## wpiea2022198-print-pdf

## Source details

**Canonical URL:** [wpiea2022198-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022198-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022198-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022198-print-pdf.pdf.json)

---

### Overview and scope
- Regional disparities among European advanced economies declined until the Great Recession and, depending on coverage, have flatlined or increased thereafter.
- For nine large EU advanced economies and the United Kingdom the trend displays a marked increase in disparities.
- The paper focuses on risks of further economic divergence across European regions resulting from the pandemic, emphasizing structural characteristics of regions stagnating prior to the crisis and likely to face stronger pandemic-related headwinds.
- Organization:
  - Section 2: dataset and concepts.
  - Section 3: evolution before COVID-19.
  - Section 4: pandemic channels (mobility, teleworkability, routinization, green transition).
  - Section 5: green/growth tradeoff.
  - Section 6: policy mix including place-based policies.
  - Section 7: conclusions.

### Data and measurement of disparities
- Country focus: Austria, Finland, France, Germany, Greece, Italy, Netherlands, Portugal, Spain, and the U.K.
- Data sources: Eurostat (GDP, employment aggregate/sector/firm-size, population at NUTS-2); occupation-customer interaction indexes (Koren and Peto 2020); occupation-teleworkability rates (Dingel and Neiman 2020); occupation-routinization indexes (Autor and Dorn 2013); Google Community Mobility Report; European Pollutant Release and Transfer Register (E-PRTR) for regional GHG emissions.
- Measure of dispersion: unweighted square of Coefficient of Variation (CV^2) for regional GDP per capita across NUTS-2 regions.
- Mathematical decomposition preserved:
  - CV(y)^2 = 1/y^2 [1/N sum_{c=1}^{C} sum_{r=1}^{N_c} (y_rc − y)^2]
  - CV(y)^2 = sum_{c=1}^{C} (y_c / y)^2 (1/N_c N) [1/y_c^2 1/N_c sum_{r=1}^{N_c} (y_rc − y_c)^2] (CV_W term) + 1/y^2 sum_{c=1}^{C} (N_c / N) (y_c − y)^2 (CV_B term)

### Evolution of regional disparities (Anatomy)
- Total disparities declined slightly through 2006, then moved up slowly post-2006, erasing earlier gains on the eve of the Great Recession.
- Divergence accelerated after the Great Recession and was mostly driven by increased between-country gaps, although within-country disparities are consistently higher than between-country disparities across the whole period.
- 2020: disparities increased by about 0.01 points in terms of the coefficient of variation, mostly driven by between-country gaps. (2020 data excludes the UK following its exit from the EU.)
- Country-level contributions:
  - Germany and Italy have the largest disparities across their regions (substantial East/West and North/South differences).
  - Increase in between-country disparity after the Great Recession driven by relatively weaker GDP per capita growth in Greece and stronger growth in Germany.

### Components of GDP per capita
- Decomposition: y = GDP/Pop = (GDP/Emp) × (Emp/LF) × (LF/WorkAgePop) × (WorkAgePop/Pop) interpreted as labor productivity × employment-to-labor-force ratio (ELR) × labor force participation × working-age population share.
- Key findings:
  - Productivity is responsible for the largest fraction of disparities and drove up disparities in GDP per capita after the Great Recession together with the ELR.
  - Dispersion in ELR declined since 2013, signaling convergence in unemployment rates across countries in this period.
  - Within-country disparities are the most important component of dispersion for each variable except the ELR.
  - Increasing ELR dispersion explains growing GDP per capita disparities after the Great Recession, driven by decreasing unemployment rates in Germany versus significant job losses in other countries (especially Greece, Spain, Portugal).

### Productivity and labor markets
- Regional productivity gaps (GDP per employee) show considerable disparities within high-productivity countries: Germany, the Netherlands, and France.
- Low productivity levels in Greece and Portugal contribute to more than half of between-country disparities.
- Portugal’s productivity has been slightly converging to the mean; Greece’s productivity has diverged sharply from the mean.
- ELR and participation:
  - Sharp increase in between-country disparities in the ELR after the Great Recession.
  - Dispersion in labor force participation increased until 2009 then mildly converged.
  - Within-country ELR disparities were higher in Italy and Germany in 2000; Germany reduced dispersion over the next 15 years while Italy’s contribution changed little.
  - Significant drops in ELR in Spain and Greece explain substantial increase in between-country ELR CV^2 after the Great Recession.

