## wp18207

## Source details

**Canonical URL:** [wp18207](https://www.imf.org/-/media/files/publications/wp/2018/wp18207.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18207.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18207.pdf.json)

---

### Introduction and core mechanisms
- Technological progress raises overall productivity and income but can create structural change that displaces and transforms jobs.
- Two specific channels emphasized:
  - Automation: machines and AI performing an increasing range of tasks previously done by humans.
  - Falling capital goods prices: cheaper capital relative to labor encourages substitution of machines for routine tasks.
- Policy role:
  - Reforms to lift growth are critical.
  - Policies to spread gains from growth (e.g., investment in human capital, redistributive measures) can boost inclusive outcomes and may increase growth if well designed.

### Key empirical facts
- Innovation fosters growth: historical technologies (steam engine, railways, electricity, combustion engine, ICT) underpinned productivity and per-capita GDP growth.
- No evidence of persistent technological unemployment over long horizons; productivity rises supported job creation and higher real wages historically.
- Adjustment triggers sectoral reallocations: employment shifted agriculture → manufacturing → services.
- Gains from reallocations uneven: routine-task-exposed jobs declined; middle-skilled “hollowing out”; rising educational wage premia in some countries.
- Technological advances can affect labor share of income: declines in labor shares occurred in periods of fast technological change, linked to capital-intensive production and falling capital goods prices.
- Technology and global integration are intertwined: cross-country adoption lags shortened (from almost 100 years in the 1800s to 20 years today).

### Quantitative statements on automation and capital prices
- McKinsey (2017) baseline: 375 million workers globally (14 percent of the global workforce) may be at risk of job losses by 2030.
- U.S.: about 23 percent of jobs potentially affected in the baseline; up to 44 percent in a fast-automation scenario (consistent with Frey and Osborne (2017) estimate of 47 percent).
- EMDE examples: about 13 percent for Mexico, 9 percent for India, 16 percent for China.
- Alternative estimates:
  - Arntz et al. (2016): 9 percent of jobs in OECD countries automatable (task-based).
  - Ahmed and Chen (2017): 1 percent automatability for Vietnam (using Arntz et al. methodology).
- Automatability declines with education level; high-skilled occupations least susceptible.
- Falling real price of capital goods observed for AEs and some EMs; ICT-driven productivity gains may further reduce capital goods prices.

### Model description and calibration
- DSGE model (Lizarazo, Peralta-Alva and Puy (2017)) features:
  - Three goods: manufacturing (M, tradable), low-skill services (L, non-tradable), high-skill services (S, non-tradable).
  - Heterogeneous agents by skill: low-skilled (u), medium-skilled (m), high-skilled (h); skill distribution exogenous and constant.
  - Capital substitutes with low- and medium-skilled labor (elasticity of substitution between capital and aggregate of middle- and low-skilled labor set to 1.5); capital complements high-skilled labor.
  - Non-homothetic preferences over goods (rising share of services as income grows), calibrated to U.S. benchmark.
- Quantitative analysis compares steady states (abstracts from transitional dynamics).

### Steady-state comparative statics: two technology shocks
- Shock (i): 20 percent drop in relative price of capital goods (broadly consistent with U.S. pace since the 1980s).
  - Long-term GDP rises by 14½ percent relative to baseline.
  - Income effects: high-skilled +16 percent; low-skilled ≈ +7 percent; middle-skilled ≈ +7 percent.
- Shock (ii): Increase in elasticity of substitution between capital and labor (automation) calibrated to produce same 14½ percent GDP rise.
  - Requires a 10 percent increase in the elasticity of substitution between capital and labor.
  - Generates more inequality than the fall in capital goods prices: almost no benefit to low-skilled workers; high-skilled consumption increases by almost 20 percent.
- General result: both shocks raise aggregate income but disproportionately benefit high-skilled workers; higher substitutability amplifies inequality.

