## wpiea2019114 — Hysteresis in Labor Markets? Evidence from Professional Long-Term Forecasts

## Source details

**Canonical URL:** [wpiea2019114 — Hysteresis in Labor Markets? Evidence from Professional Long-Term Forecasts](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019114.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019114.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019114.pdf.json)

---

### Abstract and central question
- Explores long-term impact of economic booms on labor market outcomes using revisions to professional long-term forecasts over the past 30 years for 34 advanced economies.
- Key quantitative findings:
  - Following a 1 percent unexpected increase in current-period employment, five-year-ahead employment forecasts rise by 1.6 percent on average.
  - Five-year-ahead labor force participation forecasts rise by 1.1 percent.
  - Five-year-ahead unemployment-rate forecasts fall by 0.4 percentage point.
- Interpretation:
  - Forecasters typically raise long-term employment and participation expectations by more than one-for-one after employment surprises, consistent with positive hysteresis (long-lasting supply-side gains) rather than purely transitory effects.
  - Forecast-evaluation tests indicate forecasters are, on average, unbiased in assessing these persistent effects.

### Methodology and data
- Data and sample:
  - Historical vintages of IMF World Economic Outlook (WEO) forecasts from Spring and Fall of every year from 1990 through 2018.
  - Sample covers 34 advanced economies.
  - Complementary robustness check using U.S. CBO ten-year projections (available since 2008), yielding 17 observations of forecast revisions.
- Empirical approach:
  - Estimate regressions of revisions to five-year-ahead forecast levels on contemporaneous revisions to current-period employment forecasts.
  - Baseline conceptual equation: r_t(x_{i,t+h}) = α + β_h r_t(e_{i,t}) + ε_{h,i,t} with h = 5 (log employment and analogous specifications for other labor market variables).
  - Forecast-evaluation equation: x_{i,t+h} − f_t(x_{i,t+h}) = α + θ r_t(e_{i,t}) + ε_{h,i,t} (used to test for systematic over/understatement of persistence).

### Baseline estimation results (pooled and country-level)
- Pooled 34-AE baseline estimates:
  - Five-year-ahead employment response to a 1 percent surprise rise in current employment: 1.58 (0.12); Observations: 945.
  - Five-year-ahead labor force response: 1.10 (0.13); Observations: 944.
  - Five-year-ahead unemployment-rate response: −0.42 (0.07); Observations: 946.
  - The 90 percent confidence interval for the employment and labor force responses are above 1.0; for the unemployment-rate response the interval is below zero.
- Country-level:
  - Similar qualitative results for major individual economies (United States, Canada, Japan, Euro Area), though confidence intervals are wider.
  - Pesaran, Shin, and Smith (1999) mean group estimator yields average effect of 1.6 percent across the 34 advanced markets.

### Demand-driven episodes and inflation/unemployment surprises
- Identification:
  - Episodes where real-time forecast revisions move unemployment and CPI inflation unexpectedly in opposite directions are treated as demand-dominant.
  - Alternative identification uses large unexpected shifts in government consumption (revisions above the sample median in absolute value); robustness checks use Auerbach and Gorodnichenko (2013) shocks and large unexpected changes in cyclically-adjusted overall fiscal balance.
- Perceived long-term responses to a 1 percent current employment surprise (demand-driven sample):
  - Employment (five-year-ahead): 1.69 (0.16); Observations: 405.
  - Labor Force (five-year-ahead): 1.31 (0.16); Observations: 404.
  - Unemployment Rate (five-year-ahead): −0.34 (0.09); Observations: 405.
  - Short-run inflation (near-term): forecasters expect inflation to rise by 0.2 percentage point on average (noting this is “consistent with a Phillips Curve slope of about one-half”).
  - Inflation expectations in 3–5 years are typically unaffected in these episodes (longer-term inflation expectations appear firmly anchored).

