## _wp16122

## Source details

**Canonical URL:** [_wp16122](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16122.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16122.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16122.pdf.json)

---

### Introduction and research questions
- Topic: Slow U.S. wage growth since the global financial crisis (GFC).
- Microdata time coverage: January 2000 to March 2015 (CPS MORG).
- Key research questions:
  - Is labor market repair still weighing on recent wage growth?
  - Has the relationship between labor market slack and wage growth permanently changed (has the Wage-Growth Phillips curve flattened)?
  - What is driving the decline in job-to-job mobility?

### Main empirical findings
- Aggregate and conditional wage trends:
  - Since 2010, annual hourly earnings in nominal terms have risen by 2 percent, about 1⅓ percent less than before the GFC.
  - Labor productivity averaged only ½ percent during 2013–15.
  - Real unit labor costs (labor income share) have declined steadily since the early 2000s.
- Cyclical (labor-market-repair) effects on average wages:
  - Labor market repair is still weighing on average wage growth.
  - Using regional variation in labor demand, larger post-GFC declines in local unemployment rates are associated with smaller increases in average wages.
  - Re-employment of marginal workers at low entry wages moderates aggregate average wage growth.
  - External finding cited: 70–80 percent of new job holders post-GFC earn wages below median levels (Daly and Hobijn 2016).
- Structural changes affecting wages:
  - The wage-growth–Phillips curve has flattened after 2008 for full-time, full-year employed workers; wages do not comove with local unemployment rates post-2008 whereas they did pre-2008.
    - Pre-2008: Phillips curve steepened below 5 percent unemployment.
    - Post-GFC (data up to 2014): no evidence of a similar kink.
  - Job-to-job transition rates declined well before the GFC and continued thereafter across education and age groups; declines are not explained by local labor market tightness or demographics.
  - Shift-share analysis: demographic changes cannot account for sustained decline in job-to-job transitions.
  - These structural patterns raise concerns about allocative ability of the labor market and speed of labor reallocation, which can suppress productivity gains and wage growth.

### Drivers of low wage growth (summary)
- Cyclical channels:
  - Nominal wage rigidity: the share of workers with zero nominal wage growth has increased since the GFC.
  - Hysteresis/compositional effects: long spells of non-employment reduce employability and lead to re-employment at discounted entry wages, suppressing average hourly wage growth.
- Structural channels:
  - Changes in wage bargaining and work arrangements:
    - Share of workers in alternative arrangements: 16 percent, up 6 percentage points from a decade ago (Card and Krueger 2016).
    - Share of workers with “non-compete” clauses: 18 percent (Starr, Evan and Norman Bishara, 2016).
    - Decline in unionization and adoption of “Right to work” laws have weakened labor’s bargaining position.
  - Slowing labor reallocation and business dynamism:
    - Pace of job reallocation and job churning has declined (Haltiwanger and Davis 2014; Molloy et al. 2016).
    - Reduced mobility not readily explained by better matching, household structure changes, job locks, or declines in job-switching premia.
  - Rising global labor supply and offshoring: ambiguous effects via productivity and import penetration.

### Empirical strategy and data summary
- Data source: Current Population Survey (CPS) Merged Outgoing Rotation Group (MORG), IPUMS-CPS, January 2000–March 2015.
- Main variable: real hourly wage rate (reported hourly wage deflated by the consumer price index).
- Three empirical exercises:
  - Section B: Annual cross-section — links individual wage levels to county-level unemployment rates to capture compositional entry/exit effects on average wages.
  - Section C: MORG panel (12-month apart observations) — derives annual wage growth to test whether the slope of the Wage–Phillips curve for full-employed workers has flattened; state-level unemployment rates matched to individuals.
  - Section D: Job-to-job mobility analysis — measures whether an individual obtained new employment during the last four weeks conditional on having been fully employed during the last 12 months.
- Matching strategy:
  - County unemployment rates matched to individuals for level-of-wage (composition) analysis.
  - State unemployment rates matched to individuals for wage-growth/Phillips-curve analysis.

