The Great Divide: Regional Inequality and Fiscal Policy
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
  • | 2 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: wgbohoui@imf.org; wlam@imf.org; vlledo@imf.org

Growing regional inequality within countries has raised the perception that “some places and people” are left behind. This has prompted a shift toward inward-looking policies and away from pro-growth reforms. This paper presents novel stylized facts on regional inequality for OECD countries. It shows that regional disparity in per-capita GDP is large (even after adjusting for regional price differences), persistent, and widening over time. The paper also finds that rising nationwide income inequality is associated with both rising within-region income inequality and widening average income across regions. The rise in inequality is related to declining incentives for interregional labor mobility, especially for poor households in lagging regions, which are estimated to reduce by as much as one-third in the United States. Against these facts, the paper proposes a framework to identify whether, how and by whom fiscal policies can be used to tackle regional inequality. It outlines conditions under which those policies should be spatially-targeted and illustrates how they can be complementary to conventional means-testing methods in mitigating income inequality.

Abstract

Growing regional inequality within countries has raised the perception that “some places and people” are left behind. This has prompted a shift toward inward-looking policies and away from pro-growth reforms. This paper presents novel stylized facts on regional inequality for OECD countries. It shows that regional disparity in per-capita GDP is large (even after adjusting for regional price differences), persistent, and widening over time. The paper also finds that rising nationwide income inequality is associated with both rising within-region income inequality and widening average income across regions. The rise in inequality is related to declining incentives for interregional labor mobility, especially for poor households in lagging regions, which are estimated to reduce by as much as one-third in the United States. Against these facts, the paper proposes a framework to identify whether, how and by whom fiscal policies can be used to tackle regional inequality. It outlines conditions under which those policies should be spatially-targeted and illustrates how they can be complementary to conventional means-testing methods in mitigating income inequality.

Many regions have been marginalized in the rich world” The Economist, October 2017

I. Introduction

Regional perspectives have taken a back seat in the discussion about inequality. Policy debates have focused on nationwide inequality of income and opportunity across individuals and not on inequality between regions within a country. The literature has shown that high and rising nationwide inequality has been negatively associated with growth, macroeconomic stability, and pro-growth reforms (Easterly, 2007; Ostry, Berg, and Tsangarides, 2014, Dabla-Norris and others, 2015) and discussed how fiscal policies can mitigate inequality (IMF 2017). Regional perspectives have not featured prominently, as regional inequality has been largely seen as a natural part of economic development. Regional inequality is expected to initially rise when economies grow, but would eventually decline through the mobility of labor and capital within national borders, leading regional incomes to converge (Barro and Sala-i-Martin 1991). As such, regional inequality was not expected to exert lasting impact on nationwide inequality and the macroeconomy.

Globalization and the global financial crisis have brought regional inequality to the forefront. As growth shifted jobs and industries on the map, gains were not well shared—raising the perception that some “places and people” have been left behind, including some formerly prosperous regions in advanced economies (Autor, Dorn, and Hanson 2013; Nunn, Parsons, and Shambourg 2018). The trend of rising regional inequality began in 1980s-1990s (OECD 2016; Roses and Wolf 2018) and has turned more pronounced in the years following the global financial crisis, with already-prosperous places being less affected and even seeing more robust growth and job creation. Meanwhile, many regions—urban and rural alike—have lagged behind, widening the gaps with prosperous regions (Moretti 2012). It is important to note, however, that these studies showing rising disparity of income across regions did not adjust for differences of regional prices, which usually vary significantly across regions and could lead to mis-measurement of regional disparity.

Recent politics and new evidence further rooted regional inequality in the policy debate. The growing populism prompted a shift toward more inward-looking policies and put many pro-growth reforms at risk (The Economist 2016; Yglesias 2017). Emerging evidence shows that regional inequality reflects a decline of internal factor mobility (Ganong and Shoag 2017) and may contribute to weaker long-term growth and growing populistic policies if let unaddressed (OECD 2016; Bouchet and Parilla 2017, Winkler 2018). On the policy front, the effectiveness of conventional spatially-blind fiscal policies in addressing rising inequality and stagnant regional growth have raised questions as to whether spatially-targeted policies should be part of policy tools (Galbraith and Garcilazo 2010; Gill 2010; Austin, Glaeser, and Summers 2018).

This paper contributes to the debate on regional inequality and fiscal policies on multiple fronts. First, the paper illustrates novel stylized facts on regional inequality for a sample of OECD countries. Second, it proposes a conceptual framework, grounded in the stylized facts, that identifies whether, how, and by whom fiscal policies could be used to mitigate regional inequality. Lastly, the paper also illustrates the conditions in which spatial considerations can complement national policies to mitigate income inequality.

This paper identifies several novel facts. First, our analysis overcomes a limitation in previous studies by adjusting for regional price differences and finds that the disparity of income across region remains large. At the same time, regional disparity has been persistent. Lagging regions have about a 70 percent chance to remain behind, and tend to grow slower in income and employment, suggesting signs of regional divergence. Second, our analysis links the two branches of studies on regional disparity and inequality of income distribution. We find that countries with greater regional disparity of income levels tend to have greater contributions of the between-region component to the rise of nationwide income inequality. Third, rising disparity of income across regions is related to falling labor mobility. For example, the incentives for interstate migration are estimated to have declined by as much as one-third in the United States over the past decade.

The proposed framework aims to guide fiscal policy to tackle regional inequality. It builds on the IMF’s operational guidelines for the analysis of inequality (IMF 2018). The first stage is to assess whether regional inequality is macro-critical and its main determinants. Then if regional inequality is deemed necessary to address, policymakers need to decide on appropriate strategies and policies. The final stage is to determine which government levels should design and implement policies. The framework aims to help inform the unsettled debate of whether, when, and by whom spatially-targeted policy should be used to tackle regional inequality.2 Spatial considerations are likely more important in deciding the fiscal strategies when targeted regions are populated by a dense mass of less-mobile, disadvantaged individuals, and when targeting those individuals through spatially-blind means-testing is less effective and costly.

The paper argues that regional inequality matters and fiscal policies that account for spatial dimensions can be complementary to existing policies to mitigate inequality. The rest of the paper is organized as follows. Section II illustrates the trends on regional inequality and discusses their key driving forces. Section III outlines a conceptual framework in tackling regional inequality. Section IV illustrates an exercise on how spatial considerations can be complementary to current means-testing in mitigating inequality. Section V concludes.

II. Dynamics and Driving Forces of Regional Inequality

It is urgent to address regional divergence: once you become aware of its magnitude, it is hard to ignore how it pervades our politics as well as our economics.” Financial Times, November 2018.

Regional inequality in this paper is measured across two dimensions: between- and within-region inequality. In this paper, regions refer to state or provinces, as defined in the Large Territory Regions (TL2) in OECD classification.3 The between-region inequality in the paper refers to disparities across regions on measures such as real per-capita GDP or disposable household income, as well as unemployment rates and nonworking population. An example would be differences of average per-capita GDP between Missouri and New Jersey. The within-region inequality refers to the inequality in household income distribution in a region.4 An example would be how household income is distributed in Missouri and in New Jersey. This paper focuses on the 1990–2016 sample period.

The rest of the section develops some stylized facts focusing on the dynamics of regional inequality across several dimensions, as well as identifying potential driving forces that are closely related to regional inequality.

A. Dynamics of Regional Inequality

1. Regional disparity (between-inequality) in income levels has been large even after adjusting for regional price differences.

Previous studies (Bartolini 2015; Iammarino, Rodriguez-Pose, and Storper 2018) that find large regional disparities often do not adjust for differences in regional prices. Such price differences could be significant in some countries. For example, the same income in dollar terms can buy more than 1.3 times more goods and services in Missouri than in New Jersey. To account for that, our analysis adjusts for living cost differences (including housing cost) when measuring regional disparity of per-capita GDP, based on available data and the Luxemburg Income Surveys (LIS).5 For countries without official statistics on regional prices, adjustments are made using the housing costs (rental payment and implicit cost if households own the dwellings) from LIS income surveys to proxy for price differences of household consumption.6

Adjusting for regional prices, regional disparity remains sizeable with per-capita GDP in the top quartile regions 1.3 times higher than in the bottom quartile regions (slightly less than 1½ times before regional price adjustment) on average for OECD countries during 2010–14 (Figure 1 bottom left panel). It is sometimes more severe than cross-country differences in OECD countries.

Figure 1.
Figure 1.

Large Disparity on Income Levels and Unemployment Rates across Regions in OECD Countries

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

The difference is also large for regional disparity of unemployment rates, an indicator less sensitive to prices. On average, the difference of unemployment rates between the regions of 90th percentile and 10th percentile was 5.6 percent on average across OECD countries during 1990–2016. The difference in some countries such as Slovakia, Italy, Spain and Belgium are in double-digit levels (Figure 1 bottom right panel), much larger than the cross-country differences across OECD economies.

Regional disparity indicators are often closely linked. Regions with low per-capita GDP levels, disposable household income or higher unemployment rate also have lower health access and education attainment. This strong correlation is also observed at the country level, in which countries with higher regional disparity in income and unemployment rate also tend to be those with weaker social outcomes (Table 1). The negative correlation ranged from -0.07 to -0.19 for health access and -0.34 to -0.58 for education attainment, both statistically significant at the 5 percent level.7 This reinforces the perception that “some people and places” are left behind.

