## wp18282 — Section III: Regional-level regression model and empirical results

## Source details

**Canonical URL:** [wp18282 — Section III: Regional-level regression model and empirical results](https://www.imf.org/-/media/files/publications/wp/2018/wp18282.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18282.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18282.pdf.json)

---

### I. Introduction and overview
- Spain’s unemployment rate history and regional disparities:
  - Unemployment rate above 8 percent during boom years and peaked at 26 percent in 2013.
  - Unemployment rate remained at 15 percent in 2018:Q3.
  - Structural unemployment estimated at around 12–16 percent.
  - Regional dispersion: differences between 90th and 10th percentile regions ranged from 8 percentage points (pre-crisis) to 18 percentage points (height of crisis); dispersion was 15 percentage points in 2017.
  - Regions in the Northeast: average unemployment rates ranged from 5 to 20 percent during 2006–17.
  - Canary Islands and the South: jobless rates around 12–32 percent during 2006–17 and more volatile outcomes.

### II. Data and stylized facts
- Data sources and coverage:
  - Regional bilateral migration flows: Residential Variation Statistics (RVS) from INE, aggregated from Municipal Registers, covering 1998–2016.
  - Working-age population defined as 16–64 years old; working-age migrants constitute "more than three quarters" of gross interregional migration flows.
  - Cross-country migration data: Eurostat Labour Force Survey, gross flows at NUTS 2 level, covering 2003–16 for 11 countries (Austria, Belgium, Finland, Germany, Greece, France, Portugal, Slovenia, Slovakia, Spain, UK). Panel is unbalanced.
  - Explanatory variables: INE, Eurostat; housing price-to-income index from OECD; Spain housing prices from Ministerio de Fomento.
- Key stylized facts for Spain:
  - Internal mobility rose in early 2000s boom and declined after 2008 crisis.
  - Immigrant mobility: rose from 6 percent in 1998 to 12 percent in 2002; share of foreigners among movers rose from about 4 percent in 1998 to 30 percent in 2008.
  - On average, interregional mobility among immigrants was around "three times higher" than that of Spaniards.
  - Working-age composition: "85 percent" of foreigners who moved were working-age vs "74 percent" for Spaniards.
  - Female immigrant share of internal movement: about "44 percent" during 1998–2005, rising to "48.5 percent" in 2016.
  - Geographic concentration: Catalonia, Madrid, Andalusia top senders and receivers in absolute terms; Balearic Islands, Catalonia, Canary Islands have highest internal migration rate relative to population.
  - Crisis-driven net flow changes: Madrid and Basque Country switched from net senders (pre-crisis) to net receivers (post-crisis); Valencia and Castile-La Mancha moved from net receivers to net senders.

### III. Regional-level regression model (gravity specification)
- Model structure:
  - Dependent variable: ln M_{ijt} = ln migration flows from origin i to destination j in year t.
  - Controls: origin and destination population (P_i, P_j), distance (DIST_{ij}), region fixed effects, time fixed effects, origin/destination economic variables X_{s,i,t}, X_{s,j,t}.
- Push and pull regional variables included:
  - Real GDP per capita.
  - Average house price (nominal).
  - Unemployment rate (lagged: t-1).
  - Real wage per worker (contemporaneous).
  - Share of low-educated workers (≤ lower secondary).
  - Share of temporary employment.
- Endogeneity and measurement notes:
  - Bilateral flows for working-age population unavailable; regressions reflect overall population mobility.
  - Unemployment uses one-year lag; wages contemporaneous.
  - Baseline regression sample restricted to 2002–2016 when including share of temporary employment.

### IV. Empirical findings — baseline regional regressions (Spain)
- Standard gravity variables (Table 1, Column 1):
  - Distance = -1.25***.
  - Population_o = -1.47**.
  - Population_d = 0.66 (not significant).
  - Real GDP per capita_o = -1.48***.
  - Real GDP per capita_d = 1.10***.
- Housing, labor market, and duality effects (Table 1, Column 1):
  - Average house price_d = -0.18**.
  - Average house price_o = 0.01.
  - Unemployment rate (t-1)_d = -0.22***.
  - Unemployment rate (t-1)_o = 0.05.
  - Real wage per worker_o = -0.99**.
  - Real wage per worker_d = -0.41** (noted as puzzling).
  - Share of temporary employment_o = -0.28**.
  - Share of low-educated_o = 0.36**.
  - Share of low-educated_d = -0.27.
- Robustness and interaction (Table 1, Column 2):
  - Interaction Real wage_d × Share of low-educated_d = -0.05***.
  - Real wage per worker_d becomes positive in this specification, suggesting negative destination wage effect driven by regions with large low-educated shares.

