## The Impact of Workforce Aging on European Productivity (WP/16/238)

## Source details

**Canonical URL:** [The Impact of Workforce Aging on European Productivity (WP/16/238)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16238.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16238.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16238.pdf.json)

---

### Workforce aging: scope, data and projections
- Paper identification: WP/16/238; Prepared by Shekhar Aiyar, Christian Ebeke and Xiaobo Shao; Authorized for distribution by Shekhar Aiyar; December 2016.
- Demographic trend: Europe’s population and workforce are aging due to declining fertility and increased life expectancy; immigration offsets this trend only partially.
- Workforce aging definition: shift in the composition of the workforce from relatively young to relatively old workers; the share of workers aged 55+ in the labor force is expected to increase substantially, especially in Spain, Italy, Portugal, Greece and Ireland.
- Sample and data: major EU28 countries over 1950 to 2014; workforce and population data from the OECD; output per worker data from the Penn World Table 9.0.
- Selected projected shares of workers aged 55-64 (Percent, share of workers aged 55-64 years to the labor force) — values by country for 2014 2020 2025 2030 2035:
  - Austria: 11.3 15.6 16.9 16.3 15.5
  - Belgium: 12.9 16.6 16.4 15.9 15.7
  - Czech Republic: 15.5 15.0 16.3 20.0 23.7
  - Denmark: 16.3 19.2 20.3 20.3 19.2
  - Estonia: 17.8 18.9 19.1 20.8 21.7
  - Finland: 18.2 18.9 18.3 16.7 16.8
  - France: 14.3 15.8 17.4 17.6 16.6
  - Germany: 18.2 22.0 23.7 21.7 20.1
  - Greece: 11.5 18.5 21.3 23.7 25.4
  - Hungary: 14.1 17.1 19.8 23.4 26.5
  - Ireland: 13.4 15.5 16.9 19.2 21.5
  - Italy: 14.8 19.9 23.6 25.8 25.8
  - Latvia: 16.8 18.1 18.5 19.3 19.0
  - Lithuania: 16.3 18.0 19.1 19.8 19.2
  - Luxembourg: 10.3 12.7 13.4 12.9 12.2
  - Netherlands: 16.2 18.5 20.2 19.6 17.8
  - Norway: 16.6 17.2 17.9 17.9 16.8
  - Poland: 14.5 14.2 14.4 16.7 20.4
  - Portugal: 14.9 18.4 20.4 22.5 24.1
  - Slovakia: 13.6 14.9 15.2 17.2 20.6
  - Slovenia: 11.4 17.2 18.9 19.3 20.4
  - Spain: 13.1 19.0 22.2 25.4 27.5
  - Sweden: 18.0 17.9 18.6 18.8 18.2
  - United Kingdom: 14.7 16.8 17.6 17.0 15.9
- Occupational composition (EU28, 2015): workforce concentrated into occupations where productivity decreases with age; age categories shown: 55-64, 25-54, 15-24 (percent of total workforce).

### Mechanisms: how aging affects productivity
- Two macroeconomic channels:
  - Higher dependency ratio (more retirees per worker) mechanically lowers per capita GDP and affects savings and public finances.
  - Workforce aging (changes in age composition of employed) affects average output per worker via age-productivity profiles.
- Age-productivity considerations and cited evidence:
  - Typical profile: productivity rises until workers’ 40s and declines toward end of working life.
  - Negative mechanisms: depreciation of knowledge, difficulty adapting to changing job requirements, lower innovation and technology adoption among older cohorts.
  - Evidence cited: Feyrer (2008) — median age of US innovators ~48; managers adopting new ideas ~40. Aksoy et al. (2015) — older workers (50-59) negatively impact patent applications. Jones (2010) — innovation positively affected by young and middle-aged cohorts and negatively by older cohorts.
  - Sectoral heterogeneity: occupations where productivity increases with age (managers, professionals); age-neutral occupations (clerical support, services and sales, electronic engineers); occupations where productivity decreases with age (technicians, craft workers, plant and machine operators, elementary occupations, armed forces).