Table 1.

Regional Inequality of Income and Selected Social Outcomes (2000–16)

article image

denotes 5 percent statistical significance level.

Sources: OECD Regional Database and IMF staff estimates.

2. The between-inequality on income has also been persistent and widened over time, particularly after the global financial crisis.

The rising regional disparity of per-capita GDP beginning in the 1980s (Roses and Wolf 2018) has intensified during the global financial crisis. Regional disparities in household income and unemployment rates have also widened, particularly after the global financial crisis (Figure 2 top left and right panels). Regional conditional convergence—commonly observed during 1980s and 1990s (Iammarino and others 2018)—has stalled in many OECD countries. Our analysis separates the sample of regions into two groups: lagging regions are defined as those with unemployment rates above the 75th percentile in the nationwide distribution, while leading regions are those with unemployment rates below the 25th percentile of the same nationwide distribution. Instead of catching up, lagging regions tend to grow slower in per-capita GDP, household income, and employment, both in the full sample and after the global financial crisis (Figure 2 bottom left panel; Table 2). Employment growth in the lagging regions was about one-third that of the leading regions during 2000–16. Real per-capita GDP growth was lower by about 1 percentage point over a 3-year horizon in lagging relative to leading regions.8 Similarly, growth of disposable household income was lower in lagging regions than in leading regions. Such regional differences were statistically significant at 5 percent confidence levels.

Figure 2.
Figure 2.

Rising Regional Disparity of Unemployment Rate in Many OECD Countries

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

Table 2.

Differences Between Lagging and Leading Regions1/

article image
Sources: IMF staff estimates and OECD Regional database.

Lagging regions are those regions with unemployment rates higher than 75-percentile in the country.

***, **, * denote statistical significance at 1, 5, and 10 percent levels.

The evolving trend of regional disparity differs across OECD countries. Such dynamics broadly fall into three groups. First, in a few countries (e.g., France, Greece, Italy, and Spain), the disparity has increased and remained larger than that before the global financial crisis, although it has narrowed from the peak as the economy picked up. The second group (e.g., Canada, United States, and Norway) has seen rising regional disparity during the global financial crisis but disparity across regions has scaled back to pre-2007 levels as the economy recovered. The third group, which includes Germany and Finland, continued to see a converging trend on unemployment rates across regions without being much affected by the global financial crisis, although the regional disparity of disposable income has not converged as much (Appendix Figure A.1).

Regional inequality is also highly persistent. The probability of a lagging region remaining lagging in the following year is high at 70 percent for the average OECD country and over 80 percent in Italy, New Zealand, and Canada (Figure 2 bottom right panel). This implies that lagging regions have a high likelihood of staying lagged instead of gradually catching up to the more prosperous regions in the country.

In sum, these findings suggest that the between-region inequality with respect to income level has been large, persistent, and widening over the global financial crisis, underscoring the importance of considering regional dimensions in the discussion of inequality. Now we turn to the stylized facts on the within-region inequality, which refers to the inequality of income distribution in a region within the country.

3. The within-region inequality has risen over the past decade and accounts for most of the nationwide income inequality, but its contribution has declined.

Nationwide income inequality has steadily risen across OECD countries over the last decade by about 0.03 Gini points on average (IMF 2017). The share of emerging market economies with a significant increase in Gini coefficient is higher than that in the advanced countries (IMF 2017). The rising trend remains even after adjusting for regional price differences and housing cost. In the case of the United States, for example, inequality on disposable income has risen by 0.02 Gini points (from 0.37 to 0.39) over the last decade. Adjusting for regional prices and housing cost across U.S. states, income inequality would be slightly lower by 0.01 Gini points but without affecting the rising trend (Figure 3 top left panel). This is partly because housing prices are higher in rich regions and low-income households spend relatively more on housing.

Figure 3.
Figure 3.

Household Income Inequality Has Risen over the Last Decade in Many Regions of OECD Countries

(measured by Gini-coefficient before and after tax and transfers)

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

The rising nationwide trend in income inequality was also mirrored in many OECD regions. For example, in countries such as Australia, Germany, Spain, and the United States, where income inequality rose more rapidly, about two-thirds of regions have recorded a rise in income inequality (Figure 3 top right panel). In addition, regions with initial higher income inequality recorded a higher rise in Gini points over the last decade. Income inequality in lagging regions has also been higher than that in leading regions (Figure 3 bottom left panel). The difference is statistically significant after the global financial crisis.

The nationwide income inequality can be decomposed into between-region income inequality (how average income distribution differs across regions) and within-region income inequality (how income is distributed among households in each region). In advanced OECD countries, the within-region inequality accounts for most of the nationwide income inequality (about 90 percent), but its contribution has declined by about 5–10 percent. The small between-component across regions in the nationwide income inequality is consistent with findings of large disparities of income across regions (Goesling 2001). It is because the between-region component has increased faster than the within-region component, and it is increasingly contributing to the nationwide income inequality. Consistent with the findings on regional disparities above, countries with greater regional disparity on income levels tend to have higher contributions of between-region component of income inequality to the rise of nationwide inequality (Figure 3 bottom right panel).

B. Driving Forces of Regional Inequality

The rest of the section examines potential drivers of the rising regional inequality among OECD countries. We first conduct an empirical analysis using an unbalanced panel of 18 OECD countries for the sample 1990–2016 to examine the key determinants driving regional disparity of income trends. As the empirical results show that one of the driving forces of regional disparity of income is labor mobility, the paper further analyzes the incentives for interregional labor mobility using micro-level individual household income surveys for 13 OECD countries during 2000–2016. Lastly, we extend the analysis done in IMF (2017) to look at the role fiscal redistribution may have played in affecting within-region income inequality.

Driving Forces of Regional Disparity on Income: Cross-country Panel Analysis

The panel analysis uses the OECD Fiscal Decentralization database, the OECD regional database, and the OECD subnational government finance.9 The baseline specification is:

Yit=α+βFDit+γXit+μi+ϵit

where the subscript i indicates the country, and t refers to the year. Y refers to a measure of regional disparity of income; FD corresponds to the different fiscal decentralization indicators, considered one at a time. The matrix X represents a set of explanatory and control variables which include labor mobility, population density, aggregate growth, and controls for industry structure; µ is the unobserved individual country effect and e is the error term.

Our baseline uses the real per-capita GDP (USD PPP adjusted) as a measure of regional income levels for international comparison.10 The disparity uses the top-bottom decile ratio (p90-p10) of real per-capita GDP as a measure of regional disparity. The ratio is mean independent, less sensitive to outliers, and invariant to the number of regions. However, other measures, such as the (weighted) coefficient of variation and the 25th-75th percentile (Appendix B), are also used as a robustness check and results are similar.

Fiscal decentralization measures include the subnational government share of revenue, expenditure, personal income tax, and debt in general government aggregates. In addition, we also use the vertical fiscal imbalance—an indicator of the share of spending not financed through own resources—and transfer dependency from the central government to assess the extent of fiscal decentralization. The vector of explanatory variables contains the level of real GDP per capita and its squared term (capturing potential nonlinear effects), GDP growth rate, the share of gross valued added in manufacturing sectors and in the energy sector, the population density, the net migration rate, the unemployment disparity across regions, and the availability of social services.

The theoretical literature suggests a tradeoff between fiscal decentralization and regional disparities. On one side, where subnational governments (SNGs) are more autonomous and accountable to the local electorate, greater tax and spending decentralization would tend to reduce regional disparities by allowing SNGs in lagging regions to better mobilize revenue and target spending to close the efficiency gaps and catch up with leading regions (Rodriguez-Pose and Ezcurra 2010). On the other side, greater fiscal decentralization may increase tax competition across regional SNGs and generate a ‘race to the bottom’ with inefficiently low tax rates and lower revenue. It may also lead to diseconomies of scale and raise the costs of providing goods and services (disproportionately in lagging regions) thus increasing regional disparities (Wilson 2015). Higher transfer dependency and vertical imbalances would, on one side, increase regional disparities by constraining the capacity and incentives of lagging SNGs to catch up with leading counterparts. However, endogeneity problems could arise from reverse causality because higher regional disparities could prompt policymakers to decentralize policies toward the subnational governments. On the other side, they could help equalize fiscal resources necessary for lagging regions to achieve minimum standards in provision of subnational goods and services that are conducive to regional convergence (Bartolini and others 2016).

Empirical results from the panel analysis show that key determinants of per capita GDP regional disparity include labor mobility, economic development, regional population density, and the degree of fiscal decentralization, accounting for the endogeneity arising from reverse causality (Appendix B). Out of these explanatory variables, population density and labor mobility are the most important factors in explaining disparity across regions.

  • Labor mobility. Higher net regional migration flows—a proxy for labor mobility—help to mitigate regional disparity of real per-capita GDP. For instance, a 1 standard deviation increase in net migration, a proxy of labor mobility, is statistically significant in reducing regional disparity by 4 percentage points as flexible labor mobility ensures more efficient resource allocation (Table 3 and Appendix B Table B1-B3). For example, in the United States, a growing literature has found that the declining U.S. interregional labor mobility is associated with barriers from rising housing costs, jobs mismatch from technology and globalization, and the inability to attract talents to lagging regions (Kaplan and Schulhofer-Wohl 2015; Bayoumi and Barkema, forthcoming; Autor, Dorn, and Hansen 2013; Hsieh and Moretti 2015).11 The panel results suggest that labor mobility also affect the disparity of regional incomes for OECD countries beyond the United States.

