IrpetDin. A Dynamic Microsimulation Model for Italy and the Region of Tuscany

The first event is death. We have taken death probabilities from ISTAT for the period 2008-2016, broken down by age, by gender and distinguishing between Italy and Tuscany. With regard to future ISTAT releases, we only take data on life expectancy, broken down by gender. We therefore use the Brass method (Brass, 1968; Livi Bacci, 1990) to estimate death probabilities in 2050, based on known death probabilities as at 2016 (broken down by age, by gender and by territory) and assuming a constant slope and intercept given by the difference in life expectancy between 2050 and 2016. We calculate probabilities between 2017 and 2049 by applying the average yearly variation in death probabilities between 2016 and 2050, broken down by age, gender and territory.

For survivors, we increase the age by one year and we change family relationships within the households where someone has died. For example, the wife becomes the head of the family if the husband dies, the son if both parents die and so on.

As briefly described in Walker and Davis (2013) many microsimulation models simulate the “marriage market” with a Monte Carlo method based on matrices of probabilities of marrying for each pair of the n-th men and the n-th women, usually estimated with a logistic regression by age and education. Differently, others models use retrospective partnership histories from surveys.

In IrpetDin we follow the first approach. Using the rate at which people get married based on official ISTAT data, broken down by age, by gender and by territory, we select potential candidates among single, divorced and widowed individuals aged between 18 and 59³. As Istat (2014) has described in detail, social studies broadly agree that marriage choices are based on affinity between spouses, rather than differences. In particular, marriages between partners with the same level of education are more common. Therefore, we match potential candidates by taking into account their level of education. More specifically, we calculate the rates at which women, distinguished by age classes, by level of education (low, medium, high) and by territory, marry men, distinguished by age classes, by level of education (low, medium, high) and by territory⁴. For every woman, there is a given rate according to which she could marry 27 different types of men.

For each woman, we extract a random number to give the probability of her marrying a certain type of man (randomly extracted from the 27 different types). If the random number is lower than the probability, the woman and the man get married, otherwise the procedure continues with another one of the 27 different types of men.

After potential candidates have been matched, the model creates a new household. If one or both of the spouses have children from past relationships, the model adds them to the new household. Finally, we attribute family relationships (head of the family, husband, son, etc.) to each member of the new household.

As often in microsimulation models, in IrpetDin fertility is modelled with a probabilistic approach and an alignment to official data. We simulate the birth of a child for married or cohabiting women aged between 18 and 45. Differently with respect to an other Italian model, Mazzaferro and Morciano (2012), we calculate fertility rates on administrative regional data where available. Specifically, we use birth certificates for the Region of Tuscany (RT) while ISTAT’s birth survey for Italy. Rates are distinguished by nationality, marital status, age, number of children and level of education and are aligned to the official total fertility rate predicted by ISTAT. Probabilities are aligned to official data. We then probabilistically generate the flow of newborns. The model adjusts family relationships inside households as a consequence of a new child being born.

Divorce is modelled similarly to Mazzaferro and Morciano (2012). We select potential candidates among married/cohabiting people aged between 20 and 64 and who have been married/cohabiting for at least three years⁵. We calculate the rate of divorce/dissolution by territory, by age, by gender and by nationality, based on official civil status data by ISTAT. Then, we make couples divorce, create a new household and adjust the family relationships of the members of both the original and the new household.

We then simulate single individuals leaving their household of origin. Potential candidates must be employed, unmarried and not the head of the family. The rate at which individuals aged between 18 and 49 leave home is calculated based on the ISTAT survey entitled “Famiglia e soggetti sociali” [“Families and social subjects”]⁶. As usual, if events alter the composition of households, we adjust the relationships of the members of the households involved: in this case, the family of origin and the new family.

Last but not least, the demographic module simulates migration. As IrpetDin covers both Italy and Tuscany, we simulate migration flows into/out of Italy from/to abroad, and migration flows of Italians into/out of Tuscany from/to other parts of Italy.

We start with migration into/out of Tuscany. According to official ISTAT data, Tuscany has a positive balance of registrations and cancellations, meaning more people are coming in than going out. We calculate migration rates into Tuscany from other parts of Italy by gender, by age and by level of education, based on regional administrative data. We then make some individuals (excluding foreigners) migrate to Tuscany from other parts of Italy.