### Robustness checks and alternative samples (selected exact estimates)
- Pre-2007 sample:
  - Employment: 1.56 (0.19); Observations: 538.
  - Labor Force: 1.12 (0.20); Observations: 537.
  - Unemployment Rate: −0.39 (0.10); Observations: 539.
- Trimmed sample (exclude largest 10% of current-period revisions):
  - Employment: 1.44 (0.14); Observations: 851.
  - Labor Force: 1.08 (0.14); Observations: 850.
  - Unemployment Rate: −0.35 (0.07); Observations: 852.
- Excluding contractions (expansion-only sample):
  - Employment: 1.71 (0.17); Observations: 605.
  - Labor Force: 1.29 (0.18); Observations: 605.
  - Unemployment Rate: −0.37 (0.10); Observations: 605.
- Excluding large deviations from full employment:
  - Employment: 1.64 (0.13); Observations: 624.
  - Labor Force: 1.35 (0.15); Observations: 624.
  - Unemployment Rate: −0.25 (0.07); Observations: 624.
- Controlling for country fixed effects:
  - Employment: 1.58 (0.12); Observations: 945.
  - Labor Force: 1.10 (0.15); Observations: 944.
  - Unemployment Rate: −0.42 (0.10); Observations: 946.
- Controlling for lagged revisions:
  - Employment: 1.54 (0.13); Observations: 832.
  - Labor Force: 1.11 (0.14); Observations: 832.
  - Unemployment Rate: −0.38 (0.07); Observations: 832.
- Comparison with U.S. CBO long-range forecasts:
  - CBO-based results consistent with WEO results—following a 1 percent unexpected rise in employment, ten-year-ahead employment forecasts rise by more than one-for-one and labor force participation forecasts also rise (CBO sample: 17 observations).

### Forecast-efficiency tests and implications for actual persistence
- Efficiency test specification regresses five-year-ahead forecast errors on real-time revisions to current-period employment; under rational expectations the null is θ = 0.
- Panel A. Full sample (θ estimates; h = 5 years):
  - Employment: θ = 0.51 (0.35); Constant = 0.56 (0.25); Observations = 773; R-squared = 0.004.
  - Labor Force: θ = 0.50 (0.22); Constant = 1.60 (0.18); Observations = 773; R-squared = 0.007.
  - Unemployment Rate: θ = 0.01 (0.18); Constant = 0.92 (0.12); Observations = 774; R-squared = 0.000.
- Panel B. Demand-side shocks sample (θ estimates; h = 5 years):
  - Employment: θ = −0.18 (0.42); Constant = 0.76 (0.33); Observations = 430; R-squared = 0.000.
  - Labor Force: θ = 0.24 (0.27); Constant = 1.76 (0.24); Observations = 430; R-squared = 0.001.
  - Unemployment Rate: θ = 0.39 (0.22); Constant = 0.87 (0.15); Observations = 430; R-squared = 0.010.
- Diagnostic:
  - Regressing ex-post forecast errors on real-time revisions to employment yields β = 0.99 (s.e. = 0.07), implying ex-post shocks to employment are strongly correlated with shocks perceived in real time.
- Interpretation:
  - Little evidence of systematic correlation between long-term forecast errors and current-period shocks; forecasters do not appear to systematically overstate or understate persistence.
  - Combined with persistent forecast revisions, this supports that perceived persistence likely reflects actual long-run effects.

### Mechanisms and implications for policy
- Forecasters’ cited channels for persistent positive effects during booms:
  - Tight labor markets drawing discouraged workers back into the labor force.
  - Efficiency-enhancing job switches.
  - Increased investment in human capital through on-the-job training and complementary investment in physical capital and research and development.
- Policy conclusions and recommendations:
  - If recessions have more long-lasting costs than traditionally assumed, policymakers should favor an especially timely and strong monetary policy response; “there may be little value in keeping powder dry.”
  - Monetary policy easing early on, possibly complemented with countercyclical fiscal policies, can mitigate long-lasting negative effects on labor force participation and employment prospects.
  - Positive demand-driven booms may yield longer-lasting supply-side benefits; maintaining an accommodative monetary policy stance and allowing tighter labor markets may produce positive long-term effects.
  - Policymakers must judge when “hidden slack” is exhausted and when accommodation risks generating inflation and financial stability concerns; repeated attempts to exploit the inflation–unemployment trade-off could weaken inflation anchoring.

*Source: wpiea2019114 - Sections 1–3 (IMF Working Paper WP/19/114).*

### Section 1

### Hysteresis in Labor Markets? Evidence from Professional Long-Term Forecasts (Section 1)

### Abstract and central question
- Explores long-term impact of economic booms on labor market outcomes using revisions to professional long-term forecasts over the past 30 years for 34 advanced economies.
- Key finding summary:
  - Following a 1 percent unexpected increase in current-period employment, five-year-ahead employment forecasts rise by 1.6 percent on average.
  - Five-year-ahead labor force participation forecasts rise by 1.1 percent.
  - Five-year-ahead unemployment-rate forecasts fall by 0.4 percentage point.
- Interpretation: Forecasters typically raise long-term employment and participation expectations by more than one-for-one after employment surprises, consistent with positive hysteresis (long-lasting supply-side gains) rather than purely transitory effects.
- Forecasters are, on average, unbiased in assessing these persistent effects according to forecast-evaluation tests.