  • Economic development, measured by the country per-capita GDP level, shows a statistically significant inverted-U relationship with regional disparity, similar to a Kuznet’s curve, indicating inequality first rises and later decreases as economic development takes hold. This could happen when capital and skilled labor are concentrated in a few regions in a country and those regions may have faster productivity gains than the rest of the regions in the country. This could first raise disparity across regions. As the economy further develops, higher wages and declining marginal agglomeration benefits could lead to reallocation of capital and labor to less developed regions. Together with the knowledge spillover effects, the reallocation could lead to regional convergence and therefore reduce disparity across regions.

  • Population density. Countries where the regional variation of population density is high tend to have higher regional disparity on per-capita GDP, suggesting a positive income effect of agglomeration forces.

  • Fiscal decentralization. Our empirical results extend previous studies (Lessman and Siedel 2017; Bartolini and others 2016) by separating the potential different effects of fiscal decentralization according to the intergovernmental structure. Greater revenue and expenditure decentralization reduces regional disparity in federal countries where higher accountability to local electorate and greater institutional capacity allow autonomous regional governments to mobilize more revenue and to better target spending to reflect local preferences (Figure 4 and Table 4).12 For unitary countries, the effects are opposite, possibly because SNGs are expected to implement pre-set revenue and spending policies from the central government with limited scope to match local preferences. Finally, the results also show that higher vertical fiscal imbalances contribute to greater regional inequality in both unitary and federal countries, pointing to the need to achieve a balanced fiscal structure to maximize efficiency gains (optimal vertical gap) and adequate resources to mitigate unfunded mandates (vertical fiscal imbalances) (Eyraud and Boadway 2018).13

Table 4.

Summary Statistics Across Unitary and Federal Countries

article image
Note: *, **, *** denote statistical significance of the difference at 10, 5, and 1 percent.
Table B.2.

Key Determinants of Regional Inequality

article image
Note: FE2SLS estimation; instruments are similar to those used in the regressions presented in the main test. Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1.
Figure 4.
Figure 4.

Fiscal Decentralization and Regional Disparity of Income Level

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

Sources: IMF Staff estimates.1/ Statistics in shaded color refer to coefficients not statistically significant at 10 percent.

Our empirical results also indicate that control variables such as the availability of social services (proxied by number of physicians per thousand population) and the efficiency of labor allocation across regions, proxied by unemployment gaps, have the expected signs. However, we do not find a strong correlation between regional inequality and industrial structure—proxied by valued-added share of manufacturing in regions, while the effects of the resources sector are only significant in some specifications. A robustness check points to similar results, regardless of using alternative indicators of regional disparity of income or different specifications.14

Labor Mobility and Regional Inequality

As the panel analysis points to labor mobility as a key determinant of regional inequality, the paper further explores this using the micro dataset of household surveys for 13 OECD countries available in LIS. Since households’ interregional movements within a country are not directly observed, we estimate the returns from interregional migration to infer the mobility pattern across regions and by income classes—essentially capturing the incentives of households to move across regions in a country. The available data from the OECD Regional Database include only the aggregate gross and net inflows of labor at regional levels but not the flows of each origin and destination region.15

Households are likely to relocate if returns, measured in terms of income difference between the origin and the destination regions, are positive and rise over time. The difference of regional income is one of the important factors in the relocation decision. As such, labor mobility patterns can be inferred by examining how the returns from relocation change over time and if rising regional price differences and income inequality contribute to such changes. The approach follows two steps. First, we extend Milanovic (2015) to estimate household income based on regional average income and its distribution by income classes for each of the 13 OECD countries (Appendix C). Second, we use the estimated coefficients to calculate the returns to interregional migration for households across different income classes, accounting for differences in regional prices. The estimated returns are compared across different vintages (cross sections) of household income surveys for these OECD countries to assess how returns have changed over vintages during the last decade.

When a household moves from its resident region to other regions in a country, some destination regions generate higher income (positive returns), while other destination regions generate lower income (negative returns). Mobility returns depend on differences in the average regional income level, regional income inequality, and regional price differences proxied by housing costs. These driving forces have different impacts across regions and income classes because households spend different fractions of income on housing and housing costs have risen more rapidly in more prosperous regions.

Our estimates show that the average regional income difference between the origin and destination states is a key determinant for mobility returns across all income classes. But income inequality affects labor mobility adversely for low-income households only. For example, in the United States, for the same difference in average state income level between the origin and destination regions, low-income households would benefit less if the destination regional income distribution is more unequal. A one Gini point increase would reduce household income by 2.9 percent for low-income households in the lowest income quintile in 2013 (Figure 5 left chart red solid line). On the other hand, a rise in income inequality would benefit the high-income households by 0.9 percent for the same average state income level.16 In other words, households moving to a higher-income region (e.g., Mississippi to New Jersey) on average tend to receive higher income (due to higher state income) but disproportionately across income classes because of the varying effects of income inequality. Similar patterns are observed in the estimated coefficients of average regional income and regional income distribution for selected OECD countries (Appendix C Table C.2).

Figure 5.
Figure 5.

Rising Income Inequality May Have Contributed to Falling Incentives of Interregional Labor Mobility

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

The incentives to move can be measured by the number of destination regions that generate higher income (positive returns) relative to the households’ income in the origin region. The number of destination regions that generate higher income may change over time. We measure in two different vintages—e.g., 2004 and 2013 for the United States, 2007 and 2013 for Spain, and 2004 and 2012 for Mexico—to assess how the number of destination regions that yield positive returns (incentives for household to move) have changed over the years. Our results suggest that mobility returns for interregional migration have declined across the board in the U.S. during the past decade, but more so for households in low-income deciles in low-income regions. The number of destination states that generate higher returns fell by over 30 percent over the last decade for the bottom quartile households, and by about 25 percent and 15 percent for the middle and upper-income classes, respectively (Figure 5 right chart).17 Moreover, the deterioration was sharper in regions with lower per-capita GDP. Similar results hold for many other advanced countries such as Spain, while the reduction of incentives is uniform across income classes in some emerging markets such as Mexico. Our results are corroborated by the findings of declining U.S. interstate mobility in Nunn, Parsons, and Shambaugh (2018), and slowing interregional labor mobility in the European Union (Arpaia, Kiss, Palvolgyi, and Turrini 2014; Liu 2018). Further analysis on the factors contributing to declining mobility would help assess the relative roles and interconnections among housing prices, income inequality, and the workforce skills and education levels.

Fiscal Redistribution and Within-Region Income Inequality

Redistribute fiscal policies have helped reduce, but not fully offset, rising nationwide income inequality (IMF 2017; Immervoll and Richardson 2011). For advanced OECD countries, the average redistributive effect of fiscal policy—measured by the difference in Gini coefficients on household income before and after taxes and transfers—is about one-third (from 0.49 Gini points from market income inequality to 0.31 Gini points for disposable income inequality in 2015) (Figure 6 top left panel). About three-quarters of the fiscal redistribution was achieved on the transfer side, while progressive taxation contributed the remaining one quarter. Across household income classes, benefits and transfers have helped reduce inequality more so than tax and social contributions.

Figure 6.
Figure 6.

Fiscal Redistributive Impact is Still Large but Declining in Some OECD Countries

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

The redistributive impact of fiscal policy, while still large nationwide, has declined since the mid-1990s in some OECD countries. Progressive fiscal policies provide an automatic mechanism to counter the rise of income inequality, even without active policy measures. The redistributive effects of fiscal policies seem to have flattened (in many European countries) or diminished (such as in the United States). The average redistributive effects (the change of Gini coefficients between before and after tax and transfer) have declined from 53 percent to about 50 percent in the group of selected OECD countries over the last decade (Figure 6 top right panel). This reinforces earlier findings that the fiscal redistributive role has declined over the mid-1990s to mid-2000s (Immervoll and Richardson 2011).

The impact also varies significantly across regions in some countries. At the regional level, the redistributive impact of fiscal policies is similar regardless of initial within-region inequality in many countries (Figure 6 bottom left panel). For example, fiscal redistribution in the United States has reduced income inequality on average by about one-fifth (from 0.46 to 0.37 Gini points in 2016) and the redistributive effects are similar (about 0.1 Gini point) across states independent of their initial market income inequality in 2016. Similar patterns are observed in United Kingdom, Switzerland, and Spain. The exceptions are Germany and Italy where the fiscal redistribution effects are larger in regions with higher within-region market income inequality. For example, in Germany, taxes and transfers have reduced market income inequality by over 40 percent (from 0.5 to 0.28) with stronger redistribution in states where market income inequality is higher (Figure 6 bottom right panel). In some regions in Canada and United States, tax and transfer schemes have only reduced inequality by less than 0.05 Gini points (just about 10 percent of market income inequality), relative to about one-third at the national level.

