2023 Skills Forecast

Technical Report
March 2023

Unedited Proof Copy

Note: This is not an official Cedefop publication. For any further
information please contact Cedefop Expert Ilias Livanos who has been
responsible for Cedefop Skills Forecast.
([email protected])
Foreword

This document is the Technical Report accompanying the 2023 Skills

Forecast. It outlines the key assumptions adopted and data utilised in
preparing the 2023 release. Full details on the methodologial framework
applied and the various modules developed for producing the Skills
Forecast can be found in the dedicated Methodological Report.

2
Table of contents
TABLE OF CONTENTS ...................................................................................... 3
LIST OF TABLES, FIGURES AND BOXES........................................................ 4
Tables ...................................................................................................... 4
Figures ..................................................................................................... 4
CHAPTER 1. INTRODUCTION ........................................................................ 5
CHAPTER 2. DEVELOPMENT AND REFINEMENT OF THE
METHODOLOGY ......................................................................................... 7
2.1. Introduction .................................................................................... 7
2.2. Macroeconomic modelling.............................................................. 7
2.2.1. Population projections .............................................................. 7
2.2.2. GDP projections ....................................................................... 9
2.2.3. Assumptions underpinning the European Green Deal ............ 10
2.3. Expansion demands for occupations and qualifications (EDMOD and
QUALMOD) ................................................................................................... 19
2.4. Replacement demand (RDMOD) ................................................. 26
2.4.1. Adjustments to the RDMOD.................................................... 26
2.4.2. Covid-19 impact...................................................................... 30
2.5. Modelling qualification supply (QMOD) ........................................ 30
CHAPTER 3. DATA UPDATE AND PREPARATION .................................... 33
3.1. Introduction .................................................................................. 33
3.2. E3ME update ............................................................................... 33
3.3. LFS update of Modules 2-5 .......................................................... 37
LIST OF ABBREVIATIONS .............................................................................. 39
REFERENCES ……………………………………………………… ............. ………40

3
List of tables, figures and boxes

Tables
Table 2.1 Assumptions used in the E3ME model ..................................... 14
Table 3.1 Description of updated variables ............................................... 34
Table 2.2 Main data inputs from EU-LFS microdata ................. 38

Figures
Figure 1.1 Cedefop Skills forecast modelling forecast ................................ 6
Figure 2.1 EU-27 Population projections, thousands .................................. 8
Figure 2.2 EU-27 short-term population scenarios, thousands ................... 9
Figure 2.3 Ameco’s 2022 Spring Economic Projections ........................... 10
Figure 2.4 EU27 final energy demand in 2030, million tonnes of oil-
equivalent (mtoe) ............................................................................ 16
Figure 2.5 EU27 power generation allocation, 2030 ................................. 17
Figure 2.6 Main greening assumptions and how they are reflected in E3ME
18
Figure 2.7 Comparisons with EU-LFS estimates ...................................... 22
Figure 2.8 Comparisons with EU-LFS estimates ...................................... 23
Figure 2.9 Adjustments to occupational employment shares for 2035 ...... 24
Figure 2.10 Implementation of automation .................................. 29

4
 CHAPTER 1. Introduction

Cedefop’s skill supply and demand projections provide comprehensive information

about the current structure, and future trends, in the labour market. This technical
report accompanies the release of the 2023 update to the projections.
The methodology 1 uses a modular approach, with the following main
elements:
(a) the demand side (skill needs), focusing on employment (jobs);
(b) the supply side, focusing on available skills in the labour market, such as the
number of people economically active and the qualifications they hold;
(c) imbalances, comparing the results from the demand and supply side
modules.
All modules adopt common data, methods and models for all countries (EU +
the Republic of North Macedonia, Iceland, Norway, Switzerland and Turkey) to
produce a comprehensive and consistent set of skill projections. The database
draws primarily on Eurostat sources, including demographic data, both historical
and projections, national accounts (NA) and the labour force survey (LFS).
Individual country experts are involved in peer review and results validation.
However, the use of common models and assumptions does not always allow
incorporation of local data and factors that may affect skill supply and demand.
Therefore, other country-specific information might be needed to complement the
results.
Figure 1.1 presents the Cedefop Skills Forecast methodological framework.
Module 1 of the methodology contains the Cambridge Econometrics’ E3ME
(macroeconomic) model. This is used to form projections of labour demand
(employment) at sectoral level and labour supply (the economically active labour
force) by demographic group.
Modules 2 and 3 cover employment levels and expansion demand (i.e. new
jobs created over time) for occupations and qualifications. Module 4 covers
replacement demand (i.e. job openings resulting from people exiting the labour
market or changing jobs). Modules 5 focus on the supply of skills as measured by
the highest qualification held, analysing stocks and flows. Module 6 reconciles skill
demand and supply.

1 Please visit the Skills forecast dedicated page for links to the Methodological Report:
https://www.cedefop.europa.eu/en/projects/skills-forecast

5
Figure 1.1 Cedefop Skills forecast modelling forecast

6
 CHAPTER 2. Development and refinement of
the methodology

2.1. Introduction
This chapter describes the development and refinement of the Cedefop Skills
Forecast methodological framework. Section 3.2 describes the macroeconomic
modelling and its underlying assumptions such as greening of the EU economy.
Section 3.3 describes the changes to the replacement demand methodology. The
final section discusses the changes to the other modules of the Cedefop Skills
Forecast.

2.2. Macroeconomic modelling

This section outlines the main macroeconomic assumptions underlying the
Cedefop projections. They cover:
(a) Population projections,
(b) GDP projections, both short and long-term,
(c) Incorporation of part of the EU Green Deal, in particular the new emission
reduction target (55% GHG reduction compared to 1990 levels by 2030) and
improved resource efficiency by increasing recycling rates.

The forecast horizon has also been expanded from 2030 to 2035. These
assumptions are discussed in further detail below.

2.2.1. Population projections

The latest population projections from Eurostat EUROPOP2019 have been used
in this forecasting exercise. Figure 2.1 shows how, compared to the projections
used in the previous forecasting exercise, EUROPOP2019 projections are slightly
more pessimistic in terms of population growth. Indeed, by 2030 EUROPOP2019
forecast that total population in EU-27 would be 9% lower compared to the
previous 2030 outcome (just over 4 million fewer people). In terms of working-age
population (15-64), EUROPOP2019 forecast a 0.4% increase in 2030 compared
to the previous population projections.

7
Figure 2.1 EU-27 Population projections, thousands

Source(s): E3ME and Eurostat [proj_esms]

In late 2022 Eurostat published a set of sensitivities set to capture the impact of
recent developments, such as the war in Ukraine, on EU-27 population (Figure
2.2). Interestingly, Eurostat’s short-term population projection baseline is even
more pessimistic in terms of population growth compared to Europop2019. Both
sensitivities, assume an increased number of refugees, leading to a substantial
increase in population over 2022-2025. After population is expected to decrease,
with one sensitivity reaching Europop2019 population levels by 2030 and the other
one with slightly higher levels. Over the longer term, neither sensitivities are
expected to achieve the same population levels as projected under Europop2016.
The Eurostat sensitivities are in line with discussion and reflections captured at the
Skills Workshop of 2022; such as the expectation over the short term of an
increase in population as a result of the war in Ukraine, with most refugees
returning to Ukraine once the conflict has died down. Furthermore, the majority of
the refugees from Ukraine are women and children, so it is expected that their
participation to the labour market would be limited.

