## _wp11105

## Source details

**Canonical URL:** [_wp11105](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11105.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11105.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11105.pdf.json)

---

### Major findings: crisis magnitude and regional impacts
- Aggregate employment declined by around 8½ million from the onset of the recession in December 2007 to its trough, before beginning to recover slowly in 2010.
- Job losses averaged over 750,000 per month from November to March 2009.
- The U.S. economy lost 1.2 million jobs in the fourth quarter of 2008 and 2.5 million jobs in the first quarter of 2009; total job loss in 2009 was 5.5 million.
- By October 2009, 15.6 million people were unemployed, compared with the previous peak of 12 million in 1982.
- The unemployment rate spiked to a 27-year high of 10.1 percent in late 2009 and "rests at around 8¾ percent" amid declining labor force participation.
- State-level heterogeneity in 2010: unemployment rates ranged from 3.9 percent in North Dakota to 14.9 percent in Nevada.
- Unemployment duration reached historic highs as of March 2011; around 46 percent of all unemployed had been out of employment for at least 27 weeks as of May 2010.
- The crisis disproportionately affected men, youth, low-skilled individuals, manufacturing, construction, and parts of the financial industry.

### Skill mismatches: measurement, patterns, and quantified effects
- Construction of the Skill Mismatch Index (SMI):
  - SMI constructed for each of the 50 states and the District of Columbia as the difference between state skill “demand” and state skill “supply.”
  - Skill demand: industries ranked by skill intensity using average educational attainment in 2006 Current Population Survey; sectoral employment from Current Employment Statistics grouped into high, semi-, low-skilled.
  - Skill supply: educational attainment of working-age population (25+), divided into Low-skilled (less than high school), Semi-skilled (high school to less than bachelor’s), High-skilled (bachelor’s or more).
  - Data were interpolated for 2008–10; interpolation understates the SMI since it assumes higher labor mobility than actually observed.
- Empirical patterns:
  - The SMI typically rises during recessions; in all states but the District of Columbia the index exhibits cyclical patterns with large surges during recessions.
  - Numerous states (including Arizona, Delaware, Florida, Hawaii, Michigan, Missouri, New Mexico, Ohio, and Wisconsin) have SMI near or at historic peak levels (data start in 1990).
  - Delaware and Washington State experienced a double digit percentage increase in their skill mismatch index in 2010.
- Quantified effect on structural unemployment:
  - A 17 percent increase in skill mismatches (as experienced by the average of the U.S. states since the onset of the recession) is associated with a third of the overall estimated increase in the structural unemployment rate.

### Housing market distress, regional disparities, and labor interaction
- Housing sector context and magnitudes:
  - Housing employed around 6 percent of the workforce and accounted for about 5 percent of U.S. GDP in 2007—figures halved by the crisis.
  - One in 12 mortgages are late for at least 30 days; 4.6 percent of outstanding mortgages are in foreclosure (end-2010).
  - Case-Shiller and National Association of Realtors indices are both around 30 percent below peak levels.
  - FHFA house-price indices show, on average, house prices declined by over 15 percent from their peak levels in 2007.
  - Examples of peak-to-2010 declines: almost 50 percent in Nevada, about 40 percent in California, and 36 percent in Florida.
- Foreclosure and delinquency specifics (end-2010):
  - United States: Delinquency rate 8.2; Foreclosure rate 4.6.
  - Nevada: Delinquency rate 12.0; Foreclosure rate 10.1.
  - Michigan: Delinquency rate 10.6; Foreclosure rate 4.3.
  - Indiana: Delinquency rate 10.2; Foreclosure rate 4.8.
  - Florida: Delinquency rate 10.1; Foreclosure rate 14.2.
  - Ohio: Delinquency rate 9.8; Foreclosure rate 4.9.
  - Arizona: Delinquency rate 9.5; Foreclosure rate 5.7.
  - California: Delinquency rate 9.2; Foreclosure rate 4.5.
- Interaction with labor markets:
  - Higher foreclosure rates and lower housing prices are associated with higher unemployment rates.
  - Interaction effect: increases in skill mismatches in states with worse housing market conditions (larger increases in foreclosure rates or larger house price declines) are associated with even higher unemployment rates, after controlling for cyclical factors.
  - Suggested mechanism: poor local housing conditions may slow the exodus of jobless individuals from depressed areas, raising equilibrium unemployment rates and reducing labor mobility.
- Regional dispersion:
  - Standard deviation of changes in foreclosure rates and housing prices grew dramatically during this recession relative to earlier recessions.
  - Composite index of SMI and foreclosure increases highlights states particularly hit by both shocks: California, Michigan, Ohio, Florida, Arizona.
  - Some states fared relatively better: Montana, North and South Dakota, Nebraska, Kansas, Colorado, Texas—though all states saw deterioration.