The different fiscal redistributive effects across regions are sometimes driven by national policies rather than spatially-targeted policies. The impact varies across regions largely because of the concentration of tax bases (e.g., high-income earners) and beneficiaries (e.g., poor and unemployed) in certain regions. For example, for regions where the per-capita unemployment insurance or pension benefits are the same, those with high unemployment or a large elderly population will receive a larger share of total benefits. As most benefits are targeted to low-income households and the progressive tax is usually levied on high-income households, these fiscal policies also affect the regional disparity of income levels.

III. A Policy Framework for Tackling Regional Inequality

Regional inequality is a hard problem to solve… economists need to try harder to find solutions” the Economist, December 2016.

This section outlines a conceptual framework that aims to identify whether, when, and how to use fiscal policies to tackle regional inequality. The framework consists of four steps, building on IMF operational guidelines on inequality (IMF 2018) (Figure 7). The first step is to determine whether regional inequality matters from a macroeconomic perspective. The next step is to identify the key determinants of regional inequality, followed by the third step of selecting strategies and policies to tackle regional inequality. The final step is to determine which level of government is responsible for designing and implementing the policies. The rest of this section discusses each step in more detail.

Figure 7:
Figure 7:

A Conceptual Framework for Addressing Regional Inequality

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

Sources: IMF (2018) and authors’ estimates.

A. Determine If Regional Inequality Is Macro-critical

Regional inequality may raise challenges for macroeconomic stability, long-term growth, and structural reforms. In some cases, a certain degree of regional inequality is a natural result of efficient resource allocation. In other cases, it could be driven by negative externalities and rent-seeking activity (Kline and Moretti 2014; Bastagli, Coady, and Gupta 2012), which can give rise to distortions that dampen growth or weaken the support for pro-growth reforms. Even if regional inequality is determined to be macro-critical, tackling it should not compromise long-term growth and macroeconomic stability.

Establishing the facts on regional inequality, such as its size and persistence, can help assess its macro-criticality. A starting point could be to determine how salient regional inequality is across different economic and social indicators (income levels and distribution, unemployment, health, and education), benchmark across relevant comparators, and adjust for differences in regional prices. The findings in Section II would suggest that in some countries such as Spain, Italy, and Poland, the regional disparities on income and unemployment levels are large and persistent, which might have wider macro consequences if left unaddressed. In contrast, in other countries such as Germany, the regional disparity of unemployment rates is relatively low.

B. Identify Regional Inequality Drivers

Identifying the drivers of regional inequality is essential if it is deemed relevant for policymakers to address. As highlighted in Section II, large and persistent regional inequality has been partly driven by declining labor mobility. Empirical results also suggest that fiscal decentralization would tend to reduce regional inequality to a certain extent. The literature also points out that natural resource endowments, proximity to markets, the geographical segregation of socially-disadvantaged groups, and the spatial mismatch between jobs and workers are other key drivers (Lessman and Seidel 2017).

C. Select Fiscal Strategies and Policies

The choice of strategies and fiscal policies to tackle regional inequality is country and context specific. It should depend on characteristics of the country’s lagging and leading regions and the key regional inequality drivers previously identified. Other considerations include the available policy space, the country’s political economy and implementation capacity, and society’s preference for interregional redistribution. The choice should ensure efficient allocation of resources, macroeconomic stability, and be growth-friendly.

There are a wide range of policy options to tackle regional inequality. Policymakers could scale up the size of fiscal redistribution to mitigate inequality, which could entail additional fiscal costs. Growth-friendly policies in the areas of education, healthcare, infrastructure, and affordable housing are also critical to improve the mobility of less skilled, low-earning individuals towards high earning areas thus helping to facilitate regional convergence.

Policymakers can also adopt strategies to improve policy efficiency with spatial considerations. Strategies to tackle regional inequality are broadly classified as spatially-blind or targeted. The two are not necessarily substitutes, but rather often complement each other. Broadly speaking, spatially-blind (or people-based) policies target recipients nationwide regardless of their place of residency. They can also aim at reinvigorating economic convergence by reducing barriers to factor mobility, especially for disadvantaged households, which face declining mobility as shown in Section II. Typical examples include U.S. progressive federal income tax systems and the national social security and unemployment benefits in France and Spain.18 On the other hand, spatially-targeted (or place-based) policies target households and firms in one or more regions. Such policies try to promote regional equity, growth that is spatially broad and inclusive, and to insure against regional idiosyncratic shocks such as natural disasters (Kim and Dougherty 2018). Their coverage is thus subnational and such policies rely mostly on tax incentives, subsidies, grants, and public infrastructure spending. Examples include the European Union’s Regional Development Fund.

The choice of spatially-targeted strategies depends on a few considerations:

  • Interregional mobility, population density, and regional concentration of recipients. Spatially-targeted strategies are preferable when lagging regions are densely-populated (larger agglomeration benefits) and production factors face strong barriers to move.19 In addition, in cases where targeted recipients (e.g., unemployed or people in poverty) are concentrated in certain regions, spatially-targeted policies have merits to complement other spatially-blind means-tested policies , as policies promoting labor mobility may turn out to be ineffective or too costly given the sheer size of the targeted group (Section IV).

  • Targeting challenges and capacity. Spatially-targeted interventions may also be justified when targeting people (e.g., means-testing) nationwide is faulty or difficult. In countries without the administrative capacity or political accountability to manage spatially-blind interventions such as progressive tax systems or cash transfers, a more geographically narrow version of these programs such as place-based people programs could be more effective (Coady, Grosh, and Hoddinott 2004).

  • Horizontal equity. Spatially-targeted policies may give rise to horizontal equity concerns-different treatment among equals—by excluding individuals and firms residing outside the targeted regions, even though they have similar status to those in the targeted regions. As ensuring uniform access of public goods and services across regions is not feasible (Boadway and Eyraud, 2018), an alternative would be to equalizing fiscal capacity across local governments to ensure certain standards of public goods and services with comparable tax rates may be more appropriate (Blochliger and Charbit 2008; see Section III.D).

  • Effectiveness and externalities. Spatially-targeted policies are more warranted if fiscal interventions are expected to have stronger impact in lagging regions. For example, spatially-targeted policies of hiring incentives could have a higher impact to create jobs and promote growth in regions with high unemployment (Austin, Glaeser, and Summers 2018). Moreover, spatially-targeted policies may have merits if regional disparity of income or unemployment creates adverse externalities for neighboring regions or wider adverse social outcomes (Hendrickson and others 2018; Nunn and others 2018).

Spatially-targeted interventions take many forms but often with mixed effects. Common examples include enterprise zones, industry clusters, and infrastructure (Table 5 and Appendix D). Fiscal instruments include a mix of tax incentives, subsidies, grants, public investment, and welfare spending. The effectiveness of these interventions, particularly over the long-term and countrywide, has been mixed.20 This is because (1) gains in the targeted regions such as higher employment are often temporary and offset by losses in other nontargeted regions; (2) the capacity of lagging regions to attract and retain firms and talents is usually overestimated; and (3) rent-seeking and political capture lead to underperforming results.

Table 5.

Spatially-Targeted Interventions

article image
Sources: World Bank (2009), Neumark and Simpson (2014), and authors' estimates.

A new generation of spatially-targeted interventions is trying to address the shortcomings of traditional place-based policies with a focus on the digital economy. Usually referred to as place-sensitive policies, they consider the higher efficiency in leading regions and focus on innovative industries and digital infrastructure (Iammarino, Rodriguez-Pose, and Storper 2018). Recent proposals include facilitating labor mobility to leading regions, boosting labor (digital) skills, and broadening access to broadband (Hendrickson, Muro, and Galston, 2018).

D. Assign Policies Across Government Levels

Assigning the levels of government that are responsible for policy design, implementation, and financing should be country and context specific. As a general principle, the central government would usually lead on policy design given their stronger capacity and the ability to fully internalize costs and benefits across regions.21 On the other hand, subnational governments (SNGs), by knowing better local preferences and needs, could be in a better position to implement such strategies, as long as they also have sufficient technical capacity and political accountability.22

For spatially-blind policies, the central government is usually responsible for the design and monitoring, while the implementation is sometimes delegated to SNGs. Central governments are better equipped for spatially-blind policies given their broader jurisdictional coverage. The need to ensure uniformity of access to some public goods and services (e.g. universal primary education) and horizontal equity also calls for a centralized approach. SNGs, in turn, can implement policies at preset standards. For instance, the design and monitoring of a social safety net is usually established at the central level while some decentralization of administration and implementation can be appropriate to improve accountability and leverage on information advantage of SNGs (Escolano and others 2015; Martinez-Vazquez and others 2008). On the tax side, there is a case for assigning both the design and implementation of income tax systems to the central government given incentives for local tax competition and their impact on location.

Subnational governments could be more involved in the design and implementation of spatially-targeted policies. SNGs’ jurisdiction and better knowledge over the policy-targeted region would warrant a more decentralized assignment, even when those policies have national objectives and generate an impact outside the targeted regions. Other considerations in assigning government levels to address regional inequality would include:

  • Current level of fiscal and political decentralization. The scope for SNG involvement depends on the country’s degree of fiscal and political decentralization. Higher degree of revenue and spending decentralization will imply SNGs have greater control and capacity to design and implement policies.23 Greater political decentralization would confer more political autonomy at subnational levels (e.g., local elections and institutional checks and balances) making SNGs more attuned with local preference.

