pay little attention to long-term consequences and how fertility behavior
is affected at the extensive and intensive margins. 45 Yet, long-run effects of policy changes
and the way they affect behavior of different cohorts are very important for the evaluation of
such policies. Any policy cha29 nge affects different cohorts differently, depending on where
women are in their career and fertility cycle when the policy is implemented. While young
women about to enter the labor market can adjust not only their fertility behavior but also
their occupational choices and entire career paths to the policy change, older women, having
already made most career- and savings decisions, have fewer possibilities of adjustment. The
effects of the policy will thus change over time as more women condition on it when making
their fertility and career choices. At the same time, transfer policies may affect decisions other
than fertility, such as labor supply, occupational choices, savings decisions, and human capital
investments. These “secondary” effects, despite being important for assessing the full impact
of these policies, are hardly investigated.
To help fill this void, we use our life-cycle model to evaluate the effect of a policy that
on fertility in Austria; and Kearney (2004) who studies the impact on fertility of caps on child benefits paid
for an additional child.
45Two notable exceptions are Parent and Wang (2007) and Kim (2014). Parent and Wang (2007) follow a
cohort of Canadian women over their fertility cycle and find that the long-run response is low compared to
the short-run response. Kim (2014) studies the long run impact of changes in the child allowance policy in
Quebec (see also Milligan (2005)). He finds a small or no permanent impact on fertility.
32

provides a cash transfer at birth of 6,000 euros. Policies of this type have been implemented
in different countries, as illustrated in the above mentioned literature. We show the effect of
the policy on the probability of giving birth, by age, in Figure 4a, comparing the behaviour
of women under the baseline and the policy. The difference in the probability is positive at
first and then negative, showing that the policy induces women to have their children earlier,
but it has little effect on the overall number of children per woman.
Next we investigate the aggregate effect of the policy, by computing the number of children
born every year, before and after the policy is implemented. In doing this, we leave aside
general effects. We use our model to simulate many overlapping birth cohorts of women,
between the age of 15 and 60. Each year, a new cohort enters and the oldest cohort exits.
Hence, when the policy is implemented, women are at different stages of their life-cycle. The
older ones have already made their fertility choices, while the youngest ones are still far from
their first child. However, the latter can alter their occupation, their labor supply or their
savings in response to the policy. Figure 4b plots the increase in the number of children born
every year, compared to a baseline without a cash transfer. The policy starts in year 4 and
is not anticipated. We observe a spike in the number of children born, with a 4.5 percent
increase in total fertility in the first year of the policy. This spike is what a reduced form
analysis would identify as the short-run effect of the policy. However, the effect of the policy
lasts more than a few years. The effect reduces to half that size after 8 years and is very close
to zero after 20 years. Simulating policies with various levels of benefit, we find a short-run
elasticity with respect to benefits of about 0.04.46
While Figure 4b displays the composite response to the policy for women in different
age groups, we illustrate in Table 8 the effect of the policy on women in three different age
groups at the start of the policy (15, 25, 35). For the group of women who are 15 when
the policy is implemented and who cannot only adjust fertility, but also their labor supply,
consumption- and occupation choices, the proportion of those with no children decreases by
46This elasticity is similar to the ones reported by Zhang, Quan, and van Meerbergen (1994) (0.05) and
lower than Milligan (2005) (0.10).
33

0.8%, while the proportion with two children (or more) increases by 0.2%. The policy induces
women to have their first child earlier (by 0.4 years) and leads mothers to spend about 0.1
years longer out of work, which translates into lower levels of skills (by about 0.3%). Similar
effects of cash transfers have been found by Card, Chetty, and Weber (2007) in the context
of a lump-sum severance payment. We also find a moderate increase in part-time work for
the youngest cohort. Finally, the proportion of women opting for a routine or manual job
increases respectively by 0.3 and 0.07%. Cash transfers therefore allow women to opt for less
lucrative careers, while maintaining a similar level of consumption. For older cohorts, the
responses are muted as many decisions have already been made, and women have less scope
to respond to the policy. For instance, as occupational choices are made predominantly at a
young age before training in vocational schools, there is no discernable effect on occupational
choices.
An important channel through which cash transfers affect behavior are assets. One reason
for fertility to be brought forward are that less assets need to be accumulated before a child
is born. To investigate this further, we plot in Figure 4c the change in assets due to the
policy for women who have been 15, 20, and 25 when the policy was implemented, in percent
deviation from the no policy scenario. The youngest cohort anticipates the policy and saves
less in their early twenties. When children are born, around age 27 on average, the conditional
cash transfer is saved and spread over a period lasting about 10 years. Assets then decrease
below the baseline by about 2% as the household has lower resources due to lower skill levels
and more children. The pattern for the older cohorts are similar, but these have less scope to
adjust their savings.
These results highlight the important difference not only in the short- and longer run effect
of these policies on choices other than fertility, but also stress that the impact of these policies
may be largest for cohorts that do not show immediate fertility responses, due to their younger
age. For these cohorts, such policies may have important consequences for career decisions as
well as savings decisions - aspects that are usually not investigated in the literature.
34

5 Conclusion
In this paper, we develop and estimate a model of fertility and career choice that sheds light on
the complex decisions determining fertility choices, how these interact with career decisions,
and how they determine the career costs of children. Following early work by Polachek (1981),
we consider occupational choice as an essential part of a woman’s career plan. We show that
different occupations imply not only different opportunity costs for intermittency and different
wage growth, but diverge in the amenity ”child raising value”. Moreover, the loss of skills
when interrupting work careers varies across occupations, is non-linear over the career cycle,
and are highest at around mid-career, which has potentially important implications for the
interplay between career choice and fertility.
Thus, the costs of fertility consist of a combination of occupational choice, lost earnings
due to intermittency, lost investment into skills and atrophy of skills while out of work, and
a reduction in work hours when in work. In addition, fertility plans affect career decisions
already before the first child is born, through the choice of the occupation for which training
is acquired - an aspect that is not only important for policies aimed at influencing fertility be-
havior, but also for understanding behavior of women before children are born. An important
additional aspect for the life time choices of fertility and career are savings that help women
to smooth consumption. Furthermore, fertility leads to sorting of women into work, with the
composition of the female workforce changing over the life course of a cohort of women, due
to different career- and fertility choices made by women of different ability.
These complex interdependencies between fertility and career choices imply that pro-
natalist policies have effects over and above their primary intention, something that we il-
lustrate in the simulations of our model. Moreover, the impact of any such policy is likely to
be particularly pronounced for cohorts of women that are at the beginning of their careers,
as they are able to adjust all future decisions in response, such as occupational choices and
the timing of the first birth. These women however are usually not the subject of analysis
in empirical work that evaluates these policies, the reason being that due to their young age,
35

their fertility behavior does not respond around the policy implementation. Furthermore, it
is not just fertility that may be affected, but other career decisions associated with fertility as
well. Our analysis suggests that responses of this sort may be important, leading to possibly
undesired consequences of any such policies. As DiD designs require restrictive assumptions
to interpret longer term effects to policy interventions as causal, they typically focus on short
term effects around the policy intervention. Combinations of clean designs with structural
models of the sort presented in this paper may therefore be an avenue that helps exploring
the longer term effects of policy interventions.
36