### Econometric approach, robustness, and key coefficients
- Model overview:
  - Panel-data model relating annual percentage-point changes in state-level unemployment rate (∆u_it) to:
    - Annual percent growth in state-level output (∆y_it),
    - Annual percent changes in state-level skill-mismatch index (∆m_it),
    - Housing indicator (∆h_it) measured as percent changes in housing prices or annual percentage-point changes in foreclosure rates,
    - Interaction term ∆m_it * ∆h_it,
    - State-specific dummies (S_i) and year-specific dummies (T_t).
  - Residual ε_it assumed i.i.d.
  - State dummies capture institutional differences; year dummies control for aggregate shocks.
- State-GDP measurement and chosen specification:
  - BEA state-GDP estimates subject to measurement error; two alternative constructed GDP measures proposed.
  - Reported specification uses the first alternative measure (Alternative Measure 1) of state-level GDP.
  - Selected GDP coefficients (from Table 1 and subsequent tables):
    - Log-change in real GDP (BEA estimates): -0.136***.
    - Log-change in real GDP (Alternative Measure 1): -0.443*** and -0.296*** (in different columns).
    - Log-change in real GDP (Alternative Measure 2): -0.374*** and -0.166***.
  - Adj. R-squared values reported across specifications: 0.15, 0.51, 0.50, 0.61, 0.77, 0.80.
  - Number of states, including D.C.: 51; Observations: 918, 969 as reported by column.
- Main regression findings (selected from Tables 2 and 3):
  - Log-change in skill mismatch index coefficients: 0.041***, 0.034***, 0.032*** (statistically significant).
  - Percentage-point change in foreclosure rate coefficients: 0.396***, 0.362***, 0.332***, with IV 0.666*** (2SLS using share of subprime mortgages as instrument).
  - Log-change in FHFA house price index coefficients: -0.028***, -0.024***, -0.026***, with IV larger in magnitude (e.g., -0.064*** in IV column).
  - Interaction terms:
    - log-change in skill mismatch * log-change FHFA house price index: -0.003*** (statistically significant).
    - log-change in skill mismatch * pp. change in foreclosure rate: 0.012 (p-value (0.17), not significant in one specification).
- Robustness using alternative skill-mismatch proxy:
  - Standard deviation of sectoral employment growth used as proxy; correlation with SMI = 0.48.
  - Std. deviation coefficients: 0.113***, 0.071***, 0.109*** in various specifications.
  - Interaction terms with house-price changes and foreclosure changes are significant in several specifications (e.g., interaction std. deviation * log-change FHFA house price index: -0.011***).
  - Results using sectoral dispersion are consistent with main findings, supporting robustness of SMI results.

### Simulation: contribution to structural (natural) unemployment during crisis
- State-level illustrative decompositions (using coefficients in column (4) of Table 2):
  - Florida: 5½ percentage points out of the 8 percentage points increase in unemployment explained by structural factors.
  - Nevada: 3½ percentage points out of the 10.2 percentage-point increase since 2006 explained by structural factors.
  - Delaware: 3 out of the 5 percentage-point increase explained by structural factors.
  - Arizona: 3¼ out of the 6 percentage-point increase explained by structural factors.
- Aggregate structural unemployment estimates:
  - Range of models suggests structural unemployment rose from end-2007 to 2010 by between 1 and 1½ percentage points (and by as much as 1¾ since the onset of the housing market collapse at end-2006).
  - Adding these figures to a 5 percent estimate for the structural unemployment rate in 2006 leads to a structural unemployment rate of around 6¾ percent by end-2010.
- Decomposition of the rise in NAIRU:
  - Unfavorable housing market conditions roughly explain 1 percentage point of the increase in structural unemployment.
  - Deteriorating skill mismatches explain another 1/2 percentage point.
  - Interaction of skill mismatches and foreclosures explains the remaining 1/4 percentage point.
  - The bulk of the increase occurred in 2009: NAIRU rose by over 3/4 percentage points in 2009; about one third of that rise was explained by rising skill mismatches.
- Timing and persistence:
  - Deteriorating housing market conditions were behind rising NAIRU until end-2009; thereafter housing market showed signs of stabilization.
  - Skill mismatches were solely responsible for almost one quarter percentage point increase in NAIRU in 2010.
  - Authors note limited ability to assess persistence due to SMI time-series starting in 1990; the natural rate had been on a decreasing trend since the mid-1970s.

### Policy implications and recommended priorities
- Principal policy message:
  - Complement macroeconomic stimulus (to address cyclical unemployment) with targeted policies addressing structural determinants: housing market interventions and labor-market measures to reduce skill mismatches.
- Housing-market policy options:
  - Controlling foreclosures and ensuring effective loan modifications are important given interlinkages with labor markets.
  - Administration measures cited: temporary tax incentives for home buyers, measures to mitigate the foreclosure epidemic, temporary foreclosure forbearance to the unemployed, assistance to underwater mortgage holders.
  - Authors advocate allowing mortgages to be renegotiated in courts ("cramdowns") as "the next best alternative" if existing programs continue to miss targets, to help clear housing markets and restore household balance sheets.
- Labor-market policy options:
  - Active Labor Market Policies (ALMPs) recommended: spending in training programs, targeted subsidies to net hiring, public employment services, and improved employment services and job-search assistance.
  - Subsidies to net hiring highlighted as a priority; conditioning subsidies on hiring of longer-term unemployed suggested to improve targeting and reduce displacement effects.
  - Evidence cited: subsidies to net hiring more effective than other ALMPs; employment services and job-search assistance can be cost-effective; returns to further education and training are high.
  - Concerns on program design and evaluation: WIA lacked rigorous evaluation since 1998; TAA allocations and awareness issues noted; federal job training policy described as "fragmented and inefficient."
- Fiscal and implementation considerations:
  - Complementary structural measures should be carefully designed to maximize impacts and minimize negative substitution effects.
  - Funding options for structural measures include reducing tax expenditures or enacting forceful entitlements reform.
  - Authors stress careful cost-benefit evaluation of ALMPs and housing measures given fiscal constraints.
- Social costs and long-term consequences:
  - Increases in structural unemployment raise long-term unemployment and permanent job loser shares, with long-lasting income losses and health consequences documented for displaced workers.
  - Declining labor mobility due to housing lock-in effects exacerbates these social costs.