  • The design of intergovernmental transfers. The empirical literature reviewed in s in Section II shows that fiscal decentralization may reduce regional inequality if it fosters SNG revenue mobilization efforts and better targeted spending without exacerbating fiscal gaps in some cases. The design of intergovernmental transfers will be critical to this effect. Formula-based equalization transfers—that allow SNGs to provide minimum levels of public services, while granting them full discretion on their own tax-expenditure mix—are shown to achieve these objectives (Boadway and Eyraud 2018).

  • SNG technical capacity. The capacity of SNGs to carry out effective fiscal policies is a key consideration. In the design stage, SNGs need capacity to assess the impact to decide the type, coverage, and beneficiaries of the policy, while in the implementation stage, SNGs need to have sound revenue administration and public financial management systems to monitor compliance and evaluate policy performance (IMF 2009).

Coordination is key among government levels given shared responsibilities. Considering the conditions above, strategies to tackle regional inequality are often shared across different government levels. Hence, strong coordination between government levels is necessary.

IV. Spatially-Targeted Fiscal Redistribution: An Illustrative Exercise

This section illustrates that spatial considerations can help improve the cost-effectiveness of fiscal redistribution programs. Countries often use a variety of measures such as means-tested income support programs, and universal child benefits and social pensions to mitigate inequality. While means-testing is an important criterion to ensure that policies are targeted to intended recipients, it requires administrative capacity to verify information, process applications, and deliver transfers, which if not met, could lead to under-coverage of the poor and leakage of benefits to the rich.

We illustrate whether and under what conditions spatially-targeted policies can mitigate inequality and compare this to typical means-testing methods under the same fiscal envelopes. The illustration uses the latest years of LIS household surveys for five OECD countries: France, Mexico, Poland, the United Kingdom, and the United States. Two cash transfer schemes are considered: universal or means-tested. We look at the cost-effectiveness of each of these two schemes when spatial perspectives are considered. The exercise is meant to be illustrative and does not account for changes in household behavior—including mobility—in response to the measures (details in Appendix D).

Scenario 1: Universal transfers

  • Spatially-blind. Our benchmark consists of a lump-sum cash transfer to all households nationwide, akin to the universal basic income (UBI). Transfers are calibrated at 25 percent of median per capita income nationwide similar to IMF (2017). They are additional to existing programs without considering the financing of the transfers cost. The estimated distributional impact of this program is substantial (a reduction in income inequality of about -0.03 to -0.04 Gini points). The impact is particularly stronger in countries where disposable income is more unequally distributed. Fiscal costs under the program would range between 3.7–6.8 percent of GDP.

  • Spatially-targeted. The above-defined lump-sum cash transfers are now targeted only to lagging regions. In the case of the United States, such a transfer targeted to regions in the lowest 45th percentile regional disposable income would reduce inequality to 0.39 Gini points at a fiscal cost of 0.6 percent of GDP. This would amount to 15 percent of the redistributive effects of the nationwide (spatially-blind) lump-sum cash transfers at just about 10 percent of the fiscal cost (Table 6). The greater cost-effectiveness of this spatially-targeted transfer is due to the relative concentration of poor households in poor regions. This concavity is observed across selected countries, underscoring that spatially-targeting has slightly better cost effectiveness in mitigating inequality.

Table 6.

Gross Fiscal Cost and Redistributive Impact of Regional-Targeted Transfers

article image
Sources: LIS database and Fiscal Monitor (2017)

Gross fiscal cost for country-level unversial basic income is obtained from October 2017 Fiscal Monitor.

Spatially-targeted to regions with average income less than 45-percentile of all regional incomes in the national distribution.

Scenario 2: Means-testing transfers

  • Spatially-blind. The benchmark in this scenario is a spatially-blind means-tested program, defined as a lump-sum cash transfer to low-income households in the lowest two deciles of the nationwide income distribution regardless of where they live. This means-tested program has some leakages, which are calibrated identical to IMF (2017) to the ratio of current noncontributory transfer relative to those eligible. The total transfer to those low-income households (fiscal cost) is normalized to be 1 percent of GDP for comparison purposes (Table 5).

  • Spatially-targeted. The above mean-tested program is now targeted only to lagging regions— those regions with income level below the 45-percentile average regional income—with the transfers to the lowest two deciles of the income distribution. The total transfer amount to low-income households in those regions (fiscal cost) is set to be 1 percent of GDP. The results show that the spatially-targeted means-tested program would be more cost-effective (i.e. greater decline of Gini coefficients conditional on the same fiscal cost) than typical means-testing if leakages in the spatially-blind program are large and low-income households are more concentrated in lagging regions (Table 7 and Figure 9 top left and right panels). For example, countries with low leakage such as the United Kingdom and France will find standard (spatially-blind) means-testing better than a means-tested program targeting lagging regions.

Table 7.

Illustrative Redistributive Effects of Regional-Targeted Transfers Relative to Means-Testing at Different Coverage Levels

article image

Current coverage level is obtained from October 2017 Fiscal Monitor based on LIS database.

Cells in bold indicate stronger redistributive power for regional targeted transfers relative to current coverage levels under the same fiscal cost.

Figure 9.
Figure 9.

Regional Considerations Could be Complementary to Current Means-Testing Measures in Tackling Inequality

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

In addition to coverage and concentration of the poor, how spatially-targeted redistribution is financed matters. The illustrative exercise above shows a new transfer program that incurs fiscal costs. Alternatively, the new transfer program may substitute for an existing program. In that case, whether spatially-targeting would generate net benefits would depend on the effectiveness and progressivity of the existing program. For example, replacing the current transfer programs— particularly if those are well-targeted and sufficiently progressive such as in France or the United Kingdom—could weaken the overall fiscal redistribution. If the current program is progressive (i.e., higher redistributive power for each unit of fiscal cost) then replacing such a program with a spatially-targeted variant may not be optimal from the redistributive perspective.

Scenario 3. Means-tested transfer with budget allocation proportional to regional concentration of poor.

  • Spatially-blind. The benchmark in this scenario is a spatially-blind means-tested program (same as scenario 2) to low-income households in the lowest two deciles of the nationwide income distribution, subject to imperfect coverage observed in the data.

  • Spatially-targeted. It considers the same total transfers (budget neutral) but distributed to the households in the lowest income quintile according to the regional concentration of the poor, that is, if a region has 10 percent of population in the lowest income quintile, it receives 10 percent of total transfers, split among the poor households in the area. Given that conventional means-testing has imperfect coverage for the households in the lowest income quintile, transfers that accounts for geography of where low-income households live would improve the overall fiscal redistribution effects. The improvement varies by countries, ranging from 7–15 percent, depending on the progressivity and coverage of current spatially-blind means-tested programs and the concentration of poor in those countries (Figure 9 bottom left panel). For example, the United Kingdom has good coverage and progressive programs so that the improvement from spatial considerations are expectedly small. But for countries like Mexico and Poland where the coverage is relatively low and the poor households are highly concentrated, improvements are likely larger (Figure 9 bottom right panel).

Based on the illustrative results, policies that consider spatial dimensions can complement the existing means-testing in tackling regional inequality, particularly when (1) low-income households are concentrated in lagging regions with low state income or high unemployment rates and household mobility is limited; 2) horizontal equity concerns are less prevalent; 3) targeting to poor households through means-testing is difficult subject to leakages; 4) existing programs are not very progressive or effective in fiscal redistribution.

The process for determining the level(s) of government for spatially-targeted measures should be similar to those in Section III.D. This would include the choice of government levels in the design, implementation, monitoring, and/or financing for the government measures. Finally, our analysis focuses on state or provincial levels but could be generalized to tailor lower-regional levels such as cities or counties or urban versus rural.

Caveats. Horizontal-equity concerns may arise in which households facing the same outcomes (e.g., poverty) may not receive the same benefits because of their residency. This could raise political economy issues and make the design and the adoption of spatially-targeted polices more challenging. Second, as in other means-tested transfers, spatial-targeting could bring incentive problems that encourage abuse by misreporting residency and generate distortions in locational choice. Third, programs to account for regional dimensions also depend on the government capacity, which is not considered here.

V. Conclusions

This paper illustrates that regional dimensions are important in the discussion of the nature of inequality and policy options to address it. It presents several new stylized facts. Regional inequality in many OECD countries has been large (even adjusting for regional price differences), persistent, and widening over time. This implies regional divergence in some OECD countries. Rising regional disparity in income is in part associated with declining regional labor mobility in a country, particularly for low-income households in lagging regions. These trends could lead to wider economic underperformance if left unaddressed as lagging regions fall behind their potential and possibly drag nationwide growth.

The paper provides a conceptual policy framework to address regional inequality. Identifying key determinants and determining a strategy—place-based or spatially-blind—and assigning the appropriate levels of government will be critical. In an environment where typical fiscal redistributive policies have turned less effective in tackling regional inequality, our illustrative exercise suggests that measures that account for regional dimensions could complement national means-tested transfers. Spatially-targeted fiscal redistribution would be more effective when low-income households are concentrated in lagging regions and face barriers to move, and existing means-testing faces leakages and low coverage.