Appendix
A Model Description
A.1 Probability of Marriage and Divorce
The probability of marriage is a function of age, skills and taste for children:
P(hit = 1|hit−1= 0; ageM
it , xit, f C
i) = λM
0+λM
1(ageM
it ) + λM
2xit +λM
3fC
i,(9)
where λM
1(.) is a non-linear function of the age of the woman. We define the probability of
divorce as a function of age and the number of children in the household:
P(hit = 0|hit−1= 1; ageM
it , nit) = λD
0+λD
1(ageM
it ) + λD
2nit ,(10)
where again λD
1(.) is a non-linear function of the age of the woman.
A.2 Job Offer Probability
Offers consist of an occupation ˜oand of hours of work ˜
l(either part time or full time work).
New offers arrive randomly and depends on the current occupation and hours of work. The
probability of receiving a job offer is denoted φ0(oit, lit). Conditional on having received an
offer, the probability of that offer being in occupation ˜owith hours of work ˜
lis φ1(˜o, ˜
l|oit, lit )
and depends again on current occupation and hours of work. We impose some structure on
that probability as it contains potentially many terms to be estimated. We assume that the
offer concerning hours of work depends only on prior hours of work, whereas occupation offers
depend on prior occupation and prior working status (i.e. working or out of the labor force,
but not whether the individual is in part time or full time work). Variations in the rate of
part time work across occupations in the model comes from differential fertility choices across
women and the amenity value of occupations with regards to children.
37

A.3 Utility Function
Women derive utility from their own consumption, the number of children, and leisure. We
define Ijan indicator variable taking the value of one if jis true and zero otherwise. The
utility function takes the following form for individual iin period t:
uit =(cit/¯c)(1−γC)−1
1−γC
exp γ1
P T Ilit =P T + (γ1
U+fL
i)Ilit=U+ (γ1
OLF +fL
i)Ilit=OLF exp(γN C Init >0)
+γ1
N(fC
i)Init=1 +γ2
N(fC
i)Init=2.exp (γN H Init>0&hit =1)
.exp (γU)Ilit=U.exp γOLF +γ1
A,OLF IageK
it ∈[0,3] +γ2
A,OLF IageK
it ∈[4,6] +γ3
A,OLF IageK
it ∈[7,9]Ilit =OLF
.exp 

3
X
iO=1
γio,P T Ioit =io+γ1
A,P T IageK
it ∈[0,3] +γ2
A,P T IageK
it ∈[4,6] +γ3
A,P T IageK
it ∈[7,9]

Ilit=P T
.exp 

3
X
iO=1
γio,W Ioit=io


Ilit=P T ,F T
+ηC
it bit +ηNC
it (1 −bit) (11)
The first term is the utility obtained from consumption (cit). The parameter γCis the relative
risk aversion and ¯cis a consumption scale. As in Attanasio, Low, and Sanchez-Marcos (2008)
and Blundell et al. (2013), we allow for an interaction between consumption and labor supply.
We distinguish between part time work, unemployment and being out of the labor force. We
introduce heterogeneity in the utility of leisure through the variable fL
i. We also allow the
marginal utility of consumption to differ when children are present (through the parameter
γN C ). The individual also derives utility from the number of children, which is displayed in
the second line. The parameters γ1
N(fC
i), γ2
N(fC
i) vary with the taste for children, fC
i. Finally,
we allow the utility from children to differ when a husband is present (hit = 1).
The third to sixth lines allow for the utility of children to vary with labor supply and
occupation choices. In a demanding occupation, the individual derives a lower utility from
children, as it is more difficult to spend time with them. For instance, even if part time work
is available, the woman may not be able to stay at home when the child is sick or reschedule
hours of work to attend a school performance. We also distinguish between different statuses
of non-work, as women who are unemployed may require time to search for a job. It should be
noted that, because full time work is the baseline, we do not specify a utility level associated
38

with that outcome. In lines four and five, we allow mothers who work part time to obtain
utility from leisure (relative to full-time work) dependent on the age of their youngest child.
Here, we distinguish between infancy (0 to 3 years), preschool (4 to 6 years), and primary
school (7 to 9 years). The final part of the utility function introduces iid preferences towards
conception or non conception, denoted ηC
it and ηNC
it . The shock that affects the woman depends
on whether she decides to conceive or not (indicated by the indicator variable bit. These shocks
are assumed iid and extreme value distributed.
A.4 Dynamic Choice
We now describe in more detail the dynamic choices individuals make. Table A1 in the online
appendix lists the notations used in the model. The main text describes it with the generic
Bellman equation:
Vt(Ωit) = max
{bit,cit ,oit,lit }u(cit , oit, lit ;nit , hit, ageK
it ,Υit, fi) + βEtVt+1(Ωit+1) (12)
with the state space defined as:
Ωit =nlit−1, oit−1, Ait−1, hit−1, ageM
it , xit, nit , ageK
it ,Υit, fio(13)
The Bellman equation can be decomposed into a sequence of choices, involving conditional
value functions, where the conditioning is on labor supply status and the decision to conceive
or not. We make the distinction between being in work, being unemployed or out of work
because individuals face different choice sets. For instance, individuals out of the labor force
are not eligible for unemployment benefits and cannot choose to become unemployed in the
next period. Individuals who chose to conceive have a probability of becoming pregnant,
and cannot be fired. Hence, these conditional value functions model institutional features
explicitly, which are only implicit in (12).
The individual maximizes these conditional value functions in sequence, which simplifies
the overall model as one can rely on closed-form solutions for some of the choices, given
39