### Wage-level (composition) regression — key findings
- Model: augmented log-wage (Mincer-type) with county unemployment and change in county unemployment as parameters of interest; controls include education, experience, demographics, year, industry, and county dummies; sample 2000–15.
- Main descriptive result:
  - Across specifications, higher unemployment rates (U-rate county) are associated with lower hourly real wages.
  - Column 1: a decrease in the local unemployment rate by 1 percentage point—while controlling for job and worker specific characteristics—offsets the increase in the average local hourly wage rate by 0.9 percent.
  - Splitting pre-GFC and post-GFC shows compositional effect more pronounced after 2008: offsetting wage level effect ranges between 1.3-1.5 percent per one percentage point decline in the unemployment rate.
- Interpretation: larger exits and entries of lower-skill workers after the GFC produced a stronger compositional discount on average hourly wages.

### Wage-growth Phillips curve — key findings
- Model: annual percentage change of real hourly wage for continuously employed workers regressed on lagged state unemployment and controls; sample restricted to full-time, year-round single-job workers.
- Main empirical findings (2000–2014):
  - Baseline (column 1): an increase in the unemployment rate by 1 percent lowers annual hourly real wage growth by about 0.3 percent on average.
  - Sub-periods: negative relationship strong prior to the GFC (elasticity between 0.5 and 0.8) but vanishes after the GFC.
- Non-linear specification (linear spline breakpoint at 5 percent unemployment):
  - Pre-GFC: elasticity of -1.6 below 5 percent, implying a growth boost of 1½ percent for every one-percent decline of the unemployment rate below 5 percent.
  - Post-GFC: the elasticity effect below 5 percent is not present; the Phillips curve has flattened over the whole segment of unemployment rates.
  - Caveat: limited post-GFC observations with unemployment below 5 percent could bias results downward; flatness is robust to choice of higher pivot points.

### Job-to-job mobility and wage premia
- Job-to-job moves:
  - Measurement: self-reported change of employment within the last three months (full-time year-round single-job workers).
  - Trend: steady decline in job-to-job changing rate (annualized) for 2000–2014.
- Wage premium of job switchers:
  - Median wage growth of job switchers shows a moderate premium of 1 percent for the average job changer.
  - Premium larger pre-GFC, reversed during the GFC, and post-GFC around 0–1 percent.
- Shift-share decomposition (2000–2014):
  - Decomposes aggregate decline into population share shifts and changes in group-specific turnover rates.
  - Groups by skill (HS or less; Some college but < 4 years; College or higher) and by age (<35; 35–55; 55+).
  - Results:
    - Overall job-to-job change rate declined by some 9 percent between 2000 and 2015.
    - Decline in within-group propensities explains more than the aggregate decline.
    - Population shift has a small offsetting effect because share of younger (more mobile) workers has risen.
  - Changes in propensities:
    - Broad-based decline across all age and skill groups.
    - Decline among 35–55 particularly large for unskilled workers.
    - Mobility decline largest for the millennial cohort.

### Probit analysis of job-to-job moves (multivariate)
- Specification: probit with state unemployment, individual and job characteristics, and year dummies in some specifications.
- Key marginal-effect estimates (Table 3, column 1, 2000–14, Observations = 63,577):
  - U-rate (t-12) marginal effect = -0.0321*** (standard error 0.0079).
  - Years of potential work experience marginal effect = -0.0177*** (standard error 0.0047).
  - Union membership marginal effect = -0.0918*** (standard error 0.0355).
  - Urban dweller marginal effect = 0.1094*** (standard error 0.0395).
  - Male marginal effect = 0.0642** (standard error 0.0309).
- Quantification of the common time component:
  - Difference in year-dummy marginal effects between 2013/14 and 2001/02: estimated annual turnover rate change (≈0.003*12) explains about 4 percentage points of the 9 percentage point decline in the job-to-job transition rate.
  - Marginally higher unemployment in 2013/14 versus 2001/02 explains about 1.5 percentage points of the decline.
  - Remaining decline attributed to composition changes and lower propensity of younger workers to move jobs.