### Data sources, appendix notes, and construction details
- Key data sources listed:
  - Federal Housing Finance Agency (House Price Index), S&P/Case-Shiller (Composite 20), Local Area Unemployment Statistics Database (BLS), Mortgage Bankers Association (subprime share, foreclosure rate), Bureau of Economic Analysis (Real GDP), Census Bureau (population by educational attainment), Current Employment Statistics (employment by state/industry).
- Appendix methodological notes:
  - SMI construction follows Peters (2000) with three skill categories and industry classification based on 2006 CPS skill-intensity weights.
  - Industry classification by skill-intensity (as presented):
    - Low Skilled: Mining and Logging; Construction
    - Semi Skilled: Manufacturing; Trade, Transportation, and Utilities; Leisure and Hospitality; Other Services; Government
    - High Skilled: Information; Financial Activities; Education and Health Care; Professional and Business Services
  - Alternative skill-mismatch proxy: standard deviation of sectoral employment growth across ten sectors; correlation with SMI = 0.48.
- Representative cross-state table values (preserved exactly as shown in the source table):
  - US row: 18.05, 7.82, 4.25, 8.4, 45.5, 48.7, 984.45, 5.54, 4.8, 1.3
  - Max row: 27.16, 6.24, 0.815, 15.45, 54.0, 77.1, 2033.87, 7.48, 8.5, 2.6
  - Min row: 11.33, 8.81, 5.11, 1.72, 21.23, 6.64, 32.23, 3.22, 2.8, 0.7

*Source: IMF working paper content (selected sections: Introduction, II. A Dismal Labor Market Situation, summary of empirical findings and policy discussion).*

### References .............................................................................................................

### _wp11105 - References

### Major findings: crisis magnitude and regional impacts
- Aggregate employment declined by around 8½ million from the onset of the recession in December 2007 to its trough, before beginning to recover slowly in 2010.
- The unemployment rate spiked to a 27-year high of 10.1 percent in late 2009 and "rests at around 8¾ percent" amid declining labor force participation.
- Job losses averaged over 750,000 per month from November to March 2009.
- The U.S. economy lost 1.2 million jobs in the fourth quarter of 2008 and 2.5 million jobs in the first quarter of 2009; total job loss in 2009 was 5.5 million.
- By October 2009, 15.6 million people were unemployed, compared with the previous peak of 12 million in 1982.
- State-level heterogeneity in 2010: unemployment rates ranged from 3.9 percent in North Dakota to 14.9 percent in Nevada.
- Unemployment duration reached historic highs as of March 2011; around 46 percent of all unemployed (historic high) had been out of employment for at least 27 weeks as of May 2010.
- The crisis disproportionately affected men, youth, low-skilled individuals, manufacturing, construction, and parts of the financial industry.

### Skill mismatches: measurement and effects
- The paper constructs a state-level Skill Mismatch Index (SMI) for each of the 50 states and the District of Columbia to capture mismatches between skill demand and supply.
- Skill mismatches and housing market conditions appear to be cyclical but deteriorated at a higher rate in the current episode than in past downturns.
- Skill mismatches have been more acute in states with depressed housing markets.
- Quantified effect: a 17 percent increase in skill mismatches (as experienced by the average of the U.S. states since the onset of the recession) is associated with a third of the overall estimated increase in the structural unemployment rate.

### Housing market impacts and interaction with labor
- The collapse in the housing market and declines in production had distinct regional patterns (examples cited: Nevada, Florida, Arizona, California—heavy foreclosure incidence; Ohio and Michigan—manufacturing collapse; New York and Delaware—financial restructuring; Hawaii—tourism decline).
- Higher foreclosure rates and lower housing prices are associated with higher unemployment rates.
- Interaction effect: increases in skill mismatches in states with worse housing market conditions (larger increases in foreclosure rates or larger house price declines) are associated with even higher unemployment rates, after controlling for cyclical factors.
- Mechanism suggested: poor local housing conditions may slow the exodus of jobless individuals from depressed areas, raising equilibrium unemployment rates and reducing labor mobility.

### Estimated impact on structural (natural) unemployment
- Combined impact: skill mismatches and higher foreclosure rates might have raised the natural rate of unemployment by about 1½ percentage points since 2007.
- A larger estimate: around 1¾ percentage points since the onset of the housing market collapse at end-2006.
- Even with these increases, there remains scope for reductions from the then-current 8¾ percent unemployment rate; however, job reallocation flows may slow as unemployment approaches below about 7 percent, with adverse implications for wage and price inflation.

### Empirical approach and robustness
- Uses a panel data model relating changes in state-level unemployment rates to:
  - time dummies (business cycle and macro policy effects),
  - state GDP growth (local business cycles),
  - the Skill Mismatch Index,
  - an indicator of housing hurdles (foreclosure rates, house prices).
- Results are broadly confirmed when using the standard deviation of sectoral employment growth in a state-year as an indirect measure of shocks to matching of skill demand and supply.
- Endogeneity between unemployment, skill mismatches, and housing conditions is addressed in the analysis.

### Policy implications and recommendations
- Addressing housing market strains is important given interlinkages with labor markets; effective measures include controlling foreclosures and ensuring effective loan modifications.
- On the skill-mismatch side, targeted policies to hire the long-term unemployed (for example, subsidies to net hiring and retraining) would reduce structural unemployment rates.
- Improved employment services and job-search assistance could be helpful.
- Both housing and labor-market interventions would complement macroeconomic stimulus by allowing the cyclical recovery to make deeper inroads on unemployment rates.
- Caution: effectiveness of these policies requires careful design to maximize impacts.

*Source: IMF working paper content (selected sections: Introduction, II. A Dismal Labor Market Situation, summary of empirical findings and policy discussion).*

### 11.4 percent). Unemployment rates for

### _wp11105 - 11.4 percent). Unemployment rates for

### Unemployment developments and key labor-market facts
- The unemployment rate peaked at a 27-year high.
- Average unemployment duration reached a record high.
- All measures of underutilization peaked at record highs.
- People unemployed for more than 15 weeks are at a historic level.
- The majority of the unemployed have lost their job permanently.
- Teenagers (16–19 years old) and individuals 25 year-old or older reached record post-Depression highs during this downturn.
- Around 2 million jobs were lost in the manufacturing sector since the start of the recession before hiring restarted in 2010.
- Low-skilled individuals 25 years of age or older (with at most a high-school diploma, representing 37 percent of the civilian labor force) have been particularly hit.
- One in seven individuals without a high-school diploma and one in ten high-school graduates are unemployed.