### V. Heterogeneity: Spaniards vs foreigners
- Spanish movers (Table 1, Column 3):
  - Mobility driven by destination pull factors:
    - Real GDP per capita_d = 1.47***.
    - Population_d = 1.57**.
    - Lower house prices and low unemployment at destination attract Spaniards.
- Foreign movers (Table 1, Column 4):
  - Mobility driven by origin push factors:
    - Real GDP per capita_o = -1.76***.
    - Real wage per worker_o = -1.00**.
    - Share of temporary employment_o = -0.50***.
  - Home-region fixed effects indicate “home bias” for both groups, generally stronger among foreigners.

### VI. Pre-crisis vs post-crisis dynamics (2002–07 vs 2008–16)
- Distance effect stable:
  - Distance ≈ -1.19*** (pre-crisis).
  - Distance ≈ -1.29*** (post-crisis).
- Changes in variable signs and significance:
  - Population, real GDP per capita, real wages in origin remain significant across episodes but larger marginal effects pre-crisis.
  - Some destination-region variables had puzzling pre-crisis signs (population_d = -3.86* pre-crisis) that revert post-crisis (population_d = 2.98*** post-crisis).
  - Share of low-educated_o: -0.13 (pre-crisis) and 0.45*** (post-crisis).
  - Share of temporary employment_o: becomes -0.48*** (post-crisis).
- Interaction with post-2013 (proxy for 2012 labor reforms) — Table 2, Column 3:
  - Post 2013 × Real wage per worker_d = 0.05***.
  - Post 2013 × Unemployment rate (t-1)_d = -0.05***.
  - Post 2013 × Share of temporary employment_o = 0.21***.
  - Post 2013 × Share of temporary employment_d = -0.10***.
  - Interpretation: 2012 labor market reforms may have strengthened the roles of labor market variables; post-2013 dummy is a proxy and may capture other concurrent changes.

### VII. Cross-country analysis — model and key findings (Table 3)
- Regression setup:
  - Dependent variable: ln LM_{ct} = labor mobility flows across regions (NUTS 2) within country c at year t.
  - Key national variables: ln GDP per capita (PPP), homeownership percent, house price-to-income ratio.
  - Labor-market controls: dispersion in regional unemployment rates, labor force education level, social protection benefit per inhabitant, labor market policy (LMP) expenditure.
  - Country and time fixed effects; clustered standard errors by country and year.
- Key coefficient results (selected, Table 3):
  - GDP per capita in PPP: -2.04* (1.02), -3.31** (1.25), -2.98** (1.19), -2.89 (1.65).
  - Home ownership: -0.05** (0.02), -0.01 (0.02), -0.01 (0.02), -0.01 (0.02).
  - House price to income ratio: 0.42 (0.49), 0.31 (0.55), 0.33 (0.65), 0.30 (0.47).
  - Social protection benefit: -0.76** (0.27), -0.91** (0.33), -0.95** (0.40).
  - Labor market policy expenditure: 0.67* (0.31), 0.73* (0.37), 0.93** (0.38).
  - Share of low-educated: -0.02 (0.02), -0.02 (0.02).
  - Dispersion in regional unemployment rates: 0.00 (0.01).
  - National unemployment rate: 0.01 (0.03).
- Model fit and sample:
  - Observations: 124, 109, 101, 109 across specifications.
  - Adjusted R-squared: 0.98, 0.98, 0.97, 0.98.
  - Adjusted within R-squared: 0.16, 0.20, 0.21, 0.21.
- Interpretations:
  - Homeownership associated with lower regional mobility in simple macro/housing specification; coefficient becomes insignificant when labor policy and benefits are included.
  - Social protection benefits reduce labor movement.
  - Higher LMP expenditure associated with greater interregional mobility.
  - Coefficients for labor force education level and unemployment dispersion have expected signs but are statistically insignificant.
  - Caveat: social protection benefits measured per inhabitant may not reflect generosity for countries with large unemployed populations.