VI. Appendix

A. Evolution of Regional Disparity on Unemployment Rates

Figure A.1.
Figure A.1.

Different Trends on Regional Disparity of Unemployment Rates for Selected OECD Countries (2000–17)

Citation: IMF Working Papers 2019, 088; 10.5089/9781498311625.001.A001

Sources: OECD Regional Database and IMF staff estimates.

B. Empirical Estimation on Key Determinants of Regional Inequality

Potential key determinants of regional inequality are estimated based on an unbalanced panel of 18 OECD countries over the period 1984–2016. Panel regression allows for better precision in estimation, although choosing the right panel model is a challenge. This section uses a clear specification strategy to select the suitable model (Baltagi 2005). Using the Least Square Dummy Variable (LSDV), the Wald test shows that the country specific dummies are not jointly equal to zero. The Lagrange multiplier test suggests that the individual specific errors are uncorrelated to the regressors. The Sargan-Hansen statistic concludes the Fixed Effects model (FE), relative to Random Effects (RE) models, is the most appropriate even when controlling for potential heteroskedasticity as suggested by the Wald statistics. In that context, the estimation uses the FE model while controlling for heteroskedasticity and temporal dependencies using the Huber-White Sandwich estimator following the statistical test for the presence of serial correlation in an unbalanced panel.

The estimation also accounts for the presence of endogeneity arising from reverse causality, simultaneity, or omitted variable bias. For instance, countries with higher regional inequality may have larger intergovernmental transfers (Lessmann and Siedel 2017) or persistent regional inequality may lead to greater preference towards deeper fiscal decentralization (Bartolini and others 2016). For simultaneity, potential migrants may choose destination regions based on how income inequality is, partly because new migrants are more likely at lower income class, while governments develop fiscal redistribution policy considering the impact on their immigration policy. While the fixed effect estimator partially accounts for the reverse causality concern, we also use the two-stage least-squares within estimator (FE2SLS) to address any estimation bias from endogeneity (Drinkwater 2003).24 We use lagged explanatory variables and other exogenous variables (the unemployment rate, the share of housing costs in household income, ratio, and the share of nonworking age population the dependency ratio etc.) selected on relevance and validity tests as instrumental variables.

Table B.1.

Choice of Estimation Methods

article image
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Estimation Results. The baseline result uses the FE2SLS estimator to control for endogeneity, comparing with the results using a fixed effect estimator (FE) with clustered standard errors (Table B.1). Main empirical results are discussed in the main text, with regression results showing the key determinants of regional disparity on per-capita GDP in Table 3 in the main text. The effects of fiscal decentralization on regional inequality in unitary and federal countries is listed in Tables B.2 and B.3.

Table B.2.

Effects of Fiscal Decentralization on Regional Inequality

article image
Note: FE2SLS estimation; instruments include lagged values of fiscal decentralisation, net migration, and other exogenous variables like the unemployment rate, the share of housing cost in household income, ratio, the share of non-working age population, the dependency ratio. Instruments are selected based on their relevance and validity using Kleibergen-Paap rank underidentification and Hansen J overidentification tests. Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1.
Table B.3.

Robustness: Including Year Fixed Effects

article image
Note: FE2SLS estimation; similar instruments as before. Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1.

C. Interstate Migration and Regional Inequality

Returns of interregional migration are estimated in two steps. First, the effects of location on household income is estimated using a variant of Milanovic (2015) for households over the past decade during 2004–13 using the LIS household income surveys. Second, we use the estimated coefficients to calculate the returns of interregional migration across households in different income classes.

The Milanovic (2015) specification shows that about 90 percent of individual household income can be explained by the average income level and its income distribution of the country in which the individual resides in. His estimation follows a specification: yi=c+βy¯j+γGinij+ϵi where y is the individual’s (i) household income, y¯j, is the country’s (j) average income level (per-capita GDP), and Ginij is the Gini coefficient measuring income distribution for the country.

Extending the approach, we estimate separately for 13 countries with data available in the LIS database the following:

yi,s=c+βy¯s+γGinis+εi,sforeachincomeventile(5thpecentile)(1)

where yi denote individual household disposable income; s is the state / region the household resides in; y¯s is the state average income; Qnis is the state average inequality of income distribution, measured by state Gini-coefficient; e is the residual term. The estimation is carried out individually for each income ventile (5th percentile, ranked from the poorest to the richest). The above can be rewritten with dummy variable (1˜Dν) representing each income ventile (v):

yi,s=cν+βνDνy¯s+γνDνGinis+ϵi,s(2)

Empirical results show that the state average income level and its distribution explain about 80–90 percent of the variability of individual household income in most ventiles.25 In addition, the estimated coefficients vary across income ventiles. While households benefit uniformly from higher state average income (estimated β at about 1), their income is affected differently by the state income distribution, i.e., estimated (γ) rises with income classes, ranging from -3.5 to 1.2 (Table C.1).

Table C.1.

Regression Results of Individual Household Income on Regional Income Level and Distribution

(based on 2013 Household Income Survey for the United States)

article image
Sources: LIS, Milanovic (2015), and authors’ estimates.Note: Here only shows the coefficients on regional income distribution in without other coefficients on regional income level (β).Market and disposable income refer to before and after tax and transfer, respectively. Adjustment on regional prices and housing cost is discussedin the main text (Fact 1).Total observations for the 2013 U.S. household income survey in LIS are 51,498 households.

The results imply that when low-income households would prefer a region with less income inequality when choosing between two regions of similar regional average income to relocate. A one Gini point increase in the regional inequality would reduce 3 percent of household income on average. To offset the negative effect, the household in the bottom ventile would need to live in a region with higher average state income by 3.2 percent. On the other hand, a rise of income inequality would benefit the richest by 1 percent for the same average state income level. The estimated γ is close to 0 for the upper-middle income class (75th-90th percentile), implying that the average state income level (β) is a key determinant to their income and state inequality does not matter much.

The second step will use the estimated coefficients to calculate the returns of interstate mobility, accounting for differences of regional prices and housing costs. The return of interstate migration (Rs,s’) for moving from the incumbent region (s) to a new region (s’) in the country can be expressed as:

(Rs,s)=(1hν)y^sPs(1hν)y^sPs0(1hν)(cν+βν^Dνγν^DνGinis)Ps(1hν)(cν+βν^Dνy¯s+γν^DνGinis)Ps0(3)

where hv is the share of implicit housing costs to household disposable income based on the LIS database for the vth ventile income class; Ps denotes the price level in region s, obtained from the U.S. Census Bureau of Analysis or approximated by nontradable goods consumption share (about one-third of housing costs based on Gennaioli and others 2014) for other countries that do not report regional prices. The returns across income class for the United States for 2004 and 2013 across 50 states using the LIS household income surveys (Rs,s2004);(Rs,s2013) are calculated.26

Table C.2.

Adverse Effects of Income Inequality on Individual Household Income across Income Classes

article image
Sources: LIS and IMF staff estimates.1/ Low-income households refer to those below 40th percentile of the household income in the country. Middle-income refers to household income between 40th-75th percentile, and high-income households are those with income higher than 75th percentile.

Besides the differences in the average income level and income distribution between incumbent and destination regions reflected in the first step (equations 1 and 2), two additional components (regional price levels (Ps, Ps’) and share of housing costs in disposable income (hv)) will also affect the return of interstate migration.

Caveats. Our calculation of returns is based on estimated coefficients on state average income level and income distribution. The household choice of locations may depend on individual skills and other nonpecuniary returns.

D. Spatially-Targeted (Place-based) Policies: Evidence and Lessons

Spatially-targeted policies could promote growth and employment in lagging regions:

  • 1. Spatially-targeted policies such as affordable housing in certain regions could facilitate labor mobility. Individuals in lagging regions may have low skills and face unemployment because of shrinking jobs following the exit of local industries (e.g., coal mining). They are less capable to move out because of job mismatch and lack of skills suitable in leading regions, further hindered by limited supply of affordable housing in leading regions (Neumark and Simpson, 2014).

  • 2. Pro-employment policies (e.g., hiring incentives) targeted in regions may have large impact in regions with higher share of nonworking age population. This could address the negative spillovers posed by widespread unemployment (or nonworking age population) in lagging regions, as shown in the midland regions in the United States by Austin, Glaeser, and Summers (2018). The hardship could be magnified across individuals because of labor skills becoming obsolete and nonemployment reduces local demand, which lead to further decline of labor demand (Topa and Zenou, 2014).

  • 3. Spatially-targeted policies have potentials to address externalities and noneconomic barriers. Nationwide growth may depend increasingly on tapping the growth potentials in lagging regions because the agglomeration in leading regions may lead to congestion externalities, such as long commute and pollution, which could limit the benefits. On the other hand, individuals may face noneconomic barriers from their ethnolinguistic groups and religion, and reside in remote or adverse areas (e.g. mountains and deserts) that spatially-blind policies may be difficult to cover.

However, the effectiveness of spatially-targeted policies is mixed and less conclusive. In the examples of place-based enterprise zones in the United States and France, the net gains in employment are not tied to the hiring incentives offered in the zones and much of the gains accrue to high-income households who own real estate nearby. Tax breaks in the zones mainly relocate activity from outside to inside the zones without much net gains (Neumark and Simpson, 2014). Spatially-targeted interventions are more likely to succeed if they exploit geographical advantages such as the proximity to external markets (World Bank, 2009).