### Policy responses simulated and fiscal magnitudes
- Two policy responses considered to spread GDP gains more evenly:
  - A. Higher education spending (human-capital investment)
    - Simulation: reduce share of low-skilled workers by 4 percentage points: 2 percentage points → medium-skilled; 2 percentage points → high-skilled.
    - Cost estimate (based on U.S. education costs): range from 1 to 3 percent of U.S. GDP; mid-point used = 2 percent of GDP.
    - Financing implication: requires an increase of 2.5 percentage points in the VAT rate relative to a no-education-policy-response baseline (when financed by VAT).
  - B. Differentiated income tax cuts
    - Assumption: reduce effective income tax rates by 6 percentage points for households near the median; cuts decline rapidly with income (2 percentage points for households at three times the median; no cut for incomes above 4 times the median).
    - Revenue cost equivalent to about 2 percent of GDP (equivalently a 2.5 percentage point increase in VAT rate).
- Policy impacts (steady-state results):
  - Financing by cuts in unproductive government consumption:
    - Human capital policy: boosts GDP gains from 14½ percent to 20½–21 percent.
    - Tax cut: GDP gains about 16 percent.
  - Financing by higher VAT (efficiency cost reduces gains):
    - Human capital policy: GDP gains ≈ 16 percent.
    - Tax cuts: GDP gains ≈ 12–12½ percent.
  - Distributional impacts (when human capital policy financed by higher VAT):
    - Low-skilled income gain: between 25 and 31 percent (depending on whether shock is cheaper capital or more substitutability).
    - High-skilled income gain: between 9 and 12 percent.
    - Middle-skilled income gain: 2 to 3 percent (would be neutral if financed by cuts in unproductive spending).
- Comparative assessment:
  - Education spending is more efficient: raises low-skilled incomes and aggregate GDP by increasing aggregate human capital and making low-skilled labor scarcer.
  - Tax cuts redistribute gains ex-post and can equalize incomes more readily but are less efficient for GDP and do not address low human capital.
  - Combining education spending and tax policy can be complementary: education addresses supply-side skill deficiencies; tax policy provides insurance against uncertainty.

### Limitations and interpretation
- Model is relatively simple and illustrates only two technology paths.
- Analysis uses steady-state comparisons and abstracts from transitional dynamics and associated adjustment costs.
- Results indicative; interpret with caution.

### Appendix — calibration and skills parameters
- Consumption and sectoral calibration:
  - Consumption parameter ܿ chosen so income elasticity demand for services increases with income; top quintile consumption of manufacturing and imported goods share is approx. eight percentage points larger than bottom quintile.
  - Low skilled services production: 30 percent of total GDP.
  - High skilled services production: 50 percent of total GDP.
  - Share of imports in total consumption expenditure: approx. 10 percent.
- Household labor productivity proxies:
  - Skill classification: high school degree or less = low skill; more than bachelor’s degree = high skill.
  - Average labor productivity of medium skill normalized: ߟ௫ = 1.
  - Differences in average years of education used as proxies for skill differences.
- Table 1. Skills Parameters:
  - Parameter ߤ௟ = 0.39 — Target: Population share with high school degree or less.
  - Parameter ߤ௫ = 0.48 — Target: Population with some college, no degree, associate degree or bachelor degree.
  - Parameter ߤ௛ = 0.13 — Target: Population share with a Bachelor’s degree or more.
  - Parameter ߟ௟ = 0.7 — Target: Approx. 12 years of education.
  - Parameter ߟ௫ = 1 — Target: Approx. 16 years of education.
  - Parameter ߟ௛ = 1.1 — Target: More than 16 years of education.
- Government calibration:
  - Consumption (and intermediate consumption) taxes set at 7.5%.
  - Benchmark government consumption chosen to be 16% of GDP (average for 2009–15 per World Bank data).

*Source: wp18207 - IMF Working Paper (chapter excerpt).*

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

### References

### Figures listed in this content unit
- 1. Technological Innovation Has Underpinned a Rise in Living Standards
- 2. Productivity Associated with More Employment...
- 3. Productivity Associated with Higher Wages
- 4. Technological Change Has Contributed to Sectoral Reallocation
- 5. Earnings Polarization Accompanied by...
- 6. Increasing Educational Wage Premia in Some Economies
- 7. Estimated Trends in Labor Shares by Country, 1991–2014
- 8. Technology Adoption Lags Have Decreased Over Time
- 9. Elasticity of Substitution Correlated with Degree of Task Routinization
- 10. Technological Change and Global Value Chains Contributed to Integration
- 11. Relative Price of Investment
- 12. Cheaper Capital Has Less Distributional Impact Than More Substitutability
- 13. Spending on Education Boosts Income Gains; Gains After Tax Cut Are Smaller but Sizeable
- 14. Human Capital Investment Disproportionally Benefits Low-Skilled Workers
- 15. Taxes Redistribute Income Gains But Not Equally

### Other content elements indicated
- Table

*Source: wp18207 - References (PDF chapter/section).*

### 1. Skills Parameters ...................................................................................................