### Beta-convergence (econometric evidence)
- Specification: Δy_r,t = α + β y_r,t−1 + ε_r,t. Negative β implies convergence.
- Periods: Pre-Great Recession (2000–08) and Post-Great Recession (2008–18). Estimations run without and with country-fixed effects.
- Selected estimation highlights (Table 1 excerpts preserved exactly as presented):
  - Panel A (without country-fixed effects):
    - Pre-GR GDP per capita: β = −0.008 ** (0.003) indicating β-convergence.
    - Post-GR GDP per capita: β = 0.011 *** for Δy (indicating divergence).
    - ∆prod pre-GR: β = −0.016 *** (0.004); post-GR: β = 0.0004 (0.007).
    - ∆ELR pre-GR: β = −0.082 *** (0.007); post-GR: β = 0.024 (0.017).
    - ∆was pre-GR: β = −0.044 *** (0.006); post-GR: β = −0.009 (0.006).
    - No of Obs: 168 for each regression in Panel A.
    - R^2 values vary across regressions (examples: 0.027, 0.053, 0.072, 0.0001, 0.693).
  - Panel B (with country-fixed effects):
    - Pre-GR within-country GDP per capita: β = −0.008 *** (0.003).
    - Post-GR within-country GDP per capita: β = −0.003 (0.002) (insignificant).
    - ∆prod post-GR with fixed effects: β = −0.014 *** (0.004).
    - ∆ELR pre-GR with fixed effects: β = −0.086 *** (0.007).
    - No of Obs: 168 for each regression in Panel B.
    - R^2 values with fixed effects are substantially higher (examples: 0.706, 0.805, 0.587, 0.756, 0.778, 0.821).
  - Interpretation: between-country rather than within-country divergence in the 10 countries after the crisis.
- Robustness: results robust to using 2010 rather than 2008 as reference (Appendix Table A1).

### Potential impact of COVID-19 on disparities — analytical channels
- Channels analyzed: mobility shocks (Google Community Mobility Report), teleworkability (Dingel and Neiman 2020), customer-interaction intensity (Koren and Peto 2020), exposure to routinization/automation (Autor and Dorn 2013).
- Taxonomy: regions classified into four groups based on 2003 GDP per capita relative to regional median and growth 2002–16 relative to country average:
  - Low-income/low-growth (Lagging).
  - Low-income/high-growth (Catching up).
  - High-income/low-growth (Striving).
  - High-income/high-growth (Thriving).

### Mobility as shock intensity
- Google Community Mobility Report (daily for NUTS-2 since February 2020) used as proxy for shock intensity; measures referenced: “Workplace” and “Retail Stores.”
- Finding: Low, Lagging regions experienced a larger decline in mobility during the first pandemic wave (around Retail stores and Workplaces), suggesting larger economic contraction in those regions; mobility differences largely disappeared by June 2020.
- Economic disruption from mobility changes depends heavily on regional teleworkability.

### Customer-interaction intensity
- Occupation-customer interaction index indicates regions specialized in high customer-interaction occupations face larger activity losses from declines in mobility.
- Empirical regularity: negative association between level of economic development and prevalence of occupations with higher customer interaction intensity; low-income regions tend to have a higher share of high customer-interaction jobs.

### Short-term risks: teleworkability — methods and key facts
- Occupation-level teleworkability rates from Dingel and Neiman (2020) at 2-digit ISCO 2008 occupations used; OECD PIAAC used to compute occupation shares within sectors and countries.
- Imputation assumptions:
  - Teleworkability by occupation is the same across countries (Dingel and Neiman U.S. classifications applied elsewhere).
  - Occupation-sector matrix computed at country level is the same across regions within country.
- Data caveat: Portugal not in PIAAC — Portuguese regions excluded from teleworkability, customer intensity, and routinization analyses.
- Sectoral patterns:
  - Low teleworkability: agriculture, construction, retail trade and wholesale, transportation and other services.
  - High teleworkability: information and communication; financial and insurance activities.
- Cross-region teleworkability facts:
  - Positive correlation between regional teleworkability rates and GDP per capita: richer regions have higher share of jobs performable from home.
  - Convergence in teleworkability between 2000 and 2018:
    - Teleworkability rate at the 10th percentile increased by 1.6 percent.
    - Teleworkability rate at the 90th percentile increased by 0.8 percent.
    - Teleworkability rates were 30.5 percent and 41 percent at the 10th and 90th percentiles in 2018, respectively.
  - Interpretation: poorer regions have lower teleworkability and COVID-19 may exacerbate short-run regional disparities; convergence process is slow.

### Longer-term risks: automation (routinization)
- Index of jobs at risk from automation (RTI) computed using Autor and Dorn (2013) routinization measures with crosswalks to contemporary occupational classifications.
- RTI definition: RTI_{l,t} = ln(T_{l,t}^R) − ln(T_{l,t}^M) − ln(T_{l,t}^A).
- Sectoral findings:
  - Agriculture: lowest routinization index.
  - Manufacturing: high routinization index; jobs at high risk of automation.
- Cross-region patterns:
  - Richer regions have more jobs at risk from automation (positive relationship between RTI_r and GDP per capita).
  - Temporal changes 2000–2018:
    - Negative correlation between initial routinization index in 2000 and change over 18 years: high-initial-routinization regions saw decreases; low-initial ones saw increases.
    - Greek regions showed highest increases in routinization (agriculture share decreased, manufacturing share increased).
    - Most U.K. regions and half of French, German, Italian regions saw negative change (workers shifted to sectors with lower routinization).
  - Group-level summary: ex-ante automation vulnerabilities are lower in lagging regions, though increasing routinization trends in some lagging regions may raise future vulnerabilities.

