## wpiea2019122

## Source details

**Canonical URL:** [wpiea2019122](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019122.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019122.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019122.pdf.json)

---

### Introduction: decline in U.S. long-distance migration and link to inequality
- Interstate migration of the working population halved between 1980 and 2016 (Current Population Survey).
- Decline in migration reduces labor market churning, lengthens downturns, slows recoveries, and is most marked out of poorer areas, exacerbating geographic inequality.
- Core finding: rising differences in house prices and incomes are linked to the fall in migration.
  - House price divergence plays the dominant role discouraging moves from poorer to richer metro areas.
  - Income divergence reduces moves in the opposite direction.
- Mechanistic insight:
  - Centrifugal effects of rising wealth inequality (house prices) dominate centripetal effects of rising income inequality for moves from poor to rich metro areas.
  - For moves out of prosperous areas, the centrifugal force of moving to lower incomes dominates the centripetal effect of lower house prices.
- Macro consequence: lower labor market churning leads to less efficient labor allocation, reducing productivity and output.

### Stylized facts: migration, house prices, and incomes
- Long-distance migration decline:
  - CPS Annual Geographical Mobility Rates: inter-state migration fell from 3.0 percent in 1981 to 1.5 percent in 2016.
  - Intra-state migration fell by about a quarter over comparable periods.
  - IRS address-based migration: 2.9 percent in 1990 to 2.4 percent in 2015.
  - Job-related motive prevalence: 34.3 percent of moves across counties in 2015 versus 20.2 percent of moves within a county (2015).
  - Educational gradient: in 2016, those with education beyond high school were almost twice as likely to move to another state than those with only high school.
- Rising divergence in house prices and incomes (1996–2016, Zillow):
  - House price dispersion (std dev of log median home values) widens by nearly 50 percent.
  - Income dispersion (std dev of log median incomes across 381 CBSAs) grew by 20 percent.
  - Ratio of log median house prices to median incomes increased by 39 percent.
  - Bilateral observations: 80 percent represent simultaneous upward or downward movements in both house prices and income.
- Extended series to 1981 (FHFA and county income data) show upward trends in standard deviations of log house prices and incomes contemporaneous with the decline in long-distance mobility.

### Data, sample, and measurement
- Migration data:
  - Source: Internal Revenue Service’s Statistics on migration (county-to-county migration across the United States from 1990-2015).
  - Coverage: 323 CBSAs with overlap of Zillow and income databases (35 percent of total number of CBSAs, but 82 percent by population).
  - Sample size: roughly 200,000 observations over 20 years (1996-2015 sample used for regressions).
  - Migration tracked by changes in tax return mailing address; excludes some low-income families who do not file taxes.
  - Focus on job-related migration: moves between CBSAs of over 200 miles.
  - Three-year averages used to reduce noise; adjusted for reporting cut-off change in 2013.
- Variables and normalization:
  - Gravity normalization: bilateral migration divided by the square root of the product of their respective populations.
  - HP: log median house price in destination CBSA minus log median house price in source CBSA.
  - I: log median income in destination CBSA minus log median income in source CBSA.
  - Distance: log of distance between CBSA centers.
  - Controls (X): proportion of population over 60 in source and destination; average household annual gross income (AGI) of migrants; relative regional unemployment; relative population growth; distance.
  - Instrumental variables: averages of incomes and house prices of other CBSAs within a 200-mile radius; house price and income differentials from 15 years earlier.

### Mechanisms and dynamics
- Self-reinforcing process:
  - Limited space in high house-price metros attracts skilled workers, increasing skilled share, which further raises wage and home price inequality.
  - Technology and agglomeration effects increase wage premium for skilled workers even with larger skill supply.
  - Rising house prices relative to incomes price out lower-earning households, reducing migration from poorer to more productive metro areas.
- Broader implications:
  - Divergences in returns to land (house prices) and returns to labor (earnings) create asymmetric migration responses, reducing labor reallocation where it would most enhance opportunity.
  - Lower long-distance migration contributes to falling economic dynamism and potentially lower aggregate output.