### Methodology and data
- Approach:
  - Use historical vintages of IMF World Economic Outlook (WEO) forecasts from Spring and Fall of every year from 1990 through 2018.
  - Estimate regressions of revisions to five-year-ahead forecast levels on contemporaneous revisions to current-period employment forecasts.
  - Baseline equation (conceptual): regress r_t(e_{i,t+h}) on r_t(e_{i,t}) with h = 5 (log employment and analogous specifications for other labor market variables).
- Sample:
  - 34 advanced economies (AEs).
  - WEO vintages covering Spring 1990 through Fall 2018.
  - Complementary robustness check using U.S. CBO ten-year projections (available since 2008), yielding 17 observations of forecast revisions.

### Baseline estimation results (pooled and country-level)
- Pooled 34-AE baseline estimates:
  - Five-year-ahead employment response to a 1 percent surprise rise in current employment: 1.6 percent.
  - Five-year-ahead labor force participation response: 1.1 percent.
  - Five-year-ahead unemployment-rate response: −0.4 percentage point.
  - The 90 percent confidence interval for the employment and labor force responses are above 1.0; for the unemployment-rate response the interval is below zero.
- Country-level:
  - Similar qualitative results for major individual economies (United States, Canada, Japan, Euro Area), though confidence intervals are wider.
  - Pesaran, Shin, and Smith (1999) mean group estimator yields average effect of 1.6 percent across the 34 advanced markets.

### Demand-driven episodes and inflation/unemployment surprises
- Definition: Episodes where current-period unemployment and inflation forecasts are revised in opposite directions are treated as dominated by shocks to aggregate demand.
- Findings in demand-driven expansions:
  - For each 1 percent surprise in current-period employment, five-year-ahead employment forecast rises by 1.7 percent.
  - Five-year-ahead labor force participation forecast rises by 1.3 percent.
  - Five-year-ahead unemployment-rate forecast is 0.4 percentage point lower.
  - Near-term forecast of inflation rises by 0.2 percentage point on average (noting this is “consistent with a Phillips Curve slope of about one-half”).
  - Expectations of inflation in 3-5 years are typically unaffected in these episodes.
- Similar persistent forecast responses are observed in episodes with large unexpected changes in government consumption.

### Robustness checks
- Excluding 2007-2018 observations (to address global financial crisis effects) leaves results virtually unchanged.
- Trimming largest 10 percent of current-period forecast revisions (in absolute value) yields virtually unchanged results.
- Excluding negative current-period employment revisions shows approximate symmetry between expansionary and contractionary shocks.
- Excluding periods where forecasters’ estimate of the current-period gap between unemployment and the natural rate is in the top 10 percent (absolute value) does not materially change results.
- Controlling for time- and country-fixed effects leaves results unchanged.
- Controlling for one-period lag of revision to current-period employment (to address correlated revisions or gradual learning) leaves results virtually unchanged.
- Comparison with U.S. CBO long-range forecasts: CBO-based results consistent with WEO results—following a 1 percent unexpected rise in employment, long-term (ten-year-ahead) employment forecasts rise by more than one-for-one and labor force participation forecasts also rise (CBO sample: 17 observations).

### Forecast-efficiency test and implications for actual persistence
- Test: If forecasters systematically overstate persistence, subsequent forecast errors should be negatively correlated with the initial revision (and vice versa).
- Result: No evidence of systematic correlation—forecasters do not appear to consistently overstate or understate persistence.
- Interpretation: If forecasters are, on average, unbiased, their perceived persistence provides evidence about the actual long-run effects of shocks on labor-market outcomes.

### Overall conclusion from Section 1
- Results are consistent with the positive hysteresis hypothesis: economic expansions (including demand-driven episodes) are associated with persistent increases in expected employment and labor force participation and persistent reductions in expected unemployment.
- Findings are not consistent with the hypothesis that aggregate supply is independent of changes in aggregate demand.