### Disparities and the green transition (GHG emissions analysis)
- Data: E-PRTR for large industrial facilities (available after 2007); GHG emission intensity = GHG emissions per unit of value added in Energy and Manufacturing Industry.
- Coverage note: In 2016, E-PRTR reports by the 10 countries covered about 98 percent of direct energy production and manufacturing industry emissions; minimum country coverage was 72 percent.
- Empirical findings:
  - Slightly positive correlation between GDP per capita and GHG emission levels across regions.
  - Negative relationship between income level and GHG emission intensity.
  - Changes 2007–2016: positive but insignificant relationship between GHG growth and GDP per capita; same for GHG intensity growth.
  - Convergence 2007–2016: emissions (and intensity) fell more in regions that had higher levels in 2007.
  - Group-level: Low, Catching-up regions have high GHG emissions growth and high dependence on carbon-inefficient sectors (agriculture and industry); green-transition policies may slow growth in these regions absent compensating measures to accelerate resource reallocation.

### Policy discussions — place-based policies and smoothing transitions
- Role and forms of place-based policies: business tax incentives, cash grants to businesses and households, public services (customized job training, business development centers), infrastructure investment.
- Effectiveness depends on context and region; more effective when:
  - Lagging regions are more densely populated, labor is less mobile, administrative capacity or political accountability is stronger (Gbohoui et al., 2019).
  - Regions have higher human capital and high-quality local government (Ehrlich and Overman, 2020).
  - Regions have high involuntary unemployment and large agglomeration potential (Bartik 2020).
- EU evidence:
  - Transfers foster growth in recipient low-income regions and reduce disparities on average (Becker, Egger and Ehrlich, 2010), though effectiveness might be declining due to lower labor mobility (Gbohoui et al., 2019).
  - EU funds concentrate in transport infrastructure and local public goods/services; 60 percent of total budget goes to less developed areas for transportation, R&D and business support (Ehrlich and Overman, 2020).
- Teleworkability and connectivity policy implications:
  - Future EU policies could tackle uneven connectivity; digitalization is important for growth and resilience.
  - Potential measures: tax incentives for firms to invest to ease teleworking; incentives to improve high-speed internet access.
  - Spatially targeted policies to boost digital labor skills and broadband access are recommended (Gbohoui et al. 2019).
- Smoothing green transition:
  - Trade-offs: carbon pricing can weaken growth and displace workers but can also generate “green growth” via public and private green investment; net effects depend on substitution between high-emission and low-emission activities.
  - IMF (2020) simulations: jointly implemented carbon pricing and fiscal stimulus can have positive impact on global employment, but low-income households may face labor income losses.
  - Empirical literature:
    - Dechezlepretre and Sato (2017): environmental regulation costs on firms are very small and concentrated in very energy-intensive sectors.
    - Metcalf and Stock (2020): lack of evidence of adverse macroeconomic effects from a carbon tax (31 European countries).
    - Marin and Vona (2019): climate policies favored technicians and worked against manual workers in 14 European countries (1995–2011).
  - Case study lesson (Ruhr Valley and Saarland): no single “silver bullet”; combine policies addressing unemployment and attracting investment with measures to improve infrastructure, education, research facilities and soft location factors; short-term employment gains may not yield durable diversification.
- Research gaps: more research needed on how to ensure smooth green transition across regions and on precise design of place-based policy mixes (connectivity, education, infrastructure, incentives).

### Conclusions — synthesis of main findings
- Pre-Great Recession: evidence of regional convergence in GDP per capita across analyzed NUTS-2 regions, driven by productivity, ELR, and working-age population share.
- Post-Great Recession: divergence accelerated, driven mainly by between-country gaps; within-country convergence weakened and became insignificant after the crisis.
- COVID-19 risks exacerbating regional disparities through mobility shocks, uneven teleworkability, differential customer-interaction intensity, and automation exposure.
- Green transition risks: low, catching-up regions often have high GHG emissions growth and dependence on carbon-inefficient sectors; policies must prioritize smoothing transition and place-based support to avoid leaving regions behind.
- Policy priorities highlighted:
  - Increase teleworkability across regions to smooth short-term pandemic shocks.
  - Improve regional resilience to automation to manage longer-term displacement risks.
  - Use spatially targeted measures to reduce disparities and ensure inclusive green growth.
- Additional key statistics and facts preserved from the source:
  - 2020 disparity increase: about 0.01 points in CV.
  - Teleworkability percentiles in 2018: 10th percentile = 30.5 percent; 90th percentile = 41 percent.
  - Sample regressions: Pre-GR GDP per capita β = −0.008 ** (0.003); Post-GR Δy β = 0.011 *** (Panel A); within-country Post-GR β = −0.003 (0.002) (Panel B).
  - No of Obs in regressions: 168.
  - E-PRTR coverage in 2016: about 98 percent of direct energy production and manufacturing industry emissions across the 10 countries; minimum country coverage 72 percent.

