## wpiea2021270-print-pdf

## Source details

**Canonical URL:** [wpiea2021270-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021270-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021270-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021270-print-pdf.pdf.json)

---

### 1. Introduction — key findings
- Negative linear association documented between the unemployment gap (u − u*) and the output gap (ln(y) − ln(y*)).
- Baseline pooled results by country group:
  - Sensitivity of the overall working age population’s unemployment gap to demand conditions is about twice as high in AEs as in EMDEs.
  - EMDEs: unemployment gap for the working age population rises less than 0.2 percentage points for a 1 percentage point decline in the output gap.
  - AEs: unemployment gap for the working age population rises more than 0.3 percentage points for a 1 percentage point decline in the output gap.
- Demographic heterogeneity:
  - Young men’s unemployment gap sensitivity ≈ twice that of the overall working age population in both AEs and EMDEs; young men display the highest cyclical sensitivity.
  - In AEs, women’s unemployment gap sensitivity ≈ 80 percent of men’s for both youth and adults.
  - In EMDEs, no significant gender differences in labor market sensitivity for adults or youth.
- Cyclical asymmetry:
  - Unemployment is more sensitive to demand conditions in downturns than in upturns; asymmetry driven by men and particularly young men.

### 2. Econometric methodology
- Baseline specification (gaps):
  - u_{i,t} − u^*_{i,t} = μ_i + β[ln(y_{i,t}) − ln(y^*_{i,t})] + ε_{i,t}  (Equation 1)
  - β measures short-run responsiveness of the unemployment gap to the output gap; expected negative.
- Estimation:
  - Pooled coefficient estimates by country group (AEs and EMDEs).
  - Least squares for panel data with heteroskedasticity- and autocorrelation-robust standard errors clustered at the country level.
- Interpretation benchmarks:
  - Frictionless markets with constant participation: Okun’s coefficient ≈ -1.5.
  - Empirical estimates expected to be above -1.5 in absolute value; greater informality in EMDEs expected to reduce |β|.
- Demographic breakdown:
  - Groups: working age (15–64), youth (15–24), adults (25–64); each disaggregated by gender (women, men).
- Hypotheses:
  - Youth unemployment more sensitive than adults due to fragile employment, temporary contracts, lower adjustment costs.
  - Gender differences ambiguous; weaker female attachment or procyclical female participation could reduce measured Okun coefficients for women.

### 3. Extensions and decompositions — methodology and key mechanisms
- Employment/participation decomposition:
  - Approximation: ln(E_{i,t}) − ln(L_{i,t}) ≈ −u_{i,t}.
  - Two gap regressions:
    - Employment gap: ln(E_{i,t}) − ln(E^*_{i,t}) = μ^E_i + δ[ln(y_{i,t}) − ln(y^*_{i,t})] + ε^E_{i,t}  (Equation 2)
    - Participation gap: ln(L_{i,t}) − ln(L^*_{i,t}) = μ^L_i + θ[ln(y_{i,t}) − ln(y^*_{i,t})] + ε^L_{i,t}  (Equation 3)
  - δ and θ capture cyclical sensitivities of employment and participation.
  - Finding: procyclicality of participation reduces the absolute unemployment response relative to the employment response for all groups.
- Business-cycle phase asymmetry:
  - Extended specification allows different sensitivities in expansions (d_{i,t}=1) and contractions (d_{i,t}=0) (Equation 4).
  - Finding: Okun’s Law stronger during downturns than upturns; asymmetry driven by men, especially young men.

