The Effect of Childbirth on Women’s Formal Labour Market Trajectories: Evidence from Uruguayan Administrative Data

Abstract The motherhood penalty for developed countries is well-established in the economic literature. Childbirth intensifies the traditional gender roles that affect paid and unpaid work and contributes to the persistence of the gender labour gap. However, little is known about this phenomenon for developing contexts. This paper investigates the motherhood effects on women’s formal employment and wage trajectories in Uruguay. We document significant and robust motherhood penalties in the labour market, applying an event study method to almost 20 years of social security administrative data. One year after childbirth, formal monthly labour earnings decrease by 22 per cent. This drop fails to recover over time, and ten years after the arrival of children, women’s earnings are 40 per cent below their level just before childbirth. This penalty is mainly driven by a drop in formal employment and, to a lesser extent, a wage decline for those remaining employed. Heterogeneous analysis shows that low-wage women face higher motherhood penalties than high-wage women. Interestingly, these negative effects on wages and formal employment have reduced over time, and recent mothers face lower motherhood penalties.


Introduction
Reducing labour gender gaps in developed countries is one of the most striking facts in recent history. Yet, gender gaps in participation rates and wages persist (Blau & Kahn, 2017;Goldin, 2006). In Latin America, they narrowed sharply until the 2000s but were followed by a period of stagnation (Gasparini & Marchionni, 2015). Uruguay is no exception. Even when it has one of the highest female labour force participation rates in Latin America (Gasparini & Marchionni, 2015), the gender gap is still high, and employed women receive 25 per cent less wage than their male counterparts (CEPAL, 2020).
Pioneer's studies on gender gaps are mainly based on theoretical human capital frameworks. However, motivated by the reduction in experience gaps and the reversal of educational gaps, reliability because of their administrative nature and coverage of a more extended period. Finally, this paper provides helpful evidence for the design of public policies for developing countries, particularly those aiming to redistribute childcare costs and reduce gender gaps in the labour market.
The remainder of the paper is organised as follows. Section 2 briefly presents previous research on the topic. Section 3 describes the gendered labour patterns in Uruguay and the maternity leave program. Section 4 introduces the empirical strategy and the data. Section 5 presents the empirical results of motherhood's effects estimations, the heterogeneity analysis, and the robustness checks. Finally, Section 6 concludes.

Previous work
The relation between motherhood and women's labour trajectories has been widely studied. Motherhood affects women's participation in the labour market through different mechanisms. Childbirth affects the reservation wage in two opposite directions (R€ onsen & Sundstr€ om, 1996). It increases the demand for care (raising the reservation wage) and the demand for income (decreasing the reservation wage). Besides, motherhood could also affect the expected market wage and, consequently, the participation decision. Career interruption due to maternity can lead to a loss of experience and human capital depreciation (Budig & England, 2001;Kunze, 2016;Lundberg & Rose, 2000), while the anticipation of the wage penalty can affect investment decisions in human capital even before motherhood (Mincer & Polachek, 1974). Also, the compensating wage differential model explains the wage decline after childbirth. The traditional division of labour within the household poses that maternity raises the relevance of the nonmonetary component (e.g. family-friendly jobs to reconcile with childcare) and thus forgoes better-paid jobs (Becker, 1973). Literature is not conclusive, with some papers showing that women with children make occupational choices favouring family-work balance (Goldin, 2014) but with still limited evidence about the causal relationship.  find that motherhood increases the likelihood of working in the public sector, and Fern andez-Kranz, Lacuesta, and Rodr ıguez-Planas (2013) and Berniell et al. (2021a) that increases parttime jobs. On the other hand, Felfe (2012) and Cukrowska-Torzwska (2020) provide limited evidence about this mechanism. Finally, discrimination against mothers cannot be ruled out, and there is wide literature suggesting that wage gaps are not only determined by productivity and preferences differentials (Jacobs & Steinberg, 1990). Gough and Noonan (2013) find that employers' beliefs, such as the judgment that mothers are less competent or less committed to their jobs than childless women, explain the observer discrimination. Also, Oesch, Lipps, and McDonald (2017) find that the motherhood wage gap is explained by unobserved dimensions of work productivity and discrimination against mothers. Finally, it is worth noting that cultural beliefs and social norms also help explain the motherhood penalties.
