## SECTION I: INTRODUCTION

## Source details

**Canonical URL:** [SECTION I: INTRODUCTION](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020168-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020168-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020168-print-pdf.pdf.json)

---

### Overview and recent trends
- Unemployment in the European Union (EU) rose during the global financial crisis (GFC), peaked in 2013, and fell to about 6 percent in 2019.
- Incidence of long-term unemployment (LTU)—share of people unemployed for more than one year in total unemployed—remained above the pre-GFC level (average 2005–08).
- During 2016–19, about 45 percent of the total unemployed were LTU, down from 50 percent in 2014.
- Using panel data for 25 European countries over 2000–18, higher LTU incidence is linked to skill mismatches and lower labor market matching efficiency.

### COVID-19 downside risk to LTU
- Pandemic and lockdowns caused sharp activity contraction and raised underemployment and unemployment, particularly for low-skilled workers who cannot work from home.
- Projected unemployment rate in Europe: from 6.3 percent in 2019 to 9 percent in 2020 (European Commission, 2020).
- Weak labor market and post-COVID reallocation likely to increase LTU through labor market hysteresis.

### Individual and macroeconomic impacts of LTU
- Individual-level harms:
  - Loss of income, social disconnection, mental and material stress.
  - Human-capital erosion from skill loss and higher exit probability from the labor force.
  - Re-employment after LTU often involves lower pay and demotion (Mitchell et al., 2013).
  - Youth facing protracted unemployment suffer weakened future employment prospects, psychological harm, and institutional distrust (Eurofound, 2017; Duell et al., 2016).
- Macroeconomic and distributional effects:
  - High LTU concentrates unemployment on a small group, contributing to income inequality.
  - Lack of work is a principal cause of poverty among working-age households in many European countries.
  - Integrating LTU (and discouraged persons) can expand the productive labor pool and help address demographic challenges.

### Cross-country patterns and notable statistics
- As of 2018/19 the LTU rate was highest in Greece—“almost one in five working age population” were LTU (≈20 percent).
- Greece and Spain saw LTU rates rise about fourfold after the GFC compared with the pre-crisis period.
- Czech Republic’s LTU rate fell from 3.3 percent (2005–08) to less than 1.5 percent in 2018.
- Youth LTU changes:
  - Greece: >15 percentage points increase between pre- and post-GFC.
  - Italy and Spain: about 10 percentage points increases.
  - Germany and many new member states (Poland, Czech Republic, Hungary, Slovakia): youth LTU declined.
- Incidence of LTU across groups:
  - EU average incidence rose to 46 percent post-GFC vs 43 percent during 2005–08.
  - Greece post-crisis: almost ¾ of the unemployed long-term.
  - Incidence highest among people aged 50+ at 60 percent of unemployed in that cohort.
- Exceptions: Bulgaria and Belgium saw LTU rates decline after the GFC while incidence rose.

### Key empirical approach and data
- Empirical strategy: panel regression of incidence of LTU with country and time fixed effects:
  - LTU_it = γ1 X_it + γ2 Z_it + η_i + ν_t + ε_it
  - X: macroeconomic variables (lagged GDP growth, inflation, skill mismatches, estimated matching efficiency).
  - Z: policy variables (active and passive labor market policies, tax wedge).
  - Includes one lag of incidence of LTU to capture persistence.
- Data: annual data for 25 European countries over 2000–18; sources from Eurostat.
- Matching efficiency estimated via a matching function approach (search-matching theory).
- Skill mismatch measured as sum of squared differences between shares of population and employment by education attainment.
- Summary correlations (Table 1):
  - Incidence of LTU and skill mismatch: correlation 0.41.
  - Incidence of LTU and estimated matching efficiency: correlation -0.37.
  - Incidence of LTU positively correlated with NEET.
  - Incidence of LTU negatively correlated with expenditures on labor market policies.