*Source: wpiea2022198-print-pdf - Introduction*

### Introduction ...........................................................................................................

### Introduction

### Overview
- Regional disparities among European advanced economies (aggregating both within and across) declined until the Great Recession and, depending on the coverage, have flatlined or increased thereafter.
- For nine large EU advanced economies and the United Kingdom the trend displays a marked increase in disparities.
- The COVID-19 pandemic renewed interest in the connection between large economic downturns and disparities.
- The paper (structure summarized below) offers novel insights on risks of further economic divergence across European regions resulting from the pandemic, focusing on structural characteristics of regions stagnating prior to the crisis and likely to face stronger pandemic-related headwinds.

### Scope and Organization
- First part: documents evolution of disparities in GDP per capita across NUTS-2 regions; decomposes disparities into within- and between-country components; separates drivers into employment, labor force participation, productivity, etc.
- Second part: focuses on COVID-19 pandemic impacts (short, medium, long run) through channels including mobility declines, teleworkability, routinization exposure, and the green transition.
- Final part: discusses place-based policies and policy mixes to reduce disparities and smooth green transition effects.
- Paper organization: Section 2 dataset and concepts; Section 3 evolution before COVID-19; Section 4 pandemic channels; Section 5 green/growth tradeoff; Section 6 policy mix incl. place-based policies; Section 7 conclusions.

*Source: wpiea2022198-print-pdf - Introduction*

### Key framing statements
- Regions with low GDP per capita but high growth have traditionally been dependent on carbon-inefficient sectors in production and employment; careful policy focus is needed to ensure smooth transitions.
- Post-pandemic policies should focus on increasing teleworkability of regions and improving vulnerabilities to automation.

---

### Data and Measurement of Disparities
- Data sources: Eurostat (GDP, employment aggregate/sector/firm-size, population at NUTS-2); occupation-customer interaction indexes (Koren and Peto 2020); occupation-teleworkability rates (Dingel and Neiman 2020); occupation-routinization indexes (Autor and Dorn 2013); Google Community Mobility Report (daily regional mobility); European Pollutant Release and Transfer Register (E-PRTR) for regional GHG emissions.
- Country focus: nine EU countries (Austria, Finland, France, Germany, Greece, Italy, Netherlands, Portugal, Spain) and the U.K.
- Measure of dispersion: unweighted square of Coefficient of Variation (CV^2) for regional GDP per capita across regions r (NUTS-2). CV^2 is scale-free and decomposable into within- and between-country components.

Mathematical decomposition preserved from source:
- CV(y)^2 = 1/y^2 [1/N sum_{c=1}^{C} sum_{r=1}^{N_c} (y_rc − y)^2]
- CV(y)^2 = sum_{c=1}^{C} (y_c / y)^2 (1/N_c N) [1/y_c^2 1/N_c sum_{r=1}^{N_c} (y_rc − y_c)^2] (CV_W term) + 1/y^2 sum_{c=1}^{C} (N_c / N) (y_c − y)^2 (CV_B term)

---

### Anatomy of Regional Disparities

### The Evolution of Disparities Since the Early 2000s
- Total disparities across regions declined slightly through 2006, then started moving up slowly post-2006, erasing earlier gains on the eve of the Great Recession.
- Divergence accelerated after the Great Recession and was mostly driven by increased between-country gaps, although within-country disparities are consistently higher than between-country disparities across the whole period.
- During 2020, disparities increased by about 0.01 points in terms of the coefficient of variation, mostly driven by between-country gaps. (2020 data excludes the UK following its exit from the EU.)
- Country contributions:
  - Germany and Italy have the largest disparities across their regions (substantial East/West and North/South differences).
  - Increase in between-country disparity after the Great Recession driven by relatively weaker GDP per capita growth in Greece and stronger growth in Germany.

### Components of GDP per Capita
- GDP per capita decomposed (using EUROSTAT NUTS-2 data) as:
  - y = GDP/Pop = (GDP/Emp) × (Emp/LF) × (LF/WorkAgePop) × (WorkAgePop/Pop)
  - Interpreted as: labor productivity × employment-to-labor-force ratio (ELR, or one minus unemployment rate) × labor force participation rate × working-age population share.
- Findings:
  - Productivity is responsible for the largest fraction of disparities and drove up disparities in GDP per capita after the Great Recession together with the ELR.
  - Dispersion in ELR declined since 2013, signaling convergence in unemployment rates across countries in this period.
  - Within-country disparities are the most important component of dispersion for each variable except the ELR.
  - Increasing ELR dispersion explains growing GDP per capita disparities after the Great Recession, driven by decreasing unemployment rates in Germany versus significant job losses in other countries (especially Greece, Spain, Portugal).

### Productivity
- Regional productivity gaps (GDP per employee) show considerable disparities within high-productivity countries: Germany, the Netherlands, and France.
- Low productivity levels in Greece and Portugal contribute to more than half of between-country disparities.
- Portugal’s productivity has been slightly converging to the mean; Greece’s productivity has diverged sharply from the mean.