### Skill mismatches: motivation and consequences
- The crisis created a potential wedge between available skills and labor demand.
- Unemployment rose disproportionately for low-skill workers (in terms of years of schooling), while demand for high-skill labor (which comprises a third of the U.S. civilian labor force) is on the rise.
- Sectoral shifts: housing construction likely to remain lackluster; sectors intensive in qualified labor (e.g., the export and health sectors) may see surges in investment and activity, potentially intensifying mismatch.

### Skill-Mismatch Index (SMI): construction and interpretation
- The SMI is calculated as the difference between state skill “demand” and state skill “supply.”
- State skill demand:
  - Industries ranked by skill intensity using average educational attainment in a benchmark year (2006 Current Population Survey).
  - Sectoral employment from Current Employment Statistics grouped into three skill categories (high, semi-, low-skilled).
  - Percentage of employed individuals at each skill level in a state represents state skill “demand.”
- State skill supply:
  - Determined using educational attainment data from the U.S. Census Bureau for working-age population.
- Data note: Data were interpolated for 2008–10 due to data limitations; such interpolation understates the skills mismatch index since it assumes higher labor mobility than actually observed during the current downturn.
- Formal construction: SMI for state i at time t uses j = skill level, where:
  - (percent of working-age population with skill level j at time t in state i) and
  - (percent of employees with skill level j at time t in state i).
- Interpretation:
  - Cross-state comparisons identify states facing difficulties employing their skill base.
  - Time comparisons within a state help distinguish cyclical versus structural surges in SMI.
  - High mobility across states could mitigate mismatches via migration.

### Empirical patterns of skill mismatches across states (1990–2010)
- The SMI typically rises during recessions; in all states but the District of Columbia the index exhibits cyclical patterns with large surges during recessions.
- Observations highlighted:
  - On average, skill mismatches increased nationwide during this downturn with significant dispersion across states, in contrast to the 2001 recession.
  - Numerous states (including Arizona, Delaware, Florida, Hawaii, Michigan, Missouri, New Mexico, Ohio, and Wisconsin) have SMI near or at historic peak levels (data start in 1990).
  - Nevada, New Hampshire and Vermont experienced low mismatches; New Mexico, Arkansas, West Virginia, and Delaware experienced very large mismatches.
  - Great Lakes states experience average skills mismatches— notably Michigan—while some northeast areas (e.g., District of Columbia and New York) have large skill mismatches.
  - California and other western states are in the middle of the pack; Southern states exhibit average skills mismatches (with Arizona at the higher end).
  - Increases in skill mismatches during the recent recession varied across states: Delaware (financial hub), Hawaii (tourism reliance), and Michigan (auto hub) experienced disproportionate increases.
  - Skills mismatches continued to rise in almost all states even after the recession ended; Delaware and Washington State experienced a double digit percentage increase in their skill mismatch index in 2010.
- D.C. is an outlier: large federal government presence yields less cyclical labor usage; D.C. has a persistent downward trend in low-skilled population and diminishing job opportunities for this segment. In level terms, skill mismatches in D.C. remain large relative to other states.

### Housing market distress and regional disparities
- Housing sector context:
  - Housing employed around 6 percent of the workforce and accounted for about 5 percent of U.S. GDP in 2007—figures halved by the crisis.
  - Near record-high delinquencies and foreclosures and increasing negative equity amid sharp house-price declines from peak levels could affect GDP and employment beyond housing’s direct share.
- Key national and state housing indicators (end-2010 / peak-to-2010 declines):
  - One in 12 mortgages are late for at least 30 days.
  - 4.6 percent of outstanding mortgages are in foreclosure.
  - Case-Shiller and National Association of Realtors indices are both around 30 percent below peak levels.
  - FHFA house-price indices show, on average, house prices declined by over 15 percent from their peak levels in 2007.
  - Examples of large declines: almost 50 percent in Nevada, about 40 percent in California, and 36 percent in Florida (peak-to-2010).
- Foreclosure and delinquency specifics (end-2010):
  - United States: Delinquency rate 8.2; Foreclosure rate 4.6.
  - Nevada: Delinquency rate 12.0; Foreclosure rate 10.1.
  - Michigan: Delinquency rate 10.6; Foreclosure rate 4.3.
  - Indiana: Delinquency rate 10.2; Foreclosure rate 4.8.
  - Florida: Delinquency rate 10.1; Foreclosure rate 14.2.
  - Ohio: Delinquency rate 9.8; Foreclosure rate 4.9.
  - Arizona: Delinquency rate 9.5; Foreclosure rate 5.7.
  - California: Delinquency rate 9.2; Foreclosure rate 4.5.
- Regional dispersion:
  - Standard deviation of changes in foreclosure rates and housing prices grew dramatically during this recession compared with the early 1990s and early 2000s recessions, indicating severe regional disparities.
  - Composite index interacting levels and increases in SMI and foreclosure rates shows states particularly hit by both shocks: California, Michigan, Ohio, Florida, and Arizona.
  - Some states fared relatively better: Montana, North and South Dakota, Nebraska, Kansas, Colorado, and Texas—though all states saw deterioration.