Estimating the causal relationship between motherhood and labour market outcomes is challenging because unobservable characteristics affect both dimensions. Besides, motherhood may be the result of poor labour market outcomes. Literature has dealt with endogeneity following diverse identification strategies.
One of the first empirical strategies used to estimate the motherhood effects were fixed-effects models (Bertrand et al., 2010;Budig & England, 2001;Lundberg & Rose, 2000;Waldfogel, 1997Waldfogel, , 1998Wilde, Batchelder, & Ellwood, 2010). However, fixed effect models fail to eliminate bias when unobservable variables indirectly affect employment through other personal characteristics. Additionally, we cannot avoid bias if motherhood occurs in anticipation of a drop in employment. Another strand of studies uses instrumental variables to overcome endogeneity (Angrist & Evans, 1998;Cruces & Galiani, 2007;Lundborg, Plug, & Rasmussen, 2017;Tortarolo, 2014). However, using instrumental variables is not without criticism, highlighting the difficulty of finding a valid instrument. Also, the widely used 'same-sex siblings' instrument The effect of childbirth on women's formal labour market trajectories 211 provides evidence on the intensive margin of fertility (additional child) and not on the entrance to maternity (first child). More recently, the new empirical literature has relied on the event study methodology, the empirical strategy applied in this paper. This method relies only on treated individuals, sorting out potential selection issues. It exploits the variation in the age and period in which each individual experiences the event for causal identification. Interestingly, whatever the methods, the literature is conclusive about the negative parenthood effects on women's labour outcomes, and the negligible effects on men.
Based on the event study methodology, Kuziemko, Pan, Shen, and Washington (2018) find negative effects on women's employment in the US and UK, but of a different magnitude: from 25 to 40 per cent in the US and 40 per cent in the UK. The sharpest decline in employment occurs in the year immediately after childbirth, stabilises later, and fails to recover 10 years after the event. Kleven, Landais, Posch, et al. (2019) compare motherhood's penalties in participation rates for Denmark, Sweden, Germany, Austria, the UK, and the US. Denmark and Sweden face the lowest long-term penalties (7-13%), while the UK and the US have the largest (43-44%). They also find a long-term wage motherhood penalty of 20 per cent for Denmark and further show that the component of the total gender wage gap attributable to the first child has increased from 40 per cent in 1980 to 80 per cent in 2013 . Another study for developed countries (29 European countries) by Berniell et al. (2020) finds negative motherhood effects, evidencing that women's employment decreases on average by 25 per cent. The authors show that motherhood effects are higher in countries with more conservative social norms or weak family policies.
Finally, two previous paper applies the event study approach to Latin America. Unlike us, they use survey data and analyse employment and wages, adding formality and informality outcomes. For the Chilean case, Berniell et al. (2021a) estimate a 17 per cent reduction in female employment and a consequent 20-30 per cent drop in labour income. The other study provides evidence for Chile, Mexico, Peru, and Uruguay (Berniell et al. (2021b), concluding that motherhood reduces women's employment by 17-20 per cent. Like in developed countries, the authors show that Latin American countries with more conservative gender norms and weaker family policies experience higher motherhood penalties. The Uruguayan case study of this paper uses a longitudinal survey that provides annual information about labour market outcomes of 472 mothers five years after childbirth. After five years, mothers' employment falls by 17 per cent, and among employed, informality increases by 40 per cent.

The maternity leave program
The Social Security System administers a maternity leave program for women working in the private sector that dates back to 1958. The program aimed to avoid women's earning losses during the period surrounding birth. Since 1980 the program covered salaried female workers in the private sector and some self-employed, entitling them to 12 weeks of full paid maternal leave with 100 per cent wage replacement. This program is in effect for most of our study period. Then, in 2013 there was a complete reform of family leaves. According to Salvo (2014), the debate in the Parliament prioritized the country to be in line with ILO recommendations. Besides, the legislators presented their concern with the women's childcare burden, the need to increase their participation in the labour force, and balancing family and work responsibilities. Finally, there were explicit references to formal female employment because of concerns about women's pensions in an aging society with increasing divorce. The reform widened the benefit of paternity leave and created a three-month parental subsidy after maternity leave ends, allowing the mother and the father to alternate in the use. The duration of the maternity leave increased to 14 weeks, and the program's coverage widened to some groups of self-employed women that were not covered. The social security system is responsible for the subsidy payment, while beneficiaries do not receive labour earnings. There are no eligibility requirements except to contribute to the social security system when applying for the benefit. This rule facilitates widespread coverage for formal working women. However, it encourages registration in the social security system as an anticipatory behaviour (and an eventual exit after childbirth), which could benefit women and employers, who may fear more risk of being controlled under those circumstances.
