Modelling foreign labour inflows using a dynamic microsimulation model of an ageing country - Slovakia

Population ageing is expected to set up unprecedented labour movements. While population ageing is observable across all European countries, patterns in attracting and accepting foreign labour vary substantially between European countries. Here we adopt a perspective of a small, rapidly ageing EU country, which is not the usual target country for immigrants. Slovakia has recently experienced exponential growth in the number of foreign workers. The main driver of this inflow was the lack of domestic labour able to meet the steadily growing labour demand in a situation of declining overall labour supply. A similar situation can also be found in other new EU member states, such as Slovenia, Poland, the Czech Republic, but even more remote countries such as Malta.¹

SLAMM² is a combined, micro-macro model, developed to predict the expected dynamics of labour demand in Slovakia. Its supply side forms a dynamic microsimulation model designed to quantify the demand for labour arising because of the need for replacing retiring workers, together with the need for foreign-born workers. The purpose of this text is to introduce the structure of SLAMM, with a brief example of its potential outputs focused at projecting immigration inflows in variant scenarios. The focus is at describing the structure and functions of SLAMM clearly and concisely to stimulate methodological discussion.

In the remaining part of the introductory section, we outline our motivation with the most relevant concepts identified by the literature. The second section offers a detailed description of SLAMM, its functions and connection to a macroeconomic model. In the third section, we present SLAMM projections of the dependency ratio, the replacement demand for labour and inflowing migration in alternative policy-relevant scenarios. We conclude in the final, fourth section.

We aim to predict replacement demand for labour by capturing the dynamics driven by ageing. For this purpose, we designed⁷ a microsimulation model SLAMM.⁸ In line with the typology presented in O’Donoghue (2014), SLAMM applies the cross-sectional dynamic-ageing, as individuals in the model age in discrete, annual, time-periods.

The version of SLAMM presented here runs on Slovak data from the European Union Labour Force Survey (EU LFS), which draws the most detailed picture of the population during the inter-Census period in Slovakia. Here we are presenting figures for the nowcasting variant⁹ of SLAMM - using the microdata file of Slovak EU LFS data for 2011. We use sampling weights to inflate the file to the size of the total population of Slovakia (apx. 5.4 million).

SLAMM reproduces the assumptions of a demographic prediction¹⁰ and additionally simulates lifetime decisions¹¹ on economic activity and employment at an individual level. It simulates the development of eight main attributes of individuals using a set of deterministic, as well as stochastic, functions using the dynamic information on individuals’ age combined with the information observed (or simulated) during the previous period. The eight attributes listed in Table 1 present variables with values observable at the level of all individuals in the simulated population, present the minimal required information necessary to run the model. In principle, this information can also be acquired from other statistical sampling surveys, such as the EU SILC,¹² or from the CENSUS. The list of variables with possible values used in the model has been optimised for the purpose of labour supply modelling, with respect to the available data.¹³ Since all these variables refer to attributes of individuals, no other type of entity needs to be introduced in the model,¹⁴ the only entities recognised in SLAMM are individuals.

After the individual-level information on the eight variables is imported from the initial-period database, values of the variables are subject to functions defined for the simulation period. Alike in other labour supply microsimulation models,¹⁵ functions are organised in relatively independent and subsequently simulated modules:

• Demographic module – DEMO

• A module on attaining education – EDU

• A module on decisions related to economic activity – EA

• A module on matching individuals with available jobs - EMPL

Module DEMO simulates demographic processes of fertility, mortality and ageing. Migration is assumed to be balanced at zero in this module. Population (with its age structure) is, therefore, simulated based on the in-country potential. Both fertility and mortality are implemented using exogenous age-specific probabilities,¹⁶ allocating fertility and mortality randomly in each period within a one year age cohort. For female individuals assigned to be giving birth in the particular simulation period, a new individual is generated.¹⁷ Individuals assigned to die in the specific simulation period are dropped from the population.¹⁸

Although new immigrants are not generated in this DEMO module, one of their attributes is simulated here; namely the length of stay in the country during the current period. This attribute was not present in the initially imported data; instead, we impute this information in order to match the cross-sectional picture available from the administrative records on the length of stay of employed immigrants.¹⁹ The length of stay is, thus, imputed for immigrants in the base-population imported for the initial period. For all the new immigrants arriving in the country as a result of the simulation, the length of stay ages in discrete, one- year, periods. Based on the length of stay, we can assign an individual-specific time to leave the country. After the time to leave expires, the individual leaves the population, which leads to a further increase in replacement demand. This allows us to draw scenarios of the alternative length of immigrants’ stay in the country or to calibrate the model based on this parameter.