### Migration, Beveridge curve, and labor market adjustment
- Inter-state migration has gone down during the crisis.
- Beveridge curve observations:
  - The Beveridge curve represents an inverse relationship between unemployment rate and job openings rate.
  - Typically recessions move the economy along the curve (unemployment up, job openings down).
  - This recession appears to have altered the relationship—suggesting employers are finding it harder to match job vacancies with suitable workers.
  - Estimated Beveridge curves depend on sample period:
    - Including recent crisis and recovery (2002-10) forces the curve to fit recent data and may underestimate the upward shift.
    - Using sample ending before recession (2002-07) could overestimate the upward shift.
  - Past recessions exhibit cyclical Beveridge-curve shifts that can revert after a few quarters; the recent shift seems quite sustained so far.
- Key remark:
  - An increase in the underlying structural rate of unemployment need not be associated with observed slower mobility now; high cyclical unemployment and general economic weakness can mask increases in structural unemployment.

### Is structural unemployment rising? Econometric approach
- Data and sample:
  - Panel data approach with annual information from 1991 to 2008 for 50 states plus the District of Columbia.
  - Sample in econometric analysis limited up to 2008 because state GDP data were not available for 2009–10.
- Model specification:
  - Relates annual percentage-point changes in state-level unemployment rate (∆u_it) to:
    - Annual percent growth rate in state-level output (∆y_it),
    - Annual percent changes in state-level skill-mismatch index (∆m_it),
    - Indicator for housing conditions (∆h_it) measured either as percent changes in housing prices or annual percentage-point changes in foreclosure rates,
    - Interaction term ∆m_it * ∆h_it to capture interrelation of skill mismatches and housing conditions,
    - State-specific dummies (S_i) and year-specific dummies (T_t),
    - Residual ε_it assumed i.i.d.
  - Formal representation (as in text):
    - * itS i T t Y itM itH itMH itit u S T ymhm h β β ββββ ε Δ = + + Δ + Δ + Δ + Δ Δ + (equation notation preserved from source)
- Rationale for controls:
  - State dummies capture institutional differences (e.g., unemployment insurance systems, time-invariant labor-market features).
  - Year dummies control for aggregate shocks affecting state unemployment and other regressors (e.g., interest rates, federal spending).
  - State-level GDP changes capture state-specific business cycles.
- Addressing measurement error in state GDP:
  - BEA state-GDP estimates subject to substantial measurement error.
  - Two alternative constructed measures were proposed:
    - ∆ln(GDP1_i) = ∑_j (∆ln(GDP_j) * sh1_ij) where sh1_ij = share of employment in sector j in state i.
    - ∆ln(GDP2_i) = ∆∑_j (ln(GDP_j * sh2_ij)) where sh2_ij = share of employment in sector j in state i vis-à-vis national sectoral employment.
  - Choice between measures guided by comparing Okun’s Law estimates and selecting GDP growth variable producing coefficients closest to international evidence.

*Italicized source attribution: IMF staff analysis and figures extracted from the chapter PDF _wp11105 - 11.4 percent). Unemployment rates for*

### 0.3 to -0.4 (see Batini, Estevão, and Keim (2010), which includes estimates for the United

### _wp11105 - 0.3 to -0.4 (see Batini, Estevão, and Keim (2010), which includes estimates for the United States)

### Measures of state-level GDP and Okun estimates
- Alternative state-level GDP measures are of the same order of magnitude as national-level estimates and explain a larger share of changes in the unemployment rate than the BEA official measure.
- The first alternative measure produces a -0.3 coefficient even with time dummies; the BEA measure produces a coefficient five times smaller in the specification including time dummies (Table 1, column 4).
- Reported specification choice: estimates of equation (1) use the first alternative measure of state-level GDP to capture the largest amount of local business cycle variation beyond aggregate time effects.
- Table 1 (selected coefficients and statistics):
  - Log-change in real GDP (BEA estimates) coefficient: -0.136*** (p-values reported as (0.00) in table layout for relevant columns).
  - Log-change in real GDP (Alternative Measure 1) coefficient: -0.443*** and -0.296*** (p-values (0.00)).
  - Log-change in real GDP (Alternative Measure 2) coefficient: -0.374*** and -0.166*** (p-values (0.00)).
  - Time effects: No for columns 1–3; Yes for columns 4–6.
  - Fixed state effects: Yes for all columns.
  - Adj. R-squared: 0.15, 0.51, 0.50, 0.61, 0.77, 0.80 (by column).
  - Number of states, including D.C.: 51.
  - Observations: 918, 969, 969, 918, 969, 969 (by column).

### Skill mismatches, housing conditions, and unemployment
- Estimates suggest increases in skill mismatches and deterioration in housing markets raise unemployment rates.
- Table 2 (using foreclosure rates as proxy for housing conditions) shows:
  - Log-change in real GDP (Alternative Measure 1) coefficients: -0.178***, -0.267***, -0.171***, -0.166**, -0.264*** (p-values (0.00) across reported columns).
  - Log-change in skill mismatch index coefficient: 0.041***, 0.034***, 0.032*** (p-values (0.00)).
  - Percentage-point (pp.) change in foreclosure rate coefficients: 0.396***, 0.362***, 0.332***, 0.666*** (p-values (0.00)).
  - Interaction term log-change in skill mismatch * pp. change in foreclosure rate: 0.012 (p-value (0.17), not significant).
  - Time effects: Yes; Fixed state effects: Yes.
  - Adj. R-squared: 0.77, 0.78, 0.79, 0.79 (by column).
  - Number of states, including D.C.: 51.
  - Observations: 969 across columns.
- Interpretation: higher state-level unemployment rates are associated with higher skill mismatches and worse housing market conditions even after controlling for state GDP growth, time dummies, and state dummies.
- Causal mechanism consistent with housing conditions affecting mobility: as foreclosure rates rise, individuals with skill mismatches may (i) be reluctant to move due to capital losses in housing wealth or (ii) find it difficult to move because foreclosures affect credit scores and job search capability.

