Does a Longer Work Horizon Affect Offsprings' Labour Market Outcomes?

This paper studies the eﬀect of an increase in the work horizon of middle aged workers on the school-work transition of their oﬀsprings aged 15-29 years. I exploit the variation in the parental work horizon induced by the 2012 Fornero reform in Italy that abruptly changed the age and years of social security contribution requirements for pension eligibility. Utilising a diﬀerence-in-diﬀerence strategy in a multi-valued treatment setting, the study shows that the reform-induced increase in the work horizon of mothers caused an increase in the probability of their oﬀsprings seeking their ﬁrst job and a decrease in these oﬀsprings’ probability of being a student. This eﬀect is higher on male oﬀsprings and is stronger in southern Italy. Fathers did not signiﬁcantly aﬀect the student status or any labour market outcomes of their oﬀsprings. Mechanisms behind these ﬁndings include the longer reform-induced work horizon for mothers vs. fathers, and the consequent positive eﬀect on mothers’ lifetime earnings.

contribution system. 1 While most of the existing research has centered around the effects of pension reforms on older workers who would have been retired in the absence of the reforms (direct effect), some papers have studied its effects on younger workers who have been affected by the reforms but are not at the verge of retirement (perspective effect). Recently, Carta and De Philippis (2021) have shown that the pension reform of 2012 in Italy (called the Fornero reform) which increased the Minimum Retirement Age (MRA) to be eligible for pension benefits, has led to positive and significant increases in the labour participation and employment rates for women aged 45-59. For men aged 45-64, these outcomes did not see a significant increase. However, since monthly pension income after retirement is usually lesser than the monthly income earned during the last few years under employment, this increase in work horizon due to the reform can be viewed as an increase in the disposable income and job security of the affected individuals.
In this paper, I investigate the effect of changes in the pension eligibility rules on the education and labour market outcomes of the offsprings of the affected parents. An increase in income of parents due to delayed retirement could impact offsprings' education and employment decisions through many channels. Since schooling is a normal good, an increase in lifetime earnings of parents could lead them to invest in higher education of their offsprings so that the offsprings can reap the returns to higher education achievement. This would also be consistent with the Italian culture characterised by strong cultural ties where parents prefer coresiding with their children. Previous evidence suggests that if coresiding is viewed as a "good" for parents and a "bad" for young adult children, then Italian parents are willing to "bribe" their children by trading off a part of their consumption to have them stay longer at home (Manacorda and Moretti, 2006). Italian parents also experience negative effects on health and life satisfaction as a result of children leaving the parental home (Mazzuco, 2006). Thus, an increase in parental lifetime earnings could lead them to incentivising their offsprings to stay at school longer in order to delay their nest leaving. On the other hand, if the returns on higher education is low, offsprings might want their parents to sponsor their first job search which is costly and time consuming. In Italy, as of 2011, the transition time from school to work for individuals with a high school education was 2.57 years whereas for those with tertiary education it was 1.77 years (Pastore et al., 2020). Since returns to higher education is low in Italy as opposed to other comparable EU countries (Depalo, 2017), the increase in lifetime earnings due to the longer work horizon could increase the probability of parents bearing the additional costs of sponsoring their offsprings' first job search at a lower level of educational attainment and help them transition from school to work early. Thus, the effect of a longer work horizon of parents on education and labour market outcomes of offsprings is ambiguous.
To examine the effect of an increase in parental work horizon on offsprings' education and labour market decisions, I exploit the Fornero reform of 2012 which increased the MRA of Italian workers by an average of four years. Since the reform was implemented by the Italian government independent of offsprings' education and labor market decisions, it provides a quasi-experimental variation in parental labour supply that can be exploited to estimate causal effects of an increase in parental work horizon on offsprings' outcomes. The pension eligibility requirements are defined on the basis of several factors like age, years of contribution to the social security system, gender and sector of employment (private, public or self-employed). The Bank of Italy's Survey of Household Income and Wealth (SHIW) provides detailed individual level data on each of these variables which allows for the precise calculation of an individual's increase in MRA. Using data from the 2006-2016 cohorts, I study how the increase in the work horizon of the parents due to the reform has affected education and employment outcomes of their offsprings aged 15-29. The age band of 15-29 is motivated by the definition of NEET which is usually defined for this age group and in Italy, one also becomes legally eligible to work at the age of 15. I use a multi-valued treatment difference-in-differences technique similar to Bertoni et al. (2018) and Carta and De Philippis (2021) that exploits variation in the increase in MRA within cells defined on the basis of age, gender, years of contribution and sector of employment. I show that a 1-year increase in maternal work horizon increases the probability of coresiding offsprings aged 15-29 to be a first job seeker by 1.96 percentage points (19 percent). Since the average maternal work horizon increased by 5.2 years, this translates to a 10.19 percentage point (98.8%) total increase in the average offsprings' first job seeking probability. At the same time, that there has been a 1.97 percentage point (3.53 percent) decrease in the likelihood for offsprings of being a student due to a 1-year increase in their mothers' MRA. This translates to a 10.24 percentage point (18.36%) total decrease in the probability of the average offspring to be a student. This suggests that the offsprings of the affected mothers have been leaving schooling to engage in their first job search. The effect of affected fathers on their offsprings' education and labour market outcomes has been largely insignificant. I perform simulations of lifetime earnings increases of affected working mothers and fathers due to the reform and show that the mean increase in lifetime earnings of the average mother is e82,054 which is 2.5 times that of the average father's increase of e32,491. I also show that mothers anticipate a significant increase in their working horizons on average due to the reform, while fathers do not. Thus, the likely mechanism through which the main results can be explained is that due to the low returns to education in Italy, offsprings start looking for their first jobs instead of continuing with schooling. This job search being costly and risky is helped by the increase in lifetime earnings of mothers who successfully anticipate this increase in future earnings.