### Main empirical findings (selected coefficients and significance)
- Persistence and macro effects (simple regressions):
  - Lag of LTU: coefficient ~0.813***.
  - Lag of GDP growth: coefficient ~-0.245***.
  - Inflation: not statistically significant.
- Labor market features:
  - Skill mismatches: coefficient 0.003***.
  - Matching efficiency: coefficient -0.079***.
  - NEET: coefficient 0.009***.
- Policy variables:
  - Aggregate LMP spending: coefficient -0.119***.
  - Tax wedge: not statistically significant.
- ALMP disaggregation (multivariate/simple results):
  - Training: coefficient -0.50*** (simple); -0.666* (multivariate benchmark).
  - Employment incentives: coefficient -0.63*** (simple); -0.180 (multivariate benchmark, not significant).
  - Supported employment and rehabilitation: coefficient -0.21* (simple); 0.578 (multivariate benchmark, not significant).
  - Direct job creation: coefficient -0.10 (simple, not significant); -0.280 (multivariate, not significant).
  - Start-up incentives: coefficient -1.40** (simple); -3.411** (multivariate benchmark).
  - Unemployment benefits: coefficient 0.02 (not statistically significant).
- Sample sizes vary across specifications (examples: N = 456, 453, 429, 256, 439, 362, 368; ALMP disaggregated regressions: N = 372, 370, 290, 327, 339, 136).

### Multivariate regression — key findings
- Lag of LTU: 0.562*** (standard error 0.069) — persistent dynamics.
- Lag of GDP growth: -0.164*** (standard error 0.069).
- Skill mismatches: 0.003*** (standard error 0.001).
- Matching efficiency: -0.042** (standard error 0.016).
- ALMPs (Benchmark column (1)):
  - Training: -0.666* (standard error 0.375).
  - Start-up incentives: -3.411** (standard error 1.648).
  - Employment incentives: -0.180 (standard error 0.373).
  - Supported employment and rehabilitation: 0.578 (standard error 0.582).
  - Direct job creation: -0.280 (standard error 0.797).
  - Unemployment benefits: 0.020 (standard error 0.130).
- Employment Protection (EPL): 0.005 (standard error 0.035) — not significant.

### Robustness checks and sample splits
- Dynamic panel (Arellano and Bond (1991)) yields qualitatively similar results: LTU associated with skill mismatches, matching efficiency, and some ALMPs.
- Pooled OLS produces similar results; training not significant in pooled OLS.
- New Member States (NMS) sample: qualitative results broadly unchanged; coefficients on economic growth, matching efficiency, and ALMPs become larger for the NMS.
- Dynamic estimation used to address endogeneity of lagged dependent variable.

### Evidence from literature on ALMP effectiveness and program design
- Meta-analyses and studies:
  - Card et al. (2018): long-term unemployed benefit more from ALMPs, especially human-capital improving measures.
  - Training measures: positive medium- and long-run results (Card et al. 2010, 2018), conditional on alignment with labor market demand.
  - Start-up incentives: effective in reducing unemployment and increasing employment in advanced economies (Rodriguez-Planas, 2010; Escudero, 2018); associated with a “double dividend” (Caliendo and Künn, 2011).
  - Direct job creation: generally least effective.
- Program design matters: program length, monetary incentives, individualized follow up, activity targeting, early identification and intervention, and small-scale well-targeted programs increase effectiveness (Levy et al., 2019; Meager and Evans, 1998; Duell, 2012; Caliendo and Schmidl, 2016; Kluve, 2019).

### Labor market policy interventions and spending patterns
- Categories (Eurostat): (i) labor market services (PES and “services and sanctions”); (ii) passive labor market policies (PLMPs — unemployment benefits and income supports); (iii) active labor market policies (ALMPs — training, employment incentives, start-up incentives, supported employment and rehabilitation, direct job creation).
- EU average LMP spending in 2017: 1.3 percent of GDP.
- Cross-country range in 2017: 0.1 percent of GDP in Romania to 3 percent of GDP in Denmark.
- In general, LMP spending in advanced economies was twice as large as those in the NMS.
- Allocation shares (2017, EU average):
  - PLMPs: about one-half of LMP interventions spending.
  - ALMPs: about one-third of LMP interventions spending.
  - PES: remaining share.
- Within ALMPs (2017, EU average):
  - Employment incentives: about one-third of ALMP spending.
  - Training programs: about 30 percent of ALMP spending.
  - Start-up incentives: generally less than 5 percent of ALMP spending in most countries (exceptions: Spain, Poland, Sweden, Croatia).
- Country-level variations highlighted (examples: Cyprus, Bulgaria, Hungary, Malta, Czech Republic, Denmark, Croatia, Lithuania, Luxembourg, Poland, Bulgaria, Greece, Slovenia, Austria, Germany, France, Latvia, Portugal, Finland).