### Labor Markets
- There is a sharp increase in between-country disparities in the ELR.
- Dispersion in labor force participation rates:
  - Increased steadily until 2009 in both between and within components.
  - Mild convergence followed until recently.
- Country contributions to ELR dispersion:
  - Within-country ELR disparities were higher in Italy and Germany in 2000.
  - Germany reduced dispersion across regions over the next 15 years; Italy’s contribution changed little.
  - ELR in Italy, Finland, France, and Portugal hovered around the mean across 10 countries.
  - Significant drops in ELR in Spain and Greece explain substantial increase in between-country ELR CV^2 after the Great Recession.

---

### Beta-convergence (Econometric Analysis)

### Method
- Standard Beta-convergence specification:
  - Δy_r,t = α + β y_r,t−1 + ε_r,t
  - Negative β implies regions with lower initial y experienced faster growth → convergence.
- Analysis periods: Pre-Great Recession (2000–08) and Post-Great Recession (2008–18).
- y represents GDP per capita and components; estimations run without and with country-fixed effects to distinguish between- and within-country convergence.

### Main estimation results (Table 1 highlights)
- Panel A (without country-fixed effects):
  - Pre-GR: β-convergence in GDP per capita statistically significant (β = −0.008 **; standard error (0.003)), with contributions from convergence in productivity, ELR, and working-age population share.
  - Post-GR: results point to sharp divergence in GDP per capita across regions in the 10 countries (β = 0.011 *** for Δy, indicating divergence).
  - Other entries (examples preserved exactly as in table):
    - For ∆prod pre-GR: β = −0.016 *** (0.004); post-GR: β = 0.0004 (0.007).
    - For ∆ELR pre-GR: β = −0.082 *** (0.007); post-GR: β = 0.024 (0.017).
    - For ∆was (working-age share) pre-GR: β = −0.044 *** (0.006); post-GR: β = −0.009 (0.006).
  - No of Obs: 168 for each regression in Panel A.
  - R^2 values vary across regressions (examples: 0.027, 0.053, 0.072, 0.0001, 0.693, etc. as tabulated).

- Panel B (with country-fixed effects):
  - Pre-GR: within-country convergence in per capita GDP suggested (β = −0.008 *** (0.003)).
  - Post-GR: within-country convergence in GDP per capita becomes insignificant (β = −0.003 (0.002)).
  - Examples from table:
    - ∆prod post-GR with fixed effects: β = −0.014 *** (0.004).
    - ∆ELR pre-GR with fixed effects: β = −0.086 *** (0.007).
    - No of Obs: 168 for each regression in Panel B.
    - R^2 values with fixed effects are substantially higher (examples: 0.706, 0.805, 0.587, 0.756, 0.778, 0.821, etc.).
  - Interpretation: between-country rather than within-country divergence in the 10 countries after the crisis.

### Robustness note
- Findings are robust to using 2010 rather than 2008 as a reference point (around the European Sovereign Debt Crisis rather than the Great Recession; see Appendix Table A1).

---

### Potential Impact of the COVID-19 Pandemic on Existing Disparities

### Analytical approach
- Focus on channels: mobility shocks (Google Community Mobility Report), teleworkability (Dingel and Neiman 2020), customer-interaction intensity (Koren and Peto 2020), and exposure to routinization/automation (Autor and Dorn 2013).
- Taxonomy classifies regions into four categories based on initial GDP levels and GDP growth rates between 2002 and 2016 (classification introduced here; taxonomy used to highlight differences across regions).

### Mobility as shock intensity
- Google Community Mobility Report used as proxy for shock intensity across European regions: changes in mobility reflect both lockdown restrictions and autonomous decisions to travel less.
- Economic disruption from mobility changes depends heavily on regional teleworkability.

### Customer Interaction Intensity
- Occupation-customer interaction indexes used to identify regions more exposed to reductions in face-to-face interactions and service demand.

### Short-term Risks: Teleworkability
- Regional teleworkability rates calculated following Dingel and Neiman (2020).
- Low regional teleworkability acts as a shock transmitter: where remote work was less feasible, economic consequences of mobility declines were larger.
- Post-pandemic policy implication highlighted earlier: increase teleworkability of regions to mitigate asymmetric shocks.

### Longer-term Risks: Automation
- Regional exposure to routinization/automation indexes calculated following Autor and Dorn (2013).
- Automation exposure creates longer-term risks of asymmetric impacts across regions depending on industry and occupation composition.

---

### Disparities and the Green Transition
- Close data inspection suggests a potential green/growth tradeoff for certain regions.
- Regions with low GDP per capita but high growth often depend on carbon-inefficient sectors in production and employment.
- Policy emphasis needed to ensure these regions transition smoothly to greener and more sustainable growth, avoiding leaving regions behind as climate mitigation policies induce sectoral transformation.
- Appendix C provides a micro-level case study of two German regions to analyze potential green/growth tradeoffs (case study noted in structure).