We then move on to Italians migrating abroad. This phenomenon has considerably increased in Italy over the last decade, also as a result of the economic recession. Since ISTAT does not provide detailed data about Italians who have emigrated, we calculate a single probability of migrating abroad, without distinguishing between the various characteristics, and we apply this to Italians aged between 30 and 44.

The final step is to simulate foreigners migrating into/out of Italy. Italy has a positive balance between registrations and cancellations of foreigners, meaning we need to add new individuals to the sample. For the 2009-2017 period, we have taken the inflows of foreigners to Italy and to Tuscany, based on ISTAT’s demographic balances of foreign citizens, distinguished by gender, by type of family (single, with spouse, with children, etc.), by number of children, by family size, by marital status, by age, by employment status and by level of education. From 2018 onwards, the number of new individuals added to the sample corresponds to ISTAT’s forecast. We attribute gender, age and level of education to immigrants by taking the average characteristics of foreign citizens who are resident in Italy, based on the Survey on Household Income and Wealth (SHIW) conducted by the Bank of Italy.

After simulating all migration flows, we adjust household compositions and the relationships between the respective members. We also simulate family reunions for immigrants who come to Italy in order to join their partners or relatives.

In each year $t$ of the simulation period, immediately after the education module, labour supply is composed of employed ( $E_{t-1}$ ) and unemployed individuals ( $u_{t-1}$ ) from the previous year. However, inactive people from the previous year and students who completed their education in year $t$ could become active and start to participate in the labour market.

We calculate participation rates, by age, by gender, by level of education, by role within the family (head of the family or not) and by territory (Italy or Tuscany), based on ISTAT’s Italian Labour Force Survey. We then activate inactive people from the previous year, if their inactivity has been lower than 3 years for women⁸ , as well as students who finished school or university in year $t$ , whether they obtained a diploma/degree or dropped out. If the random number is lower than the probability, then the candidate becomes active and enters the labour market as a job seeker. Labour supply in each year t (S _t ) is therefore composed of the sum of employed at $t-1,$ ( $E_{t-1}$ ), unemployed individuals at, t – 1, ( $u_{t-1}$ ), and the new flow of job seekers ( $j_t$ ) (1). The sum of $u_{t-1}$ and $j_t$ is the total number of unemployed individuals in each year t, u_t .

Dante is a structural econometric macro model, developed by IRPET, which analyses economic structural changes and the effects of policies in Italy. Since it is designed to evaluate regional policy, Dante distinguishes between three Italian macro-regions: Centre-North, South and Tuscany, all of which are linked together using an interregional trade matrix⁹.

It is based on post-Keynesian theories and may be classed as a stock-flow consistent model. The complete structure of the model contains a rather complex set of behavioural equations and accounting identities. The following are some essential features of the model.

Goods and services market. Aggregate demand is a function of private and public consumption, investments and exports. The relationship between output and factors of production is based on a Leontief production function. Factors of production are therefore used in fixed proportions, since there is no substitutability between factors in the short run¹⁰. Potential output is determined by the maximum rate of use of labour and capital¹¹. Actual output is equal to the level of aggregate demand.

Labour market. Labour supply is exogenously aligned to the participation rates predicted by ISTAT. Labour demand, expressed in standard labour units (SLU)¹² , is given by the ratio between actual output and labour productivity¹³. SLUs are converted into the number of workers through a coefficient that measures the average ratio between SLUs and numbers of workers. Labour demand, expressed in number of workers, is then compared with labour supply in order to predict unemployment or job vacancies.

Adjustment mechanisms. Unlike neo-classical models, in the short run, excess/insufficient aggregate demand compared with potential output is only adjusted through quantities and not through wages and prices. Therefore, if aggregate demand is lower than potential output, then labour demand is lower than labour supply and the model predicts an increase in unemployment. In the medium-long run, adjustments may also be made through wages and prices. If aggregate demand exceeds potential output, because, for example, there is a labour supply shortage, then wages and labour costs increase and companies in turn increase the prices of goods and services, with potentially depressing effects on aggregate demand, actual output, labour demand and employment.