### Skill mismatches and digital skills
- Skill mismatch measurement: differences between employment and working-age population by education groups.
- High skill mismatches: Belgium, Ireland, Bulgaria.
- Low skill mismatches: Netherlands, Portugal, United Kingdom.
- Post-GFC: about half of EU member states experienced improvement in skill matching; largest improvements in many NMS; deterioration in many advanced countries (Ireland, Greece, Spain, Portugal).
- Digital skills demand and gaps:
  - Almost half of vacancies related to ICT, healthcare, science, and business and retail sectors per Cedefop (2018).
  - ESJS 2014: more than 85 percent of EU employers required at least basic ICT skills.
  - Digital skill gaps (percent difference measure): all countries except the Netherlands experienced skill gaps of at least 15 percent.
  - Digital skill gaps in Bulgaria, Croatia, and Italy were as high as 50 percent.
- Policy implications to close digital skill gaps:
  - Invest in telecommunication networks and access to internet; promote digitalization of businesses to increase demand and internal training.
  - Integrate technology into education and training; country examples cited for early programming lessons, curriculum integration, national digital skills evaluation, ICT Master Plans for Education.
  - Adult training examples: Denmark, Luxembourg, Sweden, New Zealand, Singapore cooperation with industry to identify needs and promote training.

### Evolution of matching efficiency in EU and role of PES
- Most EU countries experienced declines in matching efficiency immediately after the GFC.
- Recovery patterns:
  - Improved to pre-crisis levels: Czech Republic, Estonia.
  - Continued decline after GFC: Bulgaria, Greece, Ireland, Lithuania, Romania, United Kingdom, Spain, Slovenia.
- Matching efficiency in Europe declined relative to the pre-GFC level.
- PES roles and successful features:
  - Targeted and personalized programs combining interventions (initial profiling, step-by-step training and job search assistance, employer internships with allowances).
  - Strong institutional cooperation (PES with municipalities, youth and family services, social services, employers); recommendation for a single point of contact for LTU.
  - Outsourcing specialized services to PES partners for specific groups (examples: France, Poland).
  - Post-employment follow-up support (examples: Bulgaria, Germany) with evidence of improved retention (European Commission, 2014).

### Data, measurement, and selected summary statistics
- Matching efficiency estimated using Cobb-Douglas matching function:
  - m_i = μ_i u_i^α v_i^(1−α)
  - Job finding rate f_i = μ_i θ_i^(1−α), where θ ≡ v/u.
  - Steady state: μ_i = s_i (1−u_i)/(u_i θ_i^(1−α)).
- Data sources: quarterly unemployment rate, unemployment, and job vacancy rate for 23 European countries from Eurostat; construction of job finding and separation rates using short-term unemployment and total unemployment.
- Selected summary statistics:
  - Incidence of LTU: Number of observations 494; Mean 0.425; Standard Deviation 0.132; Min 0.095; Max 0.763.
  - GDP growth rate: Number of observations 589; Mean 0.051; Standard Deviation 0.066; Min -0.304; Max 0.349.
  - NEET: Number of observations 557; Mean 14.649; Standard Deviation 5.455; Min 4.100; Max 32.500.
  - Skill mismatch: Number of observations 520; Mean 14.518; Standard Deviation 8.084; Min 0.772; Max 43.126.
  - Matching efficiency: Number of observations 280; Mean 0.680; Standard Deviation 0.528; Min 0.159; Max 4.615.
  - Labor market policies: Number of observations 446; Mean 0.371; Standard Deviation 0.289; Min 0.033; Max 2.129.
  - Training: Number of observations 450; Mean 0.048; Standard Deviation 0.052; Min 0.000; Max 0.315.
  - Start-up incentives: Number of observations 405; Mean 0.004; Standard Deviation 0.006; Min 0.000; Max 0.035.
  - Unemployment benefits: Number of observations 456; Mean 0.263; Standard Deviation 0.211; Min 0.021; Max 1.283.
- Worker flow and elasticity highlights (Table A1):
  - Job finding rates range from 11.9 percent in Greece to 57.5 percent in Finland.
  - Separation rates range from 1.4 percent in Slovakia to 5.2 percent in Spain.
  - Estimated matching elasticities with respect to unemployment mostly in range 0.5–0.7; selected country entries:
    - Belgium 2010Q1-2019Q1 — Job finding rate 29.9 — Separation rate 2.2 — Elasticity alfa 0.87
    - Bulgaria 2005Q1-2019Q1 — Job finding rate 17.4 — Separation rate 1.5 — Elasticity alfa 0.59
    - Czech Republic 2008Q1-2019Q1 — Job finding rate 28.3 — Separation rate 1.4 — Elasticity alfa 0.82
    - Finland 2003Q1-2019Q1 — Job finding rate 57.5 — Separation rate 5.1 — Elasticity alfa 0.80
    - Greece 2009Q1-2019Q1 — Job finding rate 11.9 — Separation rate 2.5 — Elasticity alfa 0.73
    - Spain 2001Q1-2019Q1 — Job finding rate 35.4 — Separation rate 5.2 — Elasticity alfa 0.30
    - Sweden 2009Q1-2019Q1 — Job finding rate 51.8 — Separation rate 4.3 — Elasticity alfa 0.95