---

### Policy Discussions

### Is There a Role for Place-based Policies?
- The paper discusses benefits of spatially-targeted (“place-based”) policies on lagging regions to reduce disparities and ameliorate adverse pandemic effects.
- Emphasis on holistic approaches to ensure no region is left behind in climate mitigation and technological transitions.

### Policy priorities highlighted
- Increase teleworkability across regions to smooth short-term pandemic shocks.
- Improve regional resilience to automation to manage longer-term displacement risks.
- Smooth effects of green transition across regions via targeted, place-based interventions and support for sectoral transformation.

---

### Conclusions
- Pre-Great Recession: evidence of regional convergence in GDP per capita across the analyzed NUTS-2 regions, driven by productivity, ELR, and working-age population share.
- Post-Great Recession: divergence accelerated, driven mainly by between-country gaps; within-country convergence weakened and became insignificant after the crisis.
- COVID-19 threatens to exacerbate regional disparities through mobility shocks, uneven teleworkability, differential customer-interaction intensity, and automation exposure.
- Green transition poses risks for low-GDP-but-high-growth regions dependent on carbon-intensive production and employment; policy must prioritize smooth transitions and place-based support.
- Post-pandemic policy mix should focus on teleworkability enhancements, automation vulnerability mitigation, and spatially-targeted measures to reduce disparities and ensure inclusive green growth.

*Source: wpiea2022198-print-pdf - Introduction*

### 1. Low-income/low-growth (Lagging) regions are defined as those whose GDP per capita in 2003 was lower

### wpiea2022198-print-pdf - 1. Low-income/low-growth (Lagging) regions are defined as those whose GDP per capita in 2003 was lower 

### Definitions of regional groups
- Low-income/low-growth (Lagging) regions: GDP per capita in 2003 was lower than the regional median and growth was lower than the country’s average from 2002–16.
- Low-income/high growth (Catching up) regions: GDP per capita below the regional median, but growth rate faster than the country’s average from 2002–16.
- High-income/low growth (Striving) regions: GDP per capita in 2003 higher than the regional median and growth lower than the country’s average from 2002–16.
- High-income/high growth (Thriving) regions: GDP per capita above the regional median, and growth rate higher than the country’s average from 2002–16.

### High-frequency mobility evidence (Google Community Mobility Index)
- Data: daily for NUTS-2 regions since February 2020; daily index compared to January 3 and February 6 period.
- Mobility measures used: “Workplace” and “Retail Stores.”
- Findings:
  - Low, Lagging regions experienced a larger decline in mobility during the first pandemic wave (around Retail stores and Workplaces), suggesting larger economic contraction in those regions.
  - The mobility difference between Low, Lagging regions and High-income regions largely disappeared by June 2020.
- Notes: Figures referenced show weekly mobility starting from February 15 for each regional group.

### Customer interaction intensity and its implications
- Measure: occupation-customer interaction index from Koren and Peto (2020).
- Mechanism: regions specialized in occupations with higher customer interaction intensity (e.g., counselors, social workers, salespersons) face larger expected activity loss due to lockdowns.
- Empirical regularity:
  - Negative association between level of economic development and prevalence of occupations with higher customer interaction intensity (Figure 10).
  - Low-income regions tend to have a higher share of high customer-interaction jobs, implying larger negative impacts from declines in mobility.
- Teleworkability evidence in next section supports these conclusions.

### Short-term risks: Teleworkability (methods and findings)
- Data sources and imputation:
  - Occupation-level teleworkability rates from Dingel and Neiman (2020) at 2-digit ISCO 2008 occupations.
  - OECD PIAAC data used to compute share of each 2-digit ISCO occupation in each 1-digit sector for each country: weights w_{l,k,c} = Emp_{l,k,c} / sum_{l} Emp_{l,k,c}.
  - Sector-country teleworkability rates: telework_{k,c} = sum_{l} w_{l,k,c} × telework_l.
  - Sector-region employment shares from EUROSTAT, w_{k,r}, combined to get regional teleworkability: teleworkable_r = sum_{k} w_{k,rk} × telework_{k,c(r)}, where w_{k,r} = Emp_{k,r} / sum_{k} Emp_{k,r}.
  - Imputation assumptions: (i) teleworkability by occupation is the same across countries (Dingel and Neiman U.S. classifications applied elsewhere); (ii) occupation-sector matrix computed at country level is the same across regions within country.
- Data caveat: Portugal not in PIAAC; analysis excludes Portuguese regions for teleworkability, customer intensity, and routinization indexes.
- Sectoral patterns (Figure 11):
  - Low teleworkability: agriculture, construction, retail trade and wholesale, transportation and other services.
  - High teleworkability: information and communication; financial and insurance activities.
- Validation: sector-country teleworkability rates slightly overpredict survey-based sectoral teleworkability but provide a successful approximation overall (Figure 12).
- Cross-region patterns:
  - Positive correlation between regional teleworkability rates and GDP per capita (Figure 13): richer regions have higher share of jobs performable from home.
  - Convergence in teleworkability between 2000 and 2018:
    - Teleworkability rate at the 10th percentile increased by 1.6 percent.
    - Teleworkability rate at the 90th percentile increased by 0.8 percent.
    - Teleworkability rates were 30.5 percent and 41 percent at the 10th and 90th percentiles in 2018, respectively.
  - Interpretation: poorer regions have lower teleworkability and COVID-19 may exacerbate short-run regional disparities; convergence process in teleworkability has a long way to go.