### Empirical design and baseline estimators
- Baseline growth regression (equation form preserved):
  - Δlog YW_it = θ1 w55_it + θ2 YADR_it + θ3 OADR_it + u_i + η_t + ε_it
    - YW: real output per worker
    - w55: share of the total workforce aged 55-64 years
    - YADR and OADR: youth and old dependency ratios
    - u_i: country fixed effects; η_t: decadal time fixed effects
- Growth-accounting decomposition (Cobb-Douglas augmented with human capital):
  - y_it = k_it^α (h_it A_it)^(1−α)  re-written as y_it = (K/Y)_it^{α/(1−α)} A_it h_it
  - Log form: log(y_it) = α/(1−α) log(K/Y)_it + log(h_it) + log(A_it)
  - Alpha (capital share) assumed around 0.3.
- Estimation strategy:
  - Standard panel techniques; baseline follows Feyrer (2007) expanded to address heterogeneity and endogeneity.
  - Decadal time effects used to account for common shocks without proliferating parameters.
  - Decomposition follows Wong (2007): regress workforce aging on each RHS variable (capital accumulation, human capital, TFP) so coefficients sum to total effect from the baseline productivity regression.

### Key empirical findings and transmission channels
- Main result: an increase in the share of workers aged 55–64 is associated with an economically and statistically significant reduction in the growth rate of labor productivity.
- Channel decomposition:
  - The adverse effect of workforce aging on growth in output per worker comes mainly through lower TFP growth.
  - Physical capital accumulation plays a statistically significant role but is economically small relative to TFP effects; possible specialization into capital-intensive technologies to complement older workers.
- Quantitative magnitudes (past and projected impacts):
  - The aging of the workforce in the euro area has lowered TFP growth by about 0.1 percentage points each year over the past two decades.
  - Projected workforce aging could reduce TFP growth by an average of 0.2 percentage points every year over the next two decades (between 2014 and 2035 for the euro area).
  - Context: EC forecasts imply most countries expected to post average TFP growth rates of less than 1 percent every year over that horizon; thus the aging-induced reduction is economically substantial — "absent the adverse impact of aging, TFP growth could be higher by about a quarter over the next two decades."

### Identification, endogeneity concerns and instrumental strategies
- Endogeneity concern: the share of an age group in the labor force depends on both cohort size and participation rates; participation rates, especially among seniors, can be volatile and policy-sensitive; participation rates of older workers are much more responsive to productivity shocks than other groups (Tables A1 and A2).
- First instrumentation strategy:
  - Instrument country share of workforce aged 55–64 with the population share of those aged 45–54 ten years previously.
  - Instrument youth and old dependency ratios with the share of population under age 4 and the population share aged 55–59 ten years ago.
  - First-stage regressions (Table A3) show instruments are strongly correlated with endogenous demographic variables; first-stage F-statistics comfortably exceed the Staiger and Stock (1997) rule of thumb of 10.
- Second (preferred) instrumentation strategy:
  - Instrument workforce aging variable and dependency ratio with lagged birth rates 40, 30, 20, and 10 years ago (following Jaimovich and Siu (2009)).
  - Rationale: excluding migration and mortality, an age group's share of the 15–64 population is determined by births 15 to 64 years prior; fertility decisions at least fifteen years ago are plausibly exogenous to current productivity growth.
  - Drawback: significant reduction in number of observations due to long lags.
  - First-stage F-statistics for birth-rate instruments show a robust and strong correlation (Table A4).
- Instrumental variables estimates produce larger (more negative) and more precisely estimated effects than OLS (Table 2).

### Robustness checks and alternative specifications
- Controlling for hours worked:
  - Normalizing real output and TFP by hours worked (using OECD data) yields a negative and statistically significant effect of share of workers aged 55+ on both output per hour (D.lnYH) and modified TFP growth (D.lnAH).
- Controlling for full age distribution:
  - Replacing w55 with shares for 30–39, 40–49, 50–54, and 55–64 (omitting 15–29 as baseline) leaves the impact of the 55–64 group negative and statistically significant (Table 4).
  - Instrumenting multiple age cohorts with IV is challenging because of loss of degrees of freedom.
- Additional checks:
  - Similar results when including a lagged dependent variable, and when excluding dependency ratios or year effects.