### Policy implications and recommendations
- Rebalance and increase spending on effective ALMPs where needed:
  - Reallocate from less effective measures (e.g., direct job creation) toward training programs and start-up incentives where appropriate.
  - Example: Bulgaria — low ALMP spending and high LTU incidence; direct job creation received the largest allocation, suggesting scope to increase overall ALMP envelope and shift toward training and start-up incentives.
- Apply coherent and comprehensive approaches:
  - Combine ALMP types (training, counselling, subsidies) to support various LTU groups, including vulnerable and discouraged workers.
  - For high LTU countries: widespread ALMPs plus additional measures for disadvantaged groups.
  - For low LTU countries: tailored programs for disadvantaged groups.
- Address skill mismatches, with emphasis on digital skills:
  - Upgrade skills and correctly identify current and future skill needs.
  - Create a digital-friendly economy (telecommunication networks, internet access, digitalization of businesses).
  - Integrate digital skills into education and adult training; utilize industry cooperation to identify needs.
- Improve labor market matching efficiency:
  - Strengthen public employment services (PES), promote policy coordination among stakeholders, offer tailored programs for the LTU, and conduct post-placement monitoring and follow-up support.
- These policies aim to limit scarring from the COVID crisis and support transition into post-COVID labor markets featuring accelerated digitalization, automation, and greater use of short-term contracts and freelance work.

*Source: SECTION I: INTRODUCTION, wpiea2020168-print-pdf*

### SECTION I: INTRODUCTION ______________________________________________ 4

### SECTION I: INTRODUCTION

### Overview and recent trends
- Unemployment in the European Union (EU) rose during the global financial crisis (GFC), peaked in 2013, and fell to about 6 percent in 2019.
- Despite the overall unemployment decline, the incidence of long-term unemployment (LTU)—the share of people unemployed for more than one year in total unemployed—remained above the pre-GFC level (average 2005–08).
- During 2016–19, about 45 percent of the total unemployed were LTU, down from 50 percent in 2014.
- Using panel data for 25 European countries over 2000–18, the paper links higher LTU incidence to skill mismatches and lower labor market matching efficiency.

### COVID-19 downside risk to LTU
- The pandemic and lockdowns caused a sharp activity contraction and raised underemployment and unemployment, particularly for low-skilled workers who cannot work from home.
- The unemployment rate in Europe is projected to rise from 6.3 percent in 2019 to 9 percent in 2020 (European Commission, 2020).
- The weak labor market and post-COVID reallocation are likely to increase the pool of long-term unemployed through labor market hysteresis.

### Individual and macroeconomic impacts of LTU
- LTU causes loss of income, social disconnection, mental and material stress, and human-capital erosion from skill loss and higher exit probability from the labor force.
- High incidence of LTU concentrates unemployment on a small group, contributing to income inequality; lack of work is a principal cause of poverty among working-age households in many European countries.
- Re-employment after LTU often involves lower pay and demotion (Mitchell et al., 2013).
- Integrating LTU (and discouraged persons) can help address demographic challenges by expanding the productive labor pool.
- Youth facing protracted unemployment suffer weakened future employment prospects, psychological harm, and institutional distrust (Eurofound, 2017; Duell et al., 2016).

### Cross-country patterns and notable statistics
- As of 2018/19 the LTU rate was highest in Greece—“almost one in five working age population” were LTU (≈20 percent).
- Greece and Spain saw LTU rates rise about fourfold after the GFC compared with the pre-crisis period.
- Czech Republic’s LTU rate fell from 3.3 percent (2005–08) to less than 1.5 percent in 2018.
- Youth LTU: Greece >15 percentage points increase between pre- and post-GFC; Italy and Spain about 10 percentage points increases; youth LTU declined in Germany and many new member states (Poland, Czech Republic, Hungary, Slovakia).
- Incidence of LTU across groups: EU average incidence rose to 46 percent post-GFC vs 43 percent during 2005–08; Greece post-crisis had almost ¾ of the unemployed long-term; incidence highest among people aged 50+ at 60 percent of unemployed in that cohort.
- Exceptions (structural concerns): Bulgaria and Belgium saw LTU rates decline after the GFC while incidence rose.