Related literature
This paper is related to the literature on the 'perspective effect' of pension reforms on middle-aged workers who are not at the brink of retirement. This literature is relatively new as compared to the literature on the 'direct effect' of pension reforms on old workers who would have retired in the absence of the pension reforms. Hairault et al. (2010) show that distance to retirement matters for workers' employment with employment rates for 55-59 year old increasing due to a French social security reform of 1993 that increased minimum retirement age from 60 to 65. Montizaan et al. (2010) exploit a Dutch pension reform of 2006 which abolished generous early retirement pension benefits and created incentives to postpone retirement for many workers to show that it increased their participation in training programs. In a similar vein, Brunello and Comi (2015) use Italian data to show that increases in MRA due to the pension reforms of late 90's and early 2000's increased training participation of Italian workers. (Grip et al., 2012) show that the aforementioned Dutch reform of 2006 led to a significant decrease in the mental health of workers affected by the reform. Bertoni et al. (2018) show that an increase in MRA due to a 2004 Italian social security reform increased health-promoting behaviours in the form of an increased likelihood of regular exercise, a decrease in the probability of being obese and an increase in the probability of reporting a high satisfaction with own health among a cohort of 41-54 year old Italian workers. Carta and De Philippis (2021) show that the Fornero reform of 2012 significantly increased labor market participation, employment and unemployment rates of 45-59 year old Italian women. Although these effects were mostly insignificant for 45-64 year old Italian men, they find that an increase in the wife's working horizon has led to a positive and significant effect in the husband's labour market participation within the family. This paper also relates to the literature on the late entry of Italians into adulthood. Pastore et al. (2020) show that the average duration of school-to-work transition for Italians aged 18-34 was 2.88 years (34.56 months) in 2017 with the duration being 46 months shorter for the highly educated compared to those with compulsory education. Bertoni and Brunello (2021) estimate that, for any 1,000 local senior workers locked into employment by the Fornero reform, local youth and prime age employment declined by 273 (-0.86%) and 199 (-0.12%) workers, and senior employment increased by 833 (+2.70%) individuals. Manacorda and Moretti (2006) argue that Italian parents are willing to transfer part of their consumption to their offsprings in order to incentivise them to stay at home. They use the 1992 Italian pension reforms which increased the retirement age of workers to show that a 10% increase in annual father's income increases the proportion of 18-30 year old men living with their parents by 10%. Mazzuco (2006) finds negative correlations between children leaving home and satisfaction indices of parents' satisfaction with their main activity, financial situation, household situation and self-reported health. Becker et al. (2010) find evidence based on 12 European Union (EU) countries including Italy that an increase in perceived job insecurity by the youth and a decrease in parental perceived job insecurity are correlated with higher youth cohabitation with their parents. Billari and Tabellini (2010) provide an overview of the literature that investigate the role of culture and economic factors on why Italians enter adulthood late.
There is a branch of literature on the intergenerational effects of increased MRA. Battistin et al. (2014) find negative effects of Italian pension reforms that increased the MRA of the grandparental generation on informal childcare supply for the next generation. They find that one additional granparent in early child-bearing years increases the number of children by 5% in close-knit families. Bratti et al. (2018) also exploit Italian pension reforms to find that grandparental availability for child care has a positive effect on the labour force participation of women with young children. They show that mothers of cohabiting children under 15 whose own mothers are eligible for retirement have a 13% higher probability of employment as opposed to those with mothers ineligible for retirement. To my knowledge, there is not any existing research on the 'perspective effect' of pension reforms on the next generation. This paper contributes to the literature by investigating and presenting novel estimates of the intergenerational perspective effects on school to work transition of the Fornero reform. With Italy having the longest school to work transition time in the EU which has been worsening over the years, the paper brings to light new evidence on how an increased in parental work horizon could lead to early school-to-work transitions of offsprings.
The paper is organized as follows: Section 3 provides an overview of the Fornero reform and how it affected minimum retirement ages. Section 4 describes the data, Section 5 presents the empirical methodology, Section 6 presents the results and Section 7 concludes.

The Italian pension system and the Fornero reform
The Italian public pension system offers two major schemes of retirement and claiming full pension benefits -the old age scheme and the seniority scheme. Under the old age scheme, people can retire after having achieved a certain age whereas under the seniority scheme, people can retire after having achieved a certain number of years of contribution into the pension system. Before 1992, under the old age scheme, Italians could achieve pension eligibility at the age of 60 for private sector employees and the self-employed and at 65 for public sector employees with at least 15 years of contributions. Under the seniority scheme, they could retire if they had at least 35 years of contributions in the private sector and 25 years in the public sector (Angelini et al., 2009;Bertoni and Brunello, 2021). The Italian pension system underwent a series of reforms in the 1990s which aimed at better financial systainability of the system by delaying the minimum retirement age and by making pension benefits less generous. In 1995, the pension benefit calculation system changed from a defined benefit to a definded contribution system.
As of 2011, the pension eligibility requirement under the old age system was an age of 60 for women and 65 for men with at least 20 years (5 years) of contributions for individuals who started working before (after) January 1, 1996. Under the seniority scheme, individuals needed to have accrued at least 40 years of contributions to be eligible for full pension benefits. A third scheme, called the 'quota system' also existed which characterised pension eligibility in terms of a combination of age and years of contribution -for example, as of 2011, a regular employee would be eligible for retirement under the quota system if the sum of her age and years of contribution would be 96 if she was a public or private sector employee and 97 if she was self-employed (she had to be at least 60 years old if an employee, 61 if self-employed and have at least 35 years of contribution). The Fornero reform was introduced in December 2011 and became effective on 1st January, 2012. The reform increased the old age retirement eligibility for all workers (both male and female) to 67 years by 2020 with at least 20 years of contributions. It also raised the minimum years of contributions required to retire under the seniority scheme from 40 to 42 for men and to 41 for women in 2012. This was further raised to 43 and 42 years of contributions for men and women in 2013 and to 44 and 43 years of contributions for men and women in 2014. The reform abolished the quota system of retirement (Carta and De Philippis, 2021). Thus, the Fornero reform induced a sudden unexpected increase in MRA that affected different individuals differently depending on age, gender, sector and accrued years of contribution. This allows for the distinction between treatment and control groups that differ on the basis of treatment intensities. Table 1 outlines the pension eligibility rules under the seniority and quota systems for private sector, public sector and self-employed individuals for the years before and after the reform.