### Instrumental-variable estimates for housing effects
- Column (5) in Table 2 uses 2SLS with the share of subprime mortgages (contemporaneous and one-lagged) as instruments for changes in foreclosure rates.
  - Rationale: subprime share reflects structural housing market condition correlated with foreclosure rates but not directly related to changes in unemployment in a particular way.
  - Result: the IV estimate produces a stronger effect of housing market conditions on unemployment rates (e.g., foreclosure coefficient 0.666*** in the reported 2SLS column).
- Similar IV strategy applied in Table 3 using subprime share as instrument for changes in FHFA housing price index also produces stronger estimates.
  - Table 3 selected coefficients:
    - Log-change in real GDP (Alternative Measure 1): -0.178***, -0.252***, -0.207***, -0.173**, -0.206** (p-values (0.00), (0.01), (0.04), (0.02) respectively).
    - Log-change in skill mismatch index: 0.041***, 0.019***, 0.030*** (p-values (0.00), (0.01), (0.00)).
    - Log-change in FHFA house price index: -0.028***, -0.024***, -0.026***, -0.064*** (p-values (0.00)).
    - Interaction term log-change in skill mismatch * log-change FHFA house price index: -0.003*** (p-value (0.00), statistically significant).
  - Time effects and fixed state effects: Yes.
  - Adj. R-squared: 0.77, 0.78, 0.78, 0.79 (by column).
  - Number of states, including D.C.: 51; Observations: 969.

### Robustness: alternative skill-mismatch proxy (sectoral dispersion)
- Alternative specification: standard deviation of changes in sectoral employment as proxy for skill mismatch shocks.
  - Formal variable: sigma_ijt = [ sum_{i=1}^{10} s_ijt (g_ijt - g_jt)^2 ]^{1/2}, where g_ijt percent change in employment in sector i, state j, time t; g_jt percent change in total employment in state j, time t; s_ijt share of employment in sector i in total employment in state j at time t.
  - Ten sectors used (mining, lodging, and construction merged into one); correlation coefficient with proposed skill-mismatch index = 0.48.
- Tables 4 and 5 results (selected):
  - Table 4 (FHFA house price index as housing proxy):
    - Log-change in real GDP (Alternative Measure 1): -0.200***, -0.252***, -0.195**, -0.186** (p-values (0.00), (0.00), (0.02), (0.03)).
    - Std. deviation of sectoral employment growth coefficients: 0.113***, 0.071***, 0.109*** (p-values (0.00)).
    - Log-change in FHFA house price index: -0.028***, -0.029***, -0.005 (p-values (0.00), (0.00), (0.61)).
    - Interaction std. deviation * log-change FHFA house price index: -0.011*** (p-value (0.00)).
    - Adj. R-squared: 0.77, 0.78, 0.78, 0.79; Number of states: 51; Observations: 969.
  - Table 5 (foreclosure rate as housing proxy):
    - Log-change in real GDP (Alternative Measure 1): -0.200**, -0.267***, -0.191**, -0.187** (p-values (0.02), (0.00), (0.02), (0.03)).
    - Std. deviation of sectoral employment growth coefficients: 0.113***, 0.090***, 0.084*** (p-values (0.00)).
    - Percentage-point change in foreclosure rate: 0.396***, 0.377***, 0.183 (p-values (0.00), (0.00), (0.13)).
    - Interaction std. dev. * pp. change in foreclosure rate: 0.052* (p-value (0.09)).
    - Adj. R-squared: 0.77, 0.78, 0.78, 0.78; Number of states: 51; Observations: 969.
- Conclusion: results using sectoral dispersion as proxy are consistent with main findings; interaction terms significant in several specifications.

### Simulation: contribution to structural unemployment during crisis
- Using coefficients in column (4) of Table 2, the model decomposes increase in state unemployment from 2007 to 2010 into structural (skill mismatches and housing) versus cyclical factors.
- Illustrative state-level examples:
  - Florida: 5½ percentage points out of the 8 percentage points increase in unemployment explained by structural factors.
  - Nevada: 3½ percentage points out of the 10.2 percentage-point increase since 2006 explained by structural factors.
  - Delaware: 3 out of the 5 percentage-point increase explained by structural factors.
  - Arizona: 3¼ out of the 6 percentage-point increase explained by structural factors.
- Aggregate structural unemployment rate estimates:
  - Range of models in Tables 2 and 3 suggests structural unemployment rose from end-2007 to 2010 by between 1 and 1½ percentage points (and by as much as 1¾ since the onset of the housing market collapse at end-2006).
  - Adding these figures to a 5 percent estimate for the structural unemployment rate in 2006 leads to a structural unemployment rate of around 6¾ percent by end-2010.
- Further decomposition of the rise in NAIRU:
  - Unfavorable housing market conditions roughly explain 1 percentage point of the increase in structural unemployment.
  - Deteriorating skill mismatches explain another 1/2 percentage point.
  - Interaction of skill mismatches and foreclosures explains the remaining 1/4 percentage point.
  - The bulk of the increase occurred in 2009, with NAIRU rising by over 3/4 percentage points in that year; about one third of that rise was explained by rising skill mismatches.
- Timing and persistence notes:
  - Deteriorating housing market conditions were behind rising NAIRU until end-2009; thereafter housing market showed signs of stabilization.
  - Skill mismatches were solely responsible for almost one quarter percentage point increase in NAIRU in 2010.
  - Authors note estimates have little to say about persistence due to data limitations (skills mismatch index only available post 1990); the natural rate has been on a decreasing trend since the mid-1970s, making persistence an important question for future research.
- Additional contextual statistics cited:
  - In 2009, the annual average unemployment rate increased by 3½ percentage points—the largest annual increase since 1948.
  - CoreLogic (2011) data: over 65 percent of outstanding mortgages in Nevada are currently "under water"; figures for Arizona, Florida, Michigan, and California are, respectively, 50.9 percent, 47.3 percent, 36.2 percent, and [text ends].