Module EDU simulates educational attainment of individuals. Level and field of the highest education attained are randomly assigned while keeping the proportions of educational attainment of the most recently observed cohort with finalised initial education (age cohort 30-35 years old in the base year). The approach taken in this module is rather simplistic, assuring the possibility to draw scenarios of alternative assumptions on the development of the educational structure of the population.

Module EA simulates decisions related to the economic activity of individuals. The approach applied in simulating economic activity is inverse to the usually used in microsimulation models. The decision to become economically active is often simulated as a function of potential working income. SLAMM does not work with information on individuals´ income, because this is not present in the EU LFS data used as the source database. In each of the simulation periods, SLAMM assigns statuses of economic inactivity based on: i. the information on economic status during the previous period; ii. the development of the age structure and; iii. alignments to gender and age-specific proportions observed in the most recent CENSUS.²⁰

The disabled, retired, early-retired and other inactive are identified in each of the simulation periods following these steps. SLAMM first calculates a probability score of being in each of the inactivity statuses for every individual in the population. Individuals in the particular inactivity status during the previous period²¹ are assigned an extremely high value of the status-score. Individuals are re-allocated in each of the inactivity statuses based on their score while keeping²² the gender and age-specific proportions of each of the inactivity statuses observed in the most recent CENSUS. The proportions are defined in age-relative to the current retirement age, while the current legal retirement age is a globally applied parameter, varying in time, by gender and potential alternative scenarios.

Analogously to the procedure applied to these inactivity statuses, also the working retired are identified. One additional restriction is applied on potential transitions between statuses, with retiring being an ultimate decision, meaning once retired individuals are not able to become economically active in any of the subsequent periods. Disability and other inactivity allow transitions in and out of these statuses.

Students are assigned deterministically by adopting an additional assumption of educational level-specific graduation age as a threshold for switching from a student to economically active.

After these functions are evaluated, individuals not assigned to any of the inactivity statuses are assigned to be economically active; thus, in the labour supply. Individuals assigned as working retired are added to the economically active population.

Scheme 1

Download asset Open asset

Module EMPL allocates those who are assigned to be economically active to employment or unemployment. For individuals in employment, it additionally decides about the sector (branch) of economic activity and the occupation of their employment. SLAMM aligns²³ the total employment in labour market segments (the combination of the sector of economic activity and occupation) based on an exogenous macroeconomic prediction. In the final step, for labour market segments with a lack of employed and potentially employable from “domestic” labour supply, foreign workers are imported.

Migration, initially skipped in the DEMO module, is thus added into the model inthe very last step. Immigration into the country, implemented in this way, is driven by the needs of the labour market. After 2013, Slovakia was experiencing an exponential increase in the total number of workers born outside the country. Based on administrative data on work permits²⁴ (and obligatory reporting of employment of EU citizens) the total stock of foreign-born workers employed in Slovakia went up from 12.3 thousand in 2013 to 70.2 thousand in 2019. The SLAMM model, running on the EU LFS data for 2011 captured the increasing trend, not only in terms of the total number of employed persons born out of the country but also in the structure; such as the level of education.

2.1. Macroeconomic assumptions

SLAMM is a microsimulation model combined with a macroeconomic model,²⁵ with an integrated equilibrium-seeking element.²⁶ Equilibrium is sought within a classical input-output macroeconomic framework (Leontief, 1966; Miller and Blair, 2009), where an output (product) of one economic sector is traceable to be inputted (consumed) in another economic sector (or exported). Expected production is modelled based on existing, in-country, as well as export-related transactions. Employment in economic sectors is predicted based on the expected level of production. The macro prediction brings a quantification of the number of persons, which need to be employed to assure expected production; it quantifies the expansion demand for labour but tells nothing about the replacement demand for labour.