### Empirical strategy and Bartik-shock evidence
- Bartik shocks:
  - Computed using 1998 industry compositions; shocks equal changes in national employment by sector each year weighted by initial CBSA employment structure.
  - Regressions of Bartik shocks on migration, house price divergence, and income divergence from 1998 with CBSA and time fixed effects.
- Key Bartik findings:
  - A Bartik shock of 1 percent of employment in one metro area relative to another:
    - Leads to an immediate and highly statistically significant increase in relative house prices of over 2 percent.
    - Leads to an immediate and highly statistically significant increase in relative income of ¾ percent.
    - Each represents around 5 percent of the typical gap across metro areas.
  - Persistence: between half and two-thirds of the impact remains after 4 years.
  - Migration response: a relative Bartik shock of 1 percent has no significant contemporaneous impact on migration; coefficient becomes significant with 4-year lags, implying bilateral migration dwindles in response to favorable relative employment shocks.
  - Interpretation: employment shocks move prices rather than people.

### Migration regressions — baseline and asymmetry
- Baseline IV results (regression 4):
  - HP coefficient: -0.0041 percentage point — a 1 percent increase in house prices lowers the proportion of the population migrating by 0.0041 percentage point (negative and highly significant).
  - I coefficient: 0.0111 percentage point — a 1 percent increase in income raises the proportion of the population migrating by 0.0111 percentage point (positive and highly significant).
  - Economic magnitude: given median bilateral migration coefficient .01206, a 1 percent increase in house price (income) dispersion leads to 0.33 percent decrease (0.92 percent increase) in bilateral migration.
  - Controls: older populations discourage mobility; higher AGIs increase migration; higher relative unemployment discourages mobility; larger divergence in population growth increases migration; longer distances discourage mobility.
- Asymmetric uphill/downhill specification (regression 5):
  - D_UH = 1 when destination house price > source house price (uphill); (1−D_UH) identifies downhill moves.
  - House price effects:
    - HP - Uphill: -0.0064 (negative, highly significant).
    - HP - Downhill: -0.0013 (negative, significant at 10 percent).
    - Interpretation: higher destination house prices strongly deter uphill moves; lower prices elsewhere provide little incentive for downhill moves.
  - Income effects:
    - Income - Uphill: 0.0063.
    - Income - Downhill: 0.0151.
    - Interpretation: income incentives are asymmetric in the opposite direction — stronger pull for downhill income gains than uphill income gains.
  - Differences between uphill and downhill coefficients are economically large and highly statistically significant.
- Explanatory power:
  - Model explains about one-third of the overall fall in migration from 1996-8 to 2014-16.
  - Explains over half of the reduction in migration from poor to rich metro areas.
  - Explains a quarter of the fall in migration within rich metro areas.
  - Explains negligible amount of the fall in migration across poor areas.

### Robustness, dynamics, and extensions
- Lags and within-CBSA inequality:
  - Adding first lags: disincentive effect of higher destination house prices is immediate; effect of relative incomes builds over time.
  - Including within-CBSA inequality measures and population density:
    - Higher population density in source and especially destination increases migration.
    - Greater house price and income inequality tends to discourage migration.
    - Uphill/downhill asymmetry remains.
  - First-stage diagnostics: F-statistics well over cut-off of 30; Kleibergen-Paap rk LM and Wald F-statistics reject under- and weak-identification.
- Absolute-poverty directional analysis (six directions: poor→rich, rich→poor, poor→less poor, poor→poorer, rich→less rich, rich→richer):
  - Asymmetry between rich and poor metro areas persists.
  - Impediment to uphill mobility from higher house prices is even larger for migration within rich and within poor metro areas.
  - Lower house prices provide little incentive to move away from more prosperous areas.
  - Extended model explains almost two-thirds of the fall in migration between rich and poor metro areas and one-third of the fall within rich areas; still fails to explain much of the fall in migration between poor areas.