Meanwhile, women working in the public sector have had their own maternity leave benefits whose payment and rules are in charge of the specific public employer institution.

Gendered patterns in the labour market
Gender gaps in the Uruguayan labour market have declined steadily since the 1980s. The female labour force participation rate has increased considerably, from 41.4 per cent in 1986 to 54.9 per cent in 2019, whereas male participation remained relatively constant. However, the participation gender gap continues at high levels, 15.2 pp in favour of men in 2019. Similar patterns describe the employment rate.
Job quality outcomes in the Uruguayan labour market have also improved during the last decades. Moreover, informal worker share decreased more sharply among women, so the gender gap in informality has vanished (OIT-MTSS, 2019). In 2019, 26 per cent of male and 23 per cent of female employment was informal. The incidence varied between jobs: 14 per cent for private salaried workers, 67 per cent for self-employed, and null for those in the public sector.
On average, women enter the labour market at the age of 18.2 2 , while the average female age at first birth is 24.7 (Nathan, 2019). Therefore, women experience motherhood when already in the labour market, potentially affecting future employment dynamics. Indeed, $40 per cent of women were not working when they had their first child, 40 per cent were employed and continued in such conditions, and 20 per cent were employed but stopped working for more than one year after childbirth. 3 Consistently, employment shows substantial heterogeneity related to the presence of children. Figure 1 shows the employment rate of household heads and spouses aged 14-49 by the number of children under 12. The childless women's employment rate is lower than men's, and this gap increases with the number of children due to the reduction of female employment.
Unsurprisingly, gender gaps in employment accompany gender gaps in home production that intensify with motherhood. Bucheli, Gonz alez, and Lara (2019) point out that the gender gap in time spent in household activities exists among single persons, increases within unions, and widens sharply with the first child's arrival. Batthy any, Genta, and Perrotta (2015) estimate that women perform 70 per cent of total unpaid work, and these gender inequalities are even higher among lower-income populations. Figure 1 also reports the informality rate showing that, on average, 20 per cent of male and female employment is informal. Informality is slightly higher among men than women when there are no children, but the gender gap reverses when there are children and increases with their number. Among women, the share of informality increases from 17 per cent when childless to 21 per cent with one child and reaches 40 per cent when three or more children are in the household. These figures suggest that women leave formal jobs when becoming mothers to exit the labour market or to remain in informality.

Methodology
We use the quasi-experimental event study approach to study the impact of motherhood on labour dynamics. 4 The event is the birth of the first child identified by the first work interruption due to maternity leave. By analysing the evolution of the outcome variable before and after the event, this approach allows estimating causal effects when all the individuals under analysis receive treatment but at random periods.
The identification of the causal effect relies on the individual variation in the age and period (month and year) in which each woman experiences the event. This variation allows us to identify temporal trends and dynamic effects separately by comparing results between mothers and not-yet-mothers. Thus, assuming that the timing of the child's birth is not determined by the labour outcomes conditional on being a mother, we deal with the endogeneity problems.
While the main assumption in identifying short-term effects is that counterfactual labour outcomes evolve smoothly over time, the identification of long-term effects relies on further assumptions. First, the smoothness assumption is no longer sufficient since we cannot fully control all non-child components of labour dynamics. Second, long-term estimates capture the effect of total fertility and not only the first-child effect for women with more than one child.