In terms of total employment, the macroeconomic part of SLAMM predicts annual increases between 1.5 and 1 percent until 2025. After 2025 we adopted a simplifying assumption of zero growth in the total employment (Figure 1). In terms of structural change, after 2025, all labour market segments expect a constant level of employment, while before 2025, particular segments change depending on the macroeconomic prediction.²⁷²⁸

Figure 1

Download asset Open asset

Multiple channels connect the microsimulation and macroeconomic parts of the model.²⁸ First, the prediction of the number of the economically active population enters the macro part of the model as the total supply of labour. In the EA module, the group of other inactive can potentially react on a substantial increase in average wage predicted by the macroeconomic model.²⁹ The macro prediction of employment in labour market segments³⁰ presents the main channel of information exchange between the two parts of the model. For each of the segments, the total number of employed presents the ultimate number determining: i. the number of individuals allocated from the group of economically active to employed; ii. the number of workers imported labour from abroad in the case of an insufficient number of suitable workers available in the economically active population simulated for the current period. Since this, channel dominantly determines the final output of the SLAMM projections; we consider the micro-macro connection to be top-down with a behavioural approach.³¹

Besides its predictive potential, SLAMM also presents a handy tool for assessing alternative developments. As work-related immigration remains subject to various regulations, it can be influenced or even shaped by multiple policies. Figure 2 displays examples of scenarios, of labour supply reactions to a change in the retirement age, length of stay of immigrants and age structure of inflowing immigrants.

Figure 2

Download asset Open asset

The baseline scenario captures the current legislation on retirement age as well as the empirically observed length of stay and the age of inflowing immigrants. The dynamic retirement age scenario manipulates the retirement age by allowing its growth at the same pace as the life expectancy. The short-stay immigration scenario assumes the working immigrants to stay in the country for a shorter time (10 percent decline in the length of stay implemented across the distribution). Younger immigration scenario assumes all inflowing working immigrants to be between 20 and 29 years old, and the Older immigration scenario assumes all inflowing working immigrants to be between 36 and 44 years old.

First, we explore the simulated development of the dependency ratio indicator. Its values are going to show the worst development in the case of the dynamic retirement age scenario, with more employment being covered by the domestic (older) population attracting less (younger) immigrant population.³² The most favourable dependency ratio is expected in the case of the older immigration scenario because households of older immigrants include less children.

Another indicator of interest is the country´s demand for foreign-born workers (Figure 3). The observed length of stay of the current immigrants is relatively short; this is driving the demand for workers born outside the country. Further shortening of the length of stay would substantially increase the need to attract the annual flow of additional workers. Age composition of inflowing immigrants does not make such a substantial difference to the baseline scenario. Dynamizing the retirement age would result in more workers from the older, domestic labour force and thus lower need for importing foreign-born labour.

Figure 3

Download asset Open asset

Finally, replacement demand for labour, quantified in the number of persons is the lowest in the case of the dynamic retirement scenario and the highest in the case of the short-stay immigration scenario. Again, we may observe that, in comparison to the length of stay of immigrants, the age structure of inflowing immigrants plays a secondary role in shaping future replacement demand for labour.

Figure 4 also displays the growing discrepancy between the size of the graduating cohort and the demand for labour arising because of the need for replacement of workers leaving for retirement or other forms of economic inactivity.

Figure 4

Download asset Open asset

Advancements in microsimulation modelling open new potential for its application in various areas of economic modelling. Although not in the main focus, manpower requirement modelling has a long tradition of combining macroeconomic models with more detailed statistical models designed to predict the development of the labour supply. In line with this tradition, we have developed a model combining a macro-level prediction of employment trends, with micro-level information on the population of Slovakia. SLAMM is a dynamic microsimulation model designed to predict the development of the supply side of the Slovak labour market. After importing individual-level information about the Slovak population, it simulates the demographic processes, educational attainment, decisions related to economic activity and entering employment.

Since SLAMM was built in line with the tradition of manpower requirement modelling, one of the central indicators of interest is the replacement demand for labour, which results from individuals leaving the labour market for retirement or other forms of economic inactivity. SLAMM quantifies replacement demand for labour as the number of persons in need of replacement on an annual basis. In comparison to methodological approaches traditionally applied in quantifying replacement demand, we can see at least three advantages of a dynamic microsimulation model.