### VIII. Spain in cross-country context (averages 2003–16)
- Comparative standings:
  - Spain had relatively higher shares of home ownership and low-educated labor force than many peer countries.
  - Spain’s spending on labor-related social benefits per capita ranked third after Belgium and Finland during 2003–16.
  - Spain’s expenditure for LMP was below the sample average.
- Policy implication: National-level policies on labor market institutions and housing market frictions are key to addressing Spain’s low regional mobility.

### IX. Policy implications and recommendations
- Spain-specific implications:
  - Housing costs: higher destination house prices discourage inflows; origin house prices can act as push factors in some periods.
  - Labor market duality: high share of temporary employment appears to hinder mobility, particularly for foreigners and post-crisis.
  - Unemployment (lagged) in destination regions consistently deters inflows.
  - Wage signals: interaction with education composition matters; destination wages attract movers when accounting for low-educated shares.
  - 2012 labor market reforms (proxied by post-2013 interactions) appear to have strengthened the roles of real wages and unemployment at destinations and altered temporary employment’s relation with migration; interpretation cautious due to proxy nature.
- Policy recommendations to promote regional labor mobility:
  - Reduce pervasiveness of labor market duality:
    - Narrow gap between costs for firms to hire permanent and temporary workers by reducing hiring costs and severance payments associated with permanent contracts.
    - Specific policy recommendations referenced in IMF Country Reports No. 17/23 and No. 17/319.
  - Facilitate wage-setting flexibility:
    - Ensure wage growth in line with productivity to promote efficient labor allocation.
    - Further decentralize wage setting and allow collective bargaining opt-out to provide flexibility.
  - Enhance incentives for geographical mobility:
    - Consider policies to provide additional incentives for workers to move (example cited: Mobility Assistance Program (MAP) in Germany effective in encouraging distant job search).
    - Consider temporary and targeted housing assistance in regions with high house price-income ratio or underdeveloped rental markets.
  - Strengthen active labor market policies (ALMPs):
    - Introduce ALMP programs for information sharing of job opportunities across regions, individualized support, demand-driven skills training, and increased employability of low-skilled and long-term unemployed.
- Cross-country policy takeaways:
  - Reducing homeownership lock-in, mitigating disincentives from unemployment-related benefits, and expanding effective LMP expenditure may support higher internal labor mobility.
  - Addressing regional disparities in labor market performance and education composition can influence mobility patterns.

*Source: Section III of wp18282 (regional-level regression model and empirical results) from the provided IMF PDF content.*

### REFERENCES  _______________________________________________________________________ 23

### REFERENCES

### I. Introduction and overview
- Spain’s unemployment has long been one of the highest in the European Union (EU).
- Even during the economic boom years, the unemployment rate stood above 8 percent, higher than the EU average (Figure 1).
- Following the double-dip recession, Spain’s jobless rate more than tripled to a peak of 26 percent in 2013.
- The unemployment rate remained elevated at 15 percent in 2018:Q3 (Figure 1), with the structural unemployment estimated at around 12–16 percent.
- Regional disparities:
  - Differences in unemployment rates between the top (90 percentile) and bottom (10 percentile) regions ranged from 8 percentage points during the pre-crisis boom years to 18 percentage points at the height of the crisis (Figure 2, left panel).
  - Dispersion declined since the crisis but remained at 15 percentage points in 2017.
  - Regions in the Northeast had average unemployment rates ranging from 5 to 20 percent during 2006–17.
  - Canary Islands and the South had consistently highest jobless rates during 2006–17 (around 12–32 percent) and more volatile labor market outcomes (Figure 2, right panel).