We estimate the following equation for individual i in period t (month/year): where Y it is the labour outcome variable. Each individual i is treated (i.e. experiences the event) at period E i , and remains treated from then on, with e it ¼ t À E i being the number of months between the period t and when the event takes place. We denote s the normalised time variable that takes the value s ¼ 0 the month of the event in which the woman enters the maternity leave program. This paper considers the time elapsed between 12 months before the event and up to 120 months after. The series of parameters b s are the event study coefficients that indicate the dynamics effects relative to 12 months just before the event (omitted variable: s ¼ À12); s < 0 refers to pre-trends and s > 0 indicates the dynamic effects of motherhood after the event. We also include age dummies to control for life-cycle trends (c j coefficients) and period dummies to control for temporal trends (a y coefficients). We do not include individual fixed effects since they synthesise the abovementioned effects. 5 We interpret the estimates as the motherhood effect relative to the counterfactual labour outcome of women without children. We call short-term the estimated effect 12 months after the event, medium-term relates to 60 months after the event, and long-term, 120 months.

Data
The analysis relies on labour administrative data of working women registered in the social security institute of Uruguay (i.e. formal workers). The dataset consists of people randomly selected from those who contributed to social security for at least one month between April 1996 and April 2015.
The data allows us to identify the women who accessed the maternity leave program, the month they did it, and the coverage duration. 6 It also reports information regarding death or retirement, and individuals have no registers after one (the first) of these events occurs.
We define the event as the month of the first entrance to the maternity leave benefit administered by the Social Security System, which has some disadvantages. As already mentioned, the program does not cover women in the public sector, so we can only identify the entrance to the maternity leave program of workers in the private sector. Although public institutions provide maternity benefits, we cannot detect their use. In addition, using maternity leave as a proxy for the first child is less reliable for older women who used the maternity leave program at the beginning of the observation period (e.g. women aged 39 whose first use of maternity leave was in April 1997). Finally, some women may have had their first child when unemployed or employed in informality (i.e. when not eligible for the maternity leave program), so their observed first entrance to maternity leave will not correspond to their first childbirth.
The sample consists of women in the private sector who used maternity leave for the first time between April 1997 and April 2015, aged between 18 and 40. We excluded women who retired or died during the analysed period. Ideally, we would not impose additional restrictions to ensure comparability with other studies. However, the sample also excludes women who did not contribute to social security during the year before childbirth to avoid including dubious cases, such as women registered in the system when they become pregnant. Thus, we keep women who contributed to social security the twelve months before the beginning of their maternity leave. The sample is composed of 89,998 cases, which represent around 40 per cent of total maternity leave beneficiaries in the private sector. We cannot follow up women whose events occurred after April 2005 for over 120 months, so we work with an unbalanced panel of monthly data.
Our outcome variables are monthly employment status, monthly labour earnings, and monthly wages conditional on employment. Because of the database, they refer only to formal jobs. Employment takes the value 1 when the woman is registered as employed in any job and 0 otherwise. The variable also takes the value 1 the months of maternity leave. The monthly labour earning is calculated as the sum of earnings of all occupations, assigning a value equal to zero for the months the woman does not register as working. Wages are equal to labour earnings but are conditional on the woman being employed. It is worth noting that earnings, employment, and wages refer to any formal job, thus including salaried public and private sector, and self-employment. Figure 2 shows the dynamic effects on women's trajectories and their 95 per cent confidence intervals from 12 months before and up to 120 months after childbirth. The estimates should be interpreted as the percentage change from the year before the event relative to women without children in the same period. Table A1 in the Supplementary Appendix reports the estimated motherhood effects for selected months.