particular distributional assumptions on the taste shocks in Υit (extreme value distribution).
We denote these conditional value functions by indexing them with Cfor conception or NC
if the individual decides not to conceive. We also index them with Wfor work (either part
time or full time, the distinction hours of work is contained in the state variable lit in Ωit), U
for unemployment and Ofor out of the labor force. Finally, we introduce two value functions
describing individuals after birth, who enter that state from work or non-employment, and
index these respectively by LWand LNW . At the beginning of a period, women take as given
their age, skills, occupation, labor supply in the previous period, the number of children, the
age of the youngest child, whether the spouse is present, and family assets. Women first
observe the income shock to their wage and to the earnings of the husband, if present, and
then decide on whether to conceive a child or not. If conception is successful, the child is born
at the beginning of the next period. Women next decide how much to consume and save.
Once fertility and consumption choices have been made, individuals observe shocks to
labor supply, which consists of layoffs (if in work) and job offers. These shocks determine the
labor status at the beginning of the next period. New offers arrive randomly and have two
features: occupation and part time or full time work. The probability of receiving a job offer
is denoted φ0(oit, lit ) and depends on the current occupation and hours of work. Conditional
on having received an offer, the probability of that offer being in occupation ˜owith hours of
work ˜
lin period t+ 1 is φ1(˜o, ˜
l|oit, lit ) and depends again on current occupation and hours of
work.
Value of working. We start with the value of working and conceiving a child. In writing the
values, we distinguish their deterministic part from the stochastic part due to the preference
shocks, which we introduce below and which enter in a linear and additive way. As the decision
to conceive has already been made, the woman has to decide how much to consume. Choices
over occupations and hours of work are taken at the end of the period. The value is written
40

as:
VW,C (Ωit) = max
cit ucit, oit , lit;nit , hit , ageK
it ,Υit, fi(14)
+π(ageM
it , f F
i)βEtVLW(ΩP
it+1)
+δ(1 −π(ageM
it , f F
i))βEtVU(Ωit+1 )
+(1 −δ)(1 −π(ageM
it , f F
i))(1 −φ0(oit, lit ))βEmax
+(1 −δ)(1 −π(ageM
it , f F
i))βφ0(oit , lit)Eg
max,
where Etis the expectation operator. The first line consists of the current utility of con-
sumption, leisure and children. The second line is the future flow of utility if conception is
successful, which occurs with a probability π(ageM
it ). As the woman is working in the current
period, she is entitled to paid maternity leave, with a flow of utility VLW(.), defined below.
This value depends on the next state space ΩP
it+1, where the subscript Pindicates that the
women is pregnant, so that the number of children is increased by one and the age of the
youngest child is set to zero. Assets and skills evolve as described in equations (1) and (2).
The last three lines describe the case when conception is unsuccessful. With a probability
δthe individual is laid off and starts next period in unemployment, with a value VU(.). If she
is not laid off, she does not get an alternative job offer with a probability 1 −φ0(oit, lit), and
has to choose between staying in work, leaving for unemployment or leaving the labor force.
We define the term Emax as:
Emax =Etmax[VW(Ωit+1 ) + ηW
it+1, V U(Ωit+1 ) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1] (15)
The ηk
it+1, k =W, U, O, are utility shocks, and we assume that they are iid and follow an
extreme value distribution, which leads to a closed form solution for the Emax operator. The
final row of equation (14) describes the case when the individual receives an alternative job
offer {˜o, ˜
l}. This happens with a probability φ1(˜o, ˜
l|oit, lit ). In this case, the individual has to
also decide whether to choose this new job. We define the continuation value as:
Eg
max =EtX
˜o6=oit,˜
l6=lit
φ1(˜o, ˜
l|oit, lit ) max[VW(Ωit+1) + ηW
it+1, V W(˜
Ωit+1) + ˜ηW
it+1,
41

VU(Ωit+1) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1] (16)
where ˜
Ωit+1 is the future state space when the individual accepts the alternative job {˜o, ˜
l}and
where ˜ηW
it is the shock associated with the alternative offer. The value of working without
conceiving is defined as:
VW,NC (Ωit ) = max
cit ucit, oit , lit;nit , hit , ageK
it ,Υit, fi(17)
+βδEtVU(Ωit+1)
+β(1 −δ)(1 −φ0(oit, lit ))Emax
+β(1 −δ)φ0(oit, lit )Eg
max
At the beginning of next period, the individual starts with an updated state space Ωit+1,
where all the state variables have been updated but the number of children. Here again, the
individual can be laid off and starts as unemployed, or has to chose next period’s labor market
status.
Value of unemployment. When unemployed, the individual can chose whether to stay
unemployed for another period, or exit the labor market altogether. If a job offer is received,
the individual then decides whether to take up the offer or to remain non-employed. The
value of being in unemployment and not conceiving is:
VU,N C (Ωi) = max
cit ucit, oit , lit;nit , hit , ageK
it ,Υit, fi(18)
+β(1 −φ0(oit, lit ))Etmax[VU(Ωit+1) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1]
+βφ0(oit , lit))EtX
˜o6=oit,˜
l6=lit
φ1(˜o, ˜
l|oit, lit )
max[VU(Ωit+1) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1, V W(˜
Ωit+1) + ˜ηW
it+1]
where we again denote with a tilda the variables involved with the alternative job, e.g. ˜
Ωit+1
is the state space for women who accepted an alternative job. The value of conceiving while
in unemployment is defined as:
VU,C (Ωit) = max
cit ucit, oit , lit;nit , hit , ageK
it ,Υit, fi(19)
42

