The Austrian labour market is characterised by a high degree of mobility between different labour market states, resulting in approximately one million entries and exits to and from unemployment annually. This represents more than twice the average number of unemployed persons registered within one year (Horvath et al., 2022). Additionally, approximately one-third of all employment episodes are terminated within their first year, excluding direct job-to-job transitions.

To account for these transitions shaping individual employment careers, our model distinguishes seven labour market states: 1.

In education/never in the labour force

After entering the labour market, individuals (aged 15 or older) can change between these labour market states (with the exception of retirement or permanent invalidity, which is assumed to be definitive). As individuals change between different labour market states, they accumulate insurance and contribution times, potentially making them eligible for early retirement. While some of the transitions between different states are based on simple decision models (such as retirement, which is automatically triggered when a person fulfils all the necessary requirements for retirement), most transitions are modelled based on a series of piecewise constant hazard rate regressions that consider personal characteristics as well as the duration of the current state (Figure 1).

For these transitions, the hazard rate takes the form: h ( t | X i ) = exp ( β 0 + β 1 q 1 + ⋯ + β k q k + β a i a g e i + β e d u e d u i + β h e a l t h h e a l t h i + β g e o g e o i ) (1) where q x denotes the number of quarters spent in the respective state; and a g e i , e d u i , g e o i and h e a l t h i denote a person’s age (grouped into six broad age categories), education (five categories), place of birth and health status (binary indicator). For women, hazard rate models additionally account for the age of their youngest child. Assuming (piecewise) constant hazard rates, the waiting time to a transition is exponentially distributed with an expected waiting time to the transition of 1 / h . In the simulation, random waiting times based on individual hazards are assigned to all individuals for each possible transition. For a given hazard h , a random waiting time w can be obtained by w = − ln ( u ) / h , where u is a uniformly distributed random variable. Following a competing risk approach, the event that happens first comes into effect and new waiting times are calculated. Whenever the hazard changes (e.g., when a person’s health status changes, or the duration of the state increases to the next quarter) a new waiting time is drawn.

In sum, our model implements 34 such hazard rate models (17 per gender), which are estimated using a combination of different administrative data sources: 1.

Longitudinal social security data containing information on individual labour market careers for all persons covered by mandatory social security. These data cover almost the entire population with respect to their employment, unemployment, parental leave, pension receipt or sick leave from 1972 onwards (Zweimüller et al., 2009).

Resulting estimations show that individual transition risks are highly correlated with personal and job characteristics (see Tables S.2 to S.5, Supplementary material). While the likelihood of transitioning from one state to another typically decreases with the duration of the state (as can be seen by the corresponding baseline hazards), the impact of the other characteristics captured in the relative risks varies depending on the respective states ⁶ . While higher education reduces the risk of moving from employment to unemployment (Table S.3), health limitations clearly increase unemployment risks. While the risk of becoming unemployed decreases with age (Table S.3), age also reduces the chances of moving from unemployment back to employment, and increases the risk of moving from unemployment to inactivity (Table S.2). For women, having younger children reduces transition rates from inactivity back to the labour market by more than half compared to having no children (see first column in Table S.2). However, with increasing age of the youngest child this negative effect declines, and vanishes as the youngest child reaches age 16.

By incorporating these estimations into our model, we can create heterogenous labour market careers that reflect the real-life heterogeneity of individuals’ employment biographies.

As noted earlier, the model can be aligned to labour force participation and unemployment rates stemming from cross-sectional imputation models. In addition, total unemployment can be set by scenario parameters. In this case, two of the transition models – namely the re-entry into the labour force and the re-entry into employment from unemployment – are being used indirectly for queuing people by their random waiting time, while the number of transitions is driven by the alignment targets.

To accurately capture the heterogeneity of individual labour market careers in our model, we need to account for the impact of personal and family characteristics, possible changes in institutional settings as well as path dependency in employment careers. Besides the obvious correlation between personal characteristics and labour force participation, our model needs to accurately capture the distribution of time spent in different labour market states, as these times are valued differently with respect to the qualification conditions for early retirement (see Section S1, Supplementary material). Simulating realistic individual labour market careers therefore requires an adequate representation of the most influential processes shaping individual employment prospects, such as migration background, education, fertility or health status. MicroDEMS models these processes, accounting for observable differences between population groups.