Data
The data for the analysis comes from the Survey of Household Income and Wealth (SHIW) which is a biennial survey conducted by the Bank of Italy which comprises about 8,000 households (20,000 individuals), distributed over 300 Italian municipalities. I use data for the most recent years, from 2004 to 2016 for women aged 45-59 with at least one coresiding offspring of the age range 15-29 and men 45-64 with at least one coresiding child of the same age range. Thus, I use the terms women and mothers or men and fathers interchangeably throughout the paper. The age band selection has been motivated by the fact that these individuals are middle aged who are not at the margin of retirement and thus allows for the estimation of a 'perspective effect', that is the effect of foreseeing a longer working horizon due to an increased MRA. The use of the SHIW dataset is advantageous because it contains detailed information on age, gender, sector of employment and the years of contributions of individuals and thus facilitates calculations of the MRA in presence and in absence of the reform under certain assumptions. It also contains useful information about expected retirement age of the individuals and whether they have children residing outside the household which are necessary to test the iden-tifying assumptions. It also provides information on individuals' expected replacement ratios after retirement which I utlise for the lifetime income change calculations due to the reform.
The change in MRA due to the reform is calculated as T q = M RA q,2014 − M RA q,2010 where q is a cell defined on the basis of 4 factors -age, gender, years of contribution and sector of employment (private, public or self-employed). M RA q,2014 is the minimum retirement age of an individual belonging to cell q in 2014 in the presence of the Fornero reform and M RA q,2010 is the minimum retirement age of the same individual in the absence of the reform according to the 2010 rules that existed just before the introduction of the reform. Thus the difference between the two represents the increase in the work horizon of the individual because of the reform which is time invariant in nature. It is time invariant as the years used for the calculation of the workers' MRA increase have been fixed to the two years of 2014 versus 2010. Note that the use of the year 2014 is only representative of a post-reform year and using any other post-reform year does not change this T q measure as the same Fornero reform rules of retirement were in place in all post-reform years under consideration. The year 2010 is the most appropriate as a pre-reform year for the calculation of T q as it represents the retirement rules that were in place just before the Fornero reform was implemented. However, for the calculation of the MRAs, an important assumption that needs to be made is that the individuals work continuously without having periods of not being employed and therefore make continuous future contributions until the time they become eligibile for pension benefits. For this reason, following Carta and De Philippis (2021), the sample has also been restricted to those who are strongly attached to the labour market. Specifically, for the women, the sample has been restricted to those with at least 10 and less than 40 years of contributions. For the men, the sample has been restricted to those with at least 20 and less than 40 years of contribution. Carta and De Philippis (2021) argue that Italian Social Security Institute records show that the discontinuous spells in individuals' careers are concentrated before the age of 35 (because of maternity leave periods or longer study paths) and after the age of 60, and the possible error generated by the assumption of continuous future contributions is minimized under this mode of sample restriction. Thus, I refer to the sample of 45-59 year old mothers with at least 10 and less than 40 years of contribution as the sample of 'eligible mothers' and the sample of 45-64 year old fathers with at least 20 and less than 40 years of contributions as the sample of 'eligible fathers' . Fig 1 shows histograms of the distribution of MRA increase for mothers and fathers. MRA increased anywhere between 2-7 years but the vast majority of mothers experienced a 7-year increase in MRA while the majority of fathers a 3-year increase. Table 2 shows some descriptive statistics of the eligible mothers and fathers in the sample for the period of 2004-2016 in terms of their demographic characteristics throughout the sample period. Mothers are about 51 years of age with 25 years of contributions and experienced an average increase of 5.2 years in their work horizon due to the Fornero reform. About 85% are married and 55% have a high school diploma. 89% are active in the labour market with 86% being employed and 4% unemployed. Fathers are 53 years old on average with 30 years of contributions and experienced an average increase of 3.7 years in their MRA due to the reform. 97% are married, 47% have a high school diploma, almost all are active in the labour market with 95% employed and 5% unemployed. Note that for those unemployed, information from the last job has been used for the sector of employment used in the analyses that follow. Histogram plots in Figure 1 shows the increase in MRA for fathers and mothers in this sample. The majority of fathers (about 47%) experienced a 3 year increase in work horizon because of the raising of required years of contributions from 40 to 43 under the seniority scheme whereas the mothers (about 54%) experienced a 7 year increase owing to the raising of the old age retirement eligibility from 60 to 67 years for women. Tables 3 and 4 show the descriptive statistics of the same variables by T q or increase in MRA for the comparison of the groups affected in different degrees due to the Fornero reform.
For the offsprings sample, I consider the offsprings of the above mentioned parents in the age group of 15-29 as this is a suitable age range for them to decide to be students or to be in the labour market. Table 5 provides descriptive statistics on these offsprings. They are about 21.6 years old, 54% are males, 39% are from Southern Italy, 55% are students, 15% are first jobseekers, 26% are employed and 3.9% are not employed. Tables 6 and 7 also provide a comparison of these offsprings by the number of years of increase in MRA of their mothers and fathers respectively.
There is a panel component in the SHIW database. However, since it only consists of about 44% of the households in the sample of interest and only 14% of the households in this sample of interest are observed both before and after the reform, I do not use the panel component for my analyses.