### Quantitative results and selected regression outcomes (selected exact values)
- Bartik and price/income responses (Table 2):
  - Bartik Shock on Migration: -0.001942
  - Bartik Shock on House Price: 2.220***
  - Bartik Shock on Income: 0.756***
  - Distance on Migration: -0.0111***
  - Observations: 266161 (Migration), 236741 (House Price), 266161 (Income)
  - R-squared: 0.287 (Migration), 0.913 (House Price), 0.965 (Income)
- Simple specification (Table 3):
  - HP: -0.00412*** (t = -12.99)
  - Income: 0.0111*** (t = 6.44)
  - Population over 60 in Destination CBSA: -0.0219*** (t = -4.19)
  - Population over 60 in Source CBSA: -0.0511*** (t = -9.33)
  - Average Adjusted Income per Migrating Household: 0.00384*** (t = 18.67)
  - Relative Unemployment: -0.0236*** (t = -9.24)
  - Relative Population Growth: 0.0526*** (t = 14.33)
  - Log of Distance between Source and Destination: -0.0110*** (t = -28.83)
  - Observations: 200190
  - R-squared: 0.313
- Basic uphill vs downhill (Table 4):
  - HP - Uphill: -0.00640*** (t = -8.39)
  - HP - Downhill: -0.00125* (t = -1.67)
  - Income - Uphill: 0.00631*** (t = 3.16)
  - Income - Downhill: 0.0154*** (t = 8.04)
  - Observations: 200190
  - R-squared: 0.318
- Explaining the Fall in Migration (Table 5, percent contributions):
  - All: House Prices 17% / Incomes 16% / Total 33% (Basic); House Prices 23% / Incomes 18% / Total 41% (Extended)
  - Poor to Rich: House Prices 72% / Incomes -10% / Total 62% (Basic); House Prices 84% / Incomes -12% / Total 72% (Extended)
  - Rich to Poor: House Prices -24% / Incomes 66% / Total 42% (Basic); House Prices -32% / Incomes 84% / Total 52% (Extended)
  - Poor to Poor - Uphill 7% / -2% / 5%; Poor to Poor - Downhill -1% / 2% / 1%; Rich to Rich - Uphill 42% / -12% / 30%; Rich to Rich - Downhill -8% / 27% / 20% (extended values retained in table)
- Dynamic and local inequality (Table 6, selected):
  - HP - Uphill: -0.00637*** (Dynamic), -0.00660*** (Local Inequality)
  - Lagged HP - Uphill: -0.000227
  - HP - Downhill: 0.00232** (Dynamic), -0.00215*** (Local Inequality)
  - Lagged HP - Downhill: -0.00409***
  - Income - Uphill: 0.00411 (Dynamic), 0.00705*** (Local Inequality)
  - Income - Downhill: 0.00884 (Dynamic), 0.0139*** (Local Inequality)
  - Population Density Source: 0.00000367** (t = 2.27)
  - Population Density Destination: 0.00000718*** (t = 3.73)
  - Gini in Source: -0.00670** (t = -2.52)
  - HP Divergence within Source: -0.00122** (t = -2.55)
  - HP Divergence within Destination: -0.00156*** (t = -2.97)
  - Observations: 189537 (Dynamic), 165668 (Local Inequality)
  - R-squared: 0.319 (Dynamic), 0.310 (Local Inequality)
- Extended absolute differences (Table 7, selected):
  - HP - Poor to Rich: -0.00660*** (t = -6.80)
  - HP - Rich to Poor: -0.000765 (t = -0.80)
  - HP - Poor to Less Poor: -0.0109*** (t = -4.04)
  - HP - Rich to Richer: -0.00980*** (t = -6.25)
  - Income - Poor to Rich: 0.00623*** (t = 2.74)
  - Income - Rich to Poor: 0.0184*** (t = 8.08)
  - Income - Poor to Poorer: 0.0195*** (t = 6.54)
  - Observations: 200190
  - R-squared: 0.318
- Robustness (Table 8, selected coefficients preserved):
  - HP - Uphill: -0.00607***, -0.00640***, -0.00846***, -0.00393***, -0.00648***, -0.0163
  - HP - Downhill: -0.00217***, -0.00125*, -0.00189, -0.00213**, -0.000945, -0.0279
  - Income - Uphill: 0.00769***, 0.00631***, 0.0105***, 0.00633**, 0.00346, -0.00731
  - Income - Downhill: 0.0162***, 0.0154***, 0.0232***, 0.0103***, 0.0142***, 0.0150
  - Observations by column: 200190, 200190, 200190, 114579, 115830, 49278
  - R-squared by column: 0.317, 0.341, 0.459, 0.300, 0.321, 0.658

### Key substantive conclusions
- Both rising house price dispersion and income dispersion contributed to declining long-distance migration, operating differently:
  - Rising house prices are the principal driver of reduced uphill migration, especially from poor to rich metro areas.
  - Rising income inequality drives part of the fall in downhill migration, particularly between rich and poor metro areas.
- Employment shocks affect relative house prices and incomes substantially but do not induce contemporaneous increases in migration; employment shocks move prices rather than people.
- Financial constraints, housing affordability, rising concentration of skilled workers, and expectations about house appreciation jointly reduce mobility from poorer to richer areas, limiting economic churning and upward mobility.
- The model accounts for a substantial share of the decline in migration between rich and poor metro areas but little of the decline in migration among poor areas, indicating other forces are at work for the latter.