### Key empirical approach and data
- Empirical strategy: panel regression of incidence of LTU with country and time fixed effects:
  - LTU_it = γ1 X_it + γ2 Z_it + η_i + ν_t + ε_it
  - X: macroeconomic variables (lagged GDP growth, inflation, skill mismatches, estimated matching efficiency)
  - Z: policy variables (active and passive labor market policies, tax wedge)
  - Includes one lag of incidence of LTU to capture persistence
- Data: annual data for 25 European countries over 2000–18; sources from Eurostat; matching efficiency estimated via a matching function approach (search-matching theory); skill mismatch measured as sum of squared differences between shares of population and employment by education attainment.
- Summary correlations (Table 1 highlights):
  - Incidence of LTU and skill mismatch: correlation 0.41.
  - Incidence of LTU and estimated matching efficiency: correlation -0.37.
  - Incidence of LTU positively correlated with NEET (youth not in employment, education, or training).
  - Incidence of LTU negatively correlated with expenditures on labor market policies.

### Main empirical findings (selected coefficients and significance)
- Persistence and macro effects:
  - Lag of LTU: coefficient ~0.813*** in simple regressions, indicating strong persistence.
  - Lag of GDP growth: coefficient ~-0.245***, showing higher growth associated with lower LTU incidence.
  - Inflation: not statistically significant in reducing LTU in these regressions.
- Labor market features:
  - Skill mismatches: coefficient 0.003*** (simple regressions) — higher skill mismatch associated with higher LTU incidence.
  - Matching efficiency: coefficient -0.079*** (simple regressions) — higher matching efficiency associated with lower LTU incidence.
  - NEET: coefficient 0.009*** — higher NEET associated with higher LTU incidence.
- Policy variables:
  - Aggregate labor market policies (LMP) spending: coefficient -0.119*** — more generous LMP spending associated with lower LTU incidence.
  - Tax wedge: not statistically significant in the core regressions.
- ALMP disaggregation (multivariate results):
  - Training: coefficient -0.50*** (significant).
  - Employment incentives: coefficient -0.63*** (significant).
  - Supported employment and rehabilitation: coefficient -0.21* (weakly significant).
  - Direct job creation: coefficient -0.10 (not statistically significant).
  - Start-up incentives: coefficient -1.40** (significant).
  - Unemployment benefits: coefficient 0.02 (not statistically significant).
- Sample sizes in regression tables vary across specifications (examples: N = 456, 453, 429, 256, 439, 362, 368 in simple regressions; N = 372, 370, 290, 327, 339, 136 in ALMP disaggregated regressions).

### Robustness and methodological contributions
- The paper estimates labor market matching efficiency for 25 EU countries using a search-matching theory rather than relying solely on Beveridge curve shifts, addressing that Beveridge shifts can reflect cyclical demand shocks rather than structural matching efficiency changes.
- Robustness checks include alternative specifications and lagged policy variables (results broadly similar).

### Policy implications and recommendations
- Rebalance and increase spending on effective ALMPs where needed:
  - No one-size-fits-all; emphasize reallocation toward effective programs (training and private sector incentives) for countries with high LTU and low ALMP effectiveness.
- Address skill mismatches, with focus on digital skills:
  - Policies should both upgrade skills and correctly identify current and future skill needs; digital skills are highlighted as increasingly critical per Cedefop’s European skills and jobs survey (ESJS).
- Improve labor market matching efficiency:
  - Matching efficiency in the EU generally declined after the GFC; strengthening public employment services (PES) is a key instrument to improve job matching, particularly for disadvantaged and long-term unemployed.
- These policies will help limit scarring from the COVID crisis and support transition into a post-COVID labor market featuring accelerated digitalization, automation, and greater use of short-term contracts and freelance work.