Dante is able to analyse economic structural changes up until 2035. IrpetDin therefore takes labour demand data from Dante, expressed in SLU and broken down by sector of activity (public or private), until 2035¹⁴. For the following years, IrpetDin assumes a growth rate in SLU equal to the average of the previous few years.

Labour demand data from Dante does not include workers receiving redundancy funds (RF). IrpetDin therefore uses administrative data from INPS, Italy’s National Social Security Institute, regarding the number of hours of redundancy funds in order to estimate the number of workers involved with the social safety net and to simulate scenarios about their evolution over time. More specifically, we estimate SLUs receiving RF by dividing the number of hours by an average standard working time. We then add the SLUs receiving RF to the labour demand, expressed in SLU, derived from Dante.

Dante makes projections about labour demand by sector of activity. IrpetDin breaks down labour demand within the various sectors by level of education, applying the proportions observed in terms of companies’ intention to hire staff, as expressed in the Excelsior survey¹⁵. However, IrpetDin allows us to hypothesise different scenarios regarding the proportions of labour demand by sector and level of education over time.

In the basic scenario, IrpetDin converts the labour demand originally expressed in SLU into workers, using the same ratio between SLUs and number of workers used in Dante. However, the model also allows us to conjecture a different evolution of the ratio in the future, depending on assumptions about working time.

Given the peculiar characteristics of the Italian labour market, IrpetDin models the matching of labour demand and supply by sector, by level of education and by nationality. Indeed, a distinctive feature of the Italian labour market is its mismatch of particularly highly skilled individuals. According to Adda et al. (2017) the incidence of over-skilling is especially high among graduates (20%) and individuals with secondary education (14%). Another distinctive feature concerns the type of migration. Unlike continental Europe or northern countries, migration in Italy is a relatively recent phenomenon and migrants are generally lower skilled. As described by the Ministero del Lavoro e delle Politiche Sociali (2019), the employment rate is higher for migrants than native Italians and migrants are typically employed in low-skilled job positions.

IrpetDin models the matching between labour demand and supply by taking these distinctive features of the Italian labour market into account. Given the high level of over-skilled graduates in Italy, we assume that unsatisfied graduates, who do not find a job requiring tertiary education, go on to compete for jobs requiring secondary or even only primary education. Furthermore, we assume that individuals with secondary education, who do not find a job requiring secondary education, go on to compete for jobs requiring just primary education. We then hypothesise a sort of priority among migrants for jobs requiring primary education.

More specifically, the matching is completed in the following way, for each time frame $t$ and sector of activity $s$ . With regard to labour supply, we initially only consider individuals who are already employed at time $t-1$ . If labour demand, $L_{t,s}$ , is lower than the number of those employed at the time $t-1$ , $E_{t-1,s}^{}$ , then the sector has an over-supply (2). Consequently, we fire workers according to the ratio between over-supply $e_{t,s}^{}$ and labour demand $L_{t,s}^{}$ .

If labour demand, $L_{t,s}$ , is higher than the number of those employed at the time $t-1$ , $E_{t-1,s}^{}$ , then the sector has an over-demand (3). We then hire $l_{t,s}^{}$ individuals among the remaining part of labour supply, composed of unemployed individuals $u_t^{},$ to the following rules.

We start from jobs requiring tertiary education (4). If labour demand for tertiary education $l_{t,s,T}^{}$ is higher than the number of unemployed graduates $u_{t,s,T}$ , then everyone finds a job with a probability equal to 1, otherwise they find a job with a probability equal to $\frac{l_{t,s,T}^{}}{u_{t,s,T}}$ . Unsatisfied graduates compete with individuals with secondary education for jobs requiring secondary education, again with a probability equal to 1 if labour demand for secondary education is higher than the sum of labour supply with secondary education and unsatisfied graduates, or equal to the ratio $\frac{l_{t,s,S}^{}}{u_{t,s,S}+{(u_{t,s,T}-l_{t,s,T}^{}}_{})}$ if labour demand is lower (5).

IrpetDin then continues with jobs requiring primary education. Initially, only foreign workers compete for low-skilled positions. If labour demand is higher than foreign labour supply, then every foreigner can find a job with a probability equal to 1 and low-skilled Italian workers compete for the remaining positions after foreigners have been hired ( $h_{t,s,P_{str}}^{}$ ) (6). If labour demand is lower than foreign labour supply, then only foreigner workers can find a job with a probability of $\frac{l_{t,s,P}^{}}{u_{t,s,P\_str}}$ (7).