### 4. Dataset and construction
- Sample span: 1990 to 2015.
- Countries: 38 AEs and 58 EMDEs (IMF WEO classification). Panel unbalanced.
- Main data sources:
  - Unemployment and participation rates: ILOSTAT (ILO).
  - Population by age and gender: United Nations population statistics.
  - Real GDP and potential GDP: IMF World Economic Outlook (WEO).
- Analytic-sample constraint: analyses use sample with both working age and youth unemployment and participation rates available to ensure constant composition.
- Construction of adult series and gaps:
  - U_{a,g,i,t} = u_{a,g,i,t} · l_{a,g,i,t} · P_{a,g,i,t}  (Equation 5)
  - L_{a,g,i,t} = l_{a,g,i,t} · P_{a,g,i,t}  (Equation 6)
  - Adult unemployment rate by gender: u_{a,g,i,t} = (U_{WAP,g,i,t} − U_{Y,g,i,t}) / (L_{WAP,g,i,t} − L_{Y,g,i,t})
  - Employment: E_{a,g,i,t} = L_{a,g,i,t} − U_{a,g,i,t}
  - Missing observations linearly interpolated; HP filter applied to interpolated series with smoothing parameter 100 for yearly data to obtain potential levels.
  - Output gap computed from log real GDP using HP filter with smoothing parameter 100; sensitivity checks use per capita GDP HP filter and IMF WEO potential output.
  - Potential level observations preceded and followed by three or more missing observations, or with unavailable original data, treated as missing.

### 5. Baseline empirical results and robustness

- Table 1: Okun’s Law Across Demographic Groups — Equation (1) estimates (β, s.e., R^2)
  - AEs (sample: 38 countries and 908 observations)
    - All working age: β = -0.31 **, s.e. = 0.05, R^2 = 0.47
    - Adult women: β = -0.22 **, s.e. = 0.04, R^2 = 0.35
    - Adult men: β = -0.30 **, s.e. = 0.05, R^2 = 0.43
    - Young women: β = -0.53 **, s.e. = 0.09, R^2 = 0.36
    - Young men: β = -0.67 **, s.e. = 0.11, R^2 = 0.44
  - EMDEs (sample: 57 countries and 751 observations)
    - All working age: β = -0.17 **, s.e. = 0.03, R^2 = 0.14
    - Adult women: β = -0.14 **, s.e. = 0.03, R^2 = 0.08
    - Adult men: β = -0.14 **, s.e. = 0.03, R^2 = 0.13
    - Young women: β = -0.25 **, s.e. = 0.06, R^2 = 0.07
    - Young men: β = -0.32 **, s.e. = 0.06, R^2 = 0.13
  - Notes: Standard errors clustered at the country level. *, and ** denote significance at the 90 percent, and 99 percent confidence level, respectively.

- Interpretation of baseline results:
  - Okun’s coefficient magnitude ≈ twice as large (in absolute value) in AEs than EMDEs across demographic groups.
  - Relationship generally stronger for men than women and for youth than adults; differences starker in AEs.
  - In AEs, young men’s Okun coefficient (≈ -0.67) ≈ three times larger in absolute value than adult women’s (≈ -0.22).
  - Youth sensitivity ≈ twice adults in EMDEs and > twice in AEs.
  - Gender differences smaller in coefficients than in explanatory power (R^2): e.g., EMDE adults have same β for men and women (-0.14) but R^2 almost twice as large for men.
- Robustness checks:
  - Inclusion of time fixed effects.
  - Alternative country classifications and excluding borderline/graduated countries.
  - Different output gap measures: HP filter on per capita GDP and IMF WEO potential output.
  - First-difference specification: Δu_{i,t} = μ_i + β[Δln(y_{i,t})] + ε_{i,t}.
  - Results similar to baseline (see Appendix Tables A1–A5); baseline specification retained.

### 6. Extensions — decomposition and business-cycle stage results

- 6.1 Decomposition Between Employment and Participation Margins — general findings
  - Both employment and participation gaps are procyclical across demographic groups.
  - Participation movements partially dampen the unemployment response relative to the employment response.
  - AEs: employment gap responds almost twice as much as the unemployment gap for young women and young men (ratios ≈ 1.8 and 1.9).
  - EMDEs: employment/unemployment response ratio ranges between 1.6 (adult women) and 1.8 (young men).
  - Participation sensitivities tend to be larger for men than for women; discouraged-worker dynamics and sectoral composition (e.g., construction) may contribute.
  - Smaller Okun coefficient for women in AEs driven by lower employment-gap response for women, not larger participation sensitivity.