### 1. Skills Parameters

### I. Introduction
- Technological progress raises overall productivity and income but can create structural change that displaces and transforms jobs, with major repercussions for parts of the population.
- Two specific channels emphasized:
  - Automation: machines and AI performing an increasing range of tasks previously done by humans.
  - Falling capital goods prices: cheaper capital relative to labor encourages substitution of machines for routine tasks.
- Policy role:
  - Reforms to lift growth are critical.
  - Policies to spread gains from growth (e.g., investment in human capital, redistributive measures) can boost inclusive outcomes and may increase growth if well designed.

### II. Literature review (selected points)
- Historical concerns about technological unemployment have recurred (e.g., Keynes (1930)); modern waves include the ICT revolution and AI advances.
- Innovation has been a key driver of productivity growth and per-capita GDP growth.
- Declines in labor shares in many Advanced Economies (AEs) between the 1980s and late 2000s have been associated with more capital-intensive production and falling prices of capital goods (IMF, 2017a); other factors include globalization and rising market power.
- Diffusion and implementation lags mean rapid technological advances may not immediately raise measured productivity (Brynjolfsson et al., 2017; Bloom et al., 2017).
- Cross-country adoption lags have shortened (Comin and Hobijn, 2010).

### III. Facts (key empirical findings from the chapter)
- Fact #1: Innovation fosters growth.
  - Technological innovations (steam engine, railways, electricity, combustion engine, ICT) have underpinned strong productivity and per-capita GDP growth.
- Fact #2: Historical fears of persistent technological unemployment have been unwarranted.
  - No evidence of persistent negative impact on overall demand for labor; productivity rises supported job creation and higher real wages over long horizons.
- Fact #3: Adjustment triggers sectoral reallocations.
  - Employment shifted from agriculture → manufacturing → services over time (U.S. example and other AEs/EMDEs).
- Fact #4: Gains from sectoral reallocations have been uneven, contributing to income polarization.
  - Routine-task-exposed jobs declined; middle-skilled “hollowing out” observed in many AEs; rising educational wage premia in some countries.
- Fact #5: Technological advances can affect labor share of income.
  - Declines in labor shares occurred in periods of fast technological change; linked to capital-intensive production and falling capital goods prices.
- Fact #6: Technological change and global integration are intertwined.
  - Technology fosters economic and financial integration, reducing cross-country adoption lags (from almost 100 years in the 1800s to 20 years today).

### IV. Two possible drivers of labor income shares and polarization
- Two interrelated drivers:
  - (A) Automation: increase in substitutability between capital and labor, making routine cognitive and manual tasks replaceable by machines/AI.
  - (B) Falling prices of capital goods: cheaper capital encourages substitution away from labor for a given elasticity of substitution.
- Key mechanisms:
  - Sectors with higher routine task intensity exhibit larger elasticity of substitution between capital and labor and therefore greater job replacement risk when capital becomes cheaper or automation advances.
  - IMF (2017a) finds this mechanism important for labor share declines in AEs.

### IV.A Automation (selected quantitative statements)
- McKinsey (2017) baseline: 375 million workers globally (14 percent of the global workforce) may be at risk of job losses by 2030.
  - U.S.: about 23 percent of jobs potentially affected in the baseline, up to 44 percent in a fast-automation scenario (consistent with Frey and Osborne (2017) estimate of 47 percent).
  - EMDE examples: about 13 percent for Mexico, 9 percent for India, 16 percent for China.
- Alternative estimates:
  - Arntz et al. (2016): 9 percent of jobs in OECD countries automatable (task-based methodology).
  - Ahmed and Chen (2017): 1 percent automatability for Vietnam (using Arntz et al. methodology).
- Automatability declines with education level; consensus that high-skilled occupations are least susceptible to automation.

### IV.B Falling prices of capital goods
- Downward trend in the real price of capital goods observed for AEs and some EMs; ICT-driven productivity gains may further reduce capital goods prices.
- For a given substitutability, cheaper capital replicates many effects of automation: more capital, fewer workers, output up, and distributional effects favoring higher-skilled labor.