*Source: IMF Working Paper WP/19/114, Section 1 (Bluedorn and Leigh), historical WEO vintages (Spring/Fall 1990–Fall 2018) for 34 advanced economies.*

### Section 2

### Section 2

### Demand-side shocks: identification and perceived long-term effects
- Definition of demand-dominant episodes: periods where real-time forecast revisions move unemployment and CPI inflation unexpectedly in opposite directions (forecasters revise current-period unemployment down while revising contemporaneous CPI inflation up, or vice versa).
- Alternative identification: periods with large unexpected shifts in government consumption (revisions above the sample median in absolute value) using WEO historical vintages; robustness checks using shocks from Auerbach and Gorodnichenko (2013) and large unexpected changes in the cyclically-adjusted overall fiscal balance yield similar results.
- Key perceived long-term responses (Table 2; based on estimated perceived impulse responses to a 1 percent shock to employment):
  - For each 1 percent surprise in current-period employment:
    - five-year-ahead forecast of employment rises by an average of 1.7 percent.
    - five-year-ahead forecast of labor force participation rises by 1.3 percent.
    - five-year-ahead forecast of the unemployment rate is 0.4 percentage point lower.
  - For each 1 percent surprise in current-period employment, forecasters expect inflation to rise by an average of 0.2 percentage point (short run), while forecasts of inflation in 3-5 years are unaffected by the labor market shock (longer-term inflation expectations appear firmly anchored in this sample).
- U.S. evidence: repeating the estimation for U.S. data yields similar results but with much wider confidence intervals given only 13 observations for such episodes.

### Forecast efficiency test and fit of forecasters’ views to actual persistence
- Forecast efficiency approach: regress five-year-ahead forecast errors on real-time revisions to current-period employment (following Blanchard and Leigh (2013) methodology). Equation estimated labeled (2) in the source; under rational expectations the null is θ = 0; a negative θ would indicate forecasters overstate persistence.
- Empirical outcomes (Table 3 and Figure 6):
  - Little evidence of a systematic correlation between long-term forecast errors and current-period shocks.
  - Estimated θ:
    - Positive but statistically indistinguishable from zero for employment forecasts.
    - Positive and statistically significant for labor force participation forecasts.
    - Near zero and statistically indistinguishable from zero for the unemployment rate.
  - Interpretation: forecasters neither systematically overstate nor understate the long-term impact of shocks on employment and the unemployment rate; long-term effects on labor force participation are somewhat stronger than forecasters expect.
- Additional diagnostic: regress ex-post forecast errors on real-time revisions to employment yields an estimate of β = 0.99 (s.e. = 0.07), implying ex-post shocks to employment are strongly correlated with shocks perceived in real time.

### Implications for persistence and mechanisms
- Combined evidence (persistence in forecast revisions and unbiasedness of perceived persistence) supports the presence of positive hysteresis: demand-driven expansions are perceived to generate longer-lasting effects on labor market outcomes than traditionally assumed (i.e., aggregate demand shocks may have effects on aggregate supply).
- Possible channels for persistent positive effects during booms noted by forecasters:
  - Tight labor markets drawing discouraged workers back into the labor force.
  - Efficiency-enhancing job switches.
  - Increased investment in human capital through on-the-job training and complementary investment in physical capital and research and development.
- Relation to literature: results for government consumption shocks broadly consistent with Fatás and Summers (2017) on permanent supply-side effects of fiscal consolidation; results align with Bluedorn and Leigh (2018) on perceived persistence for long-term real GDP forecasts.

### Policy conclusions and recommendations
- If recessions have more long-lasting costs than traditionally assumed, policymakers should favor an especially timely and strong monetary policy response; “there may be little value in keeping powder dry.” Monetary policy easing early on, possibly complemented with countercyclical fiscal policies, can mitigate long-lasting negative effects on labor force participation and employment prospects.
- Positive demand-driven booms may yield longer-lasting supply-side benefits; maintaining an accommodative monetary policy stance and allowing tighter labor markets may produce positive long-term effects.
- Policymakers face the challenge of judging when “hidden slack” is exhausted and when accommodation risks generating inflation and financial stability concerns; this trade-off is difficult and requires careful weighing of costs and benefits.
- Note on inflation expectations: in the sample of advanced economies since the 1990s, inflation expectations have remained well-anchored in the face of demand-driven booms—but this anchoring could weaken if policymakers repeatedly try to exploit the inflation–unemployment trade-off without regard to limits to positive hysteresis.

*Source: wpiea2019114 - Section 2 (IMF Working Paper, Section 2).*

### Section 3

### wpiea2019114 - Section 3

### Methodology and Units
- Equation estimated: 푟_t(푥_{i,t+ℎ}) = 훼 + 훽_ℎ 푟_t(푒_{i,t}) + 휀_{ℎ,i,t}
- Alternative equation for forecast evaluation: 푥_{i,t+ℎ} − 푓_t(푥_{i,t+ℎ}) = 훼 + 휃 푟_t(푒_{i,t}) + 휀_{ℎ,i,t}
- Note: Table results refer to ℎ = 5 years.
- Units: percent for Employment and Labor Force; percentage points for Unemployment Rate.
- Deviation from full-employment denotes forecasters’ estimates of current-period unemployment rate in deviation from forecasters’ estimates of natural unemployment rate (in same forecast vintage).
- Economy beyond full employment is defined as periods in which forecasters’ estimates of current-period unemployment rate is below forecasters’ contemporaneous estimates of natural unemployment rate.