8
Figure 2.2 EU-27 short-term population scenarios, thousands

Source(s): E3ME and Eurostat [proj_esms, proj_stp22]

2.2.2. GDP projections

The short-term GDP projections are in line with Ameco’s Spring 2022 Economic
Forecast 2, while the long-term projections are in line with the GDP projections used
in the Europop 2019 population projections, as detailed in the 2021 Ageing
Report 3. Since the 2021 Ageing Report does not contain assumptions about the
European Green Deal 4, the long-term GDP projections have been adjusted to
reflect the implementation of parts of the Green Deal, based on information from
the European Commission Fit-For-55 Impact Assessment 5. These adjustments are
discussed in more detail in Section 2.2.3. The decision to include the Green Deal
reflects recent response to the difficulties in global energy market and disruptions
caused by Russia's invasion of Ukraine. This has led the EU to accelerate the
plans for the green energy transition and thus decrease the region’s energy
dependence on fossil fuels and unreliable supply sources.

(2) Spring 2022 Economic Forecast: Russian invasion tests EU economic resilience
(europa.eu).
(3) The 2021 Ageing Report: Economic and Budgetary Projections for the EU Member
States (2019-2070) (europa.eu).
(4) https://ec.europa.eu/info/strategy/priorities-2019-2024/european-green-deal_en
(5) EU economy and society to meet climate ambitions (europa.eu).

9
 The short-term GDP projections from Ameco’s 2022 Spring Economic
forecast include assumptions on continued supply bottlenecks and increased
energy and commodity prices as a result of the war in Ukraine. This means that
while EU-27 GDP growth in 2021 was estimated at 5.4% in 2021, it is expected to
slow down to 2.7% in 2022 and 2.3% in 2023 (see Figure 2.3).

Figure 2.3 Ameco’s 2022 Spring Economic Projections

Source(s): Ameco 2022 Spring Economic Forecast

The long-term GDP projections from the 2021 Ageing report mainly reflect
assumptions about future labour working hours and labour productivity, as well as
budgetary issues arising from changes in population structure (e.g. pension
payments). The labour market assumptions (working hours as measure of how
much people work and; labour productivity how well people work) are used in the
2021 Ageing report to estimate labour’s contribution to GDP growth, compared to
capital.

2.2.3. Assumptions underpinning the European Green Deal

This section details the assumptions used in integrating part of the European
Commission Green Deal policies into the Cedefop Skills Forecast. The European
Green Deal (EGD) covers a significant portfolio of policy areas, such as agriculture,
energy and climate, environment and research and innovation, to name a few. As
there is limited quantifiable information to implement changes in all these policy

10
areas and the E3ME model captures some of these policy areas more explicitly
than others, only some of the EGD measures will be explicitly integrated in the
Cedefop Skills Forecast. For example, the Green Deal envisages changes to the
Common Agricultural Policy with policies looking at organic farming, animal
welfare, and biodiversity, and these types of policies are beyond the scope of the
E3ME model. It also looks at Horizon Europe, which can cover a broad range of
research and thus very difficult to make assumptions on and capture their
economic impact in the E3ME model.

What will be explicitly integrated in the Cedefop Skills Forecast are the Fit-For-55
trends, for which detailed information exists in the “Commission Staff Working
Document Impact Assessment Accompanying the Document Communication from
the Commission to the European Parliament, the Council, the European Economic
and Social Committee and the Committee of the Regions: Stepping up Europe’s
2030 climate ambition. Investing in a climate-neutral future for the benefit of our
people” 6 (hereafter Impact Assessment). The Fit-For-55 trends cover
developments in electricity supply, energy efficiency and other sectors (e.g.
transport). Generally, the trends show an increased use of renewables, increased
electrification and higher energy efficiency. In terms of sectoral implications, these
trends would translate into lower activity in some energy-related sectors and
increased economic activity in sectors manufacturing the renewable and energy-
efficient equipment. Explicit assumptions on circular economy (e.g. recycling
rates) have also been included in the Cedefop Skills Forecast (see European
Green Deal skills forecast scenario 7).
These assumptions and methodology are specific to the E3ME model. The
E3ME modelling framework which was used in the 2020 Cedefop Skills Forecast
reflect the “at least 40% cuts in greenhouse gas emissions (from 1990 levels)”, i.e.
the previous ambition from the 2030 climate and energy framework 8 before the
EGD was announced (hereafter called the baseline). The assumptions discussed
in this section specifically outline the move from 40% to 55% emission reduction
by 2030 in the EU MS. The 2020 E3ME baseline discussed above is consistent
with the baseline used in the Impact Assessment to assess the impacts of the
transition from a 40% cut in greenhouse gas emissions to a 55% cut.

6 https://eur-lex.europa.eu/resource.html?uri=cellar:749e04bb-f8c5-11ea-991b-
01aa75ed71a1.0001.02/DOC_2&format=PDF
7 https://www.cedefop.europa.eu/en/publications/4206
8 https://climate.ec.europa.eu/eu-action/climate-strategies-targets/2030-climate-
energy-framework_en

11
2.2.3.1. Available information
In terms of energy trends, we have detailed, publicly available, information on
energy balances for both the 40% and 55% emission reduction pathways. The
40% emissions reduction pathway is referred to as the EU Reference Scenario
2020 9. The EU Reference Scenario 2020 includes EU level policies adopted by
mid-December 2019 with respect to energy, transport and climate change. These
include the “Clean Energy for All Europeans” package, the revised EU Emissions
Trading System (EU ETS) Directive, transport policies such as the CO2 standards
for vehicles, the Directive on alternative fuels infrastructure, the Clean Vehicles
Directive, for example. The Reference Scenario 2020 also includes national MS
policies that were adopted as part of the National Energy and Climate Plans
(NECPs). These plans include coal phase-out and nuclear-related policies(EU
Reference Scenario 2020 10).
Together with the EU Reference Scenario 2020, information on three policy
scenarios is publicly available. The level of detail for the energy trends of the three
policy scenarios is the same as the EU Reference Scenario 2020. The three policy
scenarios build on the EU Reference Scenario 2020 to look at the impact of the
European Green Deal policy package. These policy scenarios are:
1. MIX – the policy scenario relies on both ETS (carbon price signal) extension
to road transport and buildings and strong intensification of energy and
transport policies to achieve emission reduction target.
2. REG – the policy scenario relies on very strong intensification of energy
and transport policies/regulations in the absence of carbon pricing in road
transport and buildings to achieve the emission target.
3. MIX-CP – the policy scenario represents a more carbon price-driven policy
mix which includes a limited revision of the Energy Efficiency Directive and
Renewables Directive, with more focus on the carbon price signals.