### V. Model (description and calibration)
- Model: DSGE model developed by Lizarazo, Peralta-Alva and Puy (2017).
- Key features:
  - Three goods: manufacturing (M, tradable), low-skill services (L, non-tradable), high-skill services (S, non-tradable).
  - Heterogeneous agents by skill: low-skilled (u), medium-skilled (m), high-skilled (h); skill distribution exogenous and constant.
  - Capital substitutes with low- and medium-skilled labor (elasticity of substitution between capital and aggregate of middle- and low-skilled labor set to 1.5); capital complements high-skilled labor.
  - Non-homothetic preferences over goods (rising share of services as income grows), calibrated to U.S. as benchmark.
- Quantitative analysis compares steady states (abstracts from transitional dynamics).

### VI. Model simulations (steady-state comparative statics)
- Two “technology shocks” modeled:
  - (i) 20 percent drop in relative price of capital goods (broadly consistent with U.S. pace since the 1980s).
    - Long-term GDP rises by 14½ percent relative to baseline.
    - Income effects: high-skilled +16 percent; low-skilled ≈ +7 percent; middle-skilled ≈ +7 percent.
  - (ii) Increase in elasticity of substitution between capital and labor (automation) calibrated to produce the same 14½ percent GDP rise.
    - Requires a 10 percent increase in the elasticity of substitution between capital and labor.
    - Generates more inequality than the fall in capital goods prices: almost no benefit to low-skilled workers; high-skilled consumption increases by almost 20 percent.
- General result: both shocks raise aggregate income but disproportionately benefit high-skilled workers; higher substitutability amplifies inequality.

### VII. Policy implications (simulated policy responses and outcomes)
- Two policy responses considered to spread GDP gains more evenly:
  - A. Higher education spending (human-capital investment)
    - Simulation: reduce share of low-skilled workers by 4 percentage points: 2 percentage points → medium-skilled; 2 percentage points → high-skilled.
    - Cost estimate (based on U.S. education costs): range from 1 to 3 percent of U.S. GDP; mid-point used = 2 percent of GDP.
    - Financing implication: requires an increase of 2.5 percentage points in the VAT rate relative to a no-education-policy-response baseline (when financed by VAT).
  - B. Differentiated income tax cuts
    - Assumption: reduce effective income tax rates by 6 percentage points for households near the median; cuts decline rapidly with income (2 percentage points for households at three times the median; no cut for incomes above 4 times the median).
    - Revenue cost equivalent to about 2 percent of GDP (equivalently a 2.5 percentage point increase in VAT rate).
- Policy impacts (steady-state results):
  - Financing by cuts in unproductive government consumption:
    - Human capital policy: boosts GDP gains from 14½ percent to 20½–21 percent (combined impact).
    - Tax cut: GDP gains about 16 percent.
  - Financing by higher VAT (efficiency cost reduces gains):
    - Human capital policy: GDP gains ≈ 16 percent.
    - Tax cuts: GDP gains ≈ 12–12½ percent.
  - Distributional impacts (when human capital policy financed by higher VAT):
    - Low-skilled income gain: between 25 and 31 percent (depending on whether shock is cheaper capital or more substitutability).
    - High-skilled income gain: between 9 and 12 percent.
    - Middle-skilled income gain: 2 to 3 percent (reflects VAT incidence; would be neutral if financed by cuts in unproductive spending).
- Comparative assessment:
  - Education spending is more efficient: raises low-skilled incomes and aggregate GDP by increasing aggregate human capital and making low-skilled labor scarcer.
  - Tax cuts redistribute gains ex-post and can equalize incomes more readily but are less efficient for GDP and do not address the root problem of low human capital.
  - Combination of education spending and tax policy can be complementary: education addresses supply-side skill deficiencies; tax policy provides insurance against policy/skill-targeting uncertainty.
- Policy cost magnitudes emphasized:
  - Education policy simulated cost: 2 percent of GDP (mid-point); VAT increase needed: 2.5 percentage points when VAT-financed.
  - Tax-cut package revenue cost: ≈ 2 percent of GDP (equivalent VAT increase: 2.5 percentage points).

### VII.D Limitations
- Model is relatively simple and illustrates only two technology paths.
- Analysis uses steady-state comparisons and abstracts from transitional dynamics and associated adjustment costs.
- Results indicative and should be interpreted with caution.