Motherhood's effect on formal labour market trajectories
Panel (a) shows the monthly formal labour earnings dynamics. The discontinuity in the curve after birth (three months) indicates the maternity leave period covered by the benefit. Women have missing data these months as they receive public allowances, not wages. A sharp The effect of childbirth on women's formal labour market trajectories 215 decline in earnings follows at the end of the period, and the trajectory illustrates a gradual and sustained reduction from then on. In the short-term-12 months-motherhood leads to a decrease of 22 per cent in monthly earnings. The negative motherhood effect fails to recover, and in the medium-term-60 months-the fall reaches 36 per cent. In the long run, 10 years after the arrival of children, earnings are 40 per cent below their level just before childbirth. It is important to remember that the long-term estimates may be affected by subsequent births that would likely further prolong the motherhood effects. Thus, as time passes, some women would suffer the penalty of more than one child. 7 Figure 2(b) shows the estimated effects on formal employment. The flatness at the beginning of the curve corresponds to the sample restrictions -all women contributed to social security during the year before the maternity leave period began-and the fact that we treat women as employed during the maternity leave. Note that the results are reported as a percentage change compared to the year before the event (mothers related to non-mothers), avoiding the potential bias caused by mothers who only contributed to social security close to birth to claim the leave benefits. The graph shows a substantial reduction in formal employment after maternity in the first year, and this penalty not only fails to reverse but continues to increase. We estimate that motherhood reduces formal employment by 23 per cent after 12 months, 37 per cent after 60 months, and 44 per cent after 120 months. Source: Own calculations based on social security administration data. Notes: Sample: women who used maternity leave for the first time between April 1997 and April 2015, aged 18-40, with a formal job during the 12 months before maternity leave and did not retire or die. The figure shows the effect relative to the year before childbirth as a percentage of the counterfactual outcome of women without children. Fixed effects by age and period (month and year) were included as control variables. Ninety-five per cent confidence intervals are based on robust standard errors.
Finally, in panel (c), we present the wage trajectory conditional on employment. Once again, the discontinuity corresponds to the period under maternity leave. In the first months, wages rise probably because of selection issues of women exiting the labour market. The rest of the curve is much flatter than the other two, and the wage loss is 8 per cent in the long run. Taken together, our findings indicate that the trajectory of monthly labour earnings is driven mainly by employment shifts, raising the question of whether the maternity leave program effectively achieves the goal of protecting employment.
The long-term motherhood penalty is high compared to those found in developed countries by Kleven, Landais, Posch, et al. (2019) for Denmark, Sweden, the UK, and the US (7-44% depending on the country), and Kuziemko et al. (2018) for the US and the UK (40%) and compared to previous evidence for Latin America (Berniell et al., 2021a). As we expect motherhood to shift from formality to informality in developing countries, we are probably overestimating the effects on total employment and wages. Indeed, the hypothesis of the entrance to informality after childbirth is supported by Berniell et al. (2021a) in developing contexts.
We estimate heterogeneous effects by industry to explore the relevance of migration to the informal sector. In highly informal industries, we assume it is easier to move to informality after birth. Estimations by the pre-birth industry show that the impact is lower for mothers in industries with low informality rates before childbirth. Employment falls by 26 and 22 per cent in education and health services after 120 months, well below the average of 44 per cent; in those sectors, informal workers' share is 5 and 8 per cent, respectively ( Table A2 in the Supplementary Appendix). On the other extreme, employment fall is 52 per cent in manufacturing and 48 per cent in commerce, industries with informality rates of around 30 per cent. These results suggest that our estimated effects overstate the motherhood penalty on earnings and employment due to a movement from formality to informality.
It is worth noting that we may also interpret the shift from formality to informality as a motherhood penalty. Despite having some advantages related to flexible working arrangements, informal workers are outside the law labour framework and regulations, including access to social security coverage of risks and minimum wages.

Heterogenous effects by wage level
This section analyses possible heterogeneous effects according to the wage level. We split women into two groups according to whether their average hourly wage the year before childbirth is above or below the median. The same model is estimated separately for each group.
In the short run, low-wage women face higher motherhood effects on total earnings (Figure 3(a)) and formal employment (panel b) than high-wage women. Their penalty almost doubles the harm suffered by high-wage women. This result is consistent with those presented in Nylin et al. (2021) for Sweden, who find that the recovery in women's share of couples' labour earnings following childbirth is more pronounced among more educated couples.
However, the gap trajectories of earnings and employment are different. The earnings gap declines and vanishes in the medium and long term, but the heterogeneous impact on formal employment remains. Ten years after childbirth, employment drop is 30 per cent for high-wage women and 50 per cent for low-wage women. These heterogeneous effects are consistent with high-wage women facing higher opportunity costs of leaving the labour market. Monthly wages conditional on employment shows different patterns, providing evidence in support of the hypothesis of stronger self-selection into formal employment among low-wage women (Figure 3(c)).