It is important to note that all policy scenarios have the same emission
reduction target of 55% relative to 1990 levels. For the purpose of this forecasting
exercise, the MIX policy scenario was chosen among the three scenarios
mentioned above since it provides a more balanced view of both regulation and
carbon pricing.

9 https://energy.ec.europa.eu/data-and-analysis/energy-modelling/eu-reference-
scenario-2020_en
10 https://op.europa.eu/en/publication-detail/-/publication/96c2ca82-e85e-11eb-93a8-
01aa75ed71a1/language-en/format-PDF/source-219903975

12
 Another relevant piece of information that will be used in this forecast is the
Climate Target Plan Impact Assessment 11. This document includes information on
the GDP and wider competitiveness impacts of implementing the policy scenarios
starting from the EU Reference Scenario 2020 presented above. In particular,
information on sectoral impacts, such as changes to output, changes to investment
and consumer expenditure have been taken into account based on information
from the Impact Assessment. The exact treatment of inputs is described in Table
2.1.
In summary, the following information form the MIX policy scenario is included
in the Cedefop Skills Forecast:
(a) Energy balances by sector, energy carrier and Member State
(b) Power generation mix by technology by Member State
(c) Emission results by Member State
(d) Carbon pricing levels and coverage
(e) Additional EU investment requirements (e.g. power plants, energy efficiency)
(f) GDP impacts. Changes to consumer expenditure, investment and trade.

The energy balances, power generation, and emission results are available
every 5-years to 2030. The economic information is mostly available for 2030.

2.2.3.2. Modelling assumptions and implementation

The implementation of the EGD assumptions is based on the information described
above and is implemented with respect to the Reference Scenario baseline in order
to maintain consistency with the Impact Assessment. Table 2.1 outlines how the
information available and summarised in Section 2.2.3.1 translates to E3ME
modelling inputs.

11 https://eur-lex.europa.eu/legal-
content/EN/TXT/HTML/?uri=CELEX:52020SC0176&from=EN

13
Table 2.1 Assumptions used in the E3ME model

Information taken What the information Where it will impact in the

from the EC provides? model?
Reference Scenario
Final energy demand Use of different fuel (e.g. Changes of energy trends.
coal, gas) which will influence Changes in production and/or
production of these fuels imports of fuel.
and/or imports. Changes to Changes to economic
energy supply sectors. production of energy supply
sectors.
Example 1: Coal is imported by
most MSs. The decrease in
coal use will also lead to a
decrease in coal imports.
Example 2: Increase in energy
efficiency may lead to a
decrease in gas use and thus
decrease the gas imports and
decreased activity in the gas
supply sector.
Power sector Changes in the power Changes in investment made
technology changes generation profile (e.g. solar, by the power generation sector,
wind) leads to changes in and subsequent changes its
investment made by the supply chain (e.g. different
power generation sector. goods are needed to build a
gas power plant compared to a
wind farm). Some economic
sectors (e.g. mechanical
engineering, electronics) will
benefit more from the transition.
Carbon pricing Increase in fuel cost based This affects prices of different
on fuel use and carbon fuels, and therefore the costs of
pricing the sectors using these fuels. It
impacts general price levels,
based on fuel consumption
levels.

14
 Additional investment Investment needed for Investment or expenditure
requirement energy efficiency (e.g. needed by households to
retrofitting), power grids, new improve efficiency. (e.g. service
fuel supply and storage (e.g. sectors to improve office energy
electricity storage or efficiency, new boilers). The
hydrogen). increased investment activity is
expected to lead to increased
demand for investment goods.
GDP and components How much GDP and Not directly entered in the
components change model. Used for comparison
compared to the baseline. and fine-tuning after all of the
above components have been
added.

Changes in final energy demand compared to the baseline provide the energy
use by sector and fuel type, and give us enough of information in terms of changes
to economic production and imports of fuels, as well as the changes in purchases
by different economic sectors of these fuel). What this means is that, based on the
changes in fuel demand, the economic module in E3ME determines the new level
of fuel needed to be extracted and/or imported. This leads to changes in domestic
extraction which will lead to changes in employment in that sector. If there are
changes in the supply, for example households and businesses demand less gas,
then economic activity in the gas supply sectors is expected to decrease and as a
result employment. The economic module also captures changes in intermediate
demand, i.e. if a sector demands less gas and more electricity, then this is reflected
in the purchases of intermediate goods of the respective sector.
The type of changes in final energy demand also have direct and indirect
impacts on other sectors in the economy. For example, increased energy efficiency
would lead to increased economic activity in construction and the sectors that
provide services and materials to construction. A switch to electric vehicles, which
are generally more fuel efficient than conventional vehicles, would impact retail
activities (because petrol stations are accounted for under retail).
Figure 2.4 summarises the changes to the 2023 Cedefop Skills Forecast
trends for final energy demand, which are: around 7% reduction in EU27 final
energy demand, mainly from reduction in use of coal oil and gas and; a relatively
large increase in electricity use.

15
Figure 2.4 EU27 final energy demand in 2030, million tonnes of oil-equivalent
(mtoe)

Sources: Impact Assessment

Power sector transition information is used to estimate the investment

needed. For example, we have information on how much solar technology will be
deployed, so we can use this information to calculate how much it will cost to build
solar plants and how much it will cost to maintain them by using capital, and
operation and maintenance unit costs already available in E3ME. Furthermore,
different power sector technologies have different goods purchasing requirements.
The building of a solar power plant will require goods from different sectors (e.g.
mechanical engineering, electronics) than a gas plant and we can estimate these
shifts in sectoral demand based on information available in E3ME (i.e. information
on goods purchases required by different type of plant).
Figure 2.5 summarises the changes to the updated green assumptions; by
2030 an increase in renewable generation shares of about 6pp would be
implemented. The increased renewable generation means that, compared to the
40% GHG reduction baseline, an increase in economic activity is expected in the
sectors that manufacture renewable technologies (e.g. solar), such as electronics
and electrical engineering, machinery and equipment, as well as sector that install
them, such as construction and service sectors in their supply chain, for example,
architecture.

16
Figure 2.5 EU27 power generation allocation, 2030

Sources: Impact Assessment

The exogeneous carbon pricing information taken from the Impact

Assessment assumptions for the MIX policy scenario is used by the E3ME model
to determine energy prices. The energy price level and energy use then feed
through to the general price level.
The additional investment requirements mainly capture any energy efficiency
investments and other infrastructure improvements needed to support the
transition. The type of investment is determined the sector making it (e.g. energy
efficiency investment is mainly one by households and offices to improve buildings
performance). The investment activity leads to increased demand for goods and
services elsewhere in the economy, as shown in Figure 2.6 below.