### II. Theory and empirical background on labor mobility
- The literature links unemployment differentials and regional labor mobility (Harris and Todaro, 1970; Blanchard and Katz, 1992).
- Interregional labor movement helps limit impacts of economic shocks and minimize labor market volatility in the short run, and reduce regional disparities and structural unemployment in the long run.
- Empirical studies find positive migration responses to region-specific labor demand shocks in Europe and the United States (Dao et al., 2014; Arpaia et al., 2014; Beyer and Smets, 2015).
- Cross-country data suggest a negative correlation between unemployment rate and internal migration rate within a country (Figure 3, left panel).

### III. Spain’s internal mobility in international comparison
- Eurostat Labour Force Survey comparison (2016):
  - Gross internal migration flows accounted for 0.3 percent of the working-age population in Spain.
  - Germany: 1.2 percent.
  - United Kingdom: 2.4 percent.
- The migration rate in Spain had not improved since the early 2000s.
- Note on data sensitivity: internal migration level is highly sensitive to data source; Eurostat LFS yields systematically lower internal migration rates for Spain than Spain’s internal data on regional mobility. Cross-country analysis uses Eurostat LFS for consistency; cited numbers are mostly for comparison purposes.

### IV. Prior literature on determinants of Spain’s low mobility
- Studies from the 1990s found Spanish internal migration was not very responsive to traditional labor market variables such as wages and unemployment (Bentolila and Blanchard, 1990; Antolin and Bover, 1997; De La Fuente, 1999), possibly due to high structural unemployment and labor market rigidities.
- More recent studies using migration data since the 2000s find increasing roles of labor market conditions, particularly after the global financial crisis (Gutiérrez-Portilla et al., 2018; Clemente et al., 2016; Mulhern and Watson, 2009).

### V. Research questions, data, and empirical approach of the paper
- Research questions:
  - What are the stylized facts of Spanish regional migration flows?
  - What are the regional-level factors that explain Spain’s low internal mobility over the past decades?
  - How do the impacts of these factors vary before and after the crisis, and among Spaniards and immigrants?
- Approach:
  - Uses regional data of bilateral migration flows for Spain during 2000–16.
  - Investigates determinants of internal labor mobility at the macro level using a cross-country panel of 11 European countries.
- Key empirical findings (summary presented in the introduction):
  - At the regional level, housing prices, labor market conditions, and the pervasiveness of labor market duality are important in explaining Spain’s internal migration flows, particularly since the onset of the global financial crisis.
  - At the country level, labor market institutions and policies are the main factors driving internal mobility in European countries.

### VI. Contributions of the paper
- Fills a literature gap by presenting a cross-country panel analysis on determinants of internal mobility in European countries, allowing comparison of Spain with peers and examination of macro and labor variables at the national level.
- Regional-level regression includes labor market variables less explored in prior studies, such as labor market duality and education level of the labor force—variables especially relevant for Spain and with policy implications.

*Source: wp18282 - REFERENCES*

### Section III outlines the regional-level regression model and summarizes the empirical results

### wp18282 - Section III outlines the regional-level regression model and summarizes the empirical results

### Data and stylized facts
- Data sources and coverage
  - Regional bilateral migration flows: Residential Variation Statistics (RVS) from Spain’s INE, aggregated from Municipal Registers, covering 1998–2016.
  - Working-age population definition: 16–64 years old. Working-age migrants constitute "more than three quarters" of gross interregional migration flows.
  - Cross-country migration data: Eurostat Labor Force Survey, gross migration flows at NUTS 2 level, covering 2003–16 for 11 countries: Austria, Belgium, Finland, Germany, Greece, France, Portugal, Slovenia, Slovakia, Spain, and the UK. Panel is unbalanced.
  - Explanatory variables: mostly from INE (Spain) and Eurostat; housing price-to-income index from OECD Analytical House Prices Indicators; Spain housing prices from Ministerio de Fomento.
- Key stylized facts for Spain
  - Internal mobility rose during the early 2000s boom and declined after the 2008 crisis.
  - Internal mobility of immigrants rose from 6 percent in 1998 to 12 percent in 2002; share of foreigners among movers rose from about 4 percent in 1998 to 30 percent in 2008.
  - On average, interregional mobility among immigrants was around "three times higher" than that of Spaniards.
  - Working-age composition: about "85 percent" of foreigners who moved were working-age vs "74 percent" for Spaniards.
  - Female immigrant share of internal movement: about "44 percent" during 1998–2005, rising to "48.5 percent" in 2016.
  - Geographic concentration: Catalonia, Madrid, and Andalusia were top senders and receivers in absolute terms; Balearic Islands, Catalonia, and Canary Islands have highest internal migration rate relative to population.
  - Crisis-driven net flow changes: e.g., Madrid and Basque Country switched from net senders (pre-crisis) to net receivers (post-crisis); Valencia and Castile-La Mancha moved from net receivers to net senders.