First, replacement demand for labour is traditionally quantified using a cohort-component³³ type of model (Willems and de Grip, 1993) (CEDEFOP, 2012). Here changes in the aggregate number of employed within age cohorts are tracked in time to quantify the probability of exiting employment, therefore the need for replacement. In a dynamic microsimulation model, such as in the case of SLAMM, individual-level probabilities of particular types of exits are implemented in every period of the simulation. Among the types of exits (inactivity statuses), we identify: the retired, early retired,³⁴ disabled, and other types of inactivity.³⁵ Detailed disaggregation of the types of exits allows capturing different dynamics related to various sub-groups of interest; it thus allows producing more detailed predictions, enabling straightforward interpretation of the expected trends.

Second, comparing the aggregate size of a cohort in time does not allow combining information from various data sources³⁶; a microsimulation model is more flexible in this respect. In our case, SLAMM runs on the EU LFS data, as it presents the most recent picture of the whole population of interest. In the same model, while working with sub-populations of particular interest (e.g. disabled or the working retired), age-specific proportions and probabilities are extracted out of a less recent, but more detailed CENSUS. Two data-sources of the same population are thus combined to yield information of higher quality.

Finally, because of the microsimulation design, SLAMM offers a wide variety of scenario-designing options. We present some of its potential also here when expecting alternative age-structure of inflowing immigrants, or the length of their stay in the country. At the same time, these types of scenarios were compared to a completely different type of scenario, adopting an assumption related to the change in the retirement age.

The potential in drawing alternative scenarios becomes especially interesting from the policy-making perspective. Due to scenario analysis, we can provide an assessment of the contribution of changes in the age-structure of inflowing immigrants relative to the length of their stay. From the perspective of assuring available labour force through immigration policy, prolonging the length of stay should be prioritised over adjusting the age structure of immigrants. The output of the model allows considering further implications, for example, for social security or pension systems. For the sake of brevity, we do not explore these alleys here. Instead, we try to document the connection of replacement demand for labour and the number of inflowing foreign workers. Such connection can be observed empirically on aggregate figures (Figure 5) or assumed while designing a microsimulation model. Here we present the experience of Slovakia, which is a rapidly ageing country, only marginally affected by the recent wave of immigration of non-EU citizens. Although Slovakia is a small country, its experience can be to some extent generalised also to others, especially central and eastern European countries.

Figure 5

Download asset Open asset

1. 1
   EU Labor Markets After Post-Enlargement Migration

   1. T Baas
   2. H Brucker
   3. A Hauptmann

   (2010)

   Labor Mobility in the Enlarged EU: Who Wins, Who Loses?, EU Labor Markets After Post-Enlargement Migration, Springer.

   • Google Scholar
2. 2
   The Demographic Dividend, a New Perspective on the Economic Consequences of Population Change. Population Matters. Monograph MR 1274

   1. DE Bloom
   2. D Canning
   3. J Sevilla

   (2003)

   Santa Monica, CA: RAND.

   • Google Scholar
3. 3
   LIAM2 User Guide - Release 0.11.0, Federal Planning Bureau

   1. G Bryon
   2. G Dekkers
   3. G De Menten

   (2015)

   http://liam2.plan.be/download/LIAM2UserGuide-0.11.0.pdf, November 24, 2015.

   • Google Scholar
4. 4
   Skills supply and demand in Europe (Methodological framework). Research paper No. 25

   1. CEDEFOP

   (2012)

   Luxembourg: Publication Office of the European Union. ISBN 978-92-896-1112-1.

   • Google Scholar
5. 5
   Skills forecast: trends and challenges to 2030

   1. Cedefop, Eurofound

   (2018)

   Luxembourg: Publications Office. Cedefop reference series; No 108.

   • Google Scholar
6. 6
   LIAM2: a new open source development tool for Discrete-Time dynamic Microsimulation models

   1. G de Menten
   2. G Dekkers
   3. G Bryon
   4. P Liégeois
   5. C O’Donoghue

   (2014)

   Journal of Artificial Societies and Social Simulation 17:9.

   https://doi.org/10.18564/jasss.2574

   • Google Scholar
7. 7
   European agenda on migration, COM 240 final, Brussels

   1. European Commission

   (2015)

   https://ec.europa.eu/transparency/regdoc/rep/1/2015/EN/1-2015-240-EN-F1-1.PDF, May 13, 2015.