### Policy implications and recommendations
- Policymakers across levels of government should prioritize tackling impediments to migration.
- Favor targeted support that allows firms to adapt and workers to gain new skills over general support.
- Recommended policy actions:
  - Modernize land-use regulations.
  - Reduce bureaucratic delays.
  - Lower economic and racial segregation to allow housing supply to respond to demand.
  - Improve transportation and public transit to widen catchment areas for prosperous metro areas.
- Warning: without policy action, continued erosion in flexibility, competitiveness, and prosperity of the United States economy is likely, with attendant economic and social strains.

*Source: IMF working paper section on migration, house prices, incomes, and labor mobility (content unit: wpiea2019122).*

### References _______________________________________________________________ 17

### wpiea2019122 - References _______________________________________________________________ 17

### Introduction: decline in U.S. long-distance migration and its link to inequality
- U.S. interstate migration of the working population halved between 1980 and 2016, according to the Current Population Survey.
- The decline in migration reduces labor market churning, lengthens downturns, slows recoveries, and has been most marked out of poorer areas, exacerbating geographic inequality.
- Core finding: rising differences in house prices and incomes are linked to the fall in migration, with house price divergence playing the dominant role discouraging moves from poorer to richer metro areas, and income divergence reducing moves in the opposite direction.
- Key mechanistic insight:
  - Centrifugal effects of rising wealth inequality (house prices) dominate centripetal effects of rising income inequality for moves from poor to rich metro areas.
  - For moves out of prosperous areas, the centrifugal force of moving to lower incomes dominates the centripetal effect of lower house prices.
- Macro consequence: lower labor market churning leads to less efficient labor allocation, reducing productivity and output.

### Literature and prior explanations
- Alternative explanations previously emphasized:
  - Rising homeownership rates (but migration fell for both homeowners and renters).
  - Shifting demographics: fraction of ages 40–59 rose from around 45 percent to nearly 60 percent from the 1980s through 2010 (but recent work questions the economic importance).
  - Collapse of the housing boom and the 2008 recession (trend predates crisis; migration has not rebounded).
- Geographic inequality evidence:
  - State-level homogeneity arguments contradicted by CBSA-level analyses showing rising within-state inequality.
  - Literature documents social and economic problems in “left behind” areas (e.g., deaths of despair, persistent employment effects).
- Drivers of rising inequality:
  - Increasing wage premium on education; widening pay gap since the 1980s and 1990s.
  - Contributing factors: technological change, trade openness, declining unionization, falling minimum wage, lower taxes for high earners, limited supply of skilled workers.
- Urban concentration of skilled premium:
  - Since 2000 wage premium for skilled workers rose sharply in urban areas but much less in rural areas.
  - Rising house prices in productive metro areas deter unskilled in-migration and may reduce incentives for skilled workers to leave.

### Stylized facts: migration, house prices, and incomes
- Stylized Fact I — Long-distance migration decline:
  - CPS Annual Geographical Mobility Rates: inter-state migration fell from 3.0 percent in 1981 to 1.5 percent in 2016.
  - Intra-state migration fell by about a quarter over comparable periods.
  - IRS address-based migration: 2.9 percent in 1990 to 2.4 percent in 2015.
  - Long-distance moves are more job-related: job-related motives explain 34.3 percent of moves across counties in 2015 versus 20.2 percent of moves within a county.
  - Educational gradient: in 2016, those with education beyond high school were almost twice as likely to move to another state than those with only high school.
- Stylized Fact II — Rising divergence in house prices and incomes:
  - Data source: Zillow Home Value Database for 571 CBSAs, 1996–2016; Zillow provides median nominal estimated house prices and median nominal incomes by CBSA (1996–2016).
  - House price dispersion: the standard deviation between the logarithm of median home values widens by nearly 50 percent between 1996 and 2016, with a peak in 2006 and a low in 2012 before rebounding toward pre-crisis levels.
  - Income dispersion: the standard deviation of median incomes in logs of 381 CBSAs grew by 20 percent over the same period; both series show thickening in the right-hand tail (superstar cities).
  - House price to income ratio: the ratio of the logarithm of median house prices to median incomes increased by 39 percent.
  - Bilateral dataset observation: 80 percent of observations represent simultaneous upward or downward movements in both house prices and income; observations with opposite movements concentrate in cases with relatively small divergences.
  - Extended series to 1981 (using FHFA House Price Index and county income data) show upward trends in standard deviations of log house prices and incomes contemporaneous with the decline in long-distance mobility.