+π(ageM
it , f F
i)βEtVLN W (ΩP
it+1)
+(1 −φ0(oit, lit ))(1 −π(ageM
it , f F
i))βEtmax[VU(Ωit+1 ) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1]
+β(1 −π(ageM
it , f F
i))φ0(oit, lit )EtX
˜o,˜
l
φ1(˜o, ˜
l|oit, lit ) max[VU(Ωit+1) + ηU
it+1,
VO(Ωit+1) + ηO
it+1, V W(˜
Ωit+1) + ˜ηW
it+1]
If conception is successful, the mother is entitled to maternity leave, but will not be entitled
to a job at the end of that spell, generating a flow of utility VLNW (.) as defined below.
Value of being out of the labor force. The value of being out of work and trying to
conceive a child is modeled as:
VO,C (Ωit) = max
cit ucit, oit , lit;nit , hit , ageK
it ,Υit, fi(20)
+π(ageM
it , f F
i)βEtVLN W (ΩP
it+1)
+(1 −φ0(oit, lit ))(1 −π(ageM
it , f F
i))βEtVO(Ωit+1 )
+φ0(oit, lit )(1 −π(ageM
it , f F
i))βEtX
˜o6=oit,˜
l6=lit
φ1(˜o, ˜
l|oit, lit )
max[VO(Ωit+1) + ηO
it+1, V W(˜
Ωit=1) + ˜ηW
it+1]
whereas the value of not conceiving is:
VO,N C (Ωit) = max
cit ucit, oit , lit;nit , hit , ageK
it ,Υit, fi(21)
+(1 −φ0(oit, lit ))βEV O(Ωt+1)
+φ0(oit, lit )βEtX
˜o6=oit,˜
l6=lit
φ1(˜o, ˜
l|oit, lit )
max[VO(Ωit+1) + ηO
it+1, V W(˜
Ωit=1) + ˜ηW
it+1]
It should be noted that individuals who are out of the labor force cannot become unemployed
and start claiming benefits.
Value of maternity leave. Maternity leave lasts for two periods during which the mother
is not working and receives maternity benefit. The amount she gets depends on her prior
43

labor market status. The value of maternity for a woman who previously worked is defined
as:
VLW(Ωit) = max
cit,cit+1
ucit, oit , lit;nit , hit , ageK
it ,Υit, fi+βu cit+1, oit, lit ;nit, hit, ageK
it+1, fi(22)
+(1 −φ0(oit, lit ))β2Etmax[VW(Ωit+1) + ηW
it+1, V U(Ωit+1 ) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1]
+φ0(oit, lit )β2EtX
˜o6=oit,˜
l6=lit
φ1(˜o, ˜
l|oit, lit ) max[VW(Ωit+1) + ηW
it+1,
VW(˜
Ωi,t+1) + ˜ηW
it+1, V U(Ωit+1 ) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1]
In this state, the women is entitled to maternity leave, which consists on a fixed transfer, and
on a variable one, which is a function of prior earnings. If the individual did not work prior to
giving birth, she is not guaranteed a job at the end of the maternity leave and receives only
the fixed transfer:
VLNW (Ωit ) = max
cit,cit+1
ucit, oit , lit;nit , hit , ageK
it ,Υit, fi+βu cit+1, oit, lit ;nit, hit, ageK
it+1, fi
+(1 −φ0(oit, lit ))β2Etmax[VU(Ωit+1) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1]
+φ0(oit, lit )β2EtX
˜o,˜
l
φ1(˜o, ˜
l|oit, lit ) max[VW(˜
Ωi,t+1) + ˜ηW
it+1,
VU(Ωit+1) + ηU
it+1, V O(Ωit+1 ) + ηO
it+1] (23)
Conception decision. The decision of whether to conceive or not, is taken as:
Vk(Ωit) = max[Vk,C (Ωit ), V k ,NC (Ωit )], k ={W, U, O}(24)
As the preference shocks towards conception and non-conception ηC
it and ηNC
it , which are
part of the state vector Ωit, are drawn from an extreme value distribution, the probability
of conception takes a logistic form, with the values of conception and non-conception as
arguments. The decision to conceive, noted bit in equation (5) in the main text is the arg max
of expression (24).
44

References
Albrecht, James, Per-Anders Edin, Marianne Sundstr¨om, and Susan Vroman. 1999. “Career
Interruptions and Subsequent Earnings: A Reexamination Using Swedish Data.” Journal
of Human Resources 34:294–312.
Altonji, Joseph G. and Lewis M. Segal. 1996. “Small-Sample Bias in GMM Estimation of
Covariance Structures.” Journal of Business and Economic Statistics 14 (3):353–366.
Altug, S and R A Miller. 1998. “The Effect of Work Experience on Female Wage and Labor
Supply.” Review of Economic Studies 65 (1):45–85.
Arroyo, C R and J Zhang. 1997. “Dynamic Microeconomic Models Of Fertility Choice: A
Survey.” Journal of Population Economics 10:23–65.
Attanasio, Orazio, Hamish Low, and Virginia Sanchez-Marcos. 2008. “Explaining Changes in
Female Labour Supply in a Life-cycle Model.” American Economic Review 98:1517–1552.
Autor, David H., Frank Levy, and Richard J. Murnane. 2003. “The Skill Content of
Recent Technological Change: An empirical exploration*.” Quarterly Journal of Eco-
nomics 118 (4):1279–1333. URL http://www.mitpressjournals.org/doi/abs/10.1162/
003355303322552801.
Becker, Gary. 1960. “An Economic Analysis Of Fertility.” In Demography And Economic
Change In Developed Countries. Universities-NBER Research Conference Series 11, Prince-
ton, 209–231.
Becker, Gary and G H Lewis. 1973. “On the Interaction between the Quantity and Quality
of Children.” Journal of Political Economy 81:279–88.
Bertrand, Marianne, Claudia Goldin, and Lawrence F. Katz. 2010. “Dynamics of the Gender
Gap for Young Professionals in the Financial and Corporate Sectors.” American Economic
Journal: Applied Economics 2:228–255.
Black, Sandra E and Alexandra Spitz-Oener. 2010. “Explaining Women’s Success: Tech-
nological Change and the Skill Content of Women’s Work.” Review of Economics and
Statistics 92 (1):187–194. URL http://www.mitpressjournals.org/doi/abs/10.1162/
rest.2009.11761.
Blackburn, McKinley L., David E. Bloom, and David Neumark. 1990. “Fertility timing, wages,
and human capital.” Journal of Population Economics 6:1–30.
Blau, David M. and Philip K. Robins. 1988. “Child-Care Costs and Family Labor Supply.”
The Review of Economics and Statistics 70:374–381.
45