- Table 2: Cyclical sensitivity of employment (δ) and labor force participation (θ)
  - Panel A. Log employment (δ s.e. R2)
    - AEs
      - All working age: 0.43** 0.06 0.28
      - Adult women: 0.25** 0.04 0.09
      - Adult men: 0.39** 0.06 0.29
      - Young women: 0.93** 0.16 0.20
      - Young men: 1.25** 0.19 0.32
    - EMDEs
      - All working age: 0.30** 0.07 0.13
      - Adult women: 0.26** 0.08 0.03
      - Adult men: 0.23** 0.05 0.05
      - Young women: 0.44** 0.16 0.04
      - Young men: 0.57** 0.13 0.07
  - Panel B. Log labor force participation (θ s.e. R2)
    - AEs
      - All working age: 0.09** 0.02 0.06
      - Adult women: 0.02 0.03 0.01
      - Adult men: 0.07** 0.02 0.03
      - Young women: 0.23* 0.09 0.02
      - Young men: 0.38** 0.08 0.07
    - EMDEs
      - All working age: 0.11* 0.06 0.12
      - Adult women: 0.08 0.07 0.02
      - Adult men: 0.08* 0.04 0.03
      - Young women: 0.11 0.11 0.03
      - Young men: 0.17* 0.09 0.03
  - Notes: Panels present estimates from Equations 2 and 3. Standard errors clustered at the country level. *, and ** denote significance at the 90 percent, and 99 percent confidence level, respectively. AEs sample: 38 countries and 908 observations. EMDEs sample: 57 countries and 751 observations.

- 6.2 The Stage of the Business Cycle — main findings
  - Okun’s Law stronger during bad times (negative output gap) than good times; differences clearer and more statistically significant in AEs for some groups (overall working age, young men, adult men).
  - Employment gap exhibits non-linearities across cycle stages; participation generally does not.
  - AEs examples:
    - Adult men employment gap: increases 0.3 percent per 1 percentage point increase in the output gap in good times and decreases 0.5 percent per 1 percentage point decrease in bad times.
    - Young men employment gap: increases 1.1 percent during upturns and decreases 1.5 during downturns; differences statistically significant at the 95 percent confidence level.
  - Women’s employment gap response does not show statistically significant differences between good and bad times.
  - Bottom line: Negative output gaps especially detrimental for men, particularly young men, in AEs.

- Table 3: Okun’s Law in good and bad states (ρ/σ s.e.   Wald R2) — selected entries
  - AEs (first line = good state; second line = bad state)
    - All working age: -0.25** 0.04 0.01 0.48  /  -0.39** 0.07  -0.20** 0.05
    - Adult women: -0.20** 0.03 0.17 0.35  /  -0.24** 0.05  -0.17** 0.05
    - Adult men: -0.23** 0.04 0.02 0.44  /  -0.39** -0.08 -0.18** -0.04
    - Young women: -0.48** 0.09 0.17 0.35  /  -0.59** -0.11 -0.36** -0.10
    - Young men: -0.54** 0.11 0.02 0.45  /  -0.86** 0.15 -0.40** 0.09
  - EMDEs (first line = good state; second line = bad state)
    - All working age: -0.15** 0.06 0.50 0.14  /  -0.20** 0.05
    - Adult women: -0.13** 0.05 0.57 0.08  /  -0.17** 0.05
    - Adult men: -0.11* 0.04 0.25 0.13  /  -0.18** -0.04
    - Young women: -0.16* 0.09 0.57 0.08  /  -0.36** -0.10
    - Young men: -0.25** 0.09 0.31 0.13  /  -0.40** 0.09