Empirical methodology
The treatment of the increase in MRA across workers is not of a uniform dosage but of different doses that vary from 2 years to 7 years. Thus, the treatment is not binary, rather a multi-valued one. I use a multi-valued treatment difference-in-difference strategy utilising a two-way fixed effects (TWFE) model to estimate the effect of the exposure of the parents to the reform on their offsprings. The TWFE model in the case of a multivalued treatment requires a stronger version of the parallel trends assumption for the estimation of the average treatment effect on the treated (ATT) or the average causal response on the treated (ACRT) to rule out the selection bias effect that results from the comparison of two different dosage groups. In this section, I explain the assumptions required for the identification of the standard DID in a binary treatment setting and how things change in the context of a DID with a multi-valued treatment. For both cases, we need the assumptions of an identically and independently distributed random sample, no units to be treated in the pre-treatment period and no anticipation. With these three assumptions and the 'standard' parallel trends assumption (i.e., the trend of the treated group in the post-treatment period in the absence of the treatment would have been parallel to the trend of the untreated group in the post-treatment period), the ATT in the binary treatment case can be represented as: where Y t (1) is the observed outcome of the treated group in the post-treatment period and the Y t (0) is the unobserved missing counterfactual of what the outcome of the treated group would have been in the post-treatment period in the absence of the treatment. Under the above four assumptions, the ATT in the binary treatment case is estimated by differencing the difference in the average of the post versus pre-treatment outcome of the control group from the difference in the average of the post versus pre-treatment outcome of the treated group.
In the multi-valued treatment case where different treated units receive different dosages of the treatment, we need to introduce the concept of ATT by dose. Here, the ATT of receiving a dose a for a treated group that actually received dose b can be represented as: We often consider the special version of this parameter, i.e., AT T (d|d) which is the average effect of dose d amongst units that actually experienced dose d.
We can also define the average treatment effect of dose d, AT ] which is the mean difference between potential outcomes under dose d relative to untreated potential outcomes across all units, not just that experienced only by dosage group d (Callaway et al., 2021). One thing to note here is that the estimation of the ATT or the ATE requires the existence of untreated units to be used as a control group.
While the treatment effects in levels just described inform us about the average effect of being treated with a particular dose, another important metric of interest is the effect of an increment in the treatment dose. This is called the causal response of the treatment. The causal response of the d th dose unit of the treatment is defined as The average causal response on the treated (ACRT) is given by: The average causal response of the d th dose unit of the treatment is given by: For discrete multi-valued treatments, ACRT (d|d) equals the difference in potential outcomes between dose level d j and the next lowest dose d j−1 whereas the ACR(d) is the overall average causal response of a unit change in dose across the entire population, not just the units that experienced dose d (Callaway et al., 2021). Note that the identification of the ACRT and ACR can be achieved even in the absense of untreated units. Only varying amounts of doses of the treatment across different units are necessary for identification. Since all workers under consideration were affected to some extent because of the Fornero reform, I focus my attention on the identification of the ACRT and ACR instead of ATT and ATE. As an aside, it is also worth noting that in a binary treatment case, this distinction between causal response on the treated and treatment effect on the treated does not exist.
For the identification of the ACR, I estimate the following TWFE model: where T q is the increase in the minimum retirement age of a parent due to the reform belonging to cell q defined as M RA q,2014 − M RA q,2010 which is a time invariant measure of treatment intensity; post2011 t is a dummy that takes the value of 1 if time period t is post reform, parent represents either father or mother, Y itq represents different outcomes for child i (students status, first jobseeker, employed, not employed), α q,parent is the parent's cell fixed effect, X itq is a vector of controls for the child which includes marital status of the parent, region-year interaction which controls for region effects over time, age-year interaction which controls for cohort effects over time, age-region interaction controlling for cohort effects by region and a T q,parent trend over time which controls for the variable effects of macro shocks over time that can affect parents belonging to different cells differently and ϵ itq is a random error term. β 1 is the parameter of interest. However, it does not represent the ACR without imposing an additional assumption. Let us look at this in detail. β 1 is a DID coefficient which represents a difference in the average change in the outcome between the group that received treatment dose d j and the group that received the treatment dose d j−1 . Mathematically, ) is a 'selection bias' which may arise if the ATT of dose d j−1 is not the same for the group that actually received dose d j and the group that actually received dose d j−1 . This brings us to the introduction of an additional assumption, called the 'strong parallel trends' assumption which is as follows: This assumption says that the average change in outcome for any given dose d is equal between those who actually experienced dose d and those who did not in reality.
Under this assumption, which represents the average causal response of the treatment across all units that received some dose of the treatment (see Callaway et al., 2021 for the proof). It is noteworthy that De Chaisemartin and d'Haultfoeuille (2020) and Callaway et al. (2021) have recently criticized the TWFE estimator as it can be represented as a weighted average of underlying ATE parameters with the weights being negative at times. An alternative to the TWFE regression strategy in this multi-valued treatment scenario is to use the estimators provided by Callaway et al. (2021), but due to the lack of practicality of the application of their estimators in complex regressions with many control variables, I use the TWFE model. Model (1) is run separately to estimate the effect of the sample of eligible mothers (45-59 years old with at least 10 and less than 40 years of contribution) on their offsprings and the effect of the sample of eligible fathers (45-64 years old with at least 20 and less than 40 years of contribution) on their offsprings. Note that in the regression estimating the effect of eligible mothers on their offsprings, the husbands of these mothers may or may not be eligible fathers. Similarly, in the regression estimating the effect of the sample of eligible fathers on their offsprings, the corresponding wives may or may not be eligible mothers. The standard errors are clustered by cell of the parent. 2 The benchmark specification has 852 clusters for the regression of the effect of mothers on offsprings and 830 clusters for the regression of the effect of fathers on offsprings due to the reform.