### VIII. Conclusion (key takeaways)
- Technology historically raised labor income and living standards but caused disruptive structural adjustments and contributed to income polarization by making routine tasks redundant.
- Two interrelated channels—automation (higher capital-labor substitutability) and falling capital goods prices—can raise productivity and aggregate income but tend to favor capital income and high-skilled labor, increasing inequality.
- Policy options:
  - Investment in human capital (targeted education and life-long learning) can both raise low-skilled incomes and increase GDP; simulated education spending can substantially increase low-skilled incomes and aggregate output even after accounting for financing costs.
  - Differentiated tax cuts can redistribute gains more evenly but are less effective at increasing GDP and do not resolve root causes.
  - Combining skill investment with redistributive tax policies can help manage uncertainty about future skill needs and provide insurance to losers from technological change.

### Appendix (model technicalities — selected points)
- Production:
  - Three sectors: Manufacturing (M, tradable), High-skill services (S, non-tradable), Low-skill services (L, non-tradable).
  - Manufacturing uses labor (low, medium, high), capital k, and intermediate inputs from manufacturing and high-skill services.
  - Capital and high-skill labor are complements; capital substitutes with medium- and low-skill labor.
  - Low-skill services produced using only low- and medium-skill labor (assumed least capital intensive).
- Households:
  - Heterogeneity via permanent skill types (low, medium, high) and transitory idiosyncratic productivity shocks.
  - Consumption vector includes: high-skill-content services, low-skill-content services, imported goods, and tradable manufactured goods.
  - Taxes: consumption and nonlinear progressive income taxes; borrowing constraints present.
- Government:
  - Consumes manufacturing and services; invests in infrastructure; levies a range of taxes; issues foreign debt; budget constraint binds.
- Benchmark calibration:
  - Calibrated to U.S. to match macro ratios and distributional statistics.
  - Consumption aggregator, preference parameters, and elasticity assumptions chosen to match average hours and labor supply elasticities in the data.
- Model computes stationary competitive equilibrium under given tax and policy parameter sequences; households have rational expectations.

*Source: IMF Working Paper (chapter excerpt provided in the source content).*

### 0. The setting the parameter ܿ

### 0. The setting the parameter ܿ

### Consumption and sectoral calibration
- Parameter ܿ ensures that income elasticity demand for services increases with income; chosen so that consumption of manufacturing and imported consumption goods share of total consumption is, as in the data, approximately eight percentage points larger for the top quintile than for the bottom quintile of income distribution.
- The share of services on total consumption ߛ1െ and low skill services in total services consumption ߛே are calibrated such way that, given the shares of high-skill intermediate inputs in the production of manufacturing and high-skill services sectors (derived from input-output data), the model matches the total share of low skilled and high skilled services production in GDP:
  - low skilled services production: 30 percent of total GDP
  - high skilled services production: 50 percent of total GDP
- The share of domestically produced goods in tradable goods consumption ߛ் is calibrated to match a share of consumption of imports in total consumption expenditure of approx. 10 percent.

### Household’s labor productivity
- The permanent component given by skill levels in the model is proxied by education attainment levels in the US population as reported by the Census Bureau (2016).
- Skill classification:
  - Individuals with high school degree or less: low skill individuals.
  - Individuals with more than bachelor’s degree: high skill individuals.
- The average labor productivity of medium skill level households is normalized to 1 (i.e., ߟ௫ = 1).
- Differences in average years of education are used as proxies for differences in skill levels.

- Table 1. Skills Parameters
  - Parameter ߤ௟ = 0.39 — Target: Population share with high school degree or less.
  - Parameter ߤ௫ = 0.48 — Target: Population with some college, no degree, associate degree or bachelor degree.
  - Parameter ߤ௛ = 0.13 — Target: Population share with a Bachelor’s degree or more.
  - Parameter ߟ௟ = 0.7 — Target: Approx. 12 years of education.
  - Parameter ߟ௫ = 1 — Target: Approx. 16 years of education.
  - Parameter ߟ௛ = 1.1 — Target: More than 16 years of education.

### Government
- Consumption (and intermediate consumption) taxes are set at 7.5%, which is the approximate mid-point for the range of state sale taxes in US.
- For the benchmark calibration government consumption is chosen to be 16% of GDP, as the average in the data for the period 2009–15 per the World Bank data.

*Source: wp18207 - 0. The setting the parameter ܿ*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18207.pdf_