The effect of childbirth on women's formal labour market trajectories 217

Time trends in motherhood penalties
To capture potential heterogeneous time trends of motherhood penalties, we extend the baseline specification to allow year-specific effects following . Figure 4 shows the average impact on the short and medium-long term by year of motherhood. The short-medium term corresponds to the estimated average impact for 1-5 years after childbirth; the medium-long term is the average effects of years 6-10.
The main findings are 2-fold. First, there is a downward trend in motherhood penalties. The patterns across childbirth cohorts indicate that more recent mothers face lower penalties than the oldest ones. We observe this trend both in the short-medium and medium-long term, unlike , who only find it in the long run. Second, the reduction over time seems higher in the medium-long than in the short-medium term. The medium-long term effect is a reduction of around 50 per cent of the formal employment for the oldest childbirths and around 30 per cent for women entering motherhood in 2009. The short-medium term penalties are around 35 per cent for the former and 25 per cent for the latter. The results further suggest that short-and long-term earning penalties converged for women entering motherhood from Source: Own calculations based on social security administration data. Notes: Sample: women who used maternity leave for the first time between April 1997 and April 2015, aged 18-40, with a formal job during the 12 months before maternity leave and did not retire or die. The figure shows the effect relative to the year before childbirth as a percentage of the counterfactual outcome of women without children. Fixed effects by age and period (month and year) were included as control variables. Ninety-five per cent confidence intervals are based on robust standard errors. 'Lower wages' include women whose average hourly wages during the year before childbirth are below the median and 'Higher wages' those above.
2002 onwards. These trends are not visible for monthly wages conditional on employment, probably because of the lower penalty. Previous studies on this topic are not conclusive regarding the evolution of motherhood's effects over time. While there is evidence of unchanged trends in the US (Avellar & Smock, 2003;Jee, Misra, & Murray-Close, 2019), in Norway, penalties have narrowed possible because of more generous family policies (Petersen, Penner, & Høgsnes, 2014). This study's results align with those finding that the child penalty has changed over time, and recent mothers experience lower adverse consequences in their labour trajectories. We may interpret the trends for Uruguay as related to institutional changes. Compared to Latin American countries, Uruguay has a long tradition of labour institutions whose enforcement has strengthened over time. Besides, there is an increasing trend of free preschool coverage and the average class hours, and recently, a National Care System has been installed. Although this paper does not address the impact of these policies on labour gender gaps, it could be a reasonable explanation behind the results.

Robustness checks
This section summarises the results of some alternative estimations to provide robustness checks on the results. We present the estimated effects in the Supplementary Appendix. Source: Own calculations based on social security administration data. Notes: Sample: women who used maternity leave for the first time between April 1997 and April 2015, aged 18-40, with a formal job during the 12 months before maternity leave and did not retire or die. The figure shows the effect relative to the year before childbirth as a percentage of the counterfactual outcome of women without children. Fixed effects by age and period (month and year) were included as control variables. Ninety-five per cent confidence intervals are based on robust standard errors. 'Short-medium' corresponds to the average effect for years 1-5 from motherhood and 'Medium-long' to years 6-10.
The effect of childbirth on women's formal labour market trajectories 219 First, we perform a robustness analysis using a placebo for the entrance date of maternity leave. The time of the event is randomly modified, so we expect to find no effects after this false event. The almost horizontal curves shown in Figure A1 of the Supplementary Appendix indicate no significant impacts of placebo maternity on labour outcomes. Second, following , we estimate an event study model using a sample of women without children. We randomly assign a childbirth event to women who did not use the maternity benefit. The assignment of the event is based on the age at first childbirth distribution by cohort in the estimation sample. The dynamic effects are estimated separately for the sample of mothers and the placebo group. The model specification has two changes. First, we work on a yearly (and not monthly) basis because the assignment of placebo events to childless women is yearly based. Second, we exclude the age-fixed effects because we use the age for assigning the placebo births. The results provide robustness to our main specification, as depicted in Figure A2 of the Supplementary Appendix. Indeed, women's labour trajectories with and without children diverge sharply after the first child. In addition, the long-term motherhood penalties computed as the percentage change from the year before the event, relative to the change for childless women in the same period, are similar to the main specification's impact.
We further estimate the model using a less restrictive sample of women than the main one. We restrict the cases to women who were formal workers the two (and not 12) months before the maternity leave period began. Figure A3 of the Supplementary Appendix shows similar patterns to the most restrictive sample, though the estimates are of a smaller magnitude.