### Quantifying past and projected impacts of aging on TFP growth
- Historical decomposition (using Table 2 column 4 bottom regression; period 1984–2007):
  - On average, workforce aging has reduced TFP by about 0.1 percentage points per annum to date.
  - Countries with larger losses (~0.2 percentage points per annum) include Latvia, Lithuania, Finland, Netherlands and Germany.
- Future projections (2014–2035):
  - Using OECD forecast of working-age population by age group and EC’s 2015 Aging Report for participation rates, projections of workforce share aged 55–64 are constructed for 2014–2035.
  - On average, workforce aging will shave off about 0.2 percentage points of TFP growth every year until 2035.
  - Context: projected average annual TFP growth in European is only about 1 percentage point per annum.
  - Interpretation: in the absence of workforce aging, TFP growth through 2035 could be about one quarter higher than the current forecast.
  - Countries expected to be worst affected (Greece, Spain, Portugal, Italy, Hungary, Slovenia, Slovakia and Ireland):
    - Average increase in share of old workers in total workforce is about 10 percentage points between 2020 and 2035.
    - Average loss ratio (ratio of projected TFP growth loss due to workforce aging over projected TFP growth) is about 0.6.
  - Many of the worst-affected countries currently face high debt burdens.

### Policy interactions: empirical design to identify mitigating policies
- Modified regression for interaction effects:
  - ΔlogA_it = (휃4 + 휃5 P_it−1) ∙ w55_it + β P_it−1 + 휃6 YADR_it + 휃7 OADR_it + u_i + η_t + ν_it    [2]
  - P denotes a conditioning policy factor; policy variables enter with a lag to reduce endogeneity.
  - Interpretation: P ameliorates the negative effect of aging if 휃4 < 0 and 휃5 > 0 (i.e., higher P reduces the marginal negative effect of w55 on TFP growth).
- Policy/conditioning factors tested (lagged where noted):
  - Health conditions and human capital accumulation:
    - Health care indicator: physician density in total population.
    - Active labor market policies (ALMPs): reform dummy = 1 when change in public spending per unemployed on ALMP training > 1 standard deviation of sample deviation. Data on ALMP spending from Eurostat.
  - Labor market flexibility:
    - Reform dummy = 1 when OECD indicator of employment protection of regular contracts declines by at least 1 standard deviation.
  - Tax wedge:
    - Reform dummy = 1 when OECD indicator of tax wedge declines by at least 1 standard deviation.
  - Innovation:
    - Lagged public and private sector spending on R&D (in GDP).

### Main empirical findings on policy mitigation (key coefficients and interpretation)
- Baseline and IV estimates for workforce aged 55-64 (selected coefficients and t-statistics):
  - Table 5 column (1): W55-64 = -1.338** (t-statistic -2.308).
  - Table 5 column (2): W55-64 = -0.590*** (t-statistic -3.610).
  - Table 5 column (3): W55-64 = -0.608*** (t-statistic -4.676).
  - Table 5 column (4): W55-64 = -0.198** (t-statistic -2.783).
  - Table 5 column (5): W55-64 = -0.656** (t-statistic -2.810).
  - Table 5 column (6): W55-64 = -0.340* (t-statistic -1.933).
- Instrumental-variable estimates (Table 2 IVs controlling for endogeneity) — larger negative estimates:
  - Workforce share aged 55-64 = -0.700*** (t-statistic -4.625) in the IV specification with births 10, 20, 30, and 40 years ago as instruments.
  - Other IV estimates include -0.823*** (t-statistic -4.003) and -0.453*** (t-statistic -4.353) in different columns.
- Hours-worked purging (Table 3 IV estimates):
  - Workforce aged 55-64 = -0.597*** (t-statistic -4.488) and -0.755*** (t-statistic -4.119) for specifications purging the effect of hours worked.
- Magnitude summary in text:
  - "An increase in the share of workers aged 55–64 by 1 percentage point leads to a decline in the growth of output per worker of between 0.25 and 0.7 percentage points."
  - "We estimate that a 1 percentage point increase in the 55–64 age cohort of the labor force is associated with a reduction in total factor productivity of about  45 ⁄ of a percentage point."
  - Extrapolation: "projected aging could reduce TFP growth by an average of 0.2 percentage points per annum over the next twenty years."
  - Countries with largest negative impact: Spain, Italy, Portugal, Greece and Ireland.