### Mechanisms and dynamics described
- Self-reinforcing process:
  - Limited space in high house-price metros attracts skilled workers, increasing the share of skilled workers, which further raises wage and home price inequality (gentrification).
  - Technology and agglomeration effects increase wage premium for skilled workers even with larger skill supply.
  - Rising house prices relative to incomes price out lower-earning households, reducing migration from poorer to more productive metro areas.
- Broader economic implications:
  - Growing divergences in returns to land (house prices) and returns to labor (earnings) create asymmetric migration responses, reducing labor reallocation where it would most enhance opportunity.
  - Lower long-distance migration contributes to falling economic dynamism and potentially lower aggregate output.

### Key empirical evidence and statistics (selected)
- Interstate migration: 3.0 percent in 1981 → 1.5 percent in 2016 (CPS).
- IRS migration: 2.9 percent in 1990 → 2.4 percent in 2015.
- Job-related motives: 34.3 percent across counties (2015); 20.2 percent within a county (2015).
- Demographic share (ages 40–59): around 45 percent → nearly 60 percent (1980s through 2010).
- Zillow house price dispersion (std dev of log median home values): widens by nearly 50 percent (1996–2016).
- Zillow income dispersion (std dev of log median incomes across 381 CBSAs): grew by 20 percent (1996–2016).
- Ratio of log median house prices to median incomes: increased by 39 percent (1996–2016).
- In bilateral CBSA observations, 80 percent show concurrent directions in house price and income movements.

*wpiea2019122 - References _______________________________________________________________ 17*

### 16. We use the Internal Revenue Service’s Statistics on migration for our empirical

### 16. We use the Internal Revenue Service’s Statistics on migration for our empirical

### Data, sample, and measurement
- Migration data:
  - Source: Internal Revenue Service’s Statistics on migration (county-to-county migration across the United States from 1990-2015).
  - Coverage: 323 CBSAs with overlap of Zillow and income databases (35 percent of total number of CBSAs, but 82 percent by population).
  - Sample size: roughly 200,000 observations over 20 years (1996-2015 sample used for regressions).
  - Migration tracked by changes in tax return mailing address; excludes some low-income families who do not file taxes.
  - To focus on job-related migration, analysis examines migration between CBSAs of over 200 miles.
  - Three-year averages used to reduce noise; adjusted for change in reporting cut-off raised from 10 to 20 in 2013.

- Variables and normalization:
  - Gravity normalization: bilateral migration divided by the square root of the product of their respective populations.
  - HP: log median house price in destination CBSA minus log median house price in source CBSA.
  - I: log median income in destination CBSA minus log median income in source CBSA.
  - Distance: log of distance between CBSA centers (calculated from county-level longitude and latitude).
  - Controls (X): proportion of population over 60 in source and destination; average household annual gross income (AGI) of migrants; relative regional unemployment; relative population growth; distance.
  - Instrumental variables: averages of incomes and house prices of other CBSAs within a 200-mile radius; house price and income differentials from 15 years earlier.

### Empirical patterns in migration rates and heterogeneity
- Aggregate trends:
  - Migration of over-200-mile moves dropped from an average of 1.25 percent in 1996-8 to 1 percent in 2014-16.
  - Decline in migration is larger for metro areas with lower median incomes:
    - Migration out of metro areas in the lowest quartile of median income fell by 25 percent.
    - Fall of only 10 percent for metro areas in the top quartile of median income.
  - Migration from metro areas within the highest income quartile to other areas in the top quintile marginally increased.
  - Migration from high-income quartile areas to lowest quartile areas has fallen by over a quarter.
  - Descriptive summary: the 1990s “smile” in labor mobility (high migration out of highest and lowest quartiles) has become a “lopsided smirk,” with elevated migration only for areas in the highest quartile.

- Sample overlap and data notes:
  - Zillow house price database: 571 CBSAs.
  - Income database: 381 CBSAs.
  - Overlap: 323 CBSAs.