### Longer-term risks: Automation (methods and findings)
- Objective: index of jobs at risk from automation at NUTS-2 level using occupation-level routinization (RTI) from Autor and Dorn (2013).
- RTI definition: RTI_{l,t} = ln(T_{l,t}^R) − ln(T_{l,t}^M) − ln(T_{l,t}^A), where T^R, T^M, T^A denote routine, manual, and abstract task inputs in occupation l.
- Methodology parallels teleworkability computation:
  - Occupation-routinization index from Autor and Dorn (2013) for 3-digit OCC1990 occupations.
  - Two crosswalks map OCC1990 → OCC2000 → SOC2010.
  - Follow steps 1–4 from teleworkability section to compute sector-country routinization indexes RTI_{k,c} and regional RTI_r.
- Sectoral findings (Figure 15):
  - Agriculture: lowest routinization index.
  - Manufacturing: high routinization index; jobs at high risk of automation.
  - Sectoral routinization indexes similar to Das and Hilgenstock (2018).
- Cross-region patterns:
  - Plot of regional RTI_r against GDP per capita (Figure 16): richer regions have more jobs at risk from automation.
  - Implication: longer-term automation risks are lower for poorer regions than for richer regions.
  - Temporal changes (2000–2018) (Figure 17):
    - Negative correlation between initial routinization index in 2000 and change over 18 years: regions with high initial routine intensity saw decreases; those with low initial saw increases.
    - Sectoral shifts drive changes: Greek regions showed highest increases (agriculture share decreased, manufacturing share increased); most U.K. regions and half of French, German, Italian regions saw negative change (workers switched to sectors with lower routinization).
  - Group-level summary (Figure 18):
    - Speed of convergence in teleworkability rate is low.
    - Ex-ante vulnerabilities from automation are lower in lagging regions, though increasing routinization trends in these regions may shape future vulnerabilities.

### Disparities and the Green Transition (GHG emissions analysis)
- Data: GHG emissions levels annually for each region using E-PRTR data (large industrial facilities), available after 2007; GHG emission intensity = GHG emissions per unit of value added in Energy and Manufacturing Industry (least carbon-efficient sectors).
- Data coverage note: In 2016, emissions reported in the E-PRTR by the 10 countries covered represented about 98 percent of GHG emissions of the direct energy production and manufacturing industries emissions; minimum country coverage was 72 percent.
- Empirical findings:
  - Weak and mixed associations between economic indicators and GHG emissions levels:
    - Slightly positive correlation between GDP per capita and GHG emission levels across regions (Figure 19, left).
    - Negative relationship between income level and GHG emission intensity (Figure 19, right).
    - Possible rationale: richer regions have high manufacturing activity (higher emissions) but higher value added leads to relatively lower GHG emissions intensity.
  - Changes 2007–2016 (Figure 20):
    - Positive but insignificant relationship between GHG growth in emissions and GDP per capita; same for GHG intensity growth.
    - No strong association between GHG emissions reductions and economic development.
  - Convergence pattern (Figure 21):
    - Between 2007 and 2016, emissions (and intensity) fell more in regions that had higher levels in 2007.
    - Possible explanations: increases in carbon taxes or higher subsidies in higher-emission regions, or base effects.
  - Group-level pattern (Figure 22):
    - Low, Catching-up regions have high GHG emissions growth and high dependence on carbon-inefficient sectors (agriculture and industry).
    - Implication: green-transition policies may slow growth in these regions unless policies include compensating mechanisms to accelerate resource reallocation.