*Source: SECTION I: INTRODUCTION, wpiea2020168-print-pdf*

### 23.      Full multivariate regression shows skill mismatches, matching efficiency, and

### 23.      Full multivariate regression shows skill mismatches, matching efficiency, and

### Multivariate regression — key findings
- Lag of LTU: 0.562*** (standard error 0.069) — persistent dynamics in LTU.
- Lag of GDP growth: -0.164*** (standard error 0.069) — higher growth associated with lower LTU.
- Skill mismatches: 0.003*** (standard error 0.001) — higher mismatches associated with higher LTU.
- Matching efficiency: -0.042** (standard error 0.016) — higher matching efficiency associated with lower LTU.
- ALMPs (Benchmark column (1)):
  - Training: -0.666* (standard error 0.375) — training appears significant in benchmark specification.
  - Start-up incentives: -3.411** (standard error 1.648) — start-up incentives significant and large in magnitude.
  - Employment incentives: -0.180 (standard error 0.373) — not significant in benchmark.
  - Supported employment and rehabilitation: 0.578 (standard error 0.582) — not significant.
  - Direct job creation: -0.280 (standard error 0.797) — not significant.
  - Unemployment benefits: 0.020 (standard error 0.130) — not statistically significant in full multivariate regression.
- Employment Protection (EPL): coefficient 0.005 (standard error 0.035) — not statistically significant.

### Robustness checks and sample splits
- Dynamic panel (Arellano and Bond (1991), Table 4 column (2)) yields similar qualitative results: LTU associated with skill mismatches, matching efficiency, and some ALMPs.
- Pooled OLS (Table 4 column (3)) produces results similar to benchmark and dynamic panel, although training is not statistically significant in pooled OLS.
- NMS sample (Table 4 column (4)): qualitative results broadly unchanged, but the size of coefficients on economic growth, matching efficiency, and ALMPs becomes larger for the NMS.
- Note on endogeneity: dynamic estimation used to account for correlation of lagged dependent variable with error term.

### Evidence from literature on ALMP effectiveness
- Meta-analysis and studies cited:
  - Card et al. (2018): long-term unemployed benefit more from ALMPs, especially those emphasizing human capital improvement.
  - Training measures: positive medium- and long-run results (Card et al. 2010, 2018), conditional on alignment with labor market demand.
  - Start-up incentives: effective in reducing unemployment and increasing employment in advanced economies, particularly for vulnerable and low-skilled people (Rodriguez-Planas, 2010; Escudero, 2018); associated with a “double dividend” when subsidized firms create more jobs in the future (Caliendo and Künn, 2011).
  - Direct job creation: generally found to be the least effective measure.
- Program design matters: program length, monetary incentives, individualized follow up, activity targeting, early identification and intervention, and small-scale well-targeted programs increase effectiveness (Levy et al., 2019; Meager and Evans, 1998; Duell, 2012; Caliendo and Schmidl, 2016; Kluve, 2019).

### Labor market policy interventions and spending patterns
- Categories (Eurostat): (i) labor market services (PES and “services and sanctions”); (ii) passive labor market policies (PLMPs — unemployment benefits and income supports); (iii) active labor market policies (ALMPs — training, employment incentives, start-up incentives, supported employment and rehabilitation, direct job creation).
- EU average LMP spending in 2017: 1.3 percent of GDP.
- Cross-country range in 2017: 0.1 percent of GDP in Romania to 3 percent of GDP in Denmark.
- In general, LMP spending in advanced economies was twice as large as those in the NMS.
- Allocation shares (2017, EU average):
  - PLMPs: about one-half of LMP interventions spending (mostly unemployment benefits).
  - ALMPs: about one-third of LMP interventions spending.
  - PES: remaining share (smaller than PLMPs and ALMPs on average).
- Within ALMPs (2017, EU average):
  - Employment incentives: about one-third of ALMP spending.
  - Training programs: about 30 percent of ALMP spending.
  - Rehabilitation and direct job creation: remaining shares (smaller).
  - Start-up incentives: generally less than 5 percent of ALMP spending in most countries, with exceptions in Spain, Poland, Sweden, and Croatia.
- Country-level variations:
  - Cyprus and Bulgaria allocated more than three-quarters of LMP resources to PLMPs.
  - Hungary and Malta: PLMP shares about 20 percent and 15 percent of total LMP expenditure, respectively.
  - Czech Republic, Denmark, Croatia, Lithuania, Luxembourg, Hungary, and Poland prioritized ALMPs.
  - Malta spent the largest share of LMP on labor market services (PES).
  - Bulgaria, Greece, Hungary, and Slovenia prioritized direct job creation within ALMPs.
  - Austria, Germany, France, Croatia, Latvia, Portugal, and Finland prioritized training programs within ALMPs.