### A. Bartik shocks — methods and findings
- Construction:
  - Bartik shocks computed using 1998 industry compositions due to a major definitional change.
  - Shocks equal changes in national employment by sector each year weighted by initial employment structure in each CBSA.
  - Regressions of Bartik shocks on migration, house price divergence, and income divergence from 1998 with CBSA and time fixed effects.

- Key findings:
  - A Bartik shock of 1 percent of employment in one metro area relative to another:
    - Leads to an immediate and highly statistically significant increase in relative house prices of over 2 percent.
    - Leads to an immediate and highly statistically significant increase in relative income of ¾ percent.
    - In both cases, this represents around 5 percent of the typical gap across metro areas.
  - Persistence: between half and two-thirds of the impact remains after 4 years.
  - Migration response: a relative Bartik shock of 1 percent has no significant contemporaneous impact on migration; the coefficient is negative and becomes significant with 4-year lags, implying bilateral migration dwindles in response to favorable relative employment shocks.
  - Interpretation: employment shocks appear to move prices rather than people.

### B. Migration results — baseline and asymmetric models
- Baseline IV results (regression 4):
  - HP coefficient: negative and highly economically and statistically significant — a 1 percent increase in house prices lowers the proportion of the population migrating by 0.0041 percentage point.
  - I coefficient: positive and highly economically and statistically significant — a 1 percent increase in income raises the proportion of the population migrating by 0.0111 percentage point.
  - Economic magnitude: given median bilateral migration coefficient .01206, a 1 percent increase in house price (income) dispersion leads to 0.33 percent decrease (0.92 percent increase) in bilateral migration.
  - Controls behave as expected: older populations discourage mobility; higher AGIs increase migration; higher relative unemployment discourages mobility; larger divergence in population growth increases migration; longer distances discourage mobility.

- Asymmetric uphill/downhill specification (regression 5):
  - Define D_UH = 1 when destination house price > source house price (uphill move); (1−D_UH) identifies downhill moves.
  - House price effects:
    - Uphill HP coefficient: -0.0064 (negative, highly significant), much larger in magnitude than basic regression.
    - Downhill HP coefficient: -0.0013 (negative, significant at 10 percent), less than a third of uphill estimate.
    - Interpretation: higher house prices in destination strongly deter uphill moves; lower house prices elsewhere provide little incentive to move downhill.
  - Income effects:
    - Uphill I coefficient: 0.0063.
    - Downhill I coefficient: 0.0151.
    - Interpretation: income incentives are asymmetric in the opposite direction — stronger pull for downhill income gains than uphill income gains.
  - Differences between uphill and downhill coefficients are economically large and highly statistically significant.
  - Robustness check: using income to calculate uphill/downhill dummies yields similar but worse-fitting results; house prices appear more fundamental.

- Model explanatory power:
  - The model explains about one-third of the overall fall in migration from 1996-8 to 2014-16.
  - It explains over half of the reduction in migration from poor to rich metro areas.
  - Explains a quarter of the fall in migration within rich metro areas.
  - Explains negligible amount of the fall in migration across poor areas.

- Mechanisms and interpretation:
  - Rising house price differences reduce affordability for people in low-house-price areas (financial constraints on moving become more binding).
  - Concentration of skilled workers and future house appreciation prospects discourage owners in high-price areas from leaving.
  - Widening wage gaps increase incentives for low-income workers to leave rich metro areas; changes in life-cycle migration and amenity valuation by high-income households also matter.

### C. Robustness and extensions
- Lags and within-CBSA inequality:
  - Adding first lags: impact of disincentive to moving to a metro area with higher house prices is immediate; effect of relative incomes builds over time.
  - Adding measures of inequality within CBSAs (84th - 16th percentile house price gap; Gini coefficients for income where available) and population density:
    - Higher population density in source and especially destination increases migration.
    - Greater house price and income inequality tends to discourage migration.
    - These additions do not materially affect the uphill/downhill asymmetry; rise in within-CBSA inequality is not a major driver of fall in between-CBSA migration.
  - First-stage diagnostics: F-statistics for first-stage regressions well over cut-off of 30; Kleibergen-Paap rk LM and Wald F-statistics reject under- and weak-identification.