The procedure that we use for the matching simulates labour market transitions-from unemployment to employment and vice versa-only when an imbalance between labour supply and demand by sector occurs. Our partial omission of churning can be a strong drawback, especially in view of the increase in temporary work that happened in the last two decades and will probably continue in the future. IrpetDin could, therefore, over-estimate stable careers, accumulation of social contributions, pension requirements and amounts.

Further, our matching between labour demand and supply does not provide any long run price/wage adjustments. We are, therefore, implicitly assuming that no kind of adjustment of the imbalances can take place except in the quantities.

Another important shortcoming is our implicit assumption of the absence of employment/unemployment lasting less than one year. This is a significant limit when we use IrpetDin to evaluate the effects of pension reforms, such as those implemented in 2019, that change pension requirements even for fractions of a year.

After hiring and firing individuals, IrpetDin predicts the unemployment rate¹⁶. Thanks to the modelling of labour dynamics by level of education and by sector, IrpetDin can also predict a sort of qualitative mismatch, whereby individuals are hired for jobs requiring a lower level of education than they have.

After attributing employment status, IrpetDin simulates types of work, careers and working time. We divide workers into self-employed individuals and employees by applying the proportion observed in the current stock of workers. We then simulate career progressions for employees, from blue-collar worker to white-collar worker, to manager and then from manager to executive, by sector of activity. Transition rates are calculated based on the Italian Labour Force Survey and are then attributed to employees. Part-time work is imputed after estimating a model for the probability of being employed part-time depending on gender, age and number of children. In order to stabilize the value we estimate the probability of being employed part-time on five waves of the Italian Labour Force survey (2009-2013). Given that in Italy part time work has been strongly affected by the economic crisis of 2009¹⁷ we estimate a second model, with the same covariates, on the two waves 2018 and 2019 of the Italian Labour Force survey. Table A4, attached hereto, shows the estimation results of the two models. We use pre-crisis probabilities until 2017 and post-crisis ones from 2018. As we will described in paragraph 10, we assume a return to pre-crisis probabilities in the long run from 2030 onwards.

The final step of the labour and income module is to attribute incomes to individuals who are already employed and new recruits. We estimate a standard Mincer equation for the logarithm of income depending on workers’ characteristics (territory, age, gender, contributory seniority, level of education, employment status, number of hours worked, citizenship), separately for self-employed individuals and employees, based on EU-SILC 2008 data (please see tables A5 and A6, attached hereto, for the estimation results). The estimated coefficients and the distribution of the error term are used to impute incomes¹⁸. For individuals who are already employed, yearly income is calculated by applying estimated coefficients to their changing characteristics (such as age, seniority, qualification, working time) and fixed characteristics (gender), and by adding the difference between their true and estimated yearly income in 2008. For new recruits, income is calculated by applying estimated coefficients to their characteristics and adding an error term, randomly extracted from the distribution of error terms estimated by EU-SILC for workers in 2008.

IrpetDin assumes that incomes grow over time in line with the nominal GDP growth rate, taken from Dante.

In Italy, the pension system has been reformed many times in order to make it financially sustainable and fair between generations. Typically, implementation of these reforms is postponed or revised, in order to make them more acceptable to current generations, which actually ends up offloading costs onto future generations.

The most important reform, the so-called Dini Reform in 1992, had precisely these characteristics. The reform, which shifted from Defined Benefit (DB) pensions to Notional Defined Contribution plans (NDC), was not applied to individuals who were retiring in those years, but rather to those who entered the labour market after 1995.

Specifically, the Dini Reform introduced three different regimes: i) the DB plan for workers with more than 18 years of contributions in 1995; ii) the NDC plan for workers who started working after 1995; and, iii) a mixed plan for workers with contributions of between 1 and 18 years in 1995. The method used to calculate the pension amount differs for the three regimes. For retirees with the DB regime, a full DB plan is applied; for retirees with the NDC regime, a full NDC is applied; for those with the mixed system, a DB plan is applied for the years before 1995 and an NDC plan is applied for the years after 1995 (the so called “pro-rata”).