### Policy effects and statistical evidence (interaction coefficients from Table 5)
- Health inputs (lagged physician density to population):
  - Interaction: W5564*Lagged physician density to population = 0.342* (t-statistic 1.890) — suggests increased physician density dampens negative effect of aging.
  - Main effect: Lagged physician density to population = -0.0774** (t-statistic -2.610).
- Public sector R&D (lagged public sector spending on R&D (in GDP)):
  - Interaction: W5564*Lagged public sector spending on R&D (in GDP) = 1.566** (t-statistic 2.113) — associated with reduced effect of aging on TFP growth.
  - Main effect: Lagged public sector spending on R&D (in GDP) = -0.133* (t-statistic -1.868).
- Private sector R&D (lagged private sector spending on R&D (in GDP)):
  - Interaction: W5564*Lagged private sector spending on R&D (in GDP) = 0.0683 (t-statistic 0.481) — statistically unclear.
  - Main effect: Lagged private sector spending on R&D (in GDP) = -0.00744 (t-statistic -0.297).
- Active labor market policies on training (ALMP reform dummy):
  - Interaction: W5564*ALMP reform dummy = 0.351** (t-statistic 2.328) — suggests ALMP training dampens negative effect.
  - Main effect: ALMP reform dummy = -0.0242 (t-statistic -1.411).
- Labor market reform dummy (Reduction in EPLR):
  - Interaction: W5564*Labor market reform dummy = 0.159 (t-statistic 0.994) — expected sign but not statistically significant.
  - Main effect: Labor market reform dummy = -0.0106 (t-statistic -0.606).
- Tax wedge reform (Reduction in tax wedge):
  - Interaction: W5564*Tax wedge reform = 0.403** (t-statistic 2.301) — suggests reducing tax wedge mitigates negative aging effect.
  - Main effect: Tax wedge reform = -0.0389* (t-statistic -2.063).

### Interpretation and policy implications
- Policies empirically associated with dampening the negative effect of workforce aging on TFP growth:
  - Improving access to health services (physician density).
  - Expanding ALMP training.
  - Reducing the tax wedge.
  - Increasing public R&D spending.
- Private sector R&D’s moderating effect is statistically unclear in this sample; authors note possible explanation: "still very low levels of private sector R&D in several sample countries."
- Labor market flexibility reforms show the expected positive interaction sign but are not statistically significant in presented estimates.
- Many of these policy measures are desirable on their own and may increase productivity growth through multiple channels; they are likely to have disproportionately large impacts in rapidly aging societies such as Europe.

*Source: IMF Working Paper WP/16/238, "The Impact of Workforce Aging on European Productivity" (Sections 1–3).*

### Section 1

### The Impact of Workforce Aging on European Productivity (WP/16/238, Section 1)

### Workforce aging: scope and projections
- Paper identification: WP/16/238; Prepared by Shekhar Aiyar, Christian Ebeke and Xiaobo Shao; Authorized for distribution by Shekhar Aiyar; December 2016.
- Demographic trend: Europe’s population and workforce are aging due to declining fertility and increased life expectancy; immigration offsets this trend only partially.
- Workforce aging definition: shift in the composition of the workforce from relatively young to relatively old workers; the share of workers aged 55+ in the labor force is expected to increase substantially, especially in Spain, Italy, Portugal, Greece and Ireland.
- Sample and data: major EU28 countries over 1950 to 2014; workforce and population data from the OECD; output per worker data from the Penn World Table 9.0.
- Selected projected shares of workers aged 55-64 (Percent, share of workers aged 55-64 years to the labor force):
  - Austria: 11.3 15.6 16.9 16.3 15.5 (2014 2020 2025 2030 2035)
  - Belgium: 12.9 16.6 16.4 15.9 15.7
  - Czech Republic: 15.5 15.0 16.3 20.0 23.7
  - Denmark: 16.3 19.2 20.3 20.3 19.2
  - Estonia: 17.8 18.9 19.1 20.8 21.7
  - Finland: 18.2 18.9 18.3 16.7 16.8
  - France: 14.3 15.8 17.4 17.6 16.6
  - Germany: 18.2 22.0 23.7 21.7 20.1
  - Greece: 11.5 18.5 21.3 23.7 25.4
  - Hungary: 14.1 17.1 19.8 23.4 26.5
  - Ireland: 13.4 15.5 16.9 19.2 21.5
  - Italy: 14.8 19.9 23.6 25.8 25.8
  - Latvia: 16.8 18.1 18.5 19.3 19.0
  - Lithuania: 16.3 18.0 19.1 19.8 19.2
  - Luxembourg: 10.3 12.7 13.4 12.9 12.2
  - Netherlands: 16.2 18.5 20.2 19.6 17.8
  - Norway: 16.6 17.2 17.9 17.9 16.8
  - Poland: 14.5 14.2 14.4 16.7 20.4
  - Portugal: 14.9 18.4 20.4 22.5 24.1
  - Slovakia: 13.6 14.9 15.2 17.2 20.6
  - Slovenia: 11.4 17.2 18.9 19.3 20.4
  - Spain: 13.1 19.0 22.2 25.4 27.5
  - Sweden: 18.0 17.9 18.6 18.8 18.2
  - United Kingdom: 14.7 16.8 17.6 17.0 15.9
- Occupational composition (EU28, 2015): workforce concentrated into occupations where productivity decreases with age; age categories shown: 55-64, 25-54, 15-24 (percent of total workforce).