### Policy implications and social costs
- Increases in structural unemployment pose significant social costs, particularly via rises in long-term unemployment and the share of permanent job losers.
- Prior evidence cited:
  - Workers displaced from long-term jobs in the early 1980s recession faced large income losses even 20 years after displacement (Von Wachter, Song, and Manchester, 2009) and serious health consequences (Sullivan and Von Wachter referenced).
  - Declining labor mobility during the recent downturn may be linked to negative housing equity lock-in effects (Ferreira et al. (2010)).
- Authors view: rising equilibrium unemployment during the recent crisis is driven in part by hampered labor market adjustment from structural factors (skill mismatches and housing problems), implying potential policy relevance to address these structural impediments.

*Italicized source attribution: Excerpt from the PDF content unit provided.*

### 31.8 percent.

### _wp11105 - 31.8 percent.

### Summary of findings on unemployment, mismatches, and housing
- Unemployment rate is "in the neighborhood of 9 percent—say, about 2¼ percentage points above our estimated equilibrium unemployment rate."
- Applying econometric estimates to the last two years of data, the aggregate equilibrium unemployment rate in the United States "is about 1¾ percentage points higher now than the 5 percent or so before the crisis."
- "With actual unemployment rates fluctuating around 8¾ percent, an economic recovery that does not address labor market mismatches or structural housing market problems would begin to produce inflationary pressures as unemployment rates falls much below 7 percent."
- Econometric evidence controlling for cyclical relations shows that "unemployment rates increase when skill mismatches and housing market conditions worsen—and often more so when these factors coincide."
- The recent crisis produced "significant sectoral and geographic dislocations" that raised mismatches between demand and supply of labor market skills, with greater effects in some states.
- The analysis does not address persistence of the increase in structural unemployment due to "data limitations in the construction of the skills mismatch index."

### Policy implications and recommended priorities
- Shorten duration of unemployment spells and reduce structural unemployment rates by:
  - Using general macroeconomic stimulus to address cyclical determinants.
  - Deploying targeted policies for structural determinants.
- There is "significant space to increase labor utilization through straightforward macro stimulus" given current unemployment relative to estimated equilibrium.
- Complement macro stimulus with targeted measures to raise hiring and clear the housing market, considering fiscal constraints.
- Funding options for structural measures: "more action to reduce structural problems could be paid for by reducing tax expenditures or enacting a forceful entitlements reform."

### Active Labor Market Policies (ALMPs)
- ALMPs defined as "spending in training programs, targeted subsidies to job creation, public employment services and other expenditures aimed at promoting employment." (Non-targeted policies to lower labor costs excluded.)
- Evidence and caveats:
  - ALMPs "have been shown to raise labor productivity and employment levels when well targeted, although cost-benefit evaluations are often elusive."
  - "Subsidies to net hiring" are highlighted as a priority, as academic research shows they are "more effective in raising employment rates than other active labor market policies."
  - Subsidies should be "well targeted to avoid redundancy and waste" and one alternative is to "condition the subsidies to the hiring of longer-term unemployed."
  - Risks: negative substitution effects via displacement of nonsubsidized workers; design should minimize these effects.
- Specific program observations:
  - Job bills enacted in response to the crisis that "extend unemployment insurance while providing subsidies to net hiring for small businesses, are welcome."
  - Evidence that "employment services and job search assistance can be cost-effective" and that "economic returns to further education and training are also high."
  - Studies: Card (1990) found firms raise hiring when short-run wage costs fall; Bartik and Bishop (2009) and Congressional Budget Office (2010) suggest tax credits can be effective in terms of employment creation per budgetary cost.
  - Estevão (2007) shows "subsidies to direct hiring by the private sector have been the best alternative" among examined ALMPs for raising employment sustainably across OECD panel.

### Training, reemployment services, and program effectiveness
- ARRA measures included "doubling the amount of money available to train and retrain workers, mostly through programs under the Workforce Investment Act" and enhancements to the Trade Adjustment Assistance (TAA) program.
- Concerns about program design and evaluation:
  - Since WIA implementation in 1998, "there has never been a rigorous evaluation of its effectiveness."
  - GAO (2007) finds TAA fund allocation "does not adequately reflect the state's current needs for training services," lacks effective performance measurements, and "less than half of the TAA eligible workers were aware or were informed of the availability of these programs."
  - GAO (2007) asserts federal job training policy remains "fragmented and inefficient," noting "40 federal programs in 1999 that spent an estimated $11.7 billion in job training or job placement assistance."
  - Independent studies indicate program benefits "modestly outweigh the costs" but (i) are "not enough by themselves to lift their target populations ... out of poverty" and (ii) benefits "probably fade after four to five years."

### Housing market interactions and policy options
- Weak housing markets and household balance-sheet stress reduce labor mobility and slow labor market adjustment; neither macro policy nor ALMPs alone sufficiently address this interaction.
- Administration measures taken include "temporary tax incentives for home buyers and measures to mitigate the foreclosure epidemic," plus Fall 2010 actions such as "temporary foreclosure forbearance to the unemployed and some assistance to underwater mortgage holders."
- IMF economists had long recommended additional measures to support housing (IMF, 2008 and 2009).
- If housing continues to depress household balance sheets, "further steps may be needed." Allowing mortgages to be renegotiated in courts ("cramdowns") is presented as "the next best alternative," as advocated in IMF (2008, 2010a), because "existing government programs have consistently missed their desired targets."
- Measures to "raise the number of mortgage modifications and 'cramdowns'" could help clear housing markets more quickly and restore household balance sheets, thereby also tackling cyclical unemployment.