17
Figure 2.6 Main greening assumptions and how they are reflected in E3ME

2.2.3.3. Assumptions on automation/digitalisation

For the purpose of this update, we are assuming the continuation of existing trends
in automation and digitalisation, with increased capacity in advanced digital skills
and technologies as well as increased automation. This approach is consistent
with the one in the Impact Assessment of the Fit-for-55 policies. Furthermore, the
assumptions used for the Impact Assessment are actually consistent with those
outlined in Section 2.2.2 above (e.g. European Commission’s 2021 Ageing Report,
the Spring release of Ameco for the short-term GDP etc) 12.

2.2.3.4. Impact on employment forecast

The impact on sectoral employment in the forecast are directly linked to changes
in sector activity and composition resulting from the implementation of the above
assumptions. For example, changes to the use of gas by the economy have
resulted in changes in economic activity in the gas extraction and gas supply
sectors leading to a decrease in employment in these sectors. Another example is
increased renewables uptake in power generation, which has resulted in increased
activity in the sectors which supply components for the renewable power plants as

12 https://op.europa.eu/en/publication-detail/-/publication/96c2ca82-e85e-11eb-93a8-
01aa75ed71a1/language-en/format-PDF/source-219903975

18
well as construction. Increased activity in these sectors has resulted in increased
demand for labour.

2.2.3.5. Conclusions
The integration of sectoral changes resulting from the European Green Deal
implementation compared to the EU Reference Scenario 2020 baseline based on
available information. The one shortcoming is that we could not capture the
impacts of financing the transition (some financing options might lead to better
employment outcomes, whiles others to worse).

2.3. Expansion demands for occupations and

qualifications (EDMOD and QUALMOD)
General considerations
Previously detailed data on occupational employment structures as well as
information on qualification patterns has been available from the EU-LFS. This
enabled the development of models to explain and project expansion demands by
occupation (EDMOD) and similar models for qualifications (QUALMOD). These
models focus on the patterns of employment within sectors, which have been
extrapolated forward to develop the projections.
Since 2015, concerns about confidentiality have led to significant restrictions
on the level of detail that has been made available to researchers from the EU-
LFS 13. The much more limited information available since then severely restricts
what is possible in terms of model development for EDMOD and QUALMOD. The
existing methodology is based on exploring the occupational and qualification
structures of employment within sectors in considerable detail. This is no longer
possible given the restrictions imposed.
Information on the latest ISCO and ISCED classifications is therefore much
more limited than was available when the original versions of EDMOD and
QUALMOD were developed. We have information on the latest ISCO for just a few
years, and with much less detail on patterns within industries than was previously
made available.
The existing methodological framework is described in detail in the
Methodological Report. The general approach is based on the idea that
occupational and qualification demands are driven by changes in the industrial
employment structure. Technological change, demography, as well as other
factors, help to determine the overall patterns of employment by industry/sector.

13 See details on the data used in Section 2.3.

19
This, in turn, drives the demand for different occupations and qualifications, which
are often industry specific.
Technological changes, as well as digitalisation, for example, also influence
the patterns of employment within sectors. Technological developments result in
shifting demands for occupations as employers find new ways of doing things (for
example the impact of information technology on the demand for secretaries and
clerical workers). These patterns are also affected by factors such as changing
relative wage levels in different occupational and qualification groups (as
employers substitute more expensive inputs for cheaper ones).
In an ideal world, it would be possible to model these influences
econometrically. In practice, there is limited information on wages and
technological indicators available to develop such models. The existing EDMOD
and QUALMOD modules therefore rely upon a simple analysis and extrapolation
of trends in such patterns over a relatively short timescale. The data no longer
enable us to analyse and update trends within sectors. Our approach to dealing
with this is to use the data which have been made available to adjust our previous
projections to take account of what information is available on recent trends. This
will inevitably reflect recent events such as the Covid-19 pandemic and current
responses. However, recent developments in communications with Eurostat may
enable the team to access more detailed data and, subject to data checks, include
more information in the projections. The following sections outline this in more
detail.

EDMOD
EDMOD deals with expansion demands (net projected employment levels) by
occupation at the 2-digit level of ISCO. QUALMOD is the corresponding module
that deals with expansion demands for the three broad educational levels.
In the existing models econometric and other techniques were used to identify
trends in such patterns within industries and extrapolate them forward. The
sectoral employment projections from the E3ME model then drive the projections
of occupational employment and employment by qualification level.
As explained above, we have therefore adopted a pragmatic approach, which
relies on the previously available detailed data by industry, being combined with
more aggregate information now available.
For each country the latest EU-LFS information across all sectors is compared
with the previous historical estimates. Adjustments are then made to the historical
information on occupational and qualification employment shares to bring them
into line with the EU-LFS estimates.

20
 Figure 2.7 illustrates the process, with an example for Associate Professionals
in Czechia. The figure has three panels. It shows employment shares for the
selected occupations, aggregated across all industries. The left-hand panel
presents the results based on the latest E3ME projections and extrapolating the
previous trends within sectors without any additional adjustments to the
projections. The historical data have been constrained to match the observed EU-
LFS estimates.
The second panel shows the results of adjusting the overall occupational
trends for 2020-35 based on the observed patterns in the EU-LFS until 2020. The
final panel shows the most recent “raw” EU-LFS data, together with a linear trend
line fitted to the data points. In this case, the projections have been adjusted quite
significantly (compare Panel 1 and 2). In others, the EU-LFS trend line extrapolated
to 2035 does not look plausible, so more modest adjustments have been imposed.

21
Figure 2.7 Comparisons with EU-LFS estimates

Figure 2.8 focuses on the results from the EU-LFS data. These have been
applied on top of the left-hand panel. The dots are the estimates from the EU-LFS
across all industries. Note that because the sectoral structure of employment in the
EU-LFS does not match the National Accounts based employment used in E3ME
model, we do not expect a perfect match. The solid lines show the unadjusted
projections, again summed over all sectors.
In this example, the database appears to be broadly consistent with the latest
EU-LFS data. However, the projected trends appear to be out of line with the most
recent historical trends.

22
Figure 2.8 Comparisons with EU-LFS estimates

We have therefore reassessed the projected occupational shares and, where

the data suggest it is appropriate, adjusted these for 2035 to be in line with the
latest EU-LFS trends. The values of the adjustments for earlier years are linearly
interpolated back to zero in the last historical year (2021). The adjusted (solid line)
and unadjusted (dashed line) projections are shown in Figure 2.9.

23
Figure 2.9 Adjustments to occupational employment shares for 2035

A revised projection of the shares to 2035 is made, based on the linear trend
extrapolation of the historical EU-LFS estimates and professional judgment. This
is shown as the dashed lines in Figure 2.9. The values of the adjustments for earlier
years are linearly interpolated back to zero in the last historical year (2020).
Extrapolated trend shares are constrained to lie between 0 and 100%. The sum of
shares across all occupations is constrained to be 100%.