Effect of the Fornero reform on offsprings
In this subsection, I discuss the results of the effect of the Fornero reform on the co-residing offsprings of the affected workers. The results are presented in Table 4 and are obtained from estimating equation (1) separately for the sample of 15-29 year old offsprings with affected mothers (Panel A) and the sample of offsprings with affected fathers (Panel B). Note that these two samples are not mutually exclusive as for many offsprings, both parents have been affected by the reform. However, in a future section investigating the robustness of these estimates, results of only the offsprings with both parents affected by the reform have also been reported.
Panel A of Table 8 reports the effects of the reform from the mothers' side on the offsprings. This sample is consisted of 7,288 offsprings. However, 125 of them have been dropped as they were singleton observations (i.e., the only observation in the cell defined by age, gender, years of contribution and sector of employment) and hence inappropriate for clustering. We see that mothers had a statistically significant decrease of 1.97 percentage point (ppt) in the probability of their offsprings being in student status. Since 55.77% of the this offspring sample were students reform the reform, this translates to a 3.53% decrease. It is to be noted that this is just the effect of a 1-year increase of the mothers' MRA. Since, the average increase in mothers' MRA was 5.2 years, the causal response of the reform was an (3.53 × 5.2) = 18.37% decrease in the probability of the offsprings of being a student. There has been a correspondingly significant increase of 1.96 ppt in the probability of the offsprings searching for their first job due to a 1-year increase in the mother's work horizon. Since only 10.3% of the offsprings were first jobseekers in the pre-reform period, this is a large effect of a 19% increase. Also, since the average mother experienced a 5.2 year increase in their work horizon, the total increase in the probability of the average offspring to be a first jobseeker is 10.19 ppts (98.8%) due to the Fornero reform. There has been no statistically significant effect on the offsprings' likelihood of being employed or not employed. These results imply that the reform has led to an increase in the transitions of offsprings from education to their first job search.
From Panel B, we see that fathers did not affect their offsprings' likelihood of being a student, searching for their first job, being employed or unemployed. This can be accounted for by the fact that fathers did not change their labour supply significantly either in the extensive or in the intensive margin (Carta and De Philippis, 2021). The validity of the underlying parallel trends assumption of these regressions has been discussed later.

Heterogeneous effects
In the previous section, I discussed that only mothers had a significant effect on their offsprings as a result of their increased work horizon due to the Fornero reform. In this section, I dig deeper into this effect by exploring some heterogeneities in the effects experienced by the offsprings in terms of gender and north versus south of Italy.

Effects by gender
I examine whether the effects of the Fornero reform on the offsprings of the affected workers have been different for male offsprings as opposed to the female ones. In Italy, the labour force participation of females is lesser compared to males. For the age group of the offsprings under consideration which is 15-29, in the pre-reform period, 52% of the males were active in the labour market as compared to 38% of the females. Thus, there is reason to expect that the effect could be different between the two genders. As mentioned in the previous section, the original sample consists of 7,288 offsprings who have an eligible mother affected by the reform. Of these, 3.874 are males and 3,289 are females. However, in Table 10, where the heterogeneous effects of the reform by male and female offsprings are reported, one would notice a small deviation in the number of observations on account of some observations having to be dropped for being singletons. The notes of the table provide further details. In Table 10, I separate mothers' effects of the reform on offsprings by gender. By comparing the coefficients in the table, we see that the mothers' effect on the offsprings for being a student is very similar between males and females. Male offsprings, however, seem to have experienced more than twice the increase in the probability of being a first jobseeker than female offsprings. Male offsprings experienced a 3.02 ppt increase in their probability to become a first jobseeker due to an increase in mothers' working horizons. Since 10% of male offsprings were seeking for their first job before the reform, a 3.02 ppt increase equates to a 30.2 percent increase in first job seekers due to a 1-year increase in maternal work horizon. Therefore, the total increase in job seeking likelihood of male offsprings due to a 5.2 year average increase in maternal MRA is 15.7 ppts (157%).

Effects by north vs. south of Italy
Previous evidence has shown that the cost of living in the south of Italy is about 16% lower than the northern regions of Italy (Cannari and Iuzzolino, 2009). Thus, one could expect the reform to have a higher effect in the south compared to the north because a per unit increase in lifetime earnings is accompanied by a relatively higher increase in purchasing power in the south. In this section, I consider how the effects of mothers' increases in MRA have been different between the northern and southern Italy. As mentioned in the previous section, the original sample consists of 7,288 offsprings who have an eligible mother affected by the reform. Of these, 4,284 are from the north and 2,879 are from the south of Italy. However, in Table 11, where the heterogeneous effects of the reform by north and south Italy are reported, one would notice a small deviation in the number of observations on account of some observations having to be dropped for being singletons. The notes of the table provide further details. Table 11 shows that the effect that mothers had on their offsprings due to an increase in their work horizons has been significantly more in the south than in the north. We see a 4.05 ppt (36.13%) increase in the likelihood of the offspring to be a first job seeker in the south as opposed to a 0.62 ppt (6.36%) increase in the north of of Italy due to a 1-year increase in maternal work horizon . A 1-year increase in maternal work horizon has also led to a 3.46 ppt (6.20%) decrease in the likelihood of being a student in the south as opposed to 0.83 ppt (1.47%) in the north.