### Mechanisms: how aging affects productivity
- Two distinct macroeconomic channels:
  - Higher dependency ratio (more retirees per worker) mechanically lowers per capita GDP and affects savings and public finances.
  - Workforce aging (changes in age composition of employed) affects average output per worker due to age-productivity profiles.
- Age-productivity considerations:
  - Typical profile: productivity rises strongly until workers are in their 40s and declines toward end of working life.
  - Possible negative mechanisms: depreciation of knowledge, difficulty adapting to changing job requirements, lower innovation and technology adoption among older cohorts.
  - Evidence cited: Feyrer (2008) — median age of US innovators ~48; managers adopting new ideas ~40. Aksoy et al. (2015) — older workers (50-59) negatively impact patent applications. Jones (2010) — innovation positively affected by young and middle-aged cohorts and negatively by older cohorts.
  - Sectoral heterogeneity: occupations where productivity increases with age (managers, professionals), age-neutral occupations (clerical support, services and sales, electronic engineers), and occupations where productivity decreases with age (technicians, craft workers, plant and machine operators, elementary occupations, armed forces). Aggregate impact depends on industrial structure.

### Empirical design and baseline estimates
- Baseline growth regression (equation form preserved):
  - Δlog YW_it = θ1 w55_it + θ2 YADR_it + θ3 OADR_it + u_i + η_t + ε_it
    - YW: real output per worker
    - w55: share of the total workforce aged 55-64 years
    - YADR and OADR: youth and old dependency ratios
    - u_i: country fixed effects; η_t: decadal time fixed effects
- Growth-accounting decomposition (Cobb-Douglas augmented with human capital):
  - y_it = k_it^α (h_it A_it)^(1−α)  re-written as y_it = (K/Y)_it^{α/(1−α)} A_it h_it
  - Log form used: log(y_it) = α/(1−α) log(K/Y)_it + log(h_it) + log(A_it)
  - Alpha (capital share) assumed around 0.3.
- Estimation strategy:
  - Standard panel techniques; baseline follows Feyrer (2007) expanded to address heterogeneity and endogeneity.
  - Decadal time effects used to account for common shocks without proliferating parameters.
  - Decomposition follows Wong (2007): regress workforce aging on each RHS variable (capital accumulation, human capital, TFP) so coefficients sum to total effect from the baseline productivity regression.

### Key empirical findings (baseline OLS results and decomposition)
- Main result: an increase in the share of workers aged 55–64 is associated with an economically and statistically significant reduction in the growth rate of labor productivity.
- Channel decomposition:
  - The adverse effect of workforce aging on growth in output per worker comes mainly through lower TFP growth.
  - Physical capital accumulation plays a statistically significant role but is economically small relative to TFP effects.
  - Interpretation: possible specialization into capital-intensive technologies to complement older workers.
- Quantitative magnitude (past and projected impacts):
  - The aging of the workforce in the euro area has lowered TFP growth by about 0.1 percentage points each year over the past two decades.
  - Projected workforce aging could reduce TFP growth by an average of 0.2 percentage points every year over the next two decades (between 2014 and 2035 for the euro area).
  - Context: EC forecasts imply most countries expected to post average TFP growth rates of less than 1 percent every year over that horizon; thus the aging-induced reduction is economically substantial — "absent the adverse impact of aging, TFP growth could be higher by about a quarter over the next two decades."

### Identification and robustness considerations
- Endogeneity concern: the share of an age group in the labor force depends on both cohort size and participation rates; participation rates, especially among seniors, can be volatile and policy-sensitive.
- Country fixed effects and decadal time fixed effects included to absorb time-invariant country factors and common shocks.
- Authors report robustness checks: similar results obtained when including a lagged dependent variable, controlling for the entire age distribution, and excluding dependency ratios or year effects.