Regarding industrial clusters, evidence from the United States suggests that targeted regions benefit from positive productivity spillovers, but such gains are localized and industry-specific. Gains are higher if industrial clusters focus on technology and have proximity and links to university research. The effectiveness of large-scale regionally-targeted infrastructure programs has been mixed. Evidence from the U.S. interstate highways and European structural funds shows that large-scale infrastructure projects can improve short-term growth and productivity but less certain if it has long-lasting impact (World Bank, 2009, Neumark and Simpson, 2014). Some regionally-targeted public investment decisions might not fully consider externality on social returns and costs (Jarowski and Kitchens, 2016, De La Fuente and Vivis 1995, Crescenzi and Rodriguez-Pose 2012, and Yu et al 2012).

E. Illustrative Exercise

The scenarios consist of giving a lump-sum cash transfer to selected households in a country.

Existing programs consist of progressive tax and transfer that brings the market income (Ymkt) to disposable income (Yd), with the corresponding inequality measure as the Gini coefficient denoted as G(Ymkt) and G(Yd), respectively. The fiscal redistribution effect is measured by the difference between Gini-coefficient before and after taxes and transfers G(Ymkt)-G(Yd)

Yd=Ymkt(1τ(Ymkt))+T(4)

where τ(Ymkt) is a progressive tax and T is the current transfers. The fiscal cost can be denoted with the function F(τ, T) with taxes and transfers as parameters.

Scenario 1. A spatially-blind universal transfer for all individual households would imply an additional transfer of T’ for all households nationwide (i.e., integrating all across households). The aggregate income will be:

YUniv=iYd(i)+Tdi(5)

The size of the transfer is set to be 25 percent of median equivalent household income after tax. The fiscal redistributive impact will therefore be expressed as G(Ymkt)G(YUniv) and fiscal cost F(τ, T, T’). The cost effectiveness of the program can be measured [G(Ymkt)-G(Yuniv)] / F(T’).

A spatially-targeted lumpsum transfer would target all households residing in low-income regions, defined as those in the two lowest bottom income deciles. Integrating across all individuals nationwide would involve:

Yreg=i|Y¯(j)BYd(i)+Tdi+i|Y¯(j)>BYd(i)di(6)

where Y(j) is the average regional income level in region j, B is the policy threshold. A spatially-targeted lumpsum transfer is given to households in regions with the lowest regional average income level less than the policy threshold B—set to be the lowest bottom two declines of income. The cost effectiveness of the transfer is measured by [G(Ymkt)-G(Yreg)]/ F(T’, B)].

Scenario 2. The spatially-blind means-testing transfer with imperfect coverage would involve a cash transfer to those who are below a certain income threshold (i.e., households in the bottom two deciles by nationwide income classes).

YMT=i|Ymkt(i)YpYd(i)+T(1α)di+i|Ymkt(i)Y20Yd(i)di(7)

where Yp is the income threshold of low-income households of the transfer program (e.g., bottom two deciles will be Y20). The imperfect coverage is denoted as the leakage (a) of the cash transfers T’. The fiscal redistributive impact is G(Ymkt)-G(YMT), while the cost effectiveness is scaled by the fiscal cost as [G(Ymkt)-G(YMT)] / F(T’, a,Y2o)].

The spatially-targeted transfer with means-testing would involve a cash transfer to those who are below a certain income threshold and reside in regions with the lowest regional average income level less than the policy thresholds.

YMTreg=i|Ymkt(i)Yp&Y¯(j)BYd(i)+Tdi+i|Ymkt(i)YporY¯(j)>BYd(i)(8)

The fiscal redistributive impact will be the difference of Gini coefficients, denoted as G(Ymkt)-G(YMTreg). The cost effectiveness can then be scaled by the fiscal cost as [G(Ymkt)-G(YMTreg)]/F(T’,B,Y20)

The exercise scales the fiscal cost in each type of transfers to 1 percent of GDP for comparison across programs that have different coverage, income thresholds, and spatial thresholds (Table 5 in the main text). There is a tradeoff between the leakage of means-testing (α) and the concentration of low-income households within the regional threshold. Higher coverage of means-testing (i.e., lower leakage) would favor spatially blind means-testing while higher concentration of poor in the low-income regions would favor spatially targeted redistribution.

Financing. If the new spatially-targeted lump-sum cash transfer is financed by substituting current programs, the net improvement or deterioration of fiscal redistribution can be obtained by G(Ymkt)-G(Yreg)]/ F(T’, B)] – [G(Ymkt)-G(Ycurrent)]/ F(.)].

Calibration. We use the latest year of individual household surveys from LIS database for five countries and calibrate the lump-sum transfer to be 25 percent of the median equivalent household income nationwide before taxes. The imperfect coverage varies by countries ranging from 46 to 83 percent observed in the data (IMF 2017 Table A.1.6) and considers variants of the parameters from 60–90 percent. For means-testing transfers, the income threshold is set at the bottom two deciles by income class (i.e., Yp = Y20) and considers regional-targeted transfers for the regions with average household income between 40th to 60th percentile of the average income distribution across regions.

VII. References

  • Arpaia, A., A. Kiss, B. Palvolgyi, and A. Turrini. 2014. “Labor Mobility and Labor Market Adjustment in the EU”, European Economy, European Commission, December 2014.

    • Search Google Scholar
    • Export Citation
  • Austin, B., E. Glaeser, and L. Summers. 2018. ‘Saving the Heartland: Place-based Policies in 21st century America’, Brookings Paper on Economic Activity BPEA Conference Draft, March 2018.

    • Search Google Scholar
    • Export Citation
  • Autor, D., and D. Dorn, and G. Hanson. 2013. ‘The China syndrome: Local Labor Market Effects of Import Competition in the United States’, American Economic Review, vol.103, no.6, pp.212168, October 2013.

    • Search Google Scholar
    • Export Citation
  • Baltgai, B. 2005. ‘Econometric Analysis of Panel Data’, Third edition, John Wiley and Sons, Ltd.

  • Barro, R. and X. Sala-i-Martin. 1991. ‘Convergence across States and Regions’, Brookings Papers on Economic Activity, 1991 No.1, pp.107182.

    • Search Google Scholar
    • Export Citation
  • Bartolini, D. 2015. “Municipal Fragmentation and Economic Performance of OECD TL2 Regions”, OECD Regional Development Working Papers, 2015/02, OECD Publishing, Paris.

    • Search Google Scholar
    • Export Citation
  • Bartolini, D., S. Stossberg, and H. Blochliger. 2016. ‘Fiscal Decentralization and Regional Disparities’, OECD Economics Department Working Papers, October 2016.

    • Search Google Scholar
    • Export Citation
  • Bastagli, F., D. Coady, and S. Gupta. 2012. ‘Income Inequality and Fiscal Policy’, IMF Staff Discussion Note, SDN/12/08, June 2012.

  • Bayoumi T. and J. Barkema. Forthcoming. ‘Tarnishing the American Dream: how Rising Inequality Suppressed U.S. Migration and Hurt Those Left Behind”.

    • Search Google Scholar
    • Export Citation
  • Boadway, R. and L. Eyraud. 2018. ‘Designing Intergovernmental Transfers to Achieve Fiscal Balance in Decentralized Countries’, IMF Working Paper WP/18/271.

    • Search Google Scholar
    • Export Citation
  • Blochliger H. and C. Charbit (2008) ‘Fiscal EqualizationOECD Economic Studies No.44, 2008/1.

  • Blochliger H. and M. Nettley (2015) ‘Sub-central Tax Autonomy’, No.20. OECD March 2015

  • Bouchet M. and J. Parilla. 2017. ‘Getting Smarket on U.S. Regional Inequality: Lessons from Europe’, Brookings the Avenue, October 2017. https://www.brookings.edu/blog/the-avenue/2017/10/27/getting-smart-on-u-s-regional-inequality-lessons-from-europe/

    • Search Google Scholar
    • Export Citation
  • Brezzi M. and P. Luongo. 2016. ‘Regional Disparities in Access to Health Care: A Multilevel Analysis in Selected OECD Countries’, OECD Regional Development Working Papers, 2016/04.

    • Search Google Scholar
    • Export Citation
  • Coady, D., M. Grosh, and J. Hoddinott. 2004. ‘Targeting Transfers in Developing Countries: Review of Lessons and Experience’, The World Bank, Washington, D.C.

    • Search Google Scholar
    • Export Citation
  • Crescenzi, R. and A. Rodriguez-Pose. 2012. ‘Infrastructure and Regional Growth in the European Union’, CEPR Discussion Paper No. DP8882, April 2012.

    • Search Google Scholar
    • Export Citation
  • Dabla-Norris, E. K. Kochhar, N. Shphaphiphat, F. Ricka, and E. Tsounta. 2015. ‘Causes and Consequences of Income Inequality: A Global Perspectives’, IMF staff Discussion Note, SDN/15/13, June 2015.

    • Search Google Scholar
    • Export Citation
  • Drinkwater, S. 2003. ‘Estimating the Willingness to Move within Great Britain: Importance and Implications’, Department of Economics Discussion Paper No.1203, University of Surrey.