### Selected exact regression estimates and descriptive statistics
- Table 1 (Log of hourly real wage rate):
  - Observations = 29,562; R-squared = 0.3478.
  - ∆ U-rate (annual difference)-county coefficient (column 1) = 0.0086** (standard error 0.0035).
  - U-rate county (t-1) coefficient (column 1) = -0.0072*** (standard error 0.0027).
  - Years of schooling coefficient (column 1) = 0.0581*** (standard error 0.0010).
  - Male coefficient (column 1) = 0.1223*** (standard error 0.0050).
  - Union member coefficient (column 1) = 0.1655*** (standard error 0.0064).
  - Works in firm > 500 empl coefficient (column 1) = 0.0808*** (standard error 0.0048).
- Table 2 (Real Hour wage annual growth rate — Wage Growth Phillips curve of full-time employed: 2000-14):
  - Sample (column 1) 2000-14: Observations = 18,893; R-squared = 0.0184.
  - U-rate (t-12) coefficient (column 1) = -0.0029* (standard error 0.0016).
  - U-rate (t-12) coefficient (column 4, 2010-14) = -0.0162*** (standard error 0.0055).
  - Years of potential work experience coefficient (column 1) = -0.0027*** (standard error 0.0006).
  - Constant (column 1) = 0.1786*** (standard error 0.0375).
- Table 3 (Incidence of job-to-job transition — Probit analysis, full-time employed: 2000-14):
  - Sample (column 1) 2000-14: Observations = 63,577.
  - U-rate (t-12) marginal effect (column 1) = -0.0321*** (standard error 0.0079).
  - Years of potential work experience marginal effect (column 1) = -0.0177*** (standard error 0.0047).
  - Union membership marginal effect (column 1) = -0.0918*** (standard error 0.0355).
  - Urban dweller marginal effect (column 1) = 0.1094*** (standard error 0.0395).
  - Male marginal effect (column 1) = 0.0642** (standard error 0.0309).
- Descriptive statistics (selected):
  - Sample II (B) Obs = 80,756:
    - Mean Age = 40.8; Std. Dev. = 11.8; Min = 15; Max = 73.
    - Years of schooling mean = 13.0; Std. Dev. = 2.2; Min = 0; Max = 18.
    - Wage per hr (nominal) mean = 15.6; Std. Dev. = 8.2; Min = 3; Max = 88.
  - Sample II (C) Obs = 18,894:
    - Mean Age = 44.3; Std. Dev. = 11.1; Min = 17; Max = 73.
    - Years of schooling mean = 13.5; Std. Dev. = 2.0; Min = 0; Max = 18.
    - Wage per hr (nominal) mean = 19.0; Std. Dev. = 9.6; Min = 3; Max = 98.
    - Wage growth rate y/y real mean = 0.1; Std. Dev. = 0.6; Min = -1; Max = 29.
    - Unemployment rate (state) mean = 6.1; Std. Dev. = 2.1; Min = 2; Max = 15.
  - Sample II (D) Obs = 63,577:
    - Mean Age = 42.8; Std. Dev. = 11.4; Min = 16; Max = 73.
    - Years of schooling mean = 13.8; Std. Dev. = 2.1; Min = 0; Max = 18.
    - Job-to-job change rate (annualized) mean = 0.15; Std. Dev. = 1.31; Min = 0; Max = 1.

### Variable definitions (as used)
- Real wage: hourly wage rate of workers paid on an hourly basis discounted by the CPI.
- Real wage growth rate: annual growth rate of real wage.
- Years of schooling: age – years of schooling needed to achieve the reported education level – 6.
- Years of potential work experience: age – years of schooling.
- Job-to-job change indicator: Binary variable = 1 if respondent with full-time–full year employment reports a new employment within the last 4 weeks.
- Job changer (Figure 4): Binary variable = 1 if respondent was full-time–full employed worker with unchanged industry and occupation codes during last 12 months.
- Source: CPS MORG; Sources: Current Population Survey and BLS.