### Regional-level regression model (gravity specification)
- Model structure (log-linear gravity specification)
  - Dependent variable: ln M_{ijt} = ln migration flows from origin i to destination j in year t.
  - Controls include: origin and destination population (P_i, P_j), distance (DIST_{ij}), region fixed effects, time fixed effects, and origin/destination economic variables X_{s,i,t} and X_{s,j,t}.
- Push and pull variables used (regional-level)
  - Real GDP per capita (income/prosperity)
  - Average house price (nominal housing prices)
  - Unemployment rate (lagged: t-1)
  - Real wage per worker (contemporaneous)
  - Share of low-educated workers (≤ lower secondary)
  - Share of temporary employment
- Notes on endogeneity and measurement
  - Bilateral flows for working-age population are not available; bilateral regressions reflect overall population mobility.
  - Endogeneity concerns acknowledged for unemployment and wages; unemployment rate uses one-year lag; wages kept contemporaneous due to lagged response rationale.
  - Baseline regression sample restricted to 2002–2016 when including share of temporary employment.

### Empirical findings — baseline regional regressions (Spain)
- Standard gravity variables
  - Distance: negative and significant (Table 1, Column 1: Distance = -1.25***).
  - Origin population: negative and significant (Population_o = -1.47** in Column 1), suggesting larger-origin populations reduce outflows.
  - Destination population: once controlling for income, destination population effect is weaker/insignificant in some specs (Population_d = 0.66 in Column 1, not significant).
  - Real GDP per capita: origin negative, destination positive (Real GDP per capita_o = -1.48***; Real GDP per capita_d = 1.10*** in Column 1).
- Housing, labor market, and duality effects (Table 1, Column 1)
  - Average house price in destination: negative effect on inflows (Average house price_d = -0.18**).
  - Average house price in origin: not significant in Column 1 (Average house price_o = 0.01).
  - Unemployment rate (t-1) in destination: negative and significant (Unemployment rate (t-1)_d = -0.22***).
  - Unemployment rate (t-1) in origin: not significant in Column 1 (0.05).
  - Real wage per worker in origin: negative and significant (Real wage per worker_o = -0.99**), indicating higher origin wages reduce outflows.
  - Real wage per worker in destination: negative in Column 1 (Real wage per worker_d = -0.41**), result described as puzzling and investigated further.
  - Share of temporary employment in origin: negative and significant (Share of temporary employment_o = -0.28**), consistent with pervasive temporary contracts hindering mobility.
  - Share of low-educated in origin: positive and significant in Column 1 (Share of low-educated_o = 0.36**); share of low-educated in destination: not significant in Column 1 (-0.27).
- Robustness and interactions
  - Introducing interaction Real wage_d × Share of low-educated_d (Table 1, Column 2) yields Real wage_d X Share of low-educated_d = -0.05*** and changes sign of Real wage per worker_d to positive in that specification, suggesting negative wage-destination effect driven by regions with large low-educated shares.

### Heterogeneity: Spaniards vs foreigners
- Main divergent patterns (Table 1, Columns 3 and 4)
  - Spaniards (Column 3): mobility driven more by destination pull factors — large destination population, higher destination income, lower house prices at destination, and low unemployment at destination attract Spaniards (e.g., Real GDP per capita_d = 1.47***; Population_d = 1.57**).
  - Foreigners (Column 4): mobility driven more by origin push factors — smaller origin population, lower origin income, higher origin house prices, and lower origin wages correlate with larger migration outflows among foreigners (e.g., Real GDP per capita_o = -1.76***; Real wage per worker_o = -1.00**).
  - Share of temporary employment effect largely reflects foreigners: Share of temporary employment_o = -0.50*** for foreigners (Column 4).
  - Home-region fixed effects indicate “home bias” for both groups, generally stronger among foreigners.