- Table 4: Cyclical sensitivity of employment and participation in good and bad states — selected entries
  - Panel A. Log employment (ρ/σ s.e. Wald R2) — first line = good state; second line = bad state
    - AEs
      - All working age: 0.36** 0.07 0.09 0.28  /  0.52** 0.09 0.29* 0.13
      - Adult men: 0.32** 0.06 0.09 0.30  /  0.49** 0.10 0.21* 0.09
      - Young men: 1.06** 0.23 0.04 0.32  /  1.52** 0.19 1.03** 0.25
    - EMDEs
      - Young men (bad state coefficient shown): 1.03** 0.25
  - Panel B. Log labor force participation (ρ/σ s.e. Wald R2) — first line = good state; second line = bad state
    - AEs
      - Young men: 0.39* 0.16 0.96 0.07  /  0.37* 0.14 0.53** 0.19
    - EMDEs
      - Young men: -0.12 -0.13 0.02 0.04  /  0.53** 0.19
  - Notes: Panels present estimates from Equation (6) using log employment gap and log participation gap as dependent variables. Standard errors clustered at the country level. *, and ** denote significance at the 90 percent, and 99 percent confidence level, respectively. ‘Wald’ reports the p-value from a Wald test for equal coefficients (H0: ρ = σ). AEs sample: 38 countries and 908 observations. EMDEs sample: 57 countries and 751 observations.

*wpiea2021270-print-pdf - 1. Introduction*

### 1. Introduction

### 1. Introduction

### Key findings on Okun’s Law by economy and demographic group
- The paper documents a negative linear association between the unemployment gap (u − u*) and the output gap (ln(y) − ln(y*)).  
- Baseline pooled results for country groups:
  - Sensitivity of the overall working age population’s unemployment gap to demand conditions is about twice as high in AEs as in EMDEs.
  - In EMDEs, the unemployment gap for the working age population rises less than 0.2 percentage points for a 1 percentage point decline in the output gap.
  - In AEs, the unemployment gap for the working age population rises more than 0.3 percentage points for a 1 percentage point decline in the output gap.
- Demographic heterogeneity:
  - Young men’s unemployment gap sensitivity is about twice as high as that of the overall working age population, in both AEs and EMDEs; young men display the highest cyclical sensitivity.
  - In AEs, women’s unemployment gap is significantly less sensitive to demand conditions than men’s—about 80 percent the magnitude for both youth and adults.
  - In EMDEs, no significant differences by gender are found—gender does not play a role in labor market sensitivity for either adults or youth.
- Cyclical asymmetry:
  - Cyclical unemployment is more sensitive to demand conditions in downturns than in upturns; this asymmetry is driven by men and particularly young men.

*Authors’ organization of the paper: Sections 2 and 3 describe econometric methodology and data; Section 4 presents baseline results and robustness checks; Section 5 discusses extensions; Section 6 concludes.*

### Literature and motivation
- Builds on Okun (1963) and subsequent empirical literature documenting a negative and relatively stable relationship between aggregate demand and unemployment.
- Prior work finds Okun’s Law holds more strongly in advanced economies (AEs) than in emerging market and developing economies (EMDEs) (Ball, Leigh, and Loungani 2017; An et al. 2019; Ball, Furceri, et al. 2019).
- Prior demographic results motivating the study:
  - Youth unemployment found to be about twice as cyclical as adults in prior studies of advanced European countries (Hutengs and Stadtmann 2013a; Banerji, Saksonovs, et al. 2014; Banerji, Lin, and Saksonovs 2015).
  - Recent work disaggregating by age and gender (Dixon, Lim, and van Ours 2017; Evans 2018) found women’s Okun coefficients tended to be lower than men’s in OECD samples.
- The paper expands earlier analyses by examining age- and gender-specific Okun relationships across a large sample of 38 AEs and 58 EMDEs over 1990–2015.