24
 In order to ensure that the data remain consistent with the other information
in the E3ME database a RAS process is employed. 14
In the case of the Czechia, there were limited comments from the ICE on
occupational trends. The main occupational trends under discussion were
Professionals, where the growth was considered too strong and Technicians and
associate professionals, where a more gradual decline was expected. As no exact
employment levels were specified, no adjustment have been made to the trends.
However, where specific indication for levels or growth rates for occupations have
been provided, these have been integrated into the forecast.

QUALMOD
Figure 2.9 focuses upon occupations. We are also interested in the qualification or
educational levels that employers are looking for. Given the limitations of the EU-
LFS we are only able to provide a broad breakdown into three levels. The highest
is broadly equivalent to a university degree, while the lowest refers to those with
the bare minimum of formal qualification or none at all. This is dealt with in the
QUALMOD module. This is described in detail in the Methodological Report. As
with occupations, this focuses upon the patterns of employment within
sectors/industries and occupations.
The more limited information now available from the LFS means that the
EDMOD module cannot be updated as previously. A similar process is therefore
adopted for qualifications/educational levels, which ensures the results from
QUALMOD match up with the latest EU-LFS historical data and trends. The results
by qualification/ educational level are then constrained to match the available
supply as indicated by QMOD (as discussed in Section 3.5 below).

Impact of Covid-19
The most recent data from the EU-LFS are for 2020. This will not reflect some of
the changes resulting from the impact of the Covid-19 pandemic. However, it
remains too early to say what longer term effects this might have.
The sectoral employment data from E3ME model include the most up-to-date
information on the impact of Covid-19 on the labour market. The effects of the
pandemic and the various lock-down measures introduced to deal with it on sectors
such as hospitality are therefore taken into account. This includes the knock-on
effect on occupations employed in such sectors. This includes information on both

14 RAS is a widely used iterative technique, which ensures that elements in a two-
dimensional data array match target row and column totals. In our case, multi-
dimensional arrays are used, but the principles are the same

25
the immediate impact of the pandemic in 2020 and the eventual recovery as the
economy and labour market returned to something like normality in 2021 and 2022.
However, with EU-LFS data only available for 2020, the full impact on
occupational employment patterns cannot be clearly identified. The results will
therefore show the sectoral impact of the pandemic but without and special
adjustment to take account of the pandemic on occupational employment patterns
within industries. If any when new evidence on such patterns becomes available,
the adjustments described above can be modified accordingly.

2.4. Replacement demand (RDMOD)

Replacement demand focuses on what might be called “permanent or semi-
permanent” withdrawals from the employed workforce. These include retirement,
family formation or other reasons for leaving the workforce, and emigration.
Previous replacement demand forecast using the RDMOD have used constant
flow rates based on available EU-LFS data over 2010-14 as more recent data are
not available anymore in the EU-LFS. However, since the period 2010-14, worker
flows have changed due to (1) technological changes, (2) increasing job mobility
and (3) pension reforms and increasing job mobility. Therefore, historical flow rates
might not reflect the current state in the labour market. For this reason, we have
conducted background research into ways to improve the replacement demand
estimates for the current forecast.

2.4.1. Adjustments to the RDMOD

To adjust the RDMOD for reforms in pension age, we qualitatively adjust the final
outflow which lies at 65 year-old to the effective retirement ages of individual
countries. Based on the underlying datasets of (i) the European Commission 2021
Ageing report (European Commission, 2020) and (ii) the OECD effective
retirement ages (OECD, 2021), we construct a more accurate age at which the
final cohort of workers will all flow out of the labour market and need to be replaced.
For the adjustment to the outflow rates of the RDMOD due to technological
changes and job mobility, we changed the method in which we quantitatively
estimated whether the 2010-14 flow rates reflect current automation trends or not.
If this is not the case, automation can be explicitly modelled within the RDMOD to
correct for this. This background study for the RDMOD adjustment is based on the
work of Hess and Montizaan (2022) for the Netherlands. They show that the
estimation errors of replacement demand can be decreased when implementing
an automation correction on the estimates of occupations measured at a very

26
detailed level of granularity (114 occupations). We tested whether the mean
absolute errors in the RDMOD for the Netherlands can be decreased using an
automation correction. We compared estimates of the RDMOD using the 2010-14
flow rates, to estimates with the automation correction from Hess and Montizaan
(2022) and compared them to the true values of the replacement demand as these
are also known for the Netherland over the period 2015-2021.
The change in methodology assumes that adjusting the Dutch outflow rates
for automation does not improve the estimates of the replacement demand
compared to the RDMOD with 2010-2014 flow rates. This is because at the
aggregation level at which the replacement demand is estimated, the
improvements in the error terms do not hold. For some occupations, the error
decreased, for others it increased without a qualitative pattern showing which
occupations can be improved and which cannot be improved. Our conclusion
therefore is that we cannot adjust the RDMOD for automation in a quantitatively
estimated model. As we do observe that automation impacts the replacement
demand and might increase estimation biases for specific occupations, we believe
that the role of the Country Experts in assessing the RDMOD-results becomes
more important with the decreasing quality of the available quantitative data from
the EU-LFS. Any indication of policies affecting participation rates (e.g. retirement
rates, increased engagement in education) has been taken into account both in
the labour supply projections as well as in the replacement demand estimates.
Limited feedback has been provided by ICEs on replacement demand, so no other
adjustments have been implemented apart from those summarised above.
But why is it important to consider automation when forecasting replacement
demand? First, automation has been shown to severely impact the labour market
in general (e.g., Acemoglu and Autor, 2010, Cedefop, 2021, among many others),
and labour demand in particular (e.g., Dixon et al., 2021). Second, replacement
demand estimates how many workers are replaced by other workers in an
occupation. By reducing the number of workers to be replaced in an occupation,
automation can have detrimental effects on the forecast accuracy. Thus,
automation can potentially be the source of substantial forecast error.
While the concern of forecast error must not be neglected, replacement
demand already incorporates occupation-specific trends in several dimensions
closely related to automation. Therefore, explicitly accounting for automation in the
replacement demand does not necessarily improve its performance. However, if
outdated flow rates do not account for the current pace or level of automation, we
likely overestimate the replacement demand which in turn deteriorates the quality
of the occupation-specific forecasts. Therefore, we want to know whether the
replacement demand already sufficiently accounts for automation of job tasks in

27
the labour market, or whether we need to re-adjust the replacement demand to
better account for such automation. As research by Cedefop (2021) shows,
automation reduces the replacement demand in European countries for
professions that are easily automatable. This mostly holds for low qualified jobs.
Hess and Montizaan (2022) test for the Netherlands whether the predictive
power of the labour market forecast model can be improved by testing automation-
induced reductions in replacement demand. As the effects for individual industries
and technologies are probably not uniform they perform the evaluation agnostically
without expectations by using two functional forms of how automation affects
labour demand to vary the impact in current years (see also Heald et al. for similar
assumptions with respect to the Cedefop Skills Forecasts). 15 First, a linear
development is assumed that produces a gradual introduction of automation
technologies by using the estimated automatability index of occupations from
Nedelkoska & Quintini (2018) and assuming that this automatability will be reached
in 2040 with a starting point in 2010. Every year an increasing share of out rotating
workers will be replaced by technologies. Figure 3.5 shows the idea of this type of
implementation. Assuming an estimated automatability of 60%, each year an
additional 2% of replacement demand will be deducted, so that we would observe
a 60% decrease in replacement demand after 30 years due to automation.
Second, an s-shaped development is assumed (logistic function), starting with
a low impact of technology in the first years, accelerating in the medium term and
decelerating at the end again. The idea is to account for the often mentioned
accelerating technological change (see, Hershbein and Kahn, 2018). The s-
shaped function reduces replacement demand less than the linear function we
implement.