### Policy implications for ALMPs (selected)
- Ensure adequate spending on effective ALMPs; reallocate from less effective measures (e.g., direct job creation) toward training programs and start-up incentives where appropriate.
  - Example: Bulgaria — low ALMP spending and high LTU incidence; direct job creation received the largest allocation, suggesting scope to increase overall ALMP envelope and shift toward training and start-up incentives.
- Apply coherent and comprehensive approaches:
  - Combine different types of ALMPs (training, counselling, subsidies) to support various groups of LTU, including vulnerable and discouraged workers.
  - For high LTU countries: widespread ALMPs plus additional measures for disadvantaged groups.
  - For low LTU countries: more tailored programs for disadvantaged groups.

### Skill mismatches and digital skills
- Measurement: mismatch indicator constructed as differences between employment and working-age population by education groups (following Estevao and Tsounta (2011)), since vacancies by education level are not available.
- Cross-country patterns:
  - High skill mismatches: Belgium, Ireland, Bulgaria.
  - Low skill mismatches: Netherlands, Portugal, United Kingdom.
  - Post-GFC: about half of EU member states experienced improvement in skill matching; largest improvements in many NMS; deterioration in many advanced countries (Ireland, Greece, Spain, Portugal).
- Digital skills:
  - Digital skills increasingly required; almost half of vacancies related to ICT, healthcare, science, and business and retail sectors per Cedefop (2018).
  - ESJS 2014: more than 85 percent of EU employers required at least basic ICT skills to perform the job.
- Digital skill gaps (measure: difference between percent of employees that need at least basic ICT skills and percent of individuals who have at least basic and overall digital skills):
  - All countries except the Netherlands experienced skill gaps of at least 15 percent.
  - Digital skill gaps in Bulgaria, Croatia, and Italy were as high as 50 percent.
- Policy implications to close digital skill gaps:
  - Create a digital-friendly economy: invest in telecommunication networks and access to internet; promote digitalization of businesses to increase demand and internal training.
  - Integrate technology into education and training:
    - Education system examples: United Kingdom programming lessons from age five; Sweden integrated digital education into compulsory subjects; Denmark integrated ICT into examinations; Norway national digital skills evaluation for students; Singapore’s ICT Master Plans for Education including infrastructure, teacher training, and curriculum integration.
    - Adult training examples: Denmark prioritized digital skills training for employees; Luxembourg offered affordable one-on-one and group training for low-income individuals and elderly; Sweden, New Zealand, and Singapore cooperated with industries to identify digital skill needs and promote training.

*Source: wpiea2020168-print-pdf - 23.      Full multivariate regression shows skill mismatches, matching efficiency, and*

### 41.      Labor market matching efficiency in EU generally declined after the

### wpiea2020168-print-pdf - 41.      Labor market matching efficiency in EU generally declined after the

### Evolution of matching efficiency in EU
- Most EU countries experienced declines in matching efficiency immediately after the financial crisis (GFC).
- Recovery patterns differed across countries:
  - Improved to pre-crisis levels: Czech Republic, Estonia.
  - Continued decline after GFC: Bulgaria, Greece, Ireland, Lithuania, Romania, United Kingdom, Spain, Slovenia.
- The matching efficiency in Europe has declined relative to the pre-GFC level.

### Role of public employment services (PES) to enhance matching efficiency
- PES can facilitate matching jobseekers and employers and help reduce long-term unemployment (LTU) by addressing structural factors and institutional inefficiencies.
- Features of successful PES programs (based on EU peer-to-peer review and G20 case studies):
  - Targeted and personalized programs combining several interventions:
    - Initial profiling to categorize jobseekers prior to designing interventions (example: Germany’s “Kodiak” skills assessment profiling tool; Denmark’s Employability Profiling Toolbox).
    - Step-by-step interventions for the LTU: strengthen basic skills and coaching → workplace-oriented training → vocational training → job search assistance.
    - Employer internships encouraged by allowances in Austria, Belgium, Portugal, and the United Kingdom for those registered with PES.
  - Strong institutional cooperation:
    - PES collaboration with municipalities, youth and family services, other social services, and employers.
    - Recommendation: establishment of a single point of contact for the long-term unemployed to provide individualized, tailored guidance (European Commission, 2016).
    - Outsourcing specialized services to PES partners for specific groups (examples: France for youth, Poland for young people under 25 who dropped out of school).
  - Post-employment support services:
    - Follow-up support after placement to sustain integration (example programs in Bulgaria and Germany with six months of follow-up services).
    - Services include addressing family-job compatibility, employer-coordinated pre-employment training, and prevention/resolution of early employment risks.
    - Controlled-experiment evaluation found retention rates significantly improved for the treatment group (European Commission, 2014).