---

### 2. Econometric methodology

### Baseline specification
- Core estimating equation (gaps specification):
  - u_{i,t} − u^*_{i,t} = μ_i + β[ln(y_{i,t}) − ln(y^*_{i,t})] + ε_{i,t}  (Equation 1)
  - u_{i,t} is the unemployment rate; y_{i,t} is real GDP; * indicates long-run levels; μ_i are country fixed effects; ε_{i,t} is iid error assumed uncorrelated with the output gap.
  - β measures short-run responsiveness of the unemployment gap to the output gap; expected to be negative.
- Estimation details:
  - Pooled coefficient estimates by country group (AEs and EMDEs).
  - Estimation by least squares for panel data with heteroskedasticity- and autocorrelation-robust standard errors clustered at the country level.
- Expected magnitudes and interpretation:
  - In frictionless markets with constant participation, Okun’s coefficient would be around -1.5; empirical estimates (e.g., Ball, Leigh, and Loungani 2017) are much higher (around -0.4 for AEs) due to labor adjustment costs and procyclical participation.
  - Authors expect β to be always above -1.5 in absolute value.
  - Greater labor market informality in EMDEs is expected to reduce the absolute value of β relative to AEs.

### Demographic breakdown and hypotheses
- Estimations are done separately for:
  - Overall working age population: ages 15 to 64 years-old.
  - Youth: ages 15-24 years-old.
  - Adults: ages 25-64 years-old.
  - Each age group further disaggregated by gender (women and men).
- A priori expectations:
  - Youth unemployment more sensitive to output fluctuations than adults due to more fragile employment conditions, higher incidence of temporary contracts, and lower adjustment costs for employers for younger workers.
  - Gender differences ambiguous: weaker female labor force attachment or more procyclical participation for women could reduce estimated Okun coefficients for women relative to men.

---

### 3. Extensions and decompositions

### Employment and participation decomposition
- Unemployment rate approximated via log differences:
  - ln(E_{i,t}) − ln(L_{i,t}) ≈ −u_{i,t}, where E is employment level and L is labor force.
- Two gap regressions estimated:
  - Employment gap: ln(E_{i,t}) − ln(E^*_{i,t}) = μ^E_i + δ[ln(y_{i,t}) − ln(y^*_{i,t})] + ε^E_{i,t}  (Equation 2)
  - Participation gap: ln(L_{i,t}) − ln(L^*_{i,t}) = μ^L_i + θ[ln(y_{i,t}) − ln(y^*_{i,t})] + ε^L_{i,t}  (Equation 3)
- δ and θ capture cyclical sensitivities of employment and participation respectively.
- Finding: procyclicality of labor force participation reduces the absolute unemployment gap response relative to the employment gap response for all groups.

### Business-cycle phase asymmetry
- Extended specification allows sensitivity to differ in expansions and contractions:
  - z_{i,t} − z^*_{i,t} = d_{i,t}·μ^z_i + μ^z_i + ρ^z{d_{i,t}·[ln(y_{i,t}) − ln(y^*_{i,t})]} + σ^z{(1 − d_{i,t})[ln(y_{i,t}) − ln(y^*_{i,t})]} + ε^z_{i,t}  (Equation 4)
  - d_{i,t} = 1 if output gap positive, 0 otherwise.
- Finding: Okun’s Law is stronger during downturns than upturns; asymmetry driven by men, especially young men.

---

### 4. Dataset and construction

### Sample and sources
- Sample span: 1990 to 2015.
- Countries: 38 AEs and 58 EMDEs (classification per IMF WEO). Panel is unbalanced.
- Main data sources:
  - Working age and youth unemployment and labor force participation rates: ILOSTAT of the ILO.
  - Population by age and gender: United Nations population statistics.
  - Real GDP, per capita and potential GDP: IMF World Economic Outlook (WEO).
- Analytic sample constraints:
  - Analyses of sensitivity of unemployment, employment and participation use the sample with both working age and youth unemployment and participation rates available to ensure a constant composition sample.