   • Google Scholar
8. 8
   The effect of population growth on economic growth: a meta-regression analysis of the Macroeconomic literature

   1. DD Headey
   2. A Hodge

   (2009)

   Population and Development Review 35:221–248.

   https://doi.org/10.1111/j.1728-4457.2009.00274.x

   • Google Scholar
9. 9
   Forecasting the labour market by occupation and education

   (1994)

   editor, Forecasting the labour market by occupation and education, Kluwer Academic Publishers, Dordrecht.

   • Google Scholar
10. 10
    Towards a transparent labour market for educational decisions

    1. H Heijke
    2. L Borghans

    (editors) (1998)

    Aldershot: Ashgate.

    • Google Scholar
11. 11
    Projecting the occupational structure of employment in OECD countries

    1. G Hughes

    (1993)

    Paris: OECD, (OECD Labour Market and Social Policy Occasional Papers No 10).

    • Google Scholar
12. 12
    International standard classification of occupations (ISCO-08), International labour organisation, Geneva, ISBN 978-92-2-125953-4

    1. ILO

    (2012)
13. 13
    Forecasting labour markets in OECD countries; Measuring and tackling mismatches

    1. L Lassnigg

    (2002)

    Projections of qualifications and occupations in Austria – short-termism, macro perspective and emphasis on the supply side, Forecasting labour markets in OECD countries; Measuring and tackling mismatches, Cheltenham, Edward Elgar.

    • Google Scholar
14. 14
    LABORsim: an Agent-Based Microsimulation of labour supply – an application to Italy

    1. R Leombruni
    2. M Richiardi

    (2006)

    Computational Economics 27:63–88.

    https://doi.org/10.1007/s10614-005-9016-0

    • Google Scholar
15. 15
    Input-Output Economics, Oxford

    1. W Leontief

    (1966)

    New York: University Press.

    • Google Scholar
16. 16
    Input-output analysis: foundations and extensions

    1. RE Miller
    2. PD Blair

    (2009)

    Cambridge university press.

    • Google Scholar
17. 17
    Forecasting labour markets in OECD countries; Measuring and tackling mismatches

    1. M Neugart
    2. K Schomann

    (editors) (2002)

    Cheltenham: Edward Elgar.

    • Google Scholar
18. 18
    Handbook of Microsimulation Modelling (Contributions to Economic Analysis)

    (2014)

    editor, Handbook of Microsimulation Modelling (Contributions to Economic Analysis), 293, Emerald Group Publishing Limited, ISBN: 978-1-78350-569-2.

    • Google Scholar
19. 19
    Employment outlook

    1. OECD

    (1994)

    Paris: OECD.

    • Google Scholar
20. 20
    Forecasting educational needs for economic and social development

    1. HS Parnes

    (1962)

    Paris: OECD.

    • Google Scholar
21. 21
    Integrated regional Econometric input output model of Slovakia

    1. M Radvanský
    2. I Lichner
    3. J Zachar

    (2019)

    EcoMod2019 : International Conference on Economic Modeling - Ponta Delgada.

    • Google Scholar
22. 22
    Integrated regional econometric+input-output modeling: Issues and opportunities," Papers in Regional Science, Springer;

    1. SJ Rey

    (2000)

    Regional Science Association International 79:271–292.

    • Google Scholar
23. 23
    (United Nations Department of Economic and Social Affairs, Population Division). World Population Prospects: The 2017 revision, key findings and advance tables

    1. UNDESA

    (2017)

    New York: UNDESA.

    • Google Scholar
24. 24
    From demographic Dividend to demographic burden

    1. N Van Der Gaag
    2. J Beer

    (2015)

    Tijdschr Econ Soc Geogr 106:94–109.

    • Google Scholar
25. 25
    Forecasting replacement demand by occupation and education

    1. EJTA Willems
    2. A de Grip

    (1993)

    International Journal of Forecasting 9:173.

    https://doi.org/10.1016/0169-2070(93)90003-6

    • Google Scholar

Author details

1. Miroslav Štefánik

   Institute of Economic Research Slovak Academy of Sciences, Bratislava, Slovakia

   For correspondence

   miroslav.stefanik@savba.sk

   Competing interests

   No competing interests reported

   "This ORCID iD identifies the author of this article:" 0000-0002-9993-7885

2. Tomáš Miklošovič

   Institute of Economic Research Slovak Academy of Sciences, Bratislava, Slovakia

   Competing interests

   No competing interests reported

   "This ORCID iD identifies the author of this article:" 0000-0003-3104-0905