- Absolute-poverty and directional analysis:
  - Specification includes dummies for poor CBSAs (house prices below median in that year) and rich CBSAs (above-median) to identify six migration directions: poor→rich, rich→poor, poor→less poor, poor→poorer, rich→less rich, rich→richer.
  - Findings (Table 7):
    - Asymmetry between rich and poor metro areas continues to hold.
    - Impediment to uphill mobility from higher house prices is even larger for migration within rich and within poor metro areas.
    - Lower house prices continue to provide little or no incentive to move away from more prosperous areas.
    - Income asymmetry persists but is smaller within rich metro areas.
    - Extended model explains almost two-thirds of the fall in migration between rich and poor metro areas and one-third of the fall within rich areas.
    - Extended model still fails to explain much of the fall in migration between poor areas.

### Key substantive conclusions
- Rising house price dispersion and income dispersion both contributed to declining long-distance migration, but they operate differently:
  - Rising house prices are the principal driver of reduced uphill migration, especially from poor to rich metro areas.
  - Rising income inequality drives part of the fall in downhill migration, particularly between rich and poor metro areas.
- Employment shocks (Bartik shocks) affect relative house prices and incomes substantially but do not induce contemporaneous increases in migration; employment shocks move prices rather than people.
- Financial constraints, housing affordability, rising concentration of skilled workers, and expectations about house appreciation jointly reduce mobility from poorer to richer areas, limiting economic churning and opportunities for upward mobility.
- Overall, the model accounts for a substantial share of the decline in migration between rich and poor metro areas but little of the decline in migration among poor areas, indicating other forces are at work for the latter.

*Source: IMF working paper section on migration, house prices, incomes, and labor mobility (content unit: wpiea2019122 - 16).*

### 32. Our finding of major asymmetries in responses to house price and income

### 32. Our finding of major asymmetries in responses to house price and income

### Robustness checks and alternative specifications
- The finding of major asymmetries in responses to house price and income differences is robust to a range of other specifications (Table 8), including:
  - defining uphill and downhill migration using relative incomes rather than relative house prices;
  - defining migration by households rather than individuals;
  - weighting the regression by population to check results are not dominated by smaller metro areas;
  - excluding the twenty largest metropolitan areas to check results are not dominated by behavior in large metro areas;
  - cutting off the sample before 2007 to ensure results were not driven by the housing bust and its aftermath;
  - running the specification on migration of under 200 miles (coefficients on relative house prices and incomes, and AGI of migrants, are not significant for under-200-mile migration).
- Table 8 coefficients (selected, exact values preserved):
  - HP - Uphill: -0.00607***, -0.00640***, -0.00846***, -0.00393***, -0.00648***, -0.0163
  - HP - Downhill: -0.00217***, -0.00125*, -0.00189, -0.00213**, -0.000945, -0.0279
  - Income - Uphill: 0.00769***, 0.00631***, 0.0105***, 0.00633**, 0.00346, -0.00731
  - Income - Downhill: 0.0162***, 0.0154***, 0.0232***, 0.0103***, 0.0142***, 0.0150
  - Observations by column: 200190, 200190, 200190, 114579, 115830, 49278
  - R-squared by column: 0.317, 0.341, 0.459, 0.300, 0.321, 0.658

### Main findings on migration, house price and income divergence
- Long-distance migration (most linked to job moves) has fallen and is clearly linked to rising house price and income inequality.
- Asymmetry in effects:
  - House price inequality:
    - Largest effect on migrants seeking to move to more prosperous metro areas (uphill moves).
    - Little to no role in prompting people to leave richer areas (downhill outflows).
  - Income divergences:
    - Discourage outflows from high-income metro areas.
    - Provide more limited incentives for inflows into high-income metro areas.
- Quantitative explanatory power:
  - The model explains up to two-thirds of the fall in long-distance migration between poor metro areas and rich ones.
- Broader implications:
  - Rising inequality has stifled labor market churning, worsening inequality and economic sclerosis, and harming those “left behind” in decaying areas with diminishing prospects.