### Pre-crisis vs post-crisis dynamics (2002–07 vs 2008–16)
- Distance effect stable: Distance ≈ -1.19*** (pre-crisis), -1.29*** (post-crisis) (Table 2, Columns 1–2).
- Variable sign and significance shifts
  - Population, real GDP per capita, and real wages in origin remain significant across episodes but with larger marginal effects pre-crisis.
  - Several destination-region variables show puzzling signs pre-crisis (e.g., population_d = -3.86* pre-crisis) but revert to expected signs post-crisis (population_d = 2.98*** post-crisis).
  - Share of low-educated in origin: negative pre-crisis (-0.13) but positive and significant post-crisis (0.45***), interpreted as low-educated workers being forced to reallocate after construction job losses.
  - Share of temporary employment becomes statistically significant and negative in the post-crisis period (Share of temporary employment_o = -0.48*** post-crisis).
- Interaction with post-2013 (proxy for 2012 labor reforms) — Table 2, Column 3 (post-crisis sample, interactions)
  - Post 2013 × Real wage per worker_d = 0.05*** (positive strengthening of destination wage effect after 2013).
  - Post 2013 × Unemployment rate (t-1)_d = -0.05*** (strengthened negative destination unemployment effect after 2013).
  - Post 2013 × Share of temporary employment_o = 0.21*** (origin temporary employment effect becomes less negative after 2013).
  - Post 2013 × Share of temporary employment_d = -0.10*** (destination temporary employment effect turns statistically significant after 2013).
  - Authors interpret results as suggesting the 2012 labor market reforms may have strengthened the roles of labor market variables, though they note the post-2013 dummy is only a proxy and could capture other changes.

### Cross-country analysis — model and key findings
- Cross-country regression setup
  - Dependent variable: ln LM_{ct} = labor mobility flows across regions (NUTS 2) within country c at year t.
  - Key national variables: ln GDP per capita (PPP), homeownership percent, index of house price to income ratio.
  - Labor-market controls: dispersions in regional unemployment rates, labor force education level, social benefits related to labor market activity (unemployment benefit), labor market policy (LMP) expenditure.
  - Country and time fixed effects included.
- Empirical findings (Table 3 summary)
  - Homeownership: statistically significant and negative effect on regional labor mobility in a simple macro/housing specification (Table 3, Column 1).
  - National housing prices: coefficient statistically insignificant in simple specification (Table 3, Column 1).
  - Adding labor-related social benefits and LMP expenditure (Table 3, Column 2)
    - Higher social benefits reduce labor movement.
    - Greater LMP expenditure supports interregional mobility.
    - Homeownership coefficient becomes insignificant when labor policy and benefits are included.
- Policy-relevant interpretations from cross-country results
  - High homeownership rates tend to be associated with lower internal labor mobility, although significance depends on controls.
  - Generous unemployment and labor-related social benefits are associated with reduced incentives to move.
  - Greater LMP expenditure is associated with higher interregional mobility.

### Policy implications and interpretations highlighted in the section
- For Spain:
  - Housing costs matter: higher destination house prices discourage inflows; origin house prices can act as push factors in some periods.
  - Labor market duality (high share of temporary employment) appears to hinder mobility, particularly for foreigners and in the post-crisis period.
  - Unemployment (lagged) in destination regions consistently deters inflows.
  - Wage signals are complex: negative destination wage effects in some specifications likely reflect interaction with low-educated shares; interactions reveal a positive destination wage effect when accounting for education composition.
- Effects of 2012 labor market reforms (proxied by post-2013 interactions)
  - Reforms may have strengthened the roles of labor market variables in shaping mobility: real wages and unemployment effects at destinations became stronger after 2013; the relation between temporary employment and migration also changed (origin effect less negative; destination effect more negative/statistically significant).
  - Caution: post-2013 dummy is a proxy and may capture other concurrent economic or policy changes; the specific channel through which reforms altered duality’s impact requires further analysis.
- Cross-country policy implications
  - Policies that reduce homeownership lock-in, mitigate disincentives from unemployment-related benefits, and expand effective LMP expenditure may support higher internal labor mobility.
  - Addressing regional disparities in labor market performance and education composition can influence mobility patterns.