### Conclusions and implications
- Labor-market repair is not complete; return of workers to the labor force is dampening wage growth.
- The wage-growth-Phillips curve appears to have flattened: tightening since 2010 has not produced the measurable acceleration of wage growth seen pre-GFC.
- Declines in job-to-job moves have continued, most pronounced for low-skilled workers.
- Near-term implication: closing of employment gap could boost wage growth.
- Medium/long-term outlook uncertain because:
  - Flattening of the wage curve could indicate new jobs are of poorer quality and have low growth prospects.
  - Lower labor market dynamism could reflect better matching, but "there is little supportive evidence in the literature so far."
  - Lower turnover could create allocative inefficiencies leading to less productivity growth, though direct evidence of economic costs is scant.
- Research priority: identify determinants of falling labor market dynamism and linkages to wage growth and productivity using matched firm-employer data.

*IMF Working Paper WP/16/122 — What’s Up with U.S. Wage Growth and Job Mobility? (Section 1–3, Stephan Danninger, June 2016)*

### Section 1

### _wp16122 - Section 1

### Introduction and research questions
- Topic: Slow U.S. wage growth since the global financial crisis (GFC).
- Time coverage of microdata used: January 2000 to March 2015 (CPS MORG).
- Key research questions posed by the paper:
  - Is labor market repair still weighing on recent wage growth?
  - Has the relationship between labor market slack and wage growth permanently changed (has the Wage-Growth Phillips curve flattened)?
  - What is driving the decline in job-to-job mobility?

### Main empirical findings
- Aggregate and conditional wage trends:
  - Since 2010, annual hourly earnings in nominal terms have risen by 2 percent, about 1⅓ percent less than before the GFC.
  - Labor productivity averaged only ½ percent during 2013–15.
  - Real unit labor costs (labor income share) have declined steadily since the early 2000s.
- Cyclical (labor-market-repair) effects on average wages:
  - Labor market repair is still weighing on average wage growth.
  - Using regional variation in labor demand, larger post-GFC declines in local unemployment rates are associated with smaller increases in average wages.
  - Interpretation: re-employment of marginal workers at low entry wages moderates aggregate average wage growth.
  - Consistent external finding cited: 70–80 percent of new job holders post-GFC earn wages below median levels (Daly and Hobijn 2016).
- Structural changes affecting wages:
  - The wage-growth–Phillips curve has flattened after 2008 for full-time, full-year employed workers; wages do not comove with local unemployment rates post-2008, whereas they did pre-2008.
    - Pre-2008 evidence: the Phillips curve steepened below 5 percent unemployment.
    - Post-GFC (data up to 2014): no evidence of a similar kink.
  - Job-to-job transition rates (job-to-job changes associated with higher wage growth) have declined well before the GFC and continued thereafter.
    - Declines in job-to-job turnover occur across all education and age groups and are not explained by local labor market tightness or by demographics (e.g., aging, education composition).
    - Shift-share analysis: demographic changes cannot account for the sustained decline in job-to-job transitions.
  - These structural patterns raise concerns about the allocative ability of the labor market and the speed of labor reallocation, which can suppress productivity gains and wage growth.

### Drivers of low wage growth (as summarized)
- Cyclical channels:
  - Nominal wage rigidity: the share of workers with zero nominal wage growth has increased since the GFC.
  - Hysteresis/compositional effects: long spells of non-employment reduce employability and lead to re-employment at discounted entry wages, suppressing average hourly wage growth.
- Structural channels:
  - Changes in wage bargaining and work arrangements:
    - Share of workers in alternative arrangements: 16 percent, up 6 percentage points from a decade ago (Card and Krueger 2016).
    - Share of workers with “non-compete” clauses: 18 percent (Starr, Evan and Norman Bishara, 2016).
    - Decline in unionization and adoption of “Right to work” laws have weakened labor’s bargaining position.
  - Slowing labor reallocation and business dynamism:
    - Pace of job reallocation and job churning has declined (Haltiwanger and Davis 2014; Molloy et al. 2016).
    - Little consensus on causes; reduced mobility is not readily explained by better matching, household structure changes, job locks, or declines in job-switching premia.
  - Rising global labor supply and offshoring: potential ambiguous effects on domestic wage growth via productivity and import penetration.