### Quantitative results and selected regression outcomes
- Table 2 (selected):
  - Bartik Shock on Migration: -0.001942
  - Bartik Shock on House Price: 2.220***
  - Bartik Shock on Income: 0.756***
  - Distance on Migration: -0.0111***
  - Observations: 266161 (Migration), 236741 (House Price), 266161 (Income)
  - R-squared: 0.287 (Migration), 0.913 (House Price), 0.965 (Income)
- Table 3 (Simple specification):
  - HP: -0.00412*** (t = -12.99)
  - Income: 0.0111*** (t = 6.44)
  - Population over 60 in Destination CBSA: -0.0219*** (t = -4.19)
  - Population over 60 in Source CBSA: -0.0511*** (t = -9.33)
  - Average Adjusted Income per Migrating Household: 0.00384*** (t = 18.67)
  - Relative Unemployment: -0.0236*** (t = -9.24)
  - Relative Population Growth: 0.0526*** (t = 14.33)
  - Log of Distance between Source and Destination: -0.0110*** (t = -28.83)
  - Observations: 200190
  - R-squared: 0.313
- Table 4 (Basic specification, uphill vs downhill):
  - HP - Uphill: -0.00640*** (t = -8.39)
  - HP - Downhill: -0.00125* (t = -1.67)
  - Income - Uphill: 0.00631*** (t = 3.16)
  - Income - Downhill: 0.0154*** (t = 8.04)
  - Observations: 200190
  - R-squared: 0.318
- Table 5 (Explaining the Fall in Migration (Percent), Basic and Extended regressions — exact percent contributions preserved):
  - All: House Prices 17% / Incomes 16% / Total 33% (Basic); House Prices 23% / Incomes 18% / Total 41% (Extended)
  - Poor to Rich: House Prices 72% / Incomes -10% / Total 62% (Basic); House Prices 84% / Incomes -12% / Total 72% (Extended)
  - Rich to Poor: House Prices -24% / Incomes 66% / Total 42% (Basic); House Prices -32% / Incomes 84% / Total 52% (Extended)
  - (Other directional rows preserved verbatim: Poor to Poor - Uphill 7% / -2% / 5%; Poor to Poor - Downhill -1% / 2% / 1%; Rich to Rich - Uphill 42% / -12% / 30%; Rich to Rich - Downhill -8% / 27% / 20%; extended values in table retained.)
- Table 6 (Dynamic and local inequality; selected coefficients):
  - HP - Uphill: -0.00637*** (Dynamic), -0.00660*** (Local Inequality)
  - Lagged HP - Uphill: -0.000227
  - HP - Downhill: 0.00232** (Dynamic), -0.00215*** (Local Inequality)
  - Lagged HP - Downhill: -0.00409***
  - Income - Uphill: 0.00411 (Dynamic), 0.00705*** (Local Inequality)
  - Income - Downhill: 0.00884 (Dynamic), 0.0139*** (Local Inequality)
  - Population Density Source: 0.00000367** (t = 2.27)
  - Population Density Destination: 0.00000718*** (t = 3.73)
  - Gini in Source: -0.00670** (t = -2.52)
  - HP Divergence within Source: -0.00122** (t = -2.55)
  - HP Divergence within Destination: -0.00156*** (t = -2.97)
  - Observations: 189537 (Dynamic), 165668 (Local Inequality)
  - R-squared: 0.319 (Dynamic), 0.310 (Local Inequality)
- Table 7 (Extended specification, absolute differences in migration; selected coefficients):
  - HP - Poor to Rich: -0.00660*** (t = -6.80)
  - HP - Rich to Poor: -0.000765 (t = -0.80)
  - HP - Poor to Less Poor: -0.0109*** (t = -4.04)
  - HP - Rich to Richer: -0.00980*** (t = -6.25)
  - Income - Poor to Rich: 0.00623*** (t = 2.74)
  - Income - Rich to Poor: 0.0184*** (t = 8.08)
  - Income - Poor to Poorer: 0.0195*** (t = 6.54)
  - Observations: 200190
  - R-squared: 0.318

### Policy implications and recommendations
- Policymakers across levels of government should prioritize tackling the impediments to migration.
- Targeted support is favored over general support:
  - Support that allows firms to adapt and workers to gain new skills is more effective.
- Additional policy actions recommended (reflecting literature and cited reports):
  - Modernize land-use regulations.
  - Reduce bureaucratic delays.
  - Lower economic and racial segregation to allow housing supply to respond to demand.
  - Improve transportation and public transit to widen catchment areas for prosperous metro areas.
- Without policy action, the paper warns of a likely continuation of gradual erosion in flexibility, competitiveness, and prosperity of the United States economy, with attendant economic and social strains.

*Source: wpiea2019122 - 32. Our finding of major asymmetries in responses to house price and income*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019122.pdf_