Finally, we estimate the effects through alternative specifications, including observable human capital variables, to control potential selection issues. If there is self-selection in moving to a formal job after childbirth or not, the main results may be biased. To address this issue, we re-estimate the main specification by adding different control variables indicating the women's wage profile during the year before childbirth: (i) a binary variable that indicates whether the women's previous hourly wages were above the median or not, (ii) the tercile of previous average hourly wages, and (iii) the average hourly wages during the year before childbirth. Table  A3 of the Supplementary Appendix shows that the main results hold.

Conclusions
The stylised fact that working mothers earn less than childless women is now well-established by the economic literature in developed countries. However, little is known of this phenomenon for developing contexts, and this paper contributes to the literature by providing evidence for Uruguay. Using an event studies approach and based on novel administrative data, we investigate motherhood's effect on women's formal employment and wage trajectories.
One year after childbirth, formal monthly labour earnings decrease by 22 per cent. These drops fail to recover over time, and 10 years after the arrival of children, women's earnings are 40 per cent below their level just before childbirth. This penalty is mainly driven by a drop in employment and, to a lesser extent, a wage decline for those remaining employed. Indeed, ten years after the first childbirth, only 44 per cent of mothers are still employed, with a wage 8 per cent lower. These results are robust to various specifications and tests. Heterogeneous analysis shows that motherhood's negative effect on employment is nearly twice for low-wage women than for high-wage women. Finally, the negative effects on earnings and employment decline over time, and recent mothers face lower motherhood penalties.
There are some drawbacks to mention. First, using the first entrance to maternity leave as a proxy for the first child imposes some limitations. We study the trajectory of women who were working in the private sector when becoming pregnant. Second, we do not know how much of the reduction in formal employment is due to exiting the labour market or becoming informal. Indeed, we provide suggestive evidence that both (negative) phenomena coexist. Thus, the estimated effects overstate the actual drop in earnings and employment. However, moving to informality should be considered a penalty for women's work trajectories. Informality leaves workers out of the labour market regulations and is a barrier to accessing social security programs (e.g. unemployment insurance and pensions). We argue that motherhood has significant adverse effects on labour outcomes -whether because of exiting the labour market or moving to informality-that may have long-lasting consequences.
The results of this study have important policy implications. Though the explicit goals of the maternity leave program are vague, we may interpret that the original aim was to cover labour earnings losses. However, more recently, legislators and public debate showed concern about women's childcare burden and balancing family and work responsibilities. Besides gender equity, these issues are linked with the possibilities of a formal job. In an aging society with increasing divorce, the need for women for formal jobs is becoming a crucial issue in the social protection of the future female elderly. In this context, in 2013 a new law widened the motherhood and fatherhood leave benefits. Besides, it created a parental leave policy that entitles parents to a three-month half-time subsidy after maternity leave finishes, allowing the mother and the father to alternate. However, in 2017, only 67 per cent of fathers used fatherhood leave, and 2 per cent of parental leave beneficiaries were men (Batthy any, Genta, & Perrotta, 2018).
Our results indicate that the motherhood leave program is not ensuring the possibility that women maintain their formal jobs after being a mother. Besides, women facing the lowest wages are more prone to exit the formal labour market, increasing their vulnerability of reaching a contributive old pension in the future. The overall picture responds to an uneven distribution of the economic costs of care. However, the decline of the motherhood penalty for cohorts of recent mothers in a period with more generous family policies seems auspicious and makes a case for further analysis. 4. For a more detailed explanation of the methodology see Sun and Abraham (2021), Borusyak and Jaravel (2016), . 5. The estimation procedure is based on codes by : https://www.openicpsr.org/ openicpsr/project/116366/version/V1/view. 6. Unfortunately, we cannot identify the case of multiple childbirths. 7. The analysis of the effect of additional events faces serious potential bias issues. Indeed, the sample of women who use the program a second time is composed of mothers with a higher probability of remaining in a formal job after the first childbirth. However, we estimated the effect of the first child separately for women with one use of the maternity leave program, two, and three or more uses, and find a significant decline in formal employment close to our estimates for the three samples (results available under request).