Effect of the reform on lifetime earnings
In this subsection, I discuss a potential mechanism to explain the effect of the reform on offsprings only through the mother's side but not through the father's. I calculate how much of an increase in income is implied by an increase in the work horizon induced by the Fornero reform for the average mother and the average father under certain assumptions and show that this increase is more than double for the mother compared to the father.
As supported by sample averages shown in Table 2 and rounding up to the nearest integer, I assume that the representative mother is 51 years old, has 25 years of contributions, would have retired at 60 in the absence of the reform and experienced a 7-year increase in MRA due to the reform (due to the increase in old age retirement from 60 to 67). Similarly, I assume that the representative father is 53 years old, has 30 years of contributions, would have retired at the age of 63 in the absence of the reform (reaching 40 years of contributions at retirement) and experienced a 3-year increase in MRA due to the reform (because of the increase in required years of contributions from 40 to 43). It should be noted that this lifetime simulation is done for the average mother and father and not for their average increase in MRAs. Also, the results of this simulation should not be confused to correspond to the marginal (1-year) effects of the reform on the average parent reported in Table 8. Rather, the simulation results are a representation of the change in lifetime earnings of an average mother who experienced a 7-year increase in MRA and the average father who experienced a 3-year increase in MRA. I also assume that they are observed in the year 2011, just before the implementation of the reform. 3 Since SHIW data is available every other year and the latest pre-reform year that I observe is 2010, I use the values of this year for the necessary parameters in the lifetime income change simulations. The mean male income in the sample in 2010 is e22,014 and the mean female income e16,969 and that is what I assume as income earned by the representatives as of 2010. I assume that real wage increases at the rate of 2.23% annually (Jappelli and Padula, 2016), that there is no inflation in the economy and a 2% discount rate. I further assume that the representative has a probability of death every year and these survival probabilities by age are obtained from the mortality.org database which provides probabilities for each gender by cohort until the age of 110. I report all monetary amounts in 2011 values.
The lifetime income of the representative individuals starting from 2011 until death can be calculated as: Net annual income from 2011 onward growing at a 2.23% real rate until retirement + part of the severance pay received due to contributions from 2011 till retirement + pension income after retirement This lifetime earning can be calculated with different parameter values under the assumption of an absence of the reform and a presence of the reform. The differential between the two gives the change in lifetime earnings due to the Fornero reform. The severance pay is calculated according to the following formula: Years of contributions×0.0691×yearly salary. The contributions for severance pay are capitalised using a 0.015+0.75π accrual rate where π is the rate of inflation. Since I assume a zero inflation rate, this capitalisation simply happens at a factor of 0.015. I calculate pension as the final year's salary times the expected replacement ratio. This pension is received until the time of death which in this simulation is until the age of 110 with a probability of survival every year conditional on having survived the previous years. For the expected replacement ratio, I use the average pre-2011 values for each gender (64.56% for females and 70.95% for males) for lifetime income calculations under the assumption of an absence of the reform and the average post-2011 values for each gender (61.49% for females and 67.43% for males) for the calculations under the assumption of a presence of the reform. Additional notes about the details of the simulation including the the exact formula used to calculated each component of the lifetime earnings can be found in the Appendix. Table 12 presents the figures for income until retirement, severance pay and pension income for each gender under the assumptions of an absence and a presence of the reform and the differences between the two. As we can see, the lifetime earnings increase for women due to the reform is e82,054, while for men, it is e32,491. 4 Thus, the lifetime earnings of the average working mother increased e49,563 more than the average working father. It is to be noted that this differential of e49,563 is a lower bound estimate because the simulation has been done by assuming average parameter values of only employed individuals. It disregards the increase in lifetime earnings achieved due to an increased participation of women who were previously out of the labour force and became active due to the introduction of the reform (see Carta and De Philippis, 2021). Since the reform induced an increase in labour force participation only in the part of females but not males, the true differential increase in lifetime earnings can be expected to be even higher between the two genders.

Effect of the reform on present income and expected retirement age
To further delve into potential mechanisms for observing an effect of the reform on offsprings only through the mother's side but not through the father's side, I estimate the effect of the reform on the current income and expected retirement age of the parents. The estimations have been done using a similar difference-in-difference strategy using the following equation: where T q is the increase in MRA of a parent belonging to cell q, X itq is a vector of control variables for parent i which are marital status and a region-year interaction, α t are year fixed effects that control for cyclical fluctuations in the economy over time, α q are cell fixed effects that control for age, years of contribution and sector of employment and ϵ itq is a random error term. β 1 is the parameter of interest that captures the effect of the intensity of the reform.
Y itq represents the outcome variables of interest which are present income and expected retirement age. Table 9 shows that the reform did not have any effect on the present income levels of either gender. However, the table also shows that females expect that their retirement age would increase significantly as a result of the reform whereas males do not.
The results of the last two subsections are important findings which suggest the following. Females are cognizant of the fact that they would be working for a much longer time period until they retire. The increase in working horizon is large (7 years for the vast majority) and the subsequent increase in lifetime earnings is also large and anticipated. This in turn affects their interaction with their coresiding offsprings and they can offer their offsprings to stay back at home and invest time in better job search rather than continuing with their studies. The fathers, on the other hand, experience a much smaller increase in lifetime earnings due to the reform and more importantly, do not even anticipate this increase as suggested by the fact that they do not expect an increase in their working horizons due to the reform. This provides an explanation why the increase in lifetime earnings from their side does not significantly affect their offsprings' labour market decisions.

Robustness checks 8.1 Sample selection concern
The analysis done in this study provides estimates for the effect of the Fornero reform on the student status and labour market outcomes of offsprings aged 15-29 co-residing with their parents who have been affected in various degrees due to the reform. It, however, does not take into account the offsprings of the affected parents living outside the household due to a lack of information on the income and years of contribution of the parents of the offsprings residing independently. Thus, a natural concern that arises is whether the reform induced a change in the living patterns of these offsprings with their parents. If this were the case, it would threaten the internal validity of the estimates due to endogenous selection into or out of the sample. However, as shown in Table 14, the reform did not affect the number of offsprings living within or outside the household from either parents' side.

Parallel trends assumption of difference-in-difference
The estimation strategy used in the study is a difference-in-difference which relies on the assumption of parallel trends of the treatment (more exposed individuals) and control groups (less exposed individuals) before the introduction of the treatment (reform). To test whether this assumption holds, I do some placebo regressions using the pre-reform years of 2004-2010 where I fictitiously introduce the reform in the years 2006, 2008 and 2010. Since the reform did not actually take place in these years, the fictitious effect of the reform should be insignificant if parallel trends hold. A significant coefficient estimate would imply a violation of the parallel trends assumption. However, it should be noted that these pre-trend tests cannot distinguish between the 'standard' and the 'strong' parallel trends (Callaway et al., 2021). Table 9 shows the results for the offsprings' outcomes of interest as a result of the fictitious treatment experienced by mothers. The estimation has been done in the same fashion as in equation (1) with the reform being fictitiously assumed for the years 2006, 2008 and 2010 and the regression being run separately each time. As we can see, all the coefficients are insignificant, implying that parallel trends hold. Appendix tables A1-A6 further show the results of placebo tests of the other DID analyses that have been discussed in this paper. As we can see that these coefficients are almost always insignificant, On the rare occasions that some of these coefficients came out to be statistically significant, the qualitative conclusions made in this paper remain unchanged.