Blau, Francine D. and Lawrence M. Kahn. 1996. “Wage Structure and Gender Earnings
Differentials: An International Comparison.” Economica 63:S29–S62.
Blundell, Richard, Monica Costa Dias, Costas Meghir, and Jonathan Shaw. 2013. “Female
Labour Supply, Human Capital and Welfare Reform.” NBER Working Paper No. 19007.
Blundell, Richard, Amanda Gosling, Hidehiko Ichimura, and Costas Meghir. 2007. “Changes
in the Distribution of Male and Female Wages Accounting for Employment Composition
Using Bounds.” Econometrica 75:323–363.
Blundell, Richard and Thomas Macurdy. 1999. “Handbook of Labor Economics.” chap. Labor
supply: A review of alternative approaches. Elsevier, 1559–1695.
Card, David, Raj Chetty, and Andrea Weber. 2007. “Cash-on-Hand and Competing Models
of Intertemporal Behavior: New Evidence from the Labor Market.” The Quarterly Journal
of Economics :1511–1560.
Catalyst. 2009. “Women in the Fortune 500.” Tech. rep., Retrieved from
http://www.catalyst.org/press-release/104/women-in-the-fortune-500.
Centers for Disease Control and Prevention. 2002. “Fertility, Family Planning, and Repro-
ductive Health of U.S. Women: Data From the 2002 National Survey of Family Growth.”
Tech. Rep. 23, U.S. Department of Health and Human Services.
Chiappori, Pierre-Andr´e, Murat Iyigun, and Yoram Weiss. 2009. “Investment in schooling
and the marriage market.” American Economic Review 99 (5):1689–1713.
Cigno, A and J Ermisch. 1989. “A Microeconomic Analysis Of The Timing Of Births.”
European Economic Review 33:737–760.
Cohen, Alma, Rajeev Dehejia, and Dmitri Romanov. 2013. “Financial Incentives and Fertil-
ity.” Review of Economics and Statistics 95 (1):1–20. URL http://dx.doi.org/10.1162/
REST_a_00342.
Cooley, Thomas F and Edward C Prescott. 1995. “Economic growth and business cycles.”
In Frontiers of business cycle research, edited by Thomas F Cooley, chap. 1. Princeton
University Press, 1–38.
Del Boca, Daniela, Christopher Flinn, and Mathew Wiswall. 2014. “Household Choices and
Child Development.” Review of Economic Studies 81:137–185.
Duesseldorfer Tabelle. 2005. “Retrieved from http://www.olg-duesseldorf.nrw.de/infos.”
URL http://www.olg-duesseldorf.nrw.de/infos/Duesseldorfer_tabelle/Tabelle_
2005/2005-07-01_duess-tab.pdf.
46

Duffie, D and K. Singleton. 1993. “Simulated Moments Estimation of Markov Models of Asset
Prices.” Econometrica 61:929–952.
Eckstein, Zvi and Osnat Lifshitz. 2014. “Household Interaction and the Labor Supply of
Married Women.” Mimeo.
Eckstein, Zvi and Kenneth I. Wolpin. 1989. “Dynamic Labour Force Participation of Married
Women and Endogenous Wage Growth.” Review of Economic Studies 56 (3):375–390.
Eisenhauer, Philipp, James J Heckman, and Stefano Mosso. 2014. “Estimation of Dynamic
Discrete Choice Models by Maximum Likelihood and the Simulated Method of Moments.”
forthcoming International Economic Review .
Fitzenberger, Bernd and Astrid Kunze. 2005. “Vocational Training and Gender: Wages and
Occupational Mobility among young Workers.” Oxford Review of Economic Policy 21:392–
415.
Francesconi, Marco. 2002. “A Joint Dynamic Model of Fertility and Work of Married Women.”
Journal of Labor Economics 20:336–380.
Gathmann, Christina and Uta Schonberg. 2010. “How General Is Human Capital? A Task-
Based Approach.” Journal of Labor Economics 28 (1):1–49.
Gayle, George-Levi and Robert A Miller. 2006. “Life-Cycle Fertility and Human Capital
Accumulation.” . Tepper School of Business. Paper 128.
Goldin, Claudia. 2014. “A Grand Gender Convergence: Its Last Chapter.” American Eco-
nomic Review 104 (4):1091–1119.
Goldin, Claudia and Lawrence Katz. 2002. “The Power of the Pill: Oral Contraceptives and
Women’s Career and Marriage Decisions.” Journal of Political Economy 110 (4):730–770.
Gronau, Reuben. 1988. “Sex-Related Wage Differentials and Women’s Interrupted Labor
Careers- the Chicken or the Egg.” Journal of Labor Economics 6 (3):277–301.
Haan, Peter and Katharina Wrohlich. 2011. “Can child care policy encourage employment
and fertility?: Evidence from a structural model.” Labour Economics 18 (4):498 – 512.
URL http://www.sciencedirect.com/science/article/pii/S0927537110001533.
Hagenaars, A, K. de Vos, and M A Zaidi. 1994. Poverty Statistics in the Late 1980s: Research
Based on Poverty Statistics in the Late 1980s: Research Based on Micro-data. Luxembourg:
Office for Official Publications of the European Communities.
Heckman, James. 1974. “Shadow Prices, Market Wages, And Labor Supply.” Econometrica
42:679–694.
47

Heckman, James and B. Singer. 1984. “A Method for Minimizing the Impact of Distributional
Assumptions in Econometric Models for Duration Data.” Econometrica 52 (2):271–320.
Heckman, James and Robert J. Willis. 1976. “Estimation of a Stochastic Model of Repro-
duction An Econometric Approach.” In Household Production and Consumption, edited by
Nestor E. Terleckyj. NBER, 99 – 146.
Heckman, James J. and Thomas E. Macurdy. 1980. “A Life Cycle Model of Female Labour
Supply.” Review of Economic Studies 47 (1):47–74.
Heckman, James J. and Stefano Mosso. 2014. “The Economics of Human Development and
Social Mobility.” Annual Review of Economics 6 (1):689–733.
Heckman, James J. and James R. Walker. 1990. “The Relationship Between Wages and
Income and the Timing and Spacing of Births: Evidence from Swedish Longitudinal Data.”
Econometrica 58 (6):1411–1441.
Hotz, Joseph and Robert Miller. 1988. “An Empirical Analysis of Life Cycle Fertility and
Female Labor Supply.” Econometrica 56 (1):91–118.
Hotz, V. Joseph and Robert A. Miller. 1993. “Conditional Choice Probabilities and the
Estimation of Dynamic Models.” Review of Economic Studies 60:497–529.
Hyslop, Dean R. 1999. “State dependence, serial correlation and heterogeneity in intertem-
poral labor force participation of married women.” Econometrica 67.
Keane, Michael and Robert Sauer. 2009. “Classification Error in Dynamic Discrete Choice
Models: Implications for Female Labor Supply Behavior.” Econometrica 77:975–991.
Keane, Michael and Kenneth I Wolpin. 2010. “The Role of Labor and Marriage Markets,
Preference Heterogeneity, and the Welfare System in the Life Cycle Decisions of Black,
Hispanic, and White Women.” International Economic Review 51:851–892.
Kearney, Melissa Schettini. 2004. “Is There an Effect of Incremental Wel-
fare Benefits on Fertility Behavior?.” Journal of Human Resources 39 (2):295
– 325. URL http://ezproxy.library.usyd.edu.au/login?url=http://search.
ebscohost.com/login.aspx?direct=true&db=buh&AN=13257899&site=ehost-live.
Kennan, John and James R. Walker. 2011. “The Effect of Expected Income on Individual
Migration Decisions.” Econometrica 79 (1):211–251.
Khatamee, M A and M S Rosenthal. 2002. The fertility Sourcebook. 3rd Edition. McGraw-Hill
Contemporary, New York.
Kim, Moon-Kak and Solomon W. Polachek. 1994. “Panel Estimates of Male-Female Earnings
Functions.” The Journal of Human Resources 29:406–428.
48