### Empirical strategy and data summary
- Data source: Current Population Survey (CPS) Merged Outgoing Rotation Group (MORG), obtained from IPUMS-CPS, January 2000–March 2015.
- Main variable: real hourly wage rate (reported hourly wage deflated by the consumer price index).
- Three empirical exercises:
  - Section B: Annual cross-section of employed workers — links individual wage levels to county-level unemployment rates to capture compositional entry/exit effects on average wages.
  - Section C: MORG panel (12-month apart observations) — derives annual wage growth to test whether the slope of the Wage–Phillips curve for full-employed workers has flattened; matches state-level unemployment rates.
  - Section D: Job-to-job mobility analysis — measures whether an individual obtained new employment during the last four weeks conditional on having been fully employed during the last 12 months (bivariate job-to-job change rate).
- Matching strategy:
  - County unemployment rates matched to individuals for level-of-wage (composition) analysis.
  - State unemployment rates matched to individuals for wage-growth/Phillips-curve analysis.

### Implications for future wage growth and open questions
- Near term: As continued job growth reduces the remaining employment gap and the re-employment of low-wage workers abates, average wage growth is expected to accelerate.
- Medium/long term: A return to sustained high wage growth is uncertain because:
  - The post-GFC flattening of the wage-Phillips curve suggests broader structural changes.
  - Declining labor market churning could slow labor reallocation and productivity improvements, dampening wage growth.
- Open questions identified (beyond scope of this section):
  - Underlying causes of the widespread decline in job turnover remain unidentified.
  - Possible contributors mentioned: weakening bargaining power enabling firms to extract rents; new work arrangements limiting skill transferability; steady decline in firm entry rates affecting productivity.

*IMF Working Paper WP/16/122 — What’s Up with U.S. Wage Growth and Job Mobility? (Section 1), Stephan Danninger, June 2016*

### Section 2

### _wp16122 - Section 2

### Augmented log-wage regression (Mincer-type)
- Model estimated: (1) ln ωit,county = f(du t/t-1 county, u t-1, county, Xit t, county).
- ωit,county: hourly earnings in year t. u t-1, county: local unemployment rate. du t/t-1, county: annual percentage point change of the county unemployment rate. Xit: worker and job characteristics (education level, years of work experience, demographic characteristics). Specifications include year, industry, and county dummies. Sample period: 2000–15.
- Wage-rate data note: Usual weekly hours/earning questions are asked only at households in their 4th and 8th interview (outgoing interviews). New households enter each month; one fourth the households are in an outgoing rotation each month. Wage rates are based on respondent’s earnings per hour in his/her current job, for workers paid an hourly wage.
- Parameter of interest: coefficient on the change in the county unemployment rate interpreted as the compositional wage offset from a change in local demand (assumes labor demand changes affect local worker composition, accelerating entry/exit of workers with below median wage rates and producing a compositional discount on average wages).
- Main empirical findings (Table 1 summary):
  - Across specifications, higher unemployment rates (U-rate county) are associated with lower hourly real wages.
  - Column 1: a decrease in the local unemployment rate by 1 percentage point—while controlling for job and worker specific characteristics—offset the increase in the average local hourly wage rate as measured above by 0.9 percent.
  - Splitting the sample pre-GFC and post-GFC (columns 2–4) shows the compositional effect has been substantially more pronounced after 2008: the offsetting wage level effect ranges between 1.3-1.5 percent per one percentage point decline in the unemployment rate.
- Interpretation: exits and entries of workers with lower skills and productivity from/to employment were larger after the GFC, affecting average hourly wages more than in the past. The dampening effect is symmetric (declines in average wages are lower in counties with larger increases in unemployment rates). During the labor market recovery, counties with larger declines in unemployment rates saw comparatively lower average wage levels, possibly from low entry wages of unemployed and non-participants.