### Construction of adult series and gaps
- Calculate levels from rates and population:
  - U_{a,g,i,t} = u_{a,g,i,t} · l_{a,g,i,t} · P_{a,g,i,t}  (Equation 5), where a ∈ {Y, WAP}, g ∈ {W, M, B}.
  - L_{a,g,i,t} = l_{a,g,i,t} · P_{a,g,i,t}  (Equation 6).
  - Adult unemployment rate by gender: u_{a,g,i,t} = (U_{WAP,g,i,t} − U_{Y,g,i,t}) / (L_{WAP,g,i,t} − L_{Y,g,i,t}).
  - Employment level: E_{a,g,i,t} = L_{a,g,i,t} − U_{a,g,i,t}.
- Potential levels and cyclical components:
  - Missing observations in original series are linearly interpolated.
  - Hodrick-Prescott (HP) filter applied to interpolated series; trend component used as potential level.
  - HP smoothing parameter set to 100 for yearly data.
  - Potential level observations preceded and followed by three or more missing observations, or with unavailable original data, are treated as missing.
  - Output gap computed from log real GDP using HP filter with smoothing parameter 100; sensitivity checks use per capita GDP HP filter and IMF WEO potential output.

---

### 5. Baseline empirical results and robustness

### Baseline estimates (Table 1: Okun’s Law Across Demographic Groups)
- Table 1 presents Equation (1) estimates. Reported coefficients (β), standard errors (s.e.), and R^2 by country group:

  - AEs (sample: 38 countries and 908 observations)
    - All working age: β = -0.31 **, s.e. = 0.05, R^2 = 0.47
    - Adult women: β = -0.22 **, s.e. = 0.04, R^2 = 0.35
    - Adult men: β = -0.30 **, s.e. = 0.05, R^2 = 0.43
    - Young women: β = -0.53 **, s.e. = 0.09, R^2 = 0.36
    - Young men: β = -0.67 **, s.e. = 0.11, R^2 = 0.44

  - EMDEs (sample: 57 countries and 751 observations)
    - All working age: β = -0.17 **, s.e. = 0.03, R^2 = 0.14
    - Adult women: β = -0.14 **, s.e. = 0.03, R^2 = 0.08
    - Adult men: β = -0.14 **, s.e. = 0.03, R^2 = 0.13
    - Young women: β = -0.25 **, s.e. = 0.06, R^2 = 0.07
    - Young men: β = -0.32 **, s.e. = 0.06, R^2 = 0.13

  - Notes: Standard errors clustered at the country level. *, and ** denote significance at the 90 percent, and 99 percent confidence level, respectively.

### Interpretation of baseline results
- Confirmation of core patterns:
  - Okun’s coefficient magnitude about twice as large (in absolute value) in AEs than EMDEs across all demographic groups.
  - Relationship generally stronger for men than women and for youth than adults; differences starker in AEs.
  - In AEs, young men’s Okun coefficient (≈ -0.67) is about three times larger in absolute value than adult women’s (≈ -0.22).
  - Youth sensitivity: roughly twice as large as adults in EMDEs and more than twice as large in AEs.
  - Gender differences in coefficients are smaller than differences in explanatory power: e.g., in EMDE adults, β is the same for men and women (-0.14) but R^2 almost twice as large for men as for women.
- Possible explanations:
  - Labor market informality in EMDEs allows easier transition to self-employment, dampening unemployment’s sensitivity to output.
  - Youth more likely on temporary contracts and face lower employment protection; employers may lay off younger workers first during downturns.
  - Women’s lower unemployment sensitivity may reflect larger flows between employment and nonparticipation for women and more procyclical participation among women.

### Robustness checks
- Performed robustness exercises on:
  - Inclusion of time fixed effects.
  - Alternative country classifications and excluding borderline or graduated countries.
  - Different output gap measures: HP filter on per capita GDP and IMF WEO potential output.
  - First-difference specification assuming constant natural unemployment rate and potential GDP growth:
    - Δu_{i,t} = μ_i + β[Δln(y_{i,t})] + ε_{i,t}
- Results from Tables A1 to A5 in the Appendix indicate estimates similar to baseline regressions; baseline specification is retained for further analysis.