### Policy implications and mitigation strategies (overview)
- Policies that can ameliorate adverse effects of workforce aging include:
  - Improving health outcomes for older workers.
  - Active labor market policy (ALMP) reforms to boost the productivity of workers.
  - Reducing the tax wedge to increase labor mobility.
  - Increasing R&D spending to spur innovation.
- Section IV of the paper empirically examines the role of these policies in dampening aging’s adverse effects.

*Source: IMF Working Paper WP/16/238, "The Impact of Workforce Aging on European Productivity" (Section 1).*

### Section 2

### _wp16238 - Section 2

### Data, measurement, and baseline variables
- Output, and human capital data are from Penn World Table 9.0.
- Human capital is an index defined in terms of average years of schooling, with the returns to primary, secondary and tertiary education based on Mincer equation estimates from around the world taken from Psacharopoulos (1994).
- TFP is computed as a residual from the log of real output per worker minus the capital intensity weighted by the factor share expression, and minus the log of human capital per worker.
- Two variants of TFP are used: (i) TFP backed-out as a residual from equation [2], and (ii) Penn World Tables’ own estimates of TFP.

### Endogeneity concerns and instrumental strategies
- Endogeneity concern: labor force participation rates of age groups may respond to productivity shocks, producing reverse causality that biases estimates of the effect of the share of older workers (w55) on TFP growth. If the participation rate of the 55–64 group (w55) is more elastic to productivity shocks, naïve estimates of coefficient 휃1 will be biased upward.
- Empirical test: regress participation rates of different age groups on (lagged) aggregate productivity shocks; results show the participation rate of older workers (55–64) is much more responsive to productivity shocks than other groups (Tables A1 and A2).
- First instrumentation strategy:
  - Instrument country share of workforce aged 55–64 with the population share of those aged 45–54 ten years previously.
  - Instrument youth and old dependency ratios with the share of population under age 4 and the population share aged 55–59 ten years ago.
  - First-stage regressions (Table A3) show instruments are strongly correlated with endogenous demographic variables; first-stage F-statistics comfortably exceed the Staiger and Stock (1997) rule of thumb of 10.
- Second (preferred) instrumentation strategy:
  - Instrument workforce aging variable and dependency ratio with lagged birth rates 40, 30, 20, and 10 years ago (following Jaimovich and Siu (2009)).
  - Rationale: excluding migration and mortality, an age group's share of the 15–64 population is determined by births 15 to 64 years prior; fertility decisions at least fifteen years ago are plausibly exogenous to current productivity growth.
  - Drawback: significant reduction in number of observations due to long lags.
  - First-stage F-statistics for birth-rate instruments show a robust and strong correlation (Table A4).

### Main empirical findings and effect sizes
- Instrumental variables estimates produce larger (more negative) and more precisely estimated effects than OLS (Table 2).
- Effect magnitude:
  - "An increase in the share of workers aged 55–64 by 1 percentage point leads to a decline in the growth of output per worker of between 0.25 and 0.7 percentage points."
- Transmission channel:
  - The bulk of the negative effect of workforce aging on labor productivity operates through its negative impact on TFP growth.
  - This dominant TFP channel is broadly similar to the findings in Feyrer (2007) and Werding (2008).
- Instrument diagnostics:
  - First-stage regressions are consistent with priors on sign and strength; diagnostic statistics (F-statistic and Shea R2) comfortably exceed conventional thresholds indicating strong instruments.

### Robustness checks
- Controlling for hours worked:
  - Normalizing real output and TFP by hours worked (using OECD data) yields a negative and statistically significant effect of share of workers aged 55+ on both output per hour (D.lnYH) and modified TFP growth (D.lnAH).
- Controlling for full age distribution:
  - Replacing w55 with shares for 30–39, 40–49, 50–54, and 55–64 (omitting 15–29 as baseline) leaves the impact of the 55–64 group negative and statistically significant (Table 4).
  - Note: instrumenting multiple age cohorts with IV as in Table 2 is challenging because of loss of degrees of freedom.