### A flattening of the wage-growth Phillips curve
- Objective: examine determinants of wage growth among continuously employed workers (about two-thirds of employed workers and responsible for over 90 percent of labor earnings); test whether cyclical response of wage growth has become weaker since the GFC.
- Model estimated: (2) Δωit,state = f(ut-1,state, Xi,t-1; dummies).
  - Δωi t,state: annual percentage change of the hourly wage rate deflated by CPI inflation in month t.
  - Sample restricted to full-time, year-round employed workers with single jobs. State-level unemployment rates matched to individuals. Controls: education, years of work experience, demographic factors, job-related information; occupation, industry, and state dummies included.
  - Compared to previous exercise, this sample is smaller, older, more educated, and has higher average hourly wages.
- Main empirical findings (Table 2 summary):
  - Baseline (column 1, period 2000–2014): an increase in the unemployment rate by 1 percent lowers the annual hourly real wage growth by about 0.3 percent on average.
  - Sub-period estimates (columns 2–4): negative relationship strong prior to the GFC (elasticity ranging between 0.5 and 0.8) but vanishes after the global financial crisis.
  - Robustness: results robust to different measures of the unemployment rate (adjusted for time varying natural unemployment rate) and lag specifications.
- Non-linear specification (linear spline with breakpoint at pre-2007 average unemployment rate of 5 percent):
  - Pre-GFC: wage growth elasticity substantially larger once unemployment slips below 5 percent; elasticity of -1.6, implying a growth boost of 1½ percent for every one-percent decline of the unemployment rate below 5 percent.
  - Post-GFC: the elasticity effect below 5 percent is not present; the Phillips curve has flattened over the whole segment of unemployment rates (Figure 4, red versus green lines).
  - Caveat: post-GFC unemployment rate has only for a short time been below the natural rate (which could now be below 5 percent), making empirical detection of a kink difficult. Footnote: "The results post-GFC could be biased downwards by the limited number of observation of unemployment rates below 5 percent. That said, the flatness of the post-GFC wage-growth Phillips curve is robust to the choice of higher Pivot points that is break points for the slope of the Phillips curve."
- Sensitivity: Several sensitivity checks performed.

### Declining job-to-job change rates
- Background: steady decline of job mobility and labor churning documented in literature (e.g., Haltiwanger 2015). Notable decline in frequency of job-to-job moves (transitions between employments without entering unemployment). Job-to-job moves linked to wage growth via better matches and reallocation across firms.
- Empirical evidence on wage premium of job switchers:
  - Using CPS earnings data for full-time employed, median wage growth of job switchers shows a moderate premium of 1 percent for the average job changer.
  - Premium was somewhat larger prior to the GFC, reversed during the global financial crisis, and post-GFC has been around 0-1 percent.
- Measurement of job-to-job move: self-reported change of employment within the last three months for full-time year-round employed workers with single jobs.
- Trend: Figure 5 replicates steady decline in job-to-job changing rate (annualized) for 2000–2014.
- Shift-share decomposition (equation 3) for 2000–2014:
  - Decomposes decline in aggregate job-to-job changing rate into (a) population share shifts (demographic effect) and (b) changes in group-specific job-to-job turnover rates (job-to-job change shift).
  - Groups: by skill (HS or less; Some college but < 4 years; College or higher) and by age (<35 "Millennials"; 35–55; 55+).
- Sample composition findings (Figures 6–7):
  - Millennials tend to be more educated relative to older cohorts (55+).
  - Due to population aging, share of older workers is declining (Figure 7 top panel).
  - Because older workers have been less skilled compared to younger cohorts, overall share of skilled workers is rising among full-time employed (Figure 7 bottom panel).
- Shift-share results (Figure 8 summary):
  - Overall job-to-job change rate declined by some 9 percent between 2000 and 2015.
  - The decline in job-to-job change rate across groups (i.e., lower propensities within groups) explains more than the entire aggregate decline.
  - Population shift (demographic effect) has a small offsetting effect because the share of younger workers (higher mobility) has risen, partially offsetting the aggregate decline.
- Changes in propensities across groups (Figure 9):
  - Broad-based decline in job-to-job transition rates across all age and skill groups.
  - Decline among middle age (35–55) particularly large for unskilled workers—hinting at less attractive employment opportunities for this group.
  - Mobility decline largest for the millennial cohort, which had the least attachment to the labor market and likely faced the lowest degree of job security.