Effect on offsprings in households where both parents have been affected by the reform
In the main specification (1) outlined in section 5, the regressions considered offsprings who had either an eligible mother or an eligible father or both. Specifically, for estimating the effect of the affected mothers on the offsprings, the fathers may or may not have been 'eligible' (see section 4). Similarly, in the regression estimating the effect of the sample of eligible fathers on their offsprings, the corresponding wives may or may not have been eligible mothers. A reasonable concern that arises in this regard is the possible omitted variable bias due to the correlation between the increases in MRAs of mothers and fathers. To address this concern, in this section, I restrict the sample only to those households that have both an eligible mother and an eligible father which allows me to jointly control for the MRA increases of both parents. This sample is comparable to the original one in terms of parents' and offsprings' characteristics. A slightly higher proportion of the parents in this sample are highly educated and employed with comparable increases in MRAs while a higher percentage of the offsprings are students compared to the original sample. I estimate the mothers' and fathers' effects of the reform on the offsprings separately using specification (1) and report the results in Panels A and B of Table 15. I also jointly estimate the effects of both parents exposed to the reform on offsprings' outcomes using the following model: Y itq = β 1 T q,mother ×post2011 t +β 2 T q,f ather ×post2011 t +β 2 X itq +α q,mother +α q,f ather +ϵ itq (3) where symbols retain their usual meanings and standard errors have been clustered at both the cells of mothers and fathers simultaneously. I report the results in Panel C of the table.
The results show that the effect on first job seeking remains stable and so does the absence of effects from fathers. Thus, we see that the main results are qualitatively robust to changes in sample selections and do not suffer from omitted variable biases.