(15) Heald, S., Smith, A., & Fouarge, D. (2020). Labour market forecasting scenarios for
automation risks: Approach and outcomes. ROA. ROA External Reports.

28
Figure 2.10 Implementation of automation

Source: ROA (2022)

Following Bakens et al. (2021), Hess and Montizaan (2022) first examine
bivariate correlations between different empirical measures that can resemble
replacement demand, and automation (as measured by Nedelkoska and Quintini
(2018) for OECD-countries). Then they perform a multivariate regression analysis
to understand which of the different variants of replacement demand can be best
predicted by the model in step 1. Finally, the predictive power of the model for the
different variants of replacement demand is evaluated by comparing the calculated
replacement demand and the replacement demand predicted by the model. The
results show that, adjusting for automation linearly over a 30-year period, the Mean
Squared Error (MSE) is way below the MSE for the model without adjustment, or
the MSE for the model with non-linear adjustment. As other correlations and
measures of goodness of fit do not change much, the error produced by outliers
can be reduced in the replacement demand of the Dutch labour market forecasts.
We used the linear automation multivariate regression analysis to predict
replacement demand for the Netherlands within the RDMOD- framework, and the
base RDMOD forecast, and compare them to the observed replacement demand
for the Netherlands until 2018. Accounting for technology worsens the results of
the RDMOD as the MSE does not decrease. This is because for some occupations
the bias decreases and for others it increases. We correlated the estimated bias
for occupations with different labour market variables such as share of temporary
contracts or the development of the wages, but there is no strong predictor of
whether occupations belong in the group of occupations for which the bias
increases or decreases. Therefore, our conclusion is that there is no quantitative
way to incorporate automation in the RDMOD that will improve the estimates.

29
Expert knowledge on how automation impacts replacement demand in specific
occupations in specific countries is a better way to deal with this issue. While some
ICEs commented on how automation may play an important role in future
developments in their country, no detailed information on trends was given with
respect to replacement demand. As such, no further adjustments were made.

2.4.2. Covid-19 impact

The quick recovery out of the pandemic in 2022 is expected to result in strong
labour demand and subsequent labour shortages for most European countries.
During the pandemic, we expected an effect of the pandemic on replacement
demand in the short run (first 2 years of the forecast, see Bakens et al., 2021) in
2021 as positions that come available due to RD will not be filled in sectors and
occupations with decreasing demand caused by the pandemic. However, evidence
for the Dutch labour market forecast in 2022 does not show any effect of the
pandemic on the short- or mid-run replacement demand due to the quick recovery
of the economy (Bakens et al., 2022). As the same situation holds for most
countries in Europe, a quick recovery from the pandemic-related lockdowns and
overall labour shortages, we do not think there is any need to adjust the
replacement demand estimates for a possible covid-impact in the coming round of
the European skills forecasts. In addition, as the past years of the world economy
have been rather unstable, and after the pandemic we now face high inflation and
high energy prices together with high geo-political instability, it is still unsure how
these will impact the economic growth and labour demand in the coming years.

2.5. Modelling qualification supply (QMOD)

Unlike EDMOD and QUALMOD, the data now available from the EULFS are
adequate to update the existing approach to modelling qualification supply. As for
the demand side we have adopted a pragmatic approach of adjusting the detailed
results to ensure that as far as possible they are consistent with the latest EU-LFS
data.
As in previous work on QMOD, a stock-flow model has been utilised for
projecting the three levels of qualifications – low, medium and high – for the 32
countries for which data are available. The detailed results are based on the latest
EU-LFS data and adjusted in aggregate to be consistent to EU country-level
population projections. Given the constraints imposed by the characteristics of the
data, discussed further below, the model works well, although it is pushed close to

30
its limits in places in making projections through as far as 2035 and by the small
sample sizes for some countries.
The initial activity was to update the EU-LFS data within the stock-flow
model 16, insofar as was necessary, bringing past data into line with the most up to
date series. The projections are heavily dependent on the latest information and,
in particular, the changes to qualifications amongst the younger age groups. Within
the context of a stock-flow model, the outcomes for older individuals are more
heavily dependent on earlier flows. Given that formal qualifications are largely
determined before age 30, the earlier years of data do not play such an important
role, but still determine the qualification levels of older individuals.
The work followed the normal course, in the first instance, that of establishing
a satisfactory working model for one or two countries. Based upon this work, the
Czechia was used to develop a template of the stock-flow model which was
subsequently used for all the countries. Applying the model to the Czechia case
first, allowed comparisons to be made with earlier work carried out for Cedefop.
Most, although not all of the issues with the modelling were identified at this stage,
prior to rolling out the model to other countries.
The model was estimated for the whole population and for the economically
active population of the Czechia. Comparisons were then made between the two
estimates and the associated activity rates were calculated. As with all elements
of the estimation process, checks were made for accuracy and that the resulting
outcomes were plausible.
Once the results were rolled-out, they were checked on a country-by-country
basis to ensure that there were no previously unforeseen problems with the data
or the modelling. The tests applied are mainly for internal consistency of the
estimates, such as non-negative proportions and values, activity rates outside of
the range zero to unity, that for most age groups activity rates are generally higher
for more qualified than less qualified individuals, etc.
A number of issues in the latest data were investigated and two in particular
caused some initial problems during the course of estimation: the small sample
sizes in the EULFS for a number of the countries (and the associated very small
cell sizes) and the treatment of the oldest individuals, when data are only available
for an open ended, 65 and over age group. The first of these problems did not
materialise in terms of the internal checks applied to the results. The second were
limited to a small number of countries and were removed by the adoption of a
slightly different treatment of the over 65 group for all countries. Any modifications

16 For details on the Stock-flow model, please read the Methodological Report.

31
or revision have been revisited and checked during the production of the final
estimates.

32
 CHAPTER 3. Data update and preparation

3.1. Introduction
This chapter presents the work on data update and preparation required in the
modelling framework so as to produce the 2023 Cedefop Skills forecast. The
chapter is structured as follows. The first section describes the update of the E3ME
model. The second sections describe the data preparation for Modules 2-5 of the
Cedefop Skills Forecast framework (as shown in Figure 1.1).