### Shift-share and probit analysis of job-to-job moves (multivariate)
- Probit specification estimated: (4) jj i,t = f( u state, t-1, X i,t-1; year i ).
  - jj i,t is an indicator = 1 if individual moves from one to another full-time job during the last month, 0 otherwise.
  - Controls: state-level unemployment rate (to capture cyclical effects) and Xit set of individual and job-specific characteristics (education, years of work experience, etc.). Year dummies included in some specifications.
- Probit results (Table 3 summary):
  - Baseline model (column 1): younger and more educated workers churn more; union members change less often; city dwellers move more often between jobs.
  - Results similar in pre- and post-GFC subsamples (columns 2–3).
  - Model with year dummies (column 4) captures a growing negative common effect over time.
- Quantification of common component:
  - Using difference in marginal effects of year dummies between 2013/14 and 2001/02: estimated annual turnover rate change (≈0.003*12) explains about 4 percentage points of the 9 percentage point decline in the job-to-job transition rate.
  - Marginally higher unemployment rate in 2013/14 versus 2001/02 explains about 1.5 percentage points of the decline.
  - Remaining decline attributed to changes in composition of employed labor force and a lower propensity of younger workers to move jobs.
- Conclusion and open questions:
  - Identifying the underlying cause of the common component of the falling job turnover rate is beyond the paper’s scope.
  - Speculation exists about whether lower market fluidity links to lower wage growth or lower productivity; establishing such links requires matched firm-employer data and is recommended as a priority for future research.

*Source: _wp16122 - Section 2*

### Section 3

### _wp16122 - Section 3

### Conclusion — main findings
- The repair of the US labor market is not complete.
- The return of workers to the labor force is dampening wage growth.
- The wage-growth-Phillips curve appears to have flattened: the tightening of the labor market since 2010 has so far had not led to a measurable acceleration of wage growth, while it did so prior to the GFC.
- Job-to-job moves—a particular form of labor churning associated with higher wage growth—have fallen, with the decline most pronounced for low-skilled workers who are moving less from job-to-job than before the GFC.
- The decline in job turnover rates has continued.

### Implications and longer-term issues
- In the near term, the closing of the employment gap could boost wage growth.
- A return to persistently higher wage growth rates is uncertain.
- The flattening of the wage curve could indicate that new jobs are of poorer quality and have low growth prospects.
- Lower labor market dynamism could reflect better job-worker matching, but "there is little supportive evidence in the literature so far."
- Lower turnover could create allocative inefficiencies leading to less productivity growth down the road, although direct evidence of economic costs is scant.
- "Further research on the determinants of labor market dynamism is urgently needed to understand its longer-term implications."

### Key empirical estimates and regression results (selected exact figures)
- Table 1 (Log of hourly real wage rate): Observations = 29,562; R-squared = 0.3478.
  - ∆ U-rate (annual difference)-county coefficient (column 1) = 0.0086** (standard error 0.0035).
  - U-rate county (t-1) coefficient (column 1) = -0.0072*** (standard error 0.0027).
  - Years of schooling coefficient (column 1) = 0.0581*** (standard error 0.0010).
  - Male coefficient (column 1) = 0.1223*** (standard error 0.0050).
  - Union member coefficient (column 1) = 0.1655*** (standard error 0.0064).
  - Works in firm > 500 empl coefficient (column 1) = 0.0808*** (standard error 0.0048).