*Italic source attribution: wpiea2021270-print-pdf - 1. Introduction*

### 5. Extensions

### 5. Extensions

### 5.1. Decomposition Between Employment and Participation Margins
- Purpose: Decompose the unemployment response into employment and participation channels; Okun coefficient approximated as difference between employment and participation gap sensitivities.
- General findings:
  - Both the participation and the employment gaps display positive coefficients across all demographic groups (procyclical).
  - For the overall working age population and adult men in AEs, participation movements slightly dampen the unemployment response (unemployment gap response is below but close to the employment gap in absolute value).
  - In AEs, the employment gap responds almost twice as much as the unemployment gap for young women and young men (ratios about 1.8 and 1.9 respectively), driven by higher participation sensitivities for youth versus adults.
  - In EMDEs, the employment/unemployment gap response ratio ranges between 1.6 (adult women) and 1.8 (young men).
  - Across groups, participation sensitivities tend to be larger for men than for women; discouraged worker dynamics and sectoral employment composition (e.g., construction) may explain this.
  - The smaller Okun coefficient for women in AEs is driven by a lower employment gap response for women than for men, not by larger participation gap sensitivity.

- Table 2: Cyclical sensitivity of employment and labor force participation rates
  - Panel A. Log employment (δ s.e. R2)
    - AEs
      - All working age: 0.43** 0.06 0.28
      - Adult women: 0.25** 0.04 0.09
      - Adult men: 0.39** 0.06 0.29
      - Young women: 0.93** 0.16 0.20
      - Young men: 1.25** 0.19 0.32
    - EMDEs
      - All working age: 0.30** 0.07 0.13
      - Adult women: 0.26** 0.08 0.03
      - Adult men: 0.23** 0.05 0.05
      - Young women: 0.44** 0.16 0.04
      - Young men: 0.57** 0.13 0.07
  - Panel B. Log labor force participation (θ s.e. R2)
    - AEs
      - All working age: 0.09** 0.02 0.06
      - Adult women: 0.02 0.03 0.01
      - Adult men: 0.07** 0.02 0.03
      - Young women: 0.23* 0.09 0.02
      - Young men: 0.38** 0.08 0.07
    - EMDEs
      - All working age: 0.11* 0.06 0.12
      - Adult women: 0.08 0.07 0.02
      - Adult men: 0.08* 0.04 0.03
      - Young women: 0.11 0.11 0.03
      - Young men: 0.17* 0.09 0.03

- Notes (as presented): Panels A and B present estimates from Equations 2 and 3. Standard errors clustered at the country level. *, and ** denote significance at the 90 percent, and 99 percent confidence level, respectively. AEs sample: 38 countries and 908 observations. EMDEs sample: 57 countries and 751 observations.

### 5.2. The Stage of the Business Cycle
- Purpose: Test whether Okun’s Law strength varies by business cycle stage (good times: positive output gap; bad times: negative output gap).
- Main findings:
  - The negative relationship between unemployment and the output gap is stronger during bad times in general; statistical significance of differences is clearer in AEs for certain groups (overall working age, young men, adult men).
  - Employment gap exhibits non-linearities across cycle stages; labor force participation generally does not.
  - In AEs, adult men employment gap: 0.3 percent higher per percentage point increase in output gap (good times) and 0.5 percent lower per percentage point decrease in output gap (bad times).
  - In AEs, young men employment gap: increases 1.1 percent during upturns and decreases 1.5 during downturns; differences statistically significant at the 95 percent confidence level.
  - Women’s employment gap response does not show statistically significant differences between good and bad times.
  - Bottom line: Negative output gaps are especially detrimental for men, particularly young men, in AEs.