3.2. E3ME update

The historical and forecast database of the Cambridge Econometrics’ E3ME
model 17 underpinning the sectoral employment and labour force projections was
updated with the latest data available in May 2022. Table 3.1 shows in detail the
model variables updated.
Information on short-term trends is used to assess current developments and
incorporate the most recent developments into the forecast macroeconomic
trends. We make the employment projections consistent with other forecasting
exercises produced by the EC. For the short term (until 2023), we refer to the
Spring Economic Forecast (AMECO) 18 of DG ECFIN. We applied the growth rates
from the AMECO forecast to total employment and compensation of employees in
each MSs. The AMECO employment forecast is used to inform short-term growth
of employment projections, while information on compensation of employees gives
us more information on the average wage profile at the start of the projection
period. These data re used to supplement the more detailed historical sectoral data
and reflect recent developments as well as expectations. AMECO unemployment
rate and real GDP were used to update both the short-term projections and
historical values for these two variables. After 2023, we applied the annual
percentage growth rates implied by the long-term GDP projections shown in DG
ECFIN Ageing Report 2021 (which provides average projected GDP growth rates
in each decade until 2070).

17 See https://www.e3me.com/ for further details on this model.

18 https://ec.europa.eu/commission/presscorner/detail/en/IP_22_3070

33
 Table 3.1 Description of updated variables
Original Why we need
Source and Download
Variable Breakdown source update it
variable code date
date
Main variable
Eurostat, National
Employment Sector (2- projected in
accounts 19/05/2022 27/04/2022
(000s persons) digit NACE) the forecasting
(nama_10_a64_e)
exercise
Measure of
Eurostat, National
Output Sector (2- eocnomic
accounts 19/05/2022 21/04/2022
(million EUR) digit NACE) activity by
(nama_10_a64)
sector
Eurostat, National Used to
Employees Sector (2-
accounts 19/05/2022 27/04/2022 calculate
(000s) digit NACE)
(nama_10_a64_e) average wages
Compensation Eurostat, National Used to
Sector (2-
of employees accounts 19/05/2022 21/04/2022 calculate
digit NACE)
(million EUR) (nama_10_a64) average wages
Using
participation
Labour Age groups rates and
Eurostat, LFS
participation (5-year 19/05/2022 27/04/2022 population,
(lfsa_agan)
rates (%) bands) labour force
(000s) is
calculated.
Age groups Used in
Population Eurostat
(5-year 19/05/2022 24/03/2022 calculating
(000s) (demo_pjangroup)
bands) labour force
Used in
calculating
labour force
Population Age groups
Eurostat projections
projections (5-year 19/05/2022 20/07/2020
(proj_19np) based on the
(000s) bands)
participation
rate
projections
Short-term
projections
Employment cosistent with
projections AMECO (Spring GDP
Total 19/05/2022 16/05/2022
(short-term, Forecast) projections
‚000s persons) Used to
inform recent
developments.
Compensation As above
of employees
AMECO (Spring
projections Total 19/05/2022 16/05/2022
Forecast)
(short-term,
millon EUR)
Unemployment As above
AMECO (Spring
rate (%, Total 19/05/2022 16/05/2022
Forecast)
historical and

34
short term
projections)
Real GDP Used to assess
(millon EUR short term
2010 prices, AMECO (Spring trends
Total 19/05/2022 16/05/2022
historical and Forecast)
short-term
projections)
Real GDP Long-term
(millon EUR economy
Ageing Report
2010 prices, Total May 2022 May 2021 trends
2021
long-term
projections)
World Bank Global
April 2022
commodities price commoditiy
(WB)
Commodity forecast prices used to
Commodities May 2022 October 2021
prices (indices) IEA World inform import
IEA)
Economic prices
Outlook 2021
Source(s): The Authors

Historical labour force data from the EU-LFS and population data are used to
compute historical participation rates by age and gender, which are projected over
the forecast horizon through econometric equations 19. The forecast participation
rates are then multiplied with Eurostat population projections to obtain forecast of
labour force. Eurostat data for labour force and population were available until
2021 for all MSs, except for a few gaps in the historical data that were interpolated.
The population projections were released by Eurostat in 2020 (i.e.
EUROPOP2019), and therefore there is a small discrepancy with the realised
population figures in 2020 and 2021. We solve this issue by using historical
population values until 2021 and applying the annual percentage growth rates in
the population projections from 2022 onward. As shown in Table 3.1, historical
data for employment, output, compensation of employees and number of
employees come from the Eurostat National Accounts database. The variables
from the National Accounts are essential to the equations in the E3ME model.
Employment by sector is they main variable of interest in the Cedefop Skills
forecast. The historical employment data from the National Accounts are used as
input in the model to produce the employment forecast by sector through

19 More information on the econometrics equations can be found in the E3ME Technical
Manual (https://www.e3me.com/wp-content/uploads/sites/3/2019/09/E3ME-
Technical-Manual-v6.1-onlineSML.pdf ).

35
econometric equations 20. In E3ME, employment is a function of output, real wage
costs, hours worked and, capital and R&D expenditure as proxies for technological
development. The equation set up is based on the representation of the theoretical
optimisation problem in which firms trying minimise costs for a given level of output
(i.e. it allows for the substitution of labour and capital based on cost pressures in
the economy). While productivity is not explicitly featured in the equation
determining employment, its impact is reflected through wages. Among the
productivity is one of the determinants of wage rates by sector and wages feed into
the employment equation.
.

The availability of National Accounts historical data varies between Member

States (MSs), both in terms of years and sectoral breakdowns. Most MSs have
data until 2020, which can be considered the last historical year for the variables
based on the National Accounts database. However, several exceptions are
highlighted below:
• Some MSs had data in 2020 only for 1-digit NACE sectors (e.g. Germany,
Spain);
• Some MSs had data for 2020 for only a subset of 2-digit NACE sectors
(e.g. France, Italy);
• A few MSs had data also for 2021 to some extent (e.g. Croatia, Iceland,
Italy, Malta) and the existing data was used in the forecast.
• Malta, Luxembourg and Switzerland did not show data for many 2-digit
sectors due to confidentiality issues, and in the case of Malta there were
some gaps in the series;
• Some sectors in certain MSs were not available due to confidentiality, such
as R&D in Sweden, so these are considered as zero in the forecast;
• Eurostat data for the Republic of North Macedonia and Turkey had to be
complemented with data from the respective statistical offices to complete
the series. In the case of Turkey, employment levels came from the LFS
instead of National Accounts, as is the norm for the other countries.

In these instances, we used all available information to impute the missing

sectors and obtain complete historical data series at the 2-digit NACE level until
2020:

20 More information on the econometrics equations can be found in the E3ME Technical
Manual (https://www.e3me.com/wp-content/uploads/sites/3/2019/09/E3ME-
Technical-Manual-v6.1-onlineSML.pdf ).