Figure 2 to Figure 4 show how strongly the extent of labour market participation depends on health, education and origin. While health limitations are generally associated with lower labour force participation, the gap increases at higher ages, especially for men. For women, on the other hand, low educational attainment and being born abroad have larger impacts on labour market participation.

In our model, education affects labour market attachment in various ways. Higher levels of education require longer periods of study, which negatively impacts labour market participation at younger ages. At the same time, higher education increases overall labour force participation at higher ages, both directly and via its positive impact on health. The direct effect implies that otherwise identical individuals (same age, sex, health status and duration of the current labour market status) have a higher probability of participating in the labour market (with a lower risk of moving out of the labour force or a higher probability of (re)entering the labour market) if they have attained a higher level of education. The indirect effect stems from the fact that the individuals with lower levels of education in our model face a greater risk of health impairment (reducing their labour market attachment) and permanent invalidity (implying permanent withdrawal from the labour market). As a result, there are large differences in labour market participation rates by education (Figure 3).

The size and composition of the Austrian population have historically been significantly influenced by migration flows. According to the assumptions underlying official population projections, migration will continue to contribute to overall population growth in the future, with an assumed net migration of around +31,000 per year from 2023 to 2080 ⁷ , resulting from 147,000 immigrants and 116,000 emigrants per year. Thus, the composition of the population will continue to change considerably in the future: while 81% of the Austrian population was born in Austria in 2022, this share will shrink to 72% in 2080. Differentiating the population by migration background, our simulations show that while 26% of the population had a migration background in 2022, this share will increase to 42% in 2080 (first generation: 27% (2022: 19%), second generation: 15% (2022: 7%); see Figure 5).

Figure 6 depicts how the population’s age and educational composition changes under the assumptions of our projection. In addition to the obvious change in the population’s age profile, the simulation clearly shows how the education composition changes, with the number of people with compulsory schooling or vocational training declining and the number of people with a high school diploma or a university degree strongly increasing.

Given these demographic changes, our model simulates the evolution of the size and structure of the labour force over time. Among the many factors impacting individual labour market biographies, changes in educational attainment and retirement rules are of particular significance, as they strongly affect labour force participation, especially at higher ages.

As more people attain higher levels of education and the retirement ages for women increase, labour force participation among the elderly is expected to increase considerably over time. As discussed in (Bittschi et al., 2024), the age-specific invalidity rates among women decrease at younger ages, while they increase at higher ages. This is because more women of higher ages will be in the labour force in the future, with a higher probability of permanent invalidity. On the other hand, the trend towards higher education levels should lead to fewer age-specific invalidity pension claims. Furthermore, the increase in the minimum retirement age for women will result in massive changes in the distribution of the different pension types for women and in women’s labour force participation rates at higher ages.

Figure 7 shows how the size and age structure of the workforce evolve over time, given the assumptions of our scenario. The workforce remains stable in size, but the number of older individuals in the workforce increases over time, surpassing the initial level by 5% in 2080.

The population structure with respect to peoples’ labour market status also changes substantially (Figure 8). The number of people outside of the labour market is projected to increase steeply ( + 1.26     million between 2022 and 2080), while the number of people in the workforce is projected to increase moderately (+112,000 up to 2080). The number of women in the labour force is projected to increase considerably (+119,000 between 2022 and 2080), while the number of men in the workforce is expected to remain stable (2022 to 2080: +3,000).

While the total size of the workforce is projected to remain fairly stable over time, large compositional changes can be observed with respect to migration background (Figure 9) and education (Figure 10). The number of people in the labour force with no migration background is projected to decline by 865,000 between 2022 and 2080, while the number of people with a first- or second-generation migrant background is expected to increase strongly (+478,000 for the first generation and +507,000 for the second generation). At the same time, the structure of education is projected to shift towards higher education levels (+437,000 with a university degree and +225,000 with a high school diploma as their highest level of education) and away from vocational education (−387,000 with an apprenticeship and −140,000 with a vocational school qualification as their highest level of education).