### Quantifying past and projected impacts of aging on TFP growth
- Historical decomposition (using point estimates from Table 2 column 4 bottom regression; period 1984–2007):
  - On average, workforce aging has reduced TFP by about 0.1 percentage points per annum to date.
  - Countries with larger losses (~0.2 percentage points per annum) include Latvia, Lithuania, Finland, Netherlands and Germany.
- Future projections (2014–2035):
  - Using OECD forecast of working-age population by age group and EC’s 2015 Aging Report for participation rates, projections of workforce share aged 55–64 are constructed for 2014–2035.
  - On average, workforce aging will shave off about 0.2 percentage points of TFP growth every year until 2035.
  - Context: projected average annual TFP growth in European is only about 1 percentage point per annum.
  - Interpretation: in the absence of workforce aging, TFP growth through 2035 could be about one quarter higher than the current forecast.
  - Countries expected to be worst affected (Greece, Spain, Portugal, Italy, Hungary, Slovenia, Slovakia and Ireland):
    - Average increase in share of old workers in total workforce is about 10 percentage points between 2020 and 2035.
    - Average loss ratio (ratio of projected TFP growth loss due to workforce aging over projected TFP growth) is about 0.6.
  - Many of the worst-affected countries currently face high debt burdens.

### Policy interactions, empirical design to identify mitigating policies
- Modified regression specification for interaction effects:
  - ΔlogA_it = (휃4 + 휃5 P_it−1) ∙ w55_it + β P_it−1 + 휃6 YADR_it + 휃7 OADR_it + u_i + η_t + ν_it    [2]
  - P denotes a conditioning policy factor; policy variables enter with a lag to reduce endogeneity.
  - Interpretation: P ameliorates the negative effect of aging if 휃4 < 0 and 휃5 > 0 (i.e., higher P reduces the marginal negative effect of w55 on TFP growth).
- Policy/conditioning factors tested:
  - Health conditions and human capital accumulation:
    - Health care indicator: physician density in total population (physician availability expected to be particularly relevant for older workers).
    - Active labor market policies (ALMPs): reform dummy = 1 when change in public spending per unemployed on ALMP training > 1 standard deviation of sample deviation. Data on ALMP spending from Eurostat.
  - Labor market flexibility:
    - Proxy: reform dummy = 1 when OECD indicator of employment protection of regular contracts declines by at least 1 standard deviation of the sample.
    - Rationale: aging may reduce voluntary mobility; lower mobility can reduce productivity through impaired reallocation.
  - Tax wedge:
    - Reform dummy = 1 when OECD indicator of tax wedge declines by at least 1 standard deviation.
    - Rationale: high tax on marginal employment and out-of-work benefits can disproportionately discourage senior workers; tax wedge reforms could incentivize delayed retirement or participation.
  - Innovation:
    - Technological innovation and adoption may mitigate negative impact of aging if they differentially benefit senior workers.

*Source: _wp16238 - Section 2*

### Section 3

### _wp16238 - Section 3

### Empirical approach and policy interactions
- The analysis tests whether the effect of aging on TFP growth is dampened by higher spending on R&D, differentiating between public and private spending on R&D as a percentage of GDP.
- Policy variables examined include access to health services (lagged physician density to population), active labor market policies focusing on training (ALMP reform dummy), labor market flexibility (labor market reform dummy — reduction in EPLR), tax wedge reform (reduction in tax wedge), lagged public sector spending on R&D (in GDP), and lagged private sector spending on R&D (in GDP).
- Estimation framework: panel regressions fitting each dependent variable on workforce aging ratio, old-age dependency ratio, youth dependency ratio, country and time-fixed effects. Sample includes all European OECD member countries. Period covers 1950 through 2014.