### Key empirical findings on LTU and labor market characteristics
- Using annual data for 25 European countries over 2000–18:
  - Macroeconomic indicators:
    - LTU is persistent and counter-cyclical; LTU tends to decline as the economy expands in booms.
  - Labor market characteristics associated with high LTU:
    - Elevated skill mismatches.
    - High share of NEET.
    - Declining labor market matching efficiency.
- ALMPs and program effectiveness:
  - Active Labor Market Policies (ALMPs) have a significant impact on LTU in the multivariate analysis; unemployment benefits do not.
  - Among ALMPs, training programs and start-up incentives are effective tools to alleviate LTU.

### Policy measures to reduce LTU (recommended)
- Ensure adequate spending on effective ALMPs:
  - The average EU spending on ALMPs has declined compared to the pre-GFC levels.
  - Successful ALMPs typically combine different activation policies (such as training and wage subsidies).
- Address digital skill gaps:
  - Technological advancement and digitalization increase demand for digital skills; COVID-19 pandemic emphasizes this need.
  - Policy measures: invest in digital infrastructure, promote digitalization of businesses, integrate digital-skill development into school curriculum and training programs.
- Promote labor market matching efficiency:
  - Strengthen the role of PES, promote policy coordination among stakeholders, offer tailored programs for the LTU, and conduct post-monitoring after job placement.

### Data and measurement of matching efficiency
- Matching efficiency estimated using a Cobb-Douglas matching function:
  - m_i = μ_i u_i^α v_i^(1−α)
  - Job finding rate f_i = μ_i θ_i^(1−α), where θ ≡ v/u.
  - In steady state: μ_i = s_i (1−u_i)/(u_i θ_i^(1−α)), allowing computation of μ after estimating α.
- Data sources and sample:
  - Quarterly unemployment rate, unemployment, and job vacancy rate for 23 European countries from Eurostat.
  - Construction of job finding and separation rates using short-term unemployment and total unemployment.
- Summary statistics (selected):
  - Incidence of LTU: Number of observations 494; Mean 0.425; Standard Deviation 0.132; Min 0.095; Max 0.763.
  - GDP growth rate: Number of observations 589; Mean 0.051; Standard Deviation 0.066; Min -0.304; Max 0.349.
  - NEET: Number of observations 557; Mean 14.649; Standard Deviation 5.455; Min 4.100; Max 32.500.
  - Skill mismatch: Number of observations 520; Mean 14.518; Standard Deviation 8.084; Min 0.772; Max 43.126.
  - Matching efficiency: Number of observations 280; Mean 0.680; Standard Deviation 0.528; Min 0.159; Max 4.615.
  - Labor market policies: Number of observations 446; Mean 0.371; Standard Deviation 0.289; Min 0.033; Max 2.129.
  - Training: Number of observations 450; Mean 0.048; Standard Deviation 0.052; Min 0.000; Max 0.315.
  - Start-up incentives: Number of observations 405; Mean 0.004; Standard Deviation 0.006; Min 0.000; Max 0.035.
  - Unemployment benefits: Number of observations 456; Mean 0.263; Standard Deviation 0.211; Min 0.021; Max 1.283.
- Worker flow and elasticity highlights (Table A1):
  - Job finding rates range from 11.9 percent in Greece to 57.5 percent in Finland.
  - Separation rates range from 1.4 percent in Slovakia to 5.2 percent in Spain.
  - Estimated matching elasticities with respect to unemployment mostly lie in the plausible range of 0.5–0.7 (cited benchmark from Petrongolo and Pissarides, 2001).
  - Selected country entries from Table A1 (Country, Sample period, Job finding rate, Separation rate, Elasticity alfa):
    - Belgium 2010Q1-2019Q1 — 29.9 — 2.2 — 0.87
    - Bulgaria 2005Q1-2019Q1 — 17.4 — 1.5 — 0.59
    - Czech Republic 2008Q1-2019Q1 — 28.3 — 1.4 — 0.82
    - Finland 2003Q1-2019Q1 — 57.5 — 5.1 — 0.80
    - Greece 2009Q1-2019Q1 — 11.9 — 2.5 — 0.73
    - Spain 2001Q1-2019Q1 — 35.4 — 5.2 — 0.30
    - Sweden 2009Q1-2019Q1 — 51.8 — 4.3 — 0.95

*Source: wpiea2020168-print-pdf - 41.      Labor market matching efficiency in EU generally declined after the; canonical URL: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020168-print-pdf.pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020168-print-pdf.pdf_