- Table 2 (Real Hour wage annual growth rate — Wage Growth Phillips curve of full-time employed: 2000-14):
  - Sample (column 1) 2000-14: Observations = 18,893; R-squared = 0.0184.
  - U-rate (t-12) coefficient (column 1) = -0.0029* (standard error 0.0016).
  - U-rate (t-12) coefficient (column 4, 2010-14) = -0.0162*** (standard error 0.0055).
  - Years of potential work experience coefficient (column 1) = -0.0027*** (standard error 0.0006).
  - Constant (column 1) = 0.1786*** (standard error 0.0375).

- Table 3 (Incidence of job-to-job transition — Probit analysis of Job-to-job change propensity, full-time employed: 2000-14):
  - Sample (column 1) 2000-14: Observations = 63,577.
  - U-rate (t-12) marginal effect (column 1) = -0.0321*** (standard error 0.0079).
  - Years of potential work experience marginal effect (column 1) = -0.0177*** (standard error 0.0047).
  - Union membership marginal effect (column 1) = -0.0918*** (standard error 0.0355).
  - Urban dweller marginal effect (column 1) = 0.1094*** (standard error 0.0395).
  - Male marginal effect (column 1) = 0.0642** (standard error 0.0309).
  - Marginal effects are sample average of individual marginal effects.

### Descriptive statistics (selected exact figures from CPS 2000-15)
- Sample section II (B) (Obs = 80,756):
  - Mean Age = 40.8; Std. Dev. = 11.8; Min = 15; Max = 73.
  - Male mean = 0.5.
  - Caucasian mean = 0.8.
  - Urban dweller mean = 0.8.
  - Married mean = 0.6.
  - Years of schooling mean = 13.0; Std. Dev. = 2.2; Min = 0; Max = 18.
  - Years of potential work experience mean = 21.8; Std. Dev. = 11.9; Min = 1; Max = 49.
  - Wage per hr (nominal) mean = 15.6; Std. Dev. = 8.2; Min = 3; Max = 88.

- Sample section II (C) (Obs = 18,894):
  - Mean Age = 44.3; Std. Dev. = 11.1; Min = 17; Max = 73.
  - Years of schooling mean = 13.5; Std. Dev. = 2.0; Min = 0; Max = 18.
  - Years of potential work experience mean = 24.7; Std. Dev. = 11.3; Min = 1; Max = 49.
  - Wage per hr (nominal) mean = 19.0; Std. Dev. = 9.6; Min = 3; Max = 98.
  - Wage growth rate y/y real mean = 0.1; Std. Dev. = 0.6; Min = -1; Max = 29.
  - Unemployment rate (state of residence) mean = 6.1; Std. Dev. = 2.1; Min = 2; Max = 15.

- Sample section II (D) (Obs = 63,577):
  - Mean Age = 42.8; Std. Dev. = 11.4; Min = 16; Max = 73.
  - Years of schooling mean = 13.8; Std. Dev. = 2.1; Min = 0; Max = 18.
  - Years of potential work experience mean = 23.0; Std. Dev. = 11.6; Min = 1; Max = 49.
  - Job-to-job change rate (annualized) mean = 0.15; Std. Dev. = 1.31; Min = 0; Max = 1.

### Variable definitions (as used in the analysis)
- Real wage: hourly wage rate of workers paid on an hourly basis discounted by the CPI.
- Real wage growth rate: annual growth rate of real wage.
- Years of schooling: age – years of schooling needed to achieve the reported education level – 6.
- Years of potential work experience: age – years of schooling.
- Job-to-job change indicator: Binary variable with value of 1 if respondent with full-time–full year employment reports a new employment within the last 4 weeks.
- Job changer (Figure 4): Binary variable with value of 1 if respondent was full-time–full employed worker with unchanged industry and occupation codes during last 12 months.
- Source: CPS MORG; Sources: Current Population Survey and BLS.

*Source: _wp16122 - Section 3 (IMF PDF)._

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16122.pdf_