### Main empirical findings
- Baseline and overall aging effect
  - Workforce aged 55-64 (W55-64) share: point estimates indicate a negative association with TFP growth across specifications (see Table 5 and earlier tables). Example coefficients:
    - Table 5 column (1): Workforce aged 55-64 (W55-64) share = -1.338** (t-statistic -2.308).
    - Table 5 column (2): W55-64 = -0.590*** (t-statistic -3.610).
    - Table 5 column (3): W55-64 = -0.608*** (t-statistic -4.676).
    - Table 5 column (4): W55-64 = -0.198** (t-statistic -2.783).
    - Table 5 column (5): W55-64 = -0.656** (t-statistic -2.810).
    - Table 5 column (6): W55-64 = -0.340* (t-statistic -1.933).
- Instrumental-variable estimates (controlling for endogeneity) show larger negative point estimates:
  - Table 2: Workforce share aged 55-64 = -0.700*** (t-statistic -4.625) in the IV specification with births 10, 20, 30, and 40 years ago as instruments; other related estimates include -0.823*** (t-statistic -4.003) and -0.453*** (t-statistic -4.353) in different columns.
- Hours-worked purging results:
  - Table 3 IV estimates: Workforce aged 55-64 = -0.597*** (t-statistic -4.488) and -0.755*** (t-statistic -4.119) for specifications purging the effect of hours worked.
- Magnitude summary noted in text:
  - "We estimate that a 1 percentage point increase in the 55–64 age cohort of the labor force is associated with a reduction in total factor productivity of about  45 ⁄ of a percentage point."
  - Extrapolation: "projected aging could reduce TFP growth by an average of 0.2 percentage points per annum over the next twenty years."
  - Countries with largest negative impact: Spain, Italy, Portugal, Greece and Ireland.

### Policy effects and mitigation (key coefficients from Table 5)
- Health inputs (lagged physician density to population)
  - W5564*Lagged physician density to population = 0.342* (t-statistic 1.890) — interaction term suggests increased physician density dampens the negative effect of aging.
  - Lagged physician density to population = -0.0774** (t-statistic -2.610).
- Public sector R&D (lagged public sector spending on R&D (in GDP))
  - W5564*Lagged public sector spending on R&D (in GDP) = 1.566** (t-statistic 2.113) — robustly associated with a reduced effect of aging on TFP growth.
  - Lagged public sector spending on R&D (in GDP) = -0.133* (t-statistic -1.868).
- Private sector R&D (lagged private sector spending on R&D (in GDP))
  - W5564*Lagged private sector spending on R&D (in GDP) = 0.0683 (t-statistic 0.481) — effect remains statistically unclear.
  - Lagged private sector spending on R&D (in GDP) = -0.00744 (t-statistic -0.297).
- Active labor market policies on training (ALMP reform dummy)
  - W5564*ALMP reform dummy (Increase in ALMP on training) = 0.351** (t-statistic 2.328) — interaction term indicates ALMP on training dampens the negative aging effect.
  - ALMP reform dummy = -0.0242 (t-statistic -1.411).
- Labor market reform dummy (Reduction in EPLR)
  - W5564*Labor market reform dummy (Reduction in EPLR) = 0.159 (t-statistic 0.994) — expected sign but not statistically significant.
  - Labor market reform dummy = -0.0106 (t-statistic -0.606).
- Tax wedge reform (Reduction in tax wedge)
  - W5564*Tax wedge reform (Reduction in tax wedge) = 0.403** (t-statistic 2.301) — interaction term suggests reducing the tax wedge mitigates aging’s negative effect.
  - Tax wedge reform = -0.0389* (t-statistic -2.063).

### Interpretation and policy implications
- Policies that improve health services availability (physician density), expand ALMP training, lower the tax wedge, and increase public R&D spending are empirically associated with a dampening of the negative effect of workforce aging on TFP growth.
- Private sector R&D’s moderating effect is statistically unclear in this sample, possibly due to "still very low levels of private sector R&D in several sample countries."
- Labor market reforms increasing flexibility (reducing protection of regular workers) show the expected positive interaction sign but are not statistically significant in the estimates presented.
- The analysis highlights that many of these policy measures are desirable on their own and may increase productivity growth through multiple channels; they are likely to have disproportionately large impacts in rapidly aging societies such as Europe.

### Additional empirical notes
- Tables covering robustness and alternative specifications:
  - Table 1: OLS estimates of effects of aging on output per worker and TFP growth; Workforce share aged 55-64 shows negative and statistically significant coefficients in several columns (e.g., -0.165** and -0.228** and -0.149***).
  - Table 2: IV estimates controlling for endogeneity using lagged population proportions and births at various lags as instruments; first-stage F-statistics reported in Appendix tables (e.g., F-stat first stage 45.51, 47.23, 174.96 for some instruments).
  - Table 4: Controlling for entire age distribution, Workforce aged 55-64 share = -0.237** (t-statistic -2.373) and other age-group coefficients shown.
  - Appendix tables A1–A4 report sensitivity of labor force participation rates by age group to productivity growth and instrumentation equations (e.g., Table A4: F-stat first stage 39.68 and 73.56).

*Source: IMF staff estimates, _wp16238 - Section 3*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16238.pdf_