The total change in the labour force from 2022 to 2080 (+120,000 persons) resulting in the baseline scenario can be divided into various components (Figure 11). These components express how changes in the size and age structure of the population (demography), changes in the population composition (including migration and education, as well as the resulting change in health) and pension reforms (raising the minimum age as well as the required insurance and contribution periods) contribute to the change in the size of the future labour force.

The demographic effect describes the change in the size of the labour force assuming constant age-specific labour force participation rates over time. As anticipated, this demographic effect negatively impacts the size of the labour force across most age groups, and particularly in the 50 to 59 age group. On the other hand, there is an increase in the size of the labour force in the youngest age group resulting from demographic changes. Overall, the labour force is projected to decline by a total of 169,000 individuals due to demographic changes.

The composition effect shows how changes in the structure of the population, such as changes in educational attainment, cohort trends and migration history, affect the number of people in the labour force. This effect is obtained by multiplying the difference in age-specific labour force participation rates in 2022 and 2080 with the age group populations in 2080. For 2080, we utilise labour force participation rates from a scenario without pension reform to separate composition and retirement effects (see below for further details). In younger age groups, the education expansion has a negative impact on the number of people in the labour force, resulting in a negative composition effect. This is due to a lock-in effect of higher educational attainment. For individuals in their prime working years, the composition effect has a positive impact on labour force participation that increases with age. In absolute terms, the largest effects can be observed in the 60 to 64 age group. Overall, the composition effect will lead to the labour force increasing by about 139,000 people in 2080.

The pension reform effect shows how the size of the labour force changes not only due to the composition effect, but also due to the increase in the early and regular retirement ages. We isolate the retirement effect by simulating a separate scenario without pension reform (see scenario 2 below). The effect is obtained by multiplying the difference in the age-specific labour force participation rates that we project for 2080 in the baseline scenario and in scenario 2 with the age group populations in 2080. The retirement effect accounts for an increase of 150,000 people in the labour force by 2080.

While demographic dynamics will clearly alter the future size and composition of the Austrian population, changes in purely demographic ratios, such as the old-age dependency ratios, may not fully capture changes in the relationship between the number of economically active individuals and the number of individuals who are inactive (European Commission, 2021; Sanderson and Scherbov, 2015). Economic dependency ratios, on the other hand, consider to some extent the age-specific economic characteristics of the population, such as length of schooling, retirement age and participation behaviour (Loichinger et al., 2017).

We therefore contrast the demographic indicator of population change computed solely as a ratio of different age groups (the population aged 20 and younger or aged 65 and older divided by the population aged 20 to 64) with an economic dependency indicator expressed as the ratio between the inactive (not in the labour force, marginally employed, in education, receiving invalidity pension or retired) and active population (employed full-time or part-time or unemployed). This indicator captures changes in the age, education and health composition of the population, as well as changes in the retirement age and individual participation behaviour.

While the demographic dependency rate (population aged 20 and younger or aged 65 + / population aged 20–64) increases from 0.63 to 0.9 between 2022 and 2080, the economic dependency ratio increases less dramatically, from 0.93 to 1.11 (Figure 12).

Comparing the simulation results with alternative scenarios demonstrates the significant impact of individual assumptions on the projection of the economic dependency ratio, and how the ratio would change under alternative assumptions. Since our model simulates individual employment careers, it is also possible to assess the impact of a wide range of stylised policy measures.

Figure 13 shows the change in the economic dependency ratio for several alternative scenarios: •

Scenario 1 shows how the dependency ratio would change assuming constant age- and gender-specific labour force participation rates. This scenario represents a simple labour force projection while neglecting all compositional or institutional changes within the simulation period.

Although the economic dependency ratio increases in all projections, there are significant differences in scope. In the base projection, the ratio increases from 0.93 in 2022 to 1.11 in 2080. Scenarios 1 and 2 have much steeper increases, growing to 1.26 (scenario 1) and 1.16 (scenario 2). If the health gaps were closed to the extent proposed in scenario 3, the economic dependency ratio would only increase to 1.19 in 2080.