Kim, Young-Il Albert. 2014. “Lifetime impact of cash transfer on fertility.” Canadian Studies
in Population 41:97–110.
Lalive, Rafael and Josef Zweim¨uller. 2009. “How Does Parental Leave Affect Fertility And Re-
turn To Work? Evidence From Two Natural Experiments.” Quarterly Journal of Economics
124:1363–1402.
Laroque, Guy and Bernard Salanie. 2014. “Identifying The Response Of Fertility To Financial
Incentives.” Journal of Applied Econometrics 29 (2):314–332. URL http://dx.doi.org/
10.1002/jae.2332.
Leung, Siu Fai. 1994. “Uncertain Lifetime, the Theory of the Consumer and the Life Cycle
Hypothesis.” Econometrica 62 (5):1233–1239.
Milligan, Kevin. 2005. “Subsidizing the Stork: New Evidence on Tax Incentives
and Fertility.” Review of Economics and Statistics 87 (3):539 – 555. URL
http://ezproxy.library.usyd.edu.au/login?url=http://search.ebscohost.com/
login.aspx?direct=true&db=eoh&AN=0809034&site=ehost-live.
Moffitt, Robert. 1984. “Profiles of Fertility, Labour Supply and Wages of Married Women: A
Complete Life-Cycle Model.” Review of Economic Studies 51 (2):263–278.
Mulligan, Casey B. and Yona Rubinstein. 2008. “Selection, investment, and women’s relative
wages over time.” Quarterly Journal of Economics 123:1061–1110.
Olivetti, Claudia and Barbara Petrongolo. 2008. “Unequal Pay or Unequal Employment? A
Cross-country Analysis of Gender Gaps.” Journal of Labor Economics 26:621–654.
Pakes, Ariel and D Pollard. 1989. “Simulation and the Aysmptotics of Optimization Estima-
tors.” Econometrica 57:1027–1057.
Parent, Daniel and Ling Wang. 2007. “Tax incentives and fertility in Canada: quantum vs
tempo effects.” Canadian Journal of Economics 40:371–400.
Polachek, Solomon. 1981. “Occupational Self-Selection: A Human Capital Approach To Sex
Differences In Occupational Structure.” Review of Economics and Statistics 63 (1):60–69.
Rosenzweig, Mark and Paul Schultz. 1983. “Estimating a Household Production Function:
Heterogeneity, the Demand for Health Inputs, and Their Effects on Birth Weight.” Journal
of Political Economy 91 (5):723–746.
———. 1985. “The Demand For And Supply Of Births: Fertility And Its Life Cycle Conse-
quences.” American Economic Review 75 (5):992–1015.
49

Sheran, Michelle. 2007. “The career and family choices of women: A dynamic analysis of
labor force participation, schooling, marriage, and fertility decisions.” Review of Economic
Dynamics 10 (3):367–399.
Sinclair, Sarah, Jonathan Boymal, and Ashton De Silva. 2012. “A Re-Appraisal of the Fertility
Response to the Australian Baby Bonus.” Economic Record 88:78–87. URL http://dx.
doi.org/10.1111/j.1475-4932.2012.00805.x.
UNICEF. 2001. “A league table of teenage births in rich nations.” Innocenti report card no.3.,
UNICEF Innocenti Research Centre, Florence.
van der Klaauw, Wilhem. 1996. “Female Labor Supply and Marital Status Decisions: A
Life-Cycle Model.” Review of Economic Studies 63 (2):199–235.
Ward, M P and W P Butz. 1980. “Completed Fertility And Its Timing.” Journal of Political
Economy 88 (5):917–940.
Weichselbaumer, D and R Winter-Ebmer. 2005. “A meta-analysis on the international gender
wage gap.” Journal of Economic Surveys 19:479–511.
Weiss, Yoram and Reuben Gronau. 1981. “Expected Interruptions in Labor Force Partic-
ipation and Sex-Related Differences in Earnings Growth.” Review of Economic Studies
48.
Willis, Robert J. 1973. “A New Approach to the Economic Theory of Fertility Behavior.”
The Journal of Political Economy 81:S14–S64.
Wolpin, Kenneth I. 1984. “An Estimable Dynamic Stochastic Model of Fertility and Child
Mortality.” Journal of Political Economy 92 (5):852–874.
Zhang, Junsen, Jason Quan, and Peter van Meerbergen. 1994. “The Effect of Tax-Transfer
Policies on Fertility in Canada, 1921-88.” The Journal of Human Resources 29 (1):181–201.
50