*Italic: Source — Section III of wp18282 (regional-level regression model and empirical results) from the provided IMF PDF content.*

### introduction of labor market variables, such as dispersion in regional unemployment rates

### wp18282 - introduction of labor market variables, such as dispersion in regional unemployment rates

### Cross-country regression findings (Table 3)
- Regression sample covers the period of 2003-16 and includes Austria, Belgium, Finland, Germany, Greece, France, Portugal, Slovenia, Slovakia, Spain, and the UK. 1/
- Key coefficient results (dependent variable: regional labor mobility flows):
  - GDP per capita in PPP: -2.04* (1.02), -3.31** (1.25), -2.98** (1.19), -2.89 (1.65)
  - Home ownership: -0.05** (0.02), -0.01 (0.02), -0.01 (0.02), -0.01 (0.02)
  - House price to income ratio: 0.42 (0.49), 0.31 (0.55), 0.33 (0.65), 0.30 (0.47)
  - Social protection benefit: -0.76** (0.27), -0.91** (0.33), -0.95** (0.40)
  - Labor market policy expenditure: 0.67* (0.31), 0.73* (0.37), 0.93** (0.38)
  - Share of low-educated in labor force: -0.02 (0.02), -0.02 (0.02)
  - Dispersion in regional unemployment rates: 0.00 (0.01)
  - National unemployment rate: 0.01 (0.03)
- Observations: 124, 109, 101, 109
- Adjusted R-squared: 0.98, 0.98, 0.97, 0.98
- Adjusted within R-squared: 0.16, 0.20, 0.21, 0.21
- Country FE: Yes; Time FE: Yes; Cluster country and year: Yes
- Robust standard errors in parentheses. Significance: *** p<0.01, ** p<0.05, * p<0.1
- Notes:
  - Coefficients for labor force education level and unemployment dispersion have the expected sign but are statistically insignificant.
  - Replacing unemployment dispersion with the average jobless rate at the national level yields similar results.
  - Labor market social protection benefit and LMP expenditure remained the main factors driving a country’s internal labor mobility during 2003–16 (Table 3, Column 4).
  - Labor-related social protection benefits per inhabitant used to capture overall benefit package; caveat that per-inhabitant measure may not accurately reflect generosity for countries with a large number of unemployed. 24

### Spain in cross-country context (Figure 7 and discussion)
- Spain comparisons (averages 2003-16):
  - Shares of home ownership and low-educated labor force were relatively higher in Spain than in many other European countries in the sample.
  - Spain’s spending on labor-related social benefits in per capita term ranked third after Belgium and Finland during 2003–16. 25
  - Spain’s expenditure for LMP (which includes LMP services, LMP measures and LMP supports) was below the sample average.
- Policy implication: To address Spain’s low regional mobility issue, national-level policy dialogues could emphasize design of labor market institutions and policies as well as frictions in the housing market.
- Footnotes:
  - 25: Given Spain’s high unemployment rate, total spending on social protection benefits per inhabitant may not reflect generosity of the benefit system.

### Main empirical findings on Spain’s internal mobility (Sections IV–V)
- Time pattern and composition:
  - Using data from 1998–2016, internal mobility rose steadily during the economic boom years but has been on a declining trend since the onset of the crisis.
  - Much of the underlying trend was driven by movement of immigrants; mobility of Spaniards has been relatively stable.
  - On average, regional mobility among foreigners was around three times higher than that of Spaniards.
  - At the aggregate level, working age population accounted for more than three quarters of mobility flows across regions.
- Regional determinants:
  - House prices, labor market conditions, and degree of labor market duality are the main regional factors explaining interregional migration flows.
  - Average house prices and unemployment rates are negatively associated with migration inflows.
  - Average real wages and the share of temporary employment are negatively related to migration outflows, particularly after the global financial crisis.
  - Evidence of potential “home bias” in migration decisions (attributable to language, cultural, and unknown institutional factors).
  - For Spaniards, pull factors from destination regions play a more important role; for foreigners, push factors in origin regions matter more.
- Pre- vs post-crisis differences:
  - Most economic factors play a more important role in explaining regional migration flows in the post-crisis period.
  - Effects of selected labor market variables became even stronger since 2013 after major labor market reforms.