### Conclusions and areas for future research
- The crisis caused "sectoral and geographic dislocations" that increased skill mismatches and unequal housing market outcomes across states.
- Recovery that does not address mismatches and structural housing problems risks inflationary pressures if unemployment falls below 7 percent, given current equilibrium estimates.
- Recommended policy mix: continue macroeconomic stimulus while complementing with targeted measures to raise hiring and clear housing markets, with careful evaluation of cost and effectiveness given fiscal challenges.
- Research limitations and future work:
  - The analysis cannot assess persistence of structural unemployment increase due to limited time series for the skills mismatch index.
  - Longer time series would allow comparisons with the early 1980s recession and other episodes when skill mismatches rose rapidly.

*Source: _wp11105 - 31.8 percent.*

### Appendix A.

### _wp11105 - Appendix A.

### Data Description — Key Indicators and Sources
- House Price Index (purchase only, SA) — Federal Housing Finance Agency
- S&P/Case-Shiller Home Price Index, Existing Single-Family Residential Homes (Composite 20, SA) — Standard & Poor's, Fiserv and MacroMarkets LLC.
- Unemployment Rate — Local Area Unemployment Statistics Database, Bureau of Labor Statistics
- Subprime Share of Mortgages — Ratio of conventional subprime mortgages serviced in percent of total mortgages. Mortgage Banker's Association
- Foreclosure Rate — Mortgage foreclosure inventory. Mortgage Banker's Association
- Real GDP — Data for pre-1997 were interpolated given a structural break in methodology from SIC to NAICS. Bureau of Economic Analysis
- Population by educational attainment — Aged 25 and older. Data were interpolated for select years (i.e., 1991-97, 2001-03, 2005, 2008), due to data limitations. Census Bureau
- Employment by state/industry — Current Employment Statistics database. Bureau of Labor Statistics
- Note: Sources include author’s calculations for some series.

### House Prices
- Indicator definitions for house price measures are provided (Federal Housing Finance Agency and S&P/Case-Shiller Composite 20), both series are described as seasonally adjusted (SA).

### Calculating the Skills Mismatch Index (SMI) — Methodology (Appendix B)
- Construction:
  - The SMI for each state i at time t follows Peters (2000).
  - Notation: j = skill level;  = percent of population with skill level j at time t in state i (skill level supply);  = percent of employees with skill level j at time t in state i (skill level demand).
- Skill Level Supply:
  - Population aged 25+ is divided into three skill levels based on educational attainment:
    - Low-skilled — having less than high school diploma
    - Semi-skilled — having a high school diploma but less than a bachelor’s degree
    - High-skilled — having at least a bachelor’s degree
  - Data were interpolated for years when direct information was not available (namely 1991–97, 2001–03, 2005, 2008–10).
- Skill Level Demand:
  - Industries in the Establishment Survey (one-digit industrial specification) are classified into three categories according to their "skill-intensity" in 2006 (base year).
  - Using 2006 Current Population Survey (Table 7) for individuals aged 18–64, calculate the proportion of employees in each industry by educational attainment (proxy for skill intensity).
  - Industries are weighted from 1 to 3 (larger weights for higher educational attainment) to compute a weighted average skill-intensity by industry.
  - Classification thresholds:
    - Low-skilled industries — weighted average lower than 80 percent of the total average
    - Semi-skilled industries — weighted average between 80–120 percent of the total average
    - High-skilled industries — weighted average above 120 percent of the total average
  - For each skill-level demand, the number of employees by industry and state from the Current Employment Statistics database is used.
- Industry classification (by skill-intensity, as presented):
  - Low Skilled: Mining and Logging; Construction
  - Semi Skilled: Manufacturing; Trade, Transportation, and Utilities; Leisure and Hospitality; Other Services; Government
  - High Skilled: Information; Financial Activities; Education and Health Care; Professional and Business Services
- Data preparation note:
  - All series are seasonally adjusted; Census Bureau’s X-12 Arima seasonal adjustment program was used for data that were not available in seasonally adjusted terms.
  - For the District of Columbia, Delaware, Hawaii, Nebraska, Maryland, South Dakota and Tennessee, the combined series "Mining, Logging, and Construction" was used to construct the demand for low-skilled workers.

### Cross-State Data and Variables (presented in tabular form)
- The appendix presents a cross-state table (average, all years) including:
  - Percent of population (25+ years) by skill level (Low Skilled, Semi Skilled, High Skilled)
  - Percent of employed (in percent of labor force) by skill level
  - Unemployment (in percent of labor force)
  - Skill Mismatch Index (SMI)
  - Delinquencies (in percent of mortgages)
  - Foreclosures (in percent of mortgages)
- The table includes state-level rows (AK, AL, AR, …, WY), a US aggregate row (US), and Max and Min rows with corresponding state identifiers where applicable.
- Example values displayed in the table (preserved exactly as shown in the source table):
  - US row: 18.05, 7.82, 4.25, 8.4, 45.5, 48.7, 984.45, 5.54, 4.8, 1.3
  - Max row: 27.16, 6.24, 0.815, 15.45, 54.0, 77.1, 2033.87, 7.48, 8.5, 2.6
  - Min row: 11.33, 8.81, 5.11, 1.72, 21.23, 6.64, 32.23, 3.22, 2.8, 0.7
  - (State-specific rows for AK, AL, AR, … WY are included in the appendix table with the full set of columns as shown.)

*Sources: Haver Analytics, U.S. Bureau of Labor Statistics, U.S. Census Bureau, and authors' calculations.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11105.pdf_