Conclusions
This paper investigates the effects of the increase in the working horizon of the Fornero reform on cohabiting offsprings aged 15-29 of the affected workers. Exploiting the increase in the work horizon induced by the reform which affected workers of different ages, genders, sectors of employment and years of contributions differently, I use a difference-in-difference strategy to show that the increase in work horizon of mothers significantly increased the probability of their offsprings to seek their first jobs while reducing their probability of being a student. This implies that offsprings are more likely to transition from education to employment after the reform. I also find that this effect is stronger on male compared to female offsprings and the effect is more pronounced in the south rather than in the north of Italy. At the same time, fathers did not affect any student or labour market outcomes of their offsprings. Using some lifetime earnings simulations from the point of the reform till death, I show that mothers experienced a substantially higher increase in lifetime earnings than fathers. I also find that only mothers successfully anticipate an increase in their work horizons due to the reform while fathers do not. Considering the relatively low returns to a mid-high level of education in Italy with respect to comparable European Union nations (Depalo, 2017), I infer that offsprings persuade their mothers who correctly anticipate an increase in lifetime earnings due to the Fornero reform to sponsor their job search as they leave or take time off from schooling. The possible explanation to the higher effect on male offsprings is the much higher labour force participation of males compared to females in Italy. The effect is much more pronounced in the south of Italy where purchasing power of parents in real terms increased more than that in the north due to the lower cost of living. The paper provides one of the first evidences on the intergenerational perspective effects of pension reforms that increase workers' work horizons.  N otes: A stands for age, C for number of years of contribution, Q = A + C is the so-called "quota". The sum of age and years of contribution must be larger than or equal to Q to reach retirement eligibility. Independent of actual age, retirement eligibility is also granted when the number of accrued years of contribution is sufficiently high (39 in 2007, 40 in the following years, 42 or 43 after the reform).                N otes: The table shows the effects of the change in the Minimum Retirement Age of parents induced by the Fornero Reform of December 2011 In Italy on the several labor market outcomes of offsprings aged 15-29 years. The data spans the years 2004-2016 of the SHIW. In the above analyses, mothers are eligible to be in the sample if they are of the ages 45-59 with at least 10 and less than 40 years of contribution. Fathers are eligible to be in the sample if they belong to the age group 45-64 and have at least 20 and less than 40 years of contribution. Control variables include marital status of the parent, the cell the parent belongs to, region-year interaction, region-age interaction, age-year interaction and a trend for the change in the MRA over time. Robust standard errors have been generated by clustering at parents' cell level and are reported in the parentheses. In Panel A, the sample of children is consisted of those who have an eligible mother, regardless of the eligibility of the father. This sample has 879 cells/clusters. The sample originally had 7,288 observations out of which 125 have been dropped as they were singleton observations. Similarly, Panel B consists of the sample of children who have an eligible father regardless of the eligibility of the mother. This sample has 863 cells/clusters. This sample originally had 9,687 observations out of which 91 have been dropped as they were singletons. "Not employed" includes the unemployed (except the first jobseekers), homemakers, well-off, job pensioners, non-job pensioners and voluntary workers. ***p<0.01, **p<0.05, *p<0.10   N otes: The table shows the effects of the change in the Minimum Retirement Age of mothers induced by the Fornero Reform of December 2011 In Italy on the several labor market outcomes of offsprings aged 15-29 years by gender. The data spans the years 2004-2016 of the SHIW. In the above analyses, mothers are eligible to be in the sample if they are of the ages 45-59 with at least 10 and less than 40 years of contribution. Control variables include marital status of the mother, the cell the mother belongs to, region-year interaction, region-age interaction, age-year interaction and work horizon increase interacted with year. Robust standard errors have been generated by clustering at mothers' cell level and are reported in the parentheses. The samples of children in both panels are consisted of those who have an eligible mother regardless of the eligibility status of the father. There are 693 cells/clusters in the male offspring sample and 628 in the female offspring sample. The male offspring sample originally had 3,938 observations and the female 3,350. However, 207 male and 255 female offsprings had to be dropped as they were singleton observations and hence inappropriate for clustering. ***p<0.01, **p<0.05, *p<0.10 N otes: The table shows the effects of the change in the Minimum Retirement Age of mothers induced by the Fornero Reform of December 2011 in Italy on the several labor market outcomes of offsprings by macro regions. The data spans the years 2004-2016 of the SHIW. In the above analyses, mothers are eligible to be in the sample if they are of the ages 45-59 with at least 10 and less than 40 years of contribution. Control variables include marital status of the mother, the cell the mother belongs to, region-year interaction, region-age interaction, age-year interaction and a trend for the change in the MRA over time. Robust standard errors have been generated by clustering at parents' cell level and are reported in the parentheses. There are 726 cells/ clusters in the Northern Italy sample and 557 in the Southern Italy sample. The Northern Italy sample originally had 4,360 observations and the Southern Italy sample 2,928. However, 172 observations from Northern Italy and 180 observations from Southern Italy had to be dropped as they were singletons and hence inappropriate for clustering. In both panels, the samples of children are consisted of those who have an eligible mother, regardless of the eligibility of the father. ***p<0.01, **p<0.05, *p<0.10 N otes: The simulation assumes the mean pre-reform income for the representative individuals which is 16,969 for the mother and 22,014 for the father. Income is assumed to grow at the real rate of 2.23% and a zero inflation rate is assumed. The discount rate is assumed to be 2% per year. The representative mother and father are assumed to be 51 and 53 years old respectively and observed just before the reform, in the year 2011. The mother is assumed to have 25 years of contributions and the father 30. Therefore, the mother experienced a 7-year increase in the minimum retirement age and the father a 3-year increase. Severance pay is calculated according to the formula years of contributions × 0.0691 × yearly salary and is assumed to be capitalised at an accrual rate of 1.5% per year. The assumed replacement ratio is 64.56% (61.50%) for the mother and 70.95% (67.43%) for the father for the lifetime income calculation under the assumption of an absence (presence) of the reform. The representative individuals are assumed to survive up to the age of 110 with a probability of survival every year conditional on surviving the previous years. The survival probabilities are obtained from mortality.org for the respective cohorts by gender. N otes: The table shows the effect of the Fornero reform on the present income and the expected age of retirement with full pension benefits for mothers between the ages 45-59 with at least 10 years and less than 40 years of contribution and for fathers between the ages 45-64 with at least 20 and less than 40 years of contribution. Data for the years 2004-2016 of SHIW have been used. The control variables include parent's marital status, parent's cell fixed effect, year fixed effect and region-year interaction. Robust standard errors clustered at the cell level have been used for inference. The income sample for females has 758 cells/clusters while that for males has 792. The expected retirement age sample for females has 604 cells/clusters while that for males has 726. ***p<0.01, **p<0.05, *p<0.10 N otes: The table shows the effect of the Fornero reform on the number of offsprings residing outside and inside the household for mothers between the ages 45-59 with at least 10 years and less than 40 years of contribution. Data for the years 2004-2016 of SHIW have been used. The control variables include the mothers's marital status, the mother's cell fixed effect, year fixed effect and region-year interaction. Robust standard errors clustered at the cell level of the parent have been used for inference. The sample has 792 cells/clusters. ***p<0.01, **p<0.05, *p<0.10 N otes: The table shows the effects of the change in the Minimum Retirement Age of parents induced by the Fornero Reform of December 2011 In Italy on the several labor market outcomes of offsprings aged 15-29 years. The data spans the years 2004-2016 of the SHIW. In the above analyses, the sample is consisted only of households that have both an eligible mother and an elible father. Mothers are eligible to be in the sample if they are of the ages 45-59 with at least 10 and less than 40 years of contribution. Fathers are eligible to be in the sample if they belong to the age group 45-64 and have at least 20 and less than 40 years of contribution. Control variables include marital status of the parent, the cell the parent belongs to, region-year interaction, region-age interaction, age-year interaction and a trend for the change in the MRA over time. Robust standard errors have been generated by clustering at parents' cell level and are reported in the parentheses. Panels A and B present the effects of mothers and fathers on offsprings' outcomes and have 703 and 646 cells/clusters respectively. Panel C presents the results from specification (3) where both the mothers' and fathers' MRA changes have been jointly incorporated in the regression and has 632 clusters. "Not employed" includes the unemployed (except the first jobseekers), homemakers, well-off, job pensioners, non-job pensioners and voluntary workers. ***p<0.01, **p<0.05, *p<0.10

Appendices Calculation of lifetime earnings
The lifetime income of the representative individuals starting from 2011 until death is be calculated as: Net annual income from 2011 onward growing at a 2.23% real rate until retirement + part of the severance pay received due to contributions from 2011 till retirement + pension income after retirement The The net annual income in 2010 is assumed to be the mean income of the average mother/father in the sample in the year 2010 in the simulation.
The severance pay is calculated according to the following formula: Years of contributions×0.0691×yearly salary capitalised using an accrual rate of 0.015+0.75π where π is the rate of inflation which is zero by assumption. Thus, the part of the severance pay received due to contributions from 2011 until retirement can be calculated as (0.0691 × 1.015) × (Net annual income in 2011 + Net annual income in 2012 + ...+ Net annual income on the year of retirement). This value can then be transformed into 2011 value by multiplying by ( 1 1+δ ) N −1 . Pension income after retirement till death (age 110 with a probability of survival each year) can be calculated using the exact same method used to calculated net annual income from 2011 until retirement explained above accounting for the replacement ratio.