The Dini Reform was followed by other reforms, whose objectives were, respectively, to increase pension eligibility requirements, to solve financial sustainability problems, or to correct the previous reforms, mainly for electoral issues. In 2011, an important reform was introduced (Italian Law no. 214/2011, the so-called “Fornero Reform”) with the aim of making the system more financially sustainable and fair between generations. The Fornero Reform increased the retirement age, linked retirement requirements automatically to increases in life expectancy and partially extended the NDC plan to all people retiring, also for workers with more than 18 years of contributions in 1995, for years after 2012. Therefore, following this reform, there are two regimes, the mixed and the NDC regime. The two regimes are applied both for old-age pensions and seniority pensions.

IrpetDin simulates the current pension rules, which were once again changed in 2019 with the so-called “quota 100” reform (Italian Law no. 26/2011)¹⁹. It starts by identifying and attributing the pension regime, either mixed or NDC, for each worker in the sample and then calculates the pension amount for both regimes.

The calculation rule for the DB plan can be represented as in (8). The pension amount is determined by multiplying the average income of the last 10 years for employees and 15 years for self-employed workers²⁰ ( $y$ ) by a rate of return ( $r)$ and by the number of working years ( $e$ ). The rate of return decreases with income classes²¹.

The calculation rule for the NDC plan can be represented as in (9). The pension amount is determined by multiplying the total amount of social contributions paid during working life, capitalised at the rate of GDP growth ( $mc$ ) by a transformation rate ( $t$ ). The transformation rate is a value that transforms the amount of contributions into annual pension income; it increases with retirement age and increases over time.

After calculating the pension amount, IrpetDin simulates pension eligibility requirements for old-age pensions and seniority pensions, as established by the law. Requirements increase over time as life expectancy increases, although in 2019 the so-called “quota 100” reform blocked any increases to the retirement age until 2026. The simulation of requirements comes after the calculation of the pension amount because Italian law provides workers in the NDC regimes and those receiving old-age pensions with a form of flexibility to retire before the retirement age, if their pension amount is higher than certain thresholds.

By default, IrpetDin simulates retirement for anyone who meets pension eligibility requirements. However, for old-age pensions, it allows us to simulate the choice of whether to retire or not. Specifically, the model calculates the replacement rate, i.e. the ratio between the pension amount and the last working income. If the replacement rate is higher than a certain level, decided by assumption, then the workers retire, otherwise they continue to work.

Finally, IrpetDin simulates survivor pensions for those whose spouse has died and indirect pensions for children whose parents have died.

Italian social security provides some forms of social assistance for the elderly and retirees. IrpetDin simulates the eligibility requirements and the amounts established by Italian law.

There is a minimum pension supplement, representing an integration of the pension amount up until a specific minimum income threshold, for retirees who live in households²² with an income below certain levels (in Italian, the so-called “integrazione al trattamento minimo”). An additional supplement (in Italian, the so-called “maggiorazione sociale” ) is provided to retirees over the age of 70, depending on their household income.

Finally, for those aged 65 and over, the Italian welfare system provides a form of non-contributory social assistance, financed through general taxation, known as the “assegno sociale”. This is a sort of minimum income to prevent poverty, subject to means testing and proven with a top-up mechanism, which, however, is only for the elderly.

IrpetDin includes a simplified simulation of the implicit subsidy that people receive from public health services (so-called “in-kind” transfers). In-kind health transfers are only attributed to people living in Tuscany, in relation to which we have detailed regional administrative data.

Two different approaches to including in-kind transfers in microsimulation models can be found in the literature: the actual consumption approach and the insurance value approach. The actual consumption approach only imputes the monetary value of in-kind transfers to individuals who actually use the service. The insurance value approach imputes an average monetary value of in-kind transfers to every individual, based on demographic characteristics (usually age and gender), without taking into account the actual use of services.

We apply the insurance value approach, widely used in literature for health services, for hospital, specialist and pharmaceutical services. We calculate average per-capita expenditure by gender, by age, by level of education and by nationality for hospital services; by gender, by age and by nationality for specialist services; and by gender and by age for pharmaceutical services.