### Policy recommendations to promote regional labor mobility in Spain
- Reduce the pervasiveness of labor market duality:
  - Narrow the gap between costs for firms to hire permanent and temporary workers by reducing hiring costs and severance payments associated with permanent contracts.
  - Specific policy recommendations referenced in IMF Country Reports No. 17/23 and No. 17/319.
- Facilitate wage setting flexibility:
  - Ensure wage growth in line with productivity development to promote efficient allocations of labor across regions and industries.
  - Further decentralize wage setting and allow collective bargaining opt-out to provide needed flexibility for firms and workers to adjust to temporary and regional-specific shocks.
- Enhance incentives for geographical mobility:
  - Consider policies to provide additional incentives for workers to move (example: Mobility Assistance Program (MAP) adopted in Germany found effective in encouraging distant job search and promoting regional labor mobility (Caliendo et al., 2017)). 26
  - Consider temporary and targeted housing assistance in regions with high house price-income ratio or underdeveloped rental markets.
- Provide support through more effective active labor market policies (ALMPs):
  - Introduce ALMP programs to facilitate information sharing of job opportunities across regions, provide individualized support and demand-driven skills training, and increase employability of low-skilled and long-term unemployed.

### Mobility data sources and measurement caveats (Annex I)
- Five mobility data series summarized (coverage and start years):
  - INE migration statistics: overall population, flows within country, inter-autonomous and inter-provinces, starts from 2008.
  - Labor mobility data from the tax authority: workers only, includes geographic and industry flows, excludes Navarre, Basque Country, Ceuta and Melilla, starts from 2005.
  - INE labor and geographical mobility data: combines Municipal Register and Labor Force Survey, flows between autonomous regions, covers population aged 16+ in main family dwellings, starts from 2010.
  - SEPE basic mobility data: from contracts registered in public employment services, mobility defined when workplace differs from previous domicile (may be permanent or temporary), covers workers with a contract in public employment services, starts from 2001.
  - INE Residential Variations Statistics: registrations and cancellations in Municipal Registers, counts number of variations (an individual may change residence more than once in a year), includes internal and external residential variations, inter-region and inter-province breakdowns, starts from 1998.
- Comparative patterns:
  - Out of the five series, four report a declining trend in regional migration rate since the global financial crisis, with one showing a small rebound in 2016 (tax authority data).
  - SEPE series displays an upward trend during 2009–17 and reports a much higher migration rate than other sources.

### Regional fixed-effects regression results (Annex tables A1)
- Panel regression dependent variable: migration (results presented for Total, Spaniards, Foreigners, Pre-crisis, Post-crisis)
- Selected region-level coefficients (examples; robust standard errors in parentheses):
  - Aragon_o: -4.04*** (0.97), -3.27** (1.41), -5.09*** (1.01), -8.47*** (3.26), -4.19*** (0.06)
  - Asturias_o: -4.66*** (1.16), -3.45** (1.68), -6.48*** (1.19), -10.07** (3.94), -4.77*** (0.07)
  - Catalonia_o: 0.47* (0.27), 0.22 (0.31), 0.30 (0.20), 1.91** (0.82), -0.02 (0.09)
  - Madrid_o: 0.52* (0.30), 0.37 (0.30), -0.03 (0.26), 1.63** (0.73), -0.05 (0.08)
- Observations: 4,080; 3,808; 3,808; 1,632; 2,448 (panels across columns)
- Adjusted R-squared: 0.90, 0.89, 0.93, 0.86, 0.93
- Time FE: Yes; Cluster region pair and year: Yes
- Significance: *** p<0.01, ** p<0.05, * p<0.1
- Note: Main dependent variables omitted in table for space; see table 1 for full results.

*Source: IMF staff analysis in wp18282 - introduction of labor market variables, such as dispersion in regional unemployment rates (PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18282.pdf_