### Policy discussions: place-based policies and smoothing the green transition
- Role for place-based policies:
  - Forms: business tax incentives, cash grants to businesses and households, public services (customized job training, business development centers, infrastructure investment).
  - Effectiveness is context- and region-dependent; more effective when:
    - Lagging regions are more densely populated, labor is less mobile, administrative capacity or political accountability is stronger (Gbohoui et al., 2019).
    - Regions have higher human capital and high-quality local government (Ehrlich and Overman, 2020).
    - Regions have high involuntary unemployment and large agglomeration potential (Bartik 2020).
  - EU evidence:
    - Becker, Egger and Ehrlich (2010): transfers foster growth in recipient low-income regions and reduce disparities on average; effectiveness might be reducing over time due to declining labor mobility (Gbohoui et al., 2019).
    - Ehrlich and Overman (2020): EU funds concentrated in transport infrastructure and local public goods/services; 60 percent of total budget goes to less developed areas for transportation, research and development and business support.
  - Policy implications given teleworkability disparities:
    - Future EU policies could tackle uneven connectivity; digitalization is important for growth and resilience.
    - Potential measures: tax incentives for businesses to invest to ease teleworking; incentives to improve high-speed internet access.
    - Spatially targeted policies boosting digital labor skills and broadband access are highlighted (Gbohoui et al. 2019).
  - Research gap: future research could analyze whether targeting some industries is beneficial in presence of disruptive events like pandemics.
- Smoothing the effects of the green transition:
  - Catching-up regions face larger sectoral shifts due to high GHG emissions growth and dependence on carbon-inefficient sectors.
  - Trade-offs of climate policies:
    - Carbon pricing can weaken growth and displace workers but also generate “green growth” through public and private green investment; net effect depends on substitution between high-emission (job losses) and low-emission (job gains) activities.
    - IMF (2020) simulations find positive impact of jointly implemented carbon pricing and fiscal stimulus on global employment, but low-income households may experience labor income losses as they are more likely employed in low-skill, carbon-intensive occupations.
  - Empirical literature:
    - Dechezlepretre and Sato (2017): costs of environmental regulations on firms are very small and concentrated on a small group of very energy-intensive industrial sectors.
    - Metcalf and Stock (2020): lack of evidence of adverse macroeconomic effects from a carbon tax (31 European countries).
    - Marin and Vona (2019): climate policies favored technicians and worked against manual workers in 14 European countries between 1995 and 2011.
  - Policy lessons from case study (Appendix C on Ruhr Valley and Saarland coal decline):
    - No single “silver bullet.”
    - Important to combine policies addressing unemployment and attracting new companies/investment with measures to improve infrastructure, education, research facilities and soft location factors.
    - Quick short-term employment gains may not lead to longer-term diversification.

*IMF Working Paper (excerpt).*

### Conclusions

### Conclusions

### Key findings on regional disparities
- Convergence had stopped between countries pre-pandemic, but was still progressing within countries (up to 2018).
- In 2020, between-country gaps were responsible for a significant increase in regional disparities (data for 2020 do not include information on the United Kingdom).
- A major part of the level of disparities across regions can be attributed to disparities in regional productivity, although the rise since the Great Recession mostly resulted from diverging unemployment rates.
- At the EA9+U.K. level, disparities in household disposable income account for about half of GDP disparities (Theil decomposition).
- In Italy and Spain, disparities in disposable income account for the largest share of GDP disparities.
- Total regional disparities in disposable income increased markedly during the Great Recession as a result of between-countries disparities, while within-countries disparities in disposable income declined.
- Over 2002–16, aggregate net transfers (defined as disposable income over market income) increased in Spain, Italy, and the U.K., were fairly stable in France, and decreased in Germany.
- The Ruhr Valley and Saarland experienced a 50-year decline in their coal industries until hard coal production in Germany stopped in 2018 with the end of subsidies.

### Pandemic channels and regional vulnerabilities
- Mobility declined everywhere, but a unit decline in mobility results in larger negative impact in lagging regions.
- The teleworkability rate is lower in lagging regions, implying greater short-term vulnerabilities.
- A lower routinization index in lagging regions suggests potentially smaller medium-term vulnerabilities relative to more advanced regions.
- Commuting and location of capital (headquarters) can bias regional measures: disparities in household market income net of depreciation appear lower than in GDP, reflecting concentration of capital in certain places where headquarters are located.

### Green transition implications
- Low, catching-up regions have relatively high GHG emissions growth and high dependence on carbon inefficient sectors (agriculture and industry).
- Policies designed to favor the green transition should incorporate provisions to smooth the potential uneven effects across regions with different sectoral compositions.
- Case study evidence from the Ruhr Valley and Saarland shows stronger declines in GHG emissions since 2007 and declines in employment shares of carbon inefficient sectors, accompanied by increases in service sector employment shares.
- Differences in regional approaches matter: Saarland’s focus on incentivizing companies (about 170 companies including the Ford Motor Company) helped absorb similar worker skills, whereas the Ruhr Valley's more difficult political and ownership context produced a slower but ultimately more diversified industrial transformation; Saarland remains more dependent on the automotive industry and thus likely less resilient to shocks.

### Policy recommendations and lessons
- EU policies could evolve to tackle uneven connectivity in an increasingly digitalized economy, for instance:
  - incentives or tax deductions for investments by firms to improve teleworkability of workers, and
  - incentives to improve high-speed internet access.
- For climate policies there is no single “silver bullet”; important elements include:
  - combine policies to address unemployment and attract new companies and investments,
  - measures to improve infrastructure, education, research facilities and soft location factors,
  - recognition that quick short-term gains in employment may not necessarily lead to durable longer-term diversification.
- Place-based policies and policies facilitating smoother green transitions may help address disparities, though more research is needed to precisely inform policy design.

### Research gaps and further work
- More research is warranted on how to ensure a smooth green transition across regions.
- Further analysis is needed to precisely inform place-based policies and the mix of measures (connectivity, education, infrastructure, incentives) that best mitigate uneven effects across regions.

*Conclusions — IMF Working Paper: Regional Disparities in Europe (WP/2022/198).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022198-print-pdf.pdf_