Table 1: Descriptive Statistics, by Occupation
Routine Abstract Manual Whole sample
Initial occupation 25.0% 44.8% 30.3% 100%
Occupation of work 25.4% 52.7% 21.9%
Annual occupational transition rates:
if in Routine last year 97.9% 1.5% 0.5%
if in Abstract last year 0.7% 99.0% 0.2%
if in Manual last year 0.9% 0.8% 98.3%
Log wage at age 20 3.598 3.742 3.470 3.634
(0.297) (0.301) (0.386) (0.337)
Log wage growth, at potential experience=5yrs 0.0485 0.0551 0.0450 0.0510
(0.187) (0.156) (0.196) (0.175)
Log wage growth, at potential experience=10yrs 0.0181 0.0240 0.0152 0.0208
(0.187) (0.206) (0.223) (0.206)
Log wage growth, at potential experience=15yrs 0.00995 0.0147 0.0127 0.0133
(0.206) (0.195) (0.211) (0.200)
Total work experience after 15yrs 11.55 12.81 12.14 12.34
(3.273) (2.624) (2.880) (2.909)
Full time work experience after 15yrs 10.32 11.92 10.86 11.29
(3.907) (3.348) (3.570) (3.617)
Part time work experience after 15yrs 1.229 0.889 1.274 1.056
(2.187) (1.828) (2.125) (1.997)
Total log wage loss, after interruption=1yrs -0.0968 -0.147 -0.105 -0.121
(0.560) (0.636) (0.633) (0.613)
Total log wage loss, after interruption=3yrs -0.152 -0.253 -0.223 -0.216
(0.604) (0.639) (0.619) (0.625)
Age at first birth 27.27 28.39 25.94 27.56
(4.138) (3.783) (3.517) (3.943)
No child (%) at age 38 14.39 20.08 14.86 17.58
(3.067) (2.544) (4.164) (1.787)
One child (%) at age 38 25.00 28.92 18.92 26.15
(3.783) (2.879) (4.584) (2.063)
Two or more children (%) at age 38 60.61 51.00 66.22 56.26
(4.269) (3.174) (5.536) (2.328)
Notes: Occupation of work is defined conditional on working. Log wage growth is defined for all
consecutive work spells after apprenticeship training. Total log wage loss after interruption is the
change in log real daily earnings between return to work and last quarter before interruption. The
total wage loss has been purged of a change in occupation, in firm size (if change of firm) and changes
in hours of work. Standard deviations in parentheses
51

Table 2: Moments Used in the Estimations
Moments Data Nb
Set Moments
Labor supply and occupational choice
Proportion of full-time work, by age and initial occupation IAB 25
Proportion of part-time work, by age and initial occupation IAB 20
Proportion of out of labor force, by age and initial occupation IAB 20
Work experience, by age IAB 5
Annual transition rate between occupation IAB 9
Transition rates between labor market status, by occupation IAB 48
Proportion work, by number of children GSOEP 15
Proportion part-time work, no child GSOEP 5
Proportion in each occupation, initial and at all ages IAB 6
Initial choice of occupation, by region and time period IAB 440
Wages
Wage by age and initial occupation IAB 21
OLS regression of log wage on experience, by occupation IAB 9
OLS regression of log wage on age, number of children, occupation GSOEP 12
OLS regression of log wage on past and future wages IAB 3
OLS regression of log wage for interrupted spells on duration and experience IAB 14
OLS regression of wage growth around interrupted work spells by occupation IAB 10
OLS regression of husbands log earnings on women’s characteristics GSOEP 6
Variance of residual of log wage on occupation, age, work hours GSOEP 1
Proportion of women with log wage residual <1 std dev GSOEP 1
Savings
OLS regressions savings rate on age, occupation, number of children EVS 24
Fertility and marriage
Proportion with no children, by age GSOEP 5
Proportion with one child, by age GSOEP 5
Centiles of age at first birth GSOEP 10
Centiles of age at second birth GSOEP 10
Number of children at age 38 GSOEP 3
Average age at first birth, by current occupation GSOEP 3
Proportion of childbirth within marriage GSOEP 1
OLS regression of fertility on age and initial occupation GSOEP 5
IV regression of fertility on age and initial occupation (instrumented) GSOEP 5
Mean of residual of number of children on age, by wage residual GSOEP 2
Proportion married, by age GSOEP 5
OLS regression marriage on age, experience, past marital status GSOEP 15
occupation and fertility residual
763
Note: IAB: Institut fuer Arbeitsmarkt-und Berufsforschung. GSOEP: German Socio-Economic Panel.
EVS: Einkommens- und Verbrauchsstichprobe. Instruments for initial occupation in IV regressions are
the interactions between region of residence at age 16 with year of birth.
52

Table 3: Occupation specific parameters
Parameter Routine Abstract Manual
A. Atrophy rates parameters (annual depreciation rates)
At 3 years of uninterrupted work experience -0.06% (1e-5%) -0.11% (2e-5%) -0.03%(2e-5%)
At 6 years of uninterrupted work experience -0.50% (0.11%) -6.90% (0.17%) -3.45%(0.24%)
At 10 years of uninterrupted work experience -0.61% (14.2%) -2.65% (0.01%) -3.08%(0.18%)
B. Wage equation parameters
Log Wage Constant 3.39 (0.0038) 3.6 (0.0054) 3.32 (0.0059)
Years of uninterrupted work experience 0.1 (3.3e-05) 0.09 (3.6e-05) 0.123 (0.0001)
Years of uninterrupted work experience, squared -0.00382 (3e-06) -0.0021 (4.1e-06) -0.00463 (6.4e-06)
C. Amenity value of occupations
Utility of work if children 0 -0.056 (0.001) -0.014 (0.0005)
Utility of PT work if children 0 -0.42 (0.003) -0.08 (0.007)
Notes: The wage equation is defined as a function of skills - which corresponds to uninterrupted work experience
- and not work experience. The former is allowed to depreciate when out of the labor force. Asymptotic standard
errors in parenthesis.
53

Table 4: Estimated parameters: consumption decision and cost of children
Parameter Estimate
Annual discount factor 0.959 (0.00028)
CRRA utility 1.98 (0.0021)
Weight of children in consumption equivalence scale 0.392 (0.00167)
Cost of working if children, age≤6 (euros per day) 31.1 (0.36)
Cost of working, if children, age >6 (euros per day) 12.6 (0.24)
Notes: Asymptotic standard errors in parenthesis.
54

Table 5: Estimated parameters: unobserved ability and utility of children
A. Individual type (ability / fertility)
Parameter LA/HC LA/LC HA/HC HA/LC
Proportion in sample 0.125 0.174 0.309 0.393
(8.05e-05) (0.0621) (0.00775) (0.0621)
Log wage intercept 0 0 0.145 0.145
- - (0.0026) (0.0026)
utility of leisure 0 0 0.257 0.257
- - (0.0032) (0.0032)
Utility of one child 0.484 0.158 0.484 0.158
(0.0056) (0.014) (0.0056) (0.014)
Utility of two children 1.28 -2.04 1.28 -2.04
(0.00026) (1.3) (0.00026) (1.3)
Corr(Ability, desired fertility) 0.02
B. Outcome by type
Total fertility 1.88 0.953 1.88 0.951
Prop in routine occupation 0.3 0.231 0.301 0.232
Prop in abstract occupation 0.404 0.509 0.407 0.508
Prop in manual occupation 0.296 0.26 0.292 0.26
Notes: LA: low ability, HA: high ability, LC: low taste for children, HC: high taste for children.
Note that we allow ability groups to have different tastes for leisure. Asymptotic standard errors
in parenthesis. Proportions in given occupation are calculated at the start of the career.
55