36
 • In the cases where 2-digit NACE data was missing for 2020, we
disaggregated 1-digit NACE values in 2020 using the shares of 2-digit
NACE in 2019;
• In some cases, it was possible to calculate the missing 2-digit NACE sector
as the difference between the parent 1-digit NACE and the other 2-digits
NACE sectors;
• In the case of Malta, Luxembourg, and Switzerland, the 2-digit NACE
sectors were calculated using the shares from previous version of the
forecast within 1-digit NACE or based on Structural Business Statistics
data, and gaps in the series were interpolated for Malta.
In the case of the Republic of North Macedonia, several assumptions were made:
• No employment data were available for 2020 from the Eurostat National
Accounts, so we applied the total growth rate in 2020 from LFS data, and
disaggregated using the 2019 sectoral employment shares;
• Data for employees were not available from Eurostat National Accounts,
so were taken from the National Statistical Office 21 and are based on a
survey of enterprises.
In the case of Turkey, the following assumptions were made:
• For employment and employees, only LFS data at 1-digit NACE are
available, so 2-digit sectoral shares were used from the previous version
of the forecast.
• Total output was taken from the National Statistical Office and
disaggregated using 2-digit sectoral shares from the previous version of
the forecast.
• Compensation of employees was available only at the 1-digit NACE
sector, so each 1-digit NACE sector was disaggregated using 2-digit
NACE shares from the previous version of the forecast.

3.3. LFS update of Modules 2-5

Due to the restrictions on the release of confidential data from EU-LFS 22, the
research team mostly relied on the anonymised LFS microdata for Modules 2-5,
which at the time the analysis is undertaken will go up only to 2020 23 (see Table
3.2). The research team used the EU-LFS microdata for the previous update of the

21 https://www.stat.gov.mk/Default_en.aspx .
22 Eurostat is no longer releasing in the scientific use files the matrix of employment by
2-digit ISCO occupation and 2- digit NACE sectors.
23 Datasets-availability-table.pdf (europa.eu).

37
Skills Forecast. The procedure described in the Methodological Report for
estimating occupational shares of employment by sector has been implemented
for the current update. A summary of the approach can be found in Section 2.3.

Table 3.2 Main data inputs from EU-LFS microdata

Variable Breakdown
Sector (NACE 1-digit); occupation (ISCO 2-digit),
Demand Employment
qualification level (3 broad levels); gender
Employment Sector (NACE 2-digit); age; gender
Gender; age group; qualification level (3 broad levels);
Supply Employment occupation (ISCO 3digit); ILOSTAT employment status
(1,2,3,4)
Population E3ME age groups; qualification level (3 broad levels); gender
Labour force E3ME age groups; qualification level (3 broad levels); gender

Firstly, the data collection and update of the raw data were based on two
datasets, both generally available until 2020:
(g) a set of data have been provided from the LFS microdata including the
dimensions: year, country, sex, education (Low/ Medium/ High), NACE Rev.
1 1-digit sector and ISCO-08 3-digit occupation; and
(h) a set of data including the dimensions: country, year, sex, age (0-4,5-9,10-14,
…, 95-99), education (Low/ Medium/ High), labour market status and ISCO-
08 3-digit occupations.
Second, to complement and cross-check the microdata an additional special
request to Eurostat was made to also obtain data for additional countries (Turkey,
North Macedonia) which are not available from the microdata. Eurostat provided
these data at ISCO 1-digit occupation level only for the first set of data in Table
3.2. Similarly, another special request to Eurostat provided the data by NACE Rev.
2 2-digit sectors, age groups and gender according to the second set of data in
Table 3.2. These data have also been made available by Eurostat including age
groups 15-24, 25-54, 55-64, 64+. With respect to the supply side data (third set of
data in Table 3.2) another special request to Eurostat was made to collect such
data; again, age groups available are only 15-24, 25-54, 55-64, 64+. These data
are generally available until 2020).
Third, data collected from Eurostat website24 (on employment, population,
labour force) are compiled complementing the ones from Eurostat special requests
as these include more detailed age groups (15-19, 20-24, …, 65-69).
All these data have been collected and prepared and made available as an
input for Modules 2-5.

24 Database - Eurostat (europa.eu).

38
List of abbreviations

LFS Labour Force Survey

VET vocational education and training

39
References
Acemoglu, D. and Autor, D. (2010) Skills, Tasks and Technologies: Implications
for Employment and Earnings. Working Paper 16082. National Bureau of
Economic Research. Available at: https://doi.org/10.3386/w16082.

Bakens et al. (2022) Update arbeidsmarktprognoses 2026. ROA, ROA Fact

Sheets No.x (forthcoming).

Bakens, J. et al. (2021) De arbeidsmarkt naar opleiding en beroep tot 2026.

WorkingPaper 005. Netherlands: ROA. Available at:
https://doi.org/10.26481/umarep.2021005.

Dixon, J., Hong, B. and Wu, L. (2021) ‘The Robot Revolution: Managerial and
Employment Consequences for Firms’, DecisionSciRN: Other Performance
Management (Sub-Topic) [Preprint]. Available at:
https://www.semanticscholar.org/paper/The-Robot-Revolution%3A-Managerial-
and-Employment-for-Dixon-
Hong/7f3a04972c7f08c1ad54e57e780b8e41415cf958.

European Commission (2020) The 2021 Ageing Report: Underlying Assumptions

& Projection Methodologies. Institutional Paper 142. Luxembourg: Publications
Office of the European Union. Available at: https://economy-
finance.ec.europa.eu/publications/2021-ageing-report-underlying-assumptions-
and-projection-methodologies_en.

European Commission (2021) The 2021 Ageing Report: Economic & Budgetary
Projections for the EU Member States (2019-2070). European Commission.
Available at: https://ec.europa.eu/info/sites/default/files/economy-
finance/ip148_en.pdf.

European Commission (2022) European Economic Forecast Spring 2022.

Available at: https://economy-finance.ec.europa.eu/publications/european-
economic-forecast-spring-2022_en.

Eurostat (2019) The EU’s population projected up to 2100, Eurostat. Available at:
https://ec.europa.eu/eurostat/web/products-eurostat-news/-/DDN-20190710-1
(Accessed: 6 February 2020).

Hershbein, B. and Kahn, L.B. (2018) ‘Do Recessions Accelerate Routine-Biased

Technological Change? Evidence from Vacancy Postings’, American Economic
Review, 108(7), pp. 1737–72. Available at: https://doi.org/10.1257/aer.20161570.

Hess, P. and Montizaan, R. (2022) The impact of automation on the replacement

demand: an evaluation study, ROA-R-2022, forthcoming.

Nedelkoska, L. and Quintini, G. (2018) ‘Automation, skills use and training’, OECD
Social, Employment and Migration Working Papers, 202.

OECD (2021) Pensions at a Glance 2021. Available at:

https://doi.org/10.1787/ca401ebd-en.

40