The last social security transfer simulated in IrpetDin is long-term care (LTC) for people who are not self-sufficient. As LTC is a strictly age-related benefit, the evolution of the number of people who are not self-sufficient could be very significant for national and regional policymakers.

We estimate the probability of being disabled based on an ISTAT survey, broken down by age, by gender and by level of education (please see table A7, attached hereto, for the estimation results) and, accordingly, we impute disability and non-self-sufficiency to all people in the sample. Public expenditure for LTC is estimated based on official data from the Italian Ministry of Finance and is then attributed to disabled people.

According to IrpetDin, the demographic dependency, measured by the ratio between the number of over 65s and those aged between 15 and 64, will increase from 35% in 2018 to 60% in 2050 (Figure 6a). Demographic dependency figures for Tuscany are even worse, where the ratio will increase from 40% in 2018 to 74% in 2050 (Figure 6b).

Figure 6

Download asset Open asset

The percentage of individuals aged between 18 and 34 with tertiary education will increase in the future, but this will not be enough to reach the level of other European countries, at least not in the next 30 years (Figure 7a). The percentage of people who only have primary education will decrease up until 2030, after which it will become stable at about 15% (Figure 7b).

Figure 7

Download asset Open asset

With regard to the labour market, the model predicts a strong decrease in labour supply (Figure 8a). In the first years of the projection period, the workforce aged between 35 and 55 decreases due to the fall in the birth rate. Elderly workers, on the other hand, do not diminish as they continue to work due to pension reforms. Starting from 2044, the drop in younger workers is no longer offset by the growth in the number of older workers. The numerous baby boomers will begin to retire and leave the labour market (after the postponement introduced by Italian Law no. 214/2011). Similar results are described for an other microsimulation model (Caretta et al., 2012). Indeed, they predict a sharp increase in employment rates for old workers (especially female) for the next 20 years until a stabilisation in around 2040.

Figure 8

Download asset Open asset

Consequently to these dynamics, IrpetDin predicts a diminishing labour force in the long run. With demand growing by around 2.5 million units during the period, according to our projections and in line with the Dante macro model, the unemployment rate will converge to a value that we may define as structural in 2040 (Figure 8b). It then reduces to zero.

Given the growth in labour demand and the diminishing labour supply, the model predicts a gradually decreasing quantitative mismatch, i.e. the difference between labour demand and supply (Figure 9a). According to IrpetDin, the qualitative mismatch will decrease until 2038 (Figure 9b). As in the baseline scenario, the proportions of labour demand by level of education will not change over time, with the reduction in the qualitative mismatch strictly depending on the shortage of labour supply with respect to labour demand.

Figure 9

Download asset Open asset

There is a high rate of people retiring between 2025 and 2044 due to the baby boom generations retiring (Figure 10a). We will have flows of retirees with the old DB plan (which became mixed after the introduction of Italian Law no. 214/2011) and the mixed plan, until 2039. The stock of retirees will increase until 2045 and will then become stable (Figure 10b).

Figure 10

Download asset Open asset

Replacement rates will strongly decline in the future, especially for self employed workers (Figure 11a). In the long run, we will therefore face a problem of intra-generational equity, with retirees with very low pensions. On the other hand, the Gini among retirees will decrease, as differences among retires shall diminish when the NDC plan fully comes into force (Figure 11b). A decreasing in the Gini among Italian retirees in the very long run was also found in Caretta et al. (2012).

Figure 11

Download asset Open asset

The pension expenditure to GDP ratio will decrease from 2045, when the number of retirees will become stable (Figure 12a). In order to evaluate inter-generational sustainability we use the ratio between the future pension amounts and the accumulated contributions amounts for the flow of retirees (the so called “net present value ratio”). If the ratio is over 1 the pension system is not fair between generations. As can be easily seen from Figure 12b, the NDC plan will make the Italian pension system more fair between generations only in the very long run.

Figure 12

Download asset Open asset

As in Marano et al. (2012), the Italian expenditure for social assistance for old people, composed by social pensions and supplements to minimum, will increase in the future (Figure 13b). However, IrpetDin predicts a decrease starting from around 2040, when supplements to minimum will not be in force anymore (Figure 13a).