Table 6: The career cost of children - percentage loss in net present value of income at age
15, with and without fertility.
Percentage loss compared to baseline
Total cost -35.3%
Oaxaca decomposition of total cost
Labor supply contribution -27%
Wage contribution -8.5%
Oaxaca decomposition of wage contributions
Contribution of atrophy -1.8%
Contribution of other factors -6.7%
Contribution of occupation -1.6%
Contribution of other factors -7%
Notes: The career costs are evaluated using simulations and comparing the estimated
model with a scenario where the woman knows ex-ante that she cannot have children.
The costs are computed as the net present value of female incomes, including all wages,
unemployment benefits and maternity benefits in the calculations. The discount factor is
set to 0.95 annually. Initial occupation is the one in the no-fertility scenario.
56

Table 7: The career cost of children: timing and spacing of birth
Age Age second birth
first birth Only 1 child 22 24 26 28 30
20 -31.4% -36.4% -36.6% -36.6% -37% -36.9%
22 -30.2% - -34.6% -34.8% -34.8% -35.2%
24 -28.1% - - -32.2% -32.3% -32.3%
26 -26.0% - - - -29.8% -29.8%
28 -24.0% - - - - -27%
Notes: The career costs are evaluated using simulations and comparing the a scenario with no
children, with a one where either one or two children are born at a given age. The costs are
computed as the net present value at age 15. The discount factor is set to 0.95 annually.
57

Table 8: Effect of Increased Child Benefits
Age at start of policy
15 25 35 45
Change, no child (in %) -0.8% -0.7% 0% 0%
Change, one child (in %) -0.08% -0.05% -0.05% 0%
Change, two children (in %) 0.2% 0.2% 0.07% 0%
Change, age at first birth (in years) -0.4 -0.1 -0.0005 0
Change, age at second birth (in years) -0.04 -0.007 0.002 0
Change, skills (in %) -0.29% -0.11% -0.049% -0.0019%
Change, number of years working -0.08 -0.03 -0.01 -0.0004
Change, number of years working PT 0.04 0.01 -0.007 -0.0003
Change, proportion routine 0.3% 0% 0% 0%
Change, proportion manual 0.07% 0% 0% 0%
Notes: The table compares two scenarios, a baseline one and one which introduces a cash
transfer at birth of 6,000 euros. Changes in fertility, skills and work experience are computed
at age 60. Changes in occupations are computed at age 15. Simulations performed over
12,000 individuals.
58

Figure 1: Savings rates and age: evidence from EVS dataset
.06 .08 .1 .12 .14
Savings Rate
20 30 40 50
Age of woman
Notes: computed from EVS data, by pooling the waves 1993-2008.
1

Figure 2: Savings rates around first and second births, model prediction
Time to birth (years)
-4 -2 0 2 4
Average saving rates
0.02
0.04
0.06
0.08
0.1
0.12
0.14
First child
Second child
Notes: Computed through simulations of the model, involving 12,000 draws.
2

Figure 3: Effect of no fertility
(a) Occupational choice at age 15
Routine Abstract Manual
% Deviation From Baseline
-4
-3
-2
-1
0
1
2
3
4
5
(b) Proportion working
Age
20 25 30 35 40 45 50 55
Proportion Working
Deviation from Baseline
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
(c) Proportion working part-time
Age
20 25 30 35 40 45 50 55
Proportion Working Part Time
Deviation from Baseline
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
(d) Wages
Age
20 25 30 35 40 45 50 55
Log Wage Difference
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
(e) Ratio of low to high ability, conditional
on working
Age
20 25 30 35 40 45 50 55
Ratio Low to High Ability
0.37
0.38
0.39
0.4
0.41
0.42
0.43
Fertility
No Fertility
(f) Effect on gender wage gap
Age
15 20 25 30 35 40
Log Wage
3.4
3.6
3.8
4
4.2
4.4
4.6
Observed
(Males)
Observed and Predicted
(Females)
Predicted Females
No Fertility
Notes: The different panels display the difference in outcomes between a baseline scenario and a
one where a woman knows that she is infertile.
3

Figure 4: Effect of child premium
(a) Difference in probability of birth by age, policy
vs baseline
Age
20 25 30 35 40 45 50 55
Difference in Probability of Birth
×10-3
-4
-3
-2
-1
0
1
2
3
4
(b) Number of children per year, compared to baseline
Year
2 4 6 8 10 12 14 16 18 20
%Increase in Number of
Children Born
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
(c) Assets by age and birth cohort, compared to base-
line
Age
20 25 30 35 40
Assets
in % deviation from no policy
-14
-12
-10
-8
-6
-4
-2
0
2
4
6
15
20
25
Age at Start
of Policy
Notes: Panel (a) shows the effect of the policy (cash transfer of 6,000 euros at birth) by age on the
probability of giving birth, comparing the policy to the baseline. In the policy scenario, women
learn at age 15 about the policy. Panel (b) depicts the aggregate effect of the policy, by year, in
an overlapping generation economy. The graph aggregates each year the behaviour of women of
age 15 to 60. Each year a new cohort of 15 year olds enters the economy and the cohort who is 60
exits. The policy starts in Year 4. Panel (c) displays the percentage change in assets as a function
of age, compared to a baseline without transfer. The birth cohort who is 15 at the start of the
policy can adjust right away their behavior. The cohorts who are 20 or 25 when the policy starts
do not anticipate the policy.
4

Figure A1: Sensitivity of the objective function with respect to model parameters
% Change in Objective Function
012345
Parameters
Wages
Labor market
transitions
Occupation choice
Utility of children
Utility of leisure
and consumption
Marriage/Divorce
Wage of Husband
Notes: Each horizontal bar depicts the percentage change in the objective function with
respect to a one percent change in a given parameter of the model. Parameters are color
coded and grouped by categories, for instance, the label “Wages” corresponds to parameters
determining female wages. The graph is truncated on the right at 3%.
5