    • Search Google Scholar
    • Export Citation
  • Easterly, W. 2007. ‘Inequality does Cause Underdevelopment: Insights from a New Instrument’, Journal of Development Economics, vol.84/2, pp.755776, November 2007.

    • Search Google Scholar
    • Export Citation
  • Escolano, J., D. Benedek, H. Jin, C. Mulas Granados, M. Nozaki, J. Pereira, G. Rota Graziosi, L. Sinn, and J. Torres. 2015. “Distribution of Fiscal Responsibilities in Federations”, Designing a European Fiscal Union: Lessons from the Experience of Fiscal Federations, edited by Cottarelli, C. and M. Guerguil, Routledge.

    • Search Google Scholar
    • Export Citation
  • Fischer, C. 2017. ‘Inequality is about Access to Public Goods, Not Income’, Boston Review, April 2017. http://bostonreview.net/class-inequality/claude-s-fischer-inequality-about-access-public-goods-not-income

    • Search Google Scholar
    • Export Citation
  • Galbraith, J. and J.E. Garcilazo. 2010. ‘Inequalities, Employment and Income Convergence in Europe: Evidence from Regional Data’, International Review of Applied Economics, No.24(3), pp.359377, May 2010.

    • Search Google Scholar
    • Export Citation
  • Ganong P. and D. Shoag. 2017. ‘Why has Regional Income Convergence in the U.S. Declined?NBER Working Paper No.23609, July 2017.

  • Gennaioli N., R. La-Porta, F. Lopez de Silanes, A. Shleifer. 2014. “Growth in Regions”, Journal of Economic Growth 19:259309.

  • Gill, I. 2010. ‘Regional Development Policies: Place-based or People-centered?Vox CEPR Policy Portal, October 2010. https://voxeu.org/article/regional-development-policies-place-based-or-people-centred

    • Search Google Scholar
    • Export Citation
  • Goerl, C.A. and M. Seiferling (2014) “Income Inequality, Fiscal Decentralization, and Transfer DependencyIMF Working Paper No.14/64.

    • Search Google Scholar
    • Export Citation
  • Goesling, Brian. (2001). Changing Income Inequalities within and between Nations: New Evidence. American Sociological Review. No. 66(5). 745, October 2001.

    • Search Google Scholar
    • Export Citation
  • Hendrickson, C., M. Muro, and W. Galston. 2018. ‘Countering the Geography of Discontent: Strategies for Left-behind places’, Brookings, November 2018 https://www.brookings.edu/research/countering-the-geography-of-discontent-strategies-for-left-behind-places/

    • Search Google Scholar
    • Export Citation
  • Hsieh, C.T. and E. Moretti. 2015. ‘Housing Constraints and Spatial Misallocation’, NBER Working Paper No. 21154.

  • Hopkins, E., F. Bastagli, and J. Hagen-Zanker. 2016. ‘Internal Migrants and Social Protection: A Review of Eligibility and Take-up’, Working Paper, April. London: ODI.

    • Search Google Scholar
    • Export Citation
  • Iammarino, S., A. Rodriguez-Pose, and M. Storper. 2018. “Regional Inequality in Europe: Evidence, Theory, and Policy Implications’, Journal of Economic Geography, April 2018.

    • Search Google Scholar
    • Export Citation
  • Immervoll, H. and L. Richardson (2011) “Redistribution Policy and Inequality Reduction in OECD Countries: What Has Changed in Two Decades?”, Institute for the Study of Labor (IZA) Discussion Paper No. 6030, December 2011.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund. 2009. IMF, 2009. ‘Macro Policy Lessons for a Sound Design of Fiscal Decentralization’, IMF Policy Paper, Washington, D.C.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund. 2017. ‘Fiscal Monitor: Tackling Inequality’, October 2017, IMF, Washington, D.C.

  • International Monetary Fund. 2018. ‘How to Operationalize Inequality Issues in Country Work’, International Monetary Fund, June 2018.

    • Search Google Scholar
    • Export Citation
  • Jaworski, T. and C. Kitchens. 2016. ‘National Policy for Regional Development: Evidence from Appalachian Highways’, NBER Working Paper No. 22073, March 2016.

    • Search Google Scholar
    • Export Citation
  • Kaplan, G. and S. Schulhofer-Wohl. 2015. ‘Understanding the Long-run Decline in Interstate Migration’, Federal Reserve Bank of Minneapolis, Working Paper No.697, October 2015.

    • Search Google Scholar
    • Export Citation
  • Kim J.H. and S. Dougherty. 2018. ‘Fiscal Decentralization and Inclusive Growth’, OECD Fiscal Federalism Studies Launch Version.

  • Kline, P. and E. Moretti (2014). ‘People, Places, and Public Policy: Some Simple Welfare Economics of Local Economic Development Program’, Annual Review of Economics, vol.6, pp.629662, August 2014.

    • Search Google Scholar
    • Export Citation
  • Lessmann, C. and A. Seidel. 2017. ‘Regional Inequality, Convergence, and its Determinants—A View from Outer Space’, European Economic Review, 2017, vol.92, issue C, pp.110132.

    • Search Google Scholar
    • Export Citation
  • Liu, L.Q. 2018. “Regional Labor Mobility in Spain”, IMF Working Paper No.18/282, Washington, D.C.

  • Lucas, R. 1988. ‘On the Mechanics of Economic Development’, Journal of Monetary Economics, 22, pp.342.

  • Martinez-Vazquez, J., B. Qiao, S. Wang, and H. F. Zou. 2008. “Expenditure Assignments in China: Challenges and Policy Optionsin Public Finance in China: Reform and Growth for a Harmonious Society, edited by J.W. Lou, S. Wang, World Bank, Washington, D.C.

    • Search Google Scholar
    • Export Citation
  • Milanovic, B. 2015. ‘Global Inequality of Opportunity: How Much of Our Income is Determined by Where We Live?’, Review of Economics and Statistics, vol.97/2, pp.452460, May 2015.

    • Search Google Scholar
    • Export Citation
  • Moretti, E. 2012. ‘The New Geography of Jobs’. Houghton Mifflin Harcourt, Boston.

  • Musgrave, R.A. 1959. ‘The Theory of Public Finance’, McGraw Hill, New York.

  • Neumark, D. and H. Simpson. 2014. ‘Place-based Policies’, NBER Working Paper No. 20049.

  • Nunn, R., J. Parsons, and J. Shambaugh. 2018. ‘Americans Aren’t Moving to Economic Opportunities’, Brookings, November 2018. https://www.brookings.edu/blog/up-front/2018/11/19/americans-arent-moving-to-economic-opportunity/

    • Search Google Scholar
    • Export Citation
  • Oates, W. 1972. ‘Fiscal Federalism’, The Harcourt Brace Jovanovich, 1972.

  • Organisation of Economic Cooperation and Development (OECD). 2009. ‘How Regions Grow: Trend and Analysis’, OECD, Paris.

  • Organisation of Economic Cooperation and Development (OECD). 2016. ‘Regions at a Glance 2016’, OECD, Paris.

  • Organisation of Economic Cooperation and Development (OECD). 2018. ‘OECD Territorial Grids”, OECD Centre for Entrepreneurship, SMEs, Regions and Cities’, OECD March 2018.

    • Search Google Scholar
    • Export Citation
  • Ostry, J., A. Berg, and C. Tsangarides. 2014. ‘Redistribution, Inequality, and Growth’, IMF Staff Discussion Note, SDN/14/02, February 2014.

    • Search Google Scholar
    • Export Citation
  • Ramirez, M., T. Liebig, C. Thoreau, and P. Veneri (2018) “The integration of migrants in OECD regions”, OECD Reginal Development Working Papers, 2018/01, OECD Publishing, Paris.

    • Search Google Scholar
    • Export Citation
  • Rodriguez-Posé, A. and R. Ezcurra (2010), “Does Decentralization Matter for Regional Disparities? A Cross-Country Analysis”, Journal of Economic Geography, Vol. 10, pp. 619644.

    • Search Google Scholar
    • Export Citation
  • Roses, J. R., and N. Wolf (2018) “Regional Economic Development in Europe, 1900–2010: A Description of the Patterns”, CEPR Discussion Paper No.17249.

    • Search Google Scholar
    • Export Citation
  • Shankar, R. and A. Shah. 2003. ‘Bridging the Economic Divide Within Countries: A Scorecard on the Performance of Regional Policies in Reducing Regional Income Disparities’, World Development, vol.31/8, pp.14211441.

    • Search Google Scholar
    • Export Citation
  • The Economist. 2016. ‘Regional Inequality is a Hard Problem to Solve’, The Economist print edition, December 1, 2016.

  • The Economist. 2017. ‘Left in the Lurch: Globalisation has Marginalized Many Regions in the Rich World’, The Economist print edition, October 21, 2017.

    • Search Google Scholar
    • Export Citation
  • Valdesalici, A. and F. Palermo. 2018. ‘Comparing Fiscal Federalism’, ed. by A. Valdesalici and F. Palermo, Brill Nijhoff, Boston.

  • Winkler, H. 2018. ‘The Effect of Income Inequality on Political Polarization: Evidence from European Regions, 2002–14’, World Bank, mimeo.