### Perceived Long-term Effect of a 1 Percent Shock to Employment (Table 1 — Robustness; baseline and alternative samples)
- Full sample (baseline)
  - Employment: 1.58 (0.12); Observations: 945
  - Labor Force: 1.10 (0.13); Observations: 944
  - Unemployment Rate: -0.42 (0.07); Observations: 946
- Pre-2007 sample
  - Employment: 1.56 (0.19); Observations: 538
  - Labor Force: 1.12 (0.20); Observations: 537
  - Unemployment Rate: -0.39 (0.10); Observations: 539
- Trimmed sample
  - Employment: 1.44 (0.14); Observations: 851
  - Labor Force: 1.08 (0.14); Observations: 850
  - Unemployment Rate: -0.35 (0.07); Observations: 852
- Excluding contractions
  - Employment: 1.71 (0.17); Observations: 605
  - Labor Force: 1.29 (0.18); Observations: 605
  - Unemployment Rate: -0.37 (0.10); Observations: 605
- Excluding large deviations from full employment
  - Employment: 1.64 (0.13); Observations: 624
  - Labor Force: 1.35 (0.15); Observations: 624
  - Unemployment Rate: -0.25 (0.07); Observations: 624
- Controlling for country fixed effects
  - Employment: 1.58 (0.12); Observations: 945
  - Labor Force: 1.10 (0.15); Observations: 944
  - Unemployment Rate: -0.42 (0.10); Observations: 946
- Controlling for lagged revisions
  - Employment: 1.54 (0.13); Observations: 832
  - Labor Force: 1.11 (0.14); Observations: 832
  - Unemployment Rate: -0.38 (0.07); Observations: 832

### Demand-side Shocks — Perceived Effects (Table 2)
- Sample with inflation and unemployment in year t moving unexpectedly in opposite directions
  - Employment: 1.69 (0.16); Observations: 405
  - Labor Force: 1.31 (0.16); Observations: 404
  - Unemployment Rate: -0.34 (0.09); Observations: 405
- Sample with inflation and unemployment in year t moving unexpectedly in opposite directions and economy beyond full employment
  - Employment: 1.70 (0.30); Observations: 98
  - Labor Force: 1.27 (0.26); Observations: 98
  - Unemployment Rate: -0.41 (0.16); Observations: 98
- WEO government consumption shocks
  - Employment: 1.76 (0.22); Observations: 199
  - Labor Force: 1.18 (0.25); Observations: 199
  - Unemployment Rate: -0.52 (0.14); Observations: 199
- Auerbach-Gorodnichenko government shocks
  - Employment: 1.42 (0.15); Observations: 229
  - Labor Force: 1.05 (0.16); Observations: 229
  - Unemployment Rate: -0.30 (0.15); Observations: 229
- WEO structural balance shocks
  - Employment: 1.70 (0.29); Observations: 179
  - Labor Force: 0.96 (0.28); Observations: 179
  - Unemployment Rate: -0.64 (0.18); Observations: 179

### Forecast Evaluation: Estimation Results (Table 3)
- Panel A. Full Sample (θ estimates; ℎ = 5 years)
  - Employment: θ = 0.51 (0.35)
  - Labor Force: θ = 0.50 (0.22)
  - Unemployment Rate: θ = 0.01 (0.18)
  - Constants:
    - Employment constant = 0.56 (0.25)
    - Labor Force constant = 1.60 (0.18)
    - Unemployment Rate constant = 0.92 (0.12)
  - Observations:
    - Employment: 773
    - Labor Force: 773
    - Unemployment Rate: 774
  - R-squared:
    - Employment: 0.004
    - Labor Force: 0.007
    - Unemployment Rate: 0.000
- Panel B. Sample with Demand-side Shocks (θ estimates; ℎ = 5 years)
  - Long-term forecast error section (reported values)
    - Employment: θ = -0.18 (0.42); Constant = 0.76 (0.33); Observations = 430; R-squared = 0.000
    - Labor Force: θ = 0.24 (0.27); Constant = 1.76 (0.24); Observations = 430; R-squared = 0.001
    - Unemployment Rate: θ = 0.39 (0.22); Constant = 0.87 (0.15); Observations = 430; R-squared = 0.010

*Source: wpiea2019114 - Section 3*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019114.pdf_