Figure 13

Download asset Open asset

Finally, IrpetDin predicts a linear but not sharp increase in expenditure for in-kind transfer, composed by specialist, hospital, pharmaceutical services and long term care (Figure 13b).

In this paragraph we compare the distributive effects on retirees between the basic and the official scenario. As explained in the previous paragraph, the official scenario differs with respect to the basic one in the assumptions on economics trends. As can be seen from Table 3, in the official scenario the unemployment rate is lower until 2032 and higher from 2033 onwords than the basic one. In order to evaluate the distributive effect on retirees of this change we measure the 25th and the 75th percentile of the of replacement rates for employees and employed and the Gini coefficient on retirees (see Table 3).

Not surprisingly, in the first period, 2018-2032, there are not differences in replacement rates. The impact on pensions can occur at least with a certain temporal lag with respect to labour market changes. Those that occur during the last ten years, 2040-2050, will have effects only after our observation period (beyond 2050).

Indeed, some differences emerge only after 2033 and specifically the model predicts a mild increase in replacement rates, both at the top and the bottom of pensions distribution, under the official scenario. This effect depends on the lower unemployment rate in the short run (2018-2032) assumed in the official scenario compared to the basic one, that increases earnings and pension contributions and, consequently, the pension amount. Probably, given that replacements rates increase both at the top and the bottom of the distribution, we do not observe any impact on the Gini index.

Conversely, we can not observe in our observation period, ending in 2050, the effects on pensions of the higher unemployment rate assumed in the official scenario compared to the basic one for the medium and, most of all, for the long run (2043-2050).

We removed a part of observations who are 0 years old, otherwise the adjustmet to real data was not feasible.

The individual with the minimum sample weight is consequently not repeated.

According to official data by ISTAT, even if the age at which people get married is increasing, the percentage of those with more than 60 years old are residual, expecially among women.

We have overall 729 rates for each territory.

According to official data by ISTAT, the percentage of divorces before three years of marriage is not residual (about 10% of total dissolutions), but we chose this threshold for possibile issues of sample representativeness.

According to the survey, the majority of single individuals leaves home within 50 years old.

Mobility between sectors and qualifications are usually not modelled.

According to Cataldo and Tosi (2012), women may be discouraged from participating in the labour market because of the lack of adequate educational services for children and depending on their working experience. They estimate an higher probability to become inactive in a given year if a woman was unemployed or employed with non-standard contracts in the previous year. Pacelli et al. (2012) demonstrate that the probability of exiting from employment depends on age, having children, income, firm size. In our model we assume the ultimately exit from the labour after 3 years of inactivity.

Dante considers 30 industries and 59 commodities.

In the long run proportions can change for several reasons, among which technological progress.

It takes into account also the average unemployment rate of the last few years.

Standard labour units measure the number of job positions that correspond to standard workers, i.e. full time workers.

Productivity is endogenously determined within the model.

IrpetDin also takes from Dante real and nominal GDP growth rates.

Unioncamere – Minister of Labour, Excelsior survey.

A comparison of the predicted unemployment rate with the one estimated by Dante gives similar results. In some past experiences we linked IrpetDin and Dante with iterations. Dante computed labour demand as input for IrpetDin. IrpetDin predicted participation rates as input for Dante. Dante computed again labour demand as input for IrpetDin, an so on, until unemployment rates predicted by Dante and IrpetDin converge.

As clearly stated by Istat, Ministero del Lavoro, Inps, Anpal, Inail (2019) part time work strongly increased after the great recession of 2009 , owing to involuntary part time, used by firms as a strategy to tackle the economic recession. Precisely, part time increased starting from 2011 until it stabilizes from 2018. As a result, in 2018, the share of part time work in Italy is more in line to the european average (18,6% vs 20,1%) but involuntary part time is some three times higher (64,1% vs 23,4%).

For simplicity in our Mincer equation we do not correct for selection bias. Consequently, we could over-estimate the returns to education.

This last reform made less stringent pension eligibility requirements.

For working years before 1992 the average of the last 5 and 10 years is respectively considered for employees and self-employed.

Income classes change over time, according to the law.

Only the spouse’s income, if present, is considered.

No specific funding for this article is reported.

Not applicable.

1. Version of Record published: December 31, 2020 (version 1)

© 2020, Maitino et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.