- Table 3: Okun’s Law in good and bad states (ρ/σ s.e.   Wald R2)
  - AEs (first line = good state; second line = bad state in each row)
    - All working age: -0.25** 0.04 0.01 0.48  /  -0.39** 0.07  -0.20** 0.05
    - Adult women: -0.20** 0.03 0.17 0.35  /  -0.24** 0.05  -0.17** 0.05
    - Adult men: -0.23** 0.04 0.02 0.44  /  -0.39** -0.08 -0.18** -0.04
    - Young women: -0.48** 0.09 0.17 0.35  /  -0.59** -0.11 -0.36** -0.10
    - Young men: -0.54** 0.11 0.02 0.45  /  -0.86** 0.15 -0.40** 0.09
  - EMDEs (first line = good state; second line = bad state in each row)
    - All working age: -0.15** 0.06 0.50 0.14  /  -0.20** 0.05
    - Adult women: -0.13** 0.05 0.57 0.08  /  -0.17** 0.05
    - Adult men: -0.11* 0.04 0.25 0.13  /  -0.18** -0.04
    - Young women: -0.16* 0.09 0.57 0.08  /  -0.36** -0.10
    - Young men: -0.25** 0.09 0.31 0.13  /  -0.40** 0.09

- Table 4: Cyclical sensitivity of employment and labor force participation in good and bad states
  - Panel A. Log employment (ρ/σ s.e. Wald R2) — first line = good state; second line = bad state
    - AEs
      - All working age: 0.36** 0.07 0.09 0.28  /  0.52** 0.09 0.29* 0.13
      - Adult women: 0.23** 0.06 0.60 0.10  /  0.29** 0.08 0.21 0.14
      - Adult men: 0.32** 0.06 0.09 0.30  /  0.49** 0.10 0.21* 0.09
      - Young women: 0.91** 0.21 0.60 0.10  /  0.95** 0.22 0.67* 0.27
      - Young men: 1.06** 0.23 0.04 0.32  /  1.52** 0.19 1.03** 0.25
    - EMDEs
      - All working age: 0.31** 0.1 0.91 0.13  /  0.29* 0.13
      - Adult women: 0.29* 0.12 0.69 0.03  /  0.21 0.14
      - Adult men: 0.25** 0.09 0.81 0.05  /  0.21* 0.09
      - Young women: 0.26 0.25 0.69 0.03  /  0.67* 0.27
      - Young men: 0.22 0.20 0.03 0.08  /  1.03** 0.25  (EMDEs bad-state coefficient shown: 1.03** 0.25)
  - Panel B. Log labor force participation (ρ/σ s.e. Wald R2) — first line = good state; second line = bad state
    - AEs
      - All working age: 0.09* 0.04 0.91 0.06  /  0.08 0.05 0.06 0.10
      - Adult women: 0.02 0.05 0.92 0.01  /  0.01 0.07 0.01 0.13
      - Adult men: 0.08* 0.04 0.69 0.03  /  0.05 0.03 0.00 0.08
      - Young women: 0.30* 0.14 0.92 0.01  /  0.14 0.18 0.21 0.19
      - Young men: 0.39* 0.16 0.96 0.07  /  0.37* 0.14 0.53** 0.19
    - EMDEs
      - All working age: 0.14* 0.07 0.54 0.12  /  0.06 0.10
      - Adult women: 0.15 0.09 0.33 0.02  /  0.01 0.13
      - Adult men: 0.14* 0.06 0.21 0.03  /  0.00 0.08
      - Young women: 0.03 0.17 0.33 0.02  /  0.21 0.19
      - Young men: -0.12 -0.13 0.02 0.04  /  0.53** 0.19

- Notes (as presented): Panels present estimates from Equation (6), using the log employment gap and the log labor force participation gap as dependent variables. Standard errors clustered at the country level. *, and ** denote significance at the 90 percent, and 99 percent confidence level, respectively. The columns ‘Wald’ report the p-value from a Wald test for equal coefficients (H0: ρ = σ). AEs sample: 38 countries and 908 observations. EMDEs sample: 57 countries and 751 observations.

*Sources: Authors’ estimation based on ILO Key Indicators of the Labour Market and IMF World Economic Outlook.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021270-print-pdf.pdf_