## sdnea2026001

## Source details

**Canonical URL:** [sdnea2026001](https://www.imf.org/-/media/files/publications/sdn/2026/english/sdnea2026001.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/sdn/2026/english/sdnea2026001.pdf.md)
- [Structured JSON version](/-/media/files/publications/sdn/2026/english/sdnea2026001.pdf.json)

---

### Executive summary — scope and key patterns
- New skills have grown substantially since the early 2010s: in Lightcast data, roughly 1 in 10 job postings requires at least one new skill in advanced economies, and the incidence is about half of that for emerging market economies.
- New skills appear first in advanced economies—particularly the United States—and then spread to other countries; diffusion is often fast (many skills diffuse within a year).
- Demand concentration:
  - Professional, technical, and managerial occupations.
  - Information technology (IT) skills account for more than half of new skills, with a growing share linked to artificial intelligence (AI).
  - Sector-specific skills (for example, in health care) are also increasing.
- Data and coverage notes:
  - Core demand: Lightcast vacancy data; complementary sources include ACS, German administrative records, Compustat, ILO, OECD, Lightcast worker profiles.
  - Limitations: Lightcast overrepresents formal, professional, technical, and higher-skill occupations (bias larger in emerging markets); country coverage varies (US postings from 2010; Brazil and South Africa from 2020); vacancy data may understate new skill demand in economies relying on internal labor markets.

### Individual and economywide returns to new skills
- Individual (posted wage) returns:
  - Presence of a new skill in a vacancy associated with about 3 to 3.4 percent higher wages (United States and United Kingdom).
  - Wage premium increases with multiple new skills:
    - Four or more new skills: 15.1 percent in the United Kingdom.
    - Four or more new skills: 8.5 percent in the United States.
  - By skill category, highest premiums for Information technology; Business and data analysis; Engineering.
  - AI-specific posted wage returns:
    - United Kingdom: AI-developer and AI-user skills associated with posted wage premiums of about 7.5–8 percent within occupations.
    - United States: AI-developer skills above 8 percent; AI-user skills close to 2 percent; new non-AI skills about 2.5 percent.
- Economywide local-labor-market effects (United States):
  - A 1 percentage point increase in the share of job postings with new skills → average hourly wages up by 2.3 percent and employment up by 1.3 percent.
  - Observed increase in new skill postings over the period: 3.3 percentage points → predicted wage gain about 7.6 percent and predicted employment rise about 4.3 percent.
  - Estimated effect on population: 5.3 percent.
  - Occupational distribution: highest gains for high-skill occupations; low-skill workers also gain (likely via higher consumption demand); middle-skilled workers see no significant benefits (reinforcing polarization).
- Germany (local effects):
  - A 1 percentage point increase in vacancies requiring new skills raises average wages by 0.9 percent; employment effects statistically insignificant.
  - Regions with union coverage below the 25th percentile: new-skill adoption associated with higher wages but lower employment (both statistically significant).
  - Regions with stronger union presence: wage effect slightly decreases (marginally significant) while employment effect increases significantly.
- Cross-country macro result:
  - Task creation robustly predicts stronger wage growth; one standard deviation increase in task creation correlates with approximately 18 percent higher wages (macro-regression, 2000–2023 country panel).

### AI-related new skills — composition, diffusion, and employment impacts
- Trends and composition:
  - AI skills appeared in fewer than 1 percent of US postings before 2015 and in almost 5 percent by 2025.
  - In the United States (2024) composition of AI-related postings: roughly half mention only AI-user skills; one-fourth only AI-developer skills; one-fourth both types.
  - Cross-country prevalence (2024): Denmark and the United States highest (AI-user dominant); Brazil and South Africa below 2 percent.
  - Examples of AI-developer skills: Python for machine learning; TensorFlow/PyTorch; model evaluation; ML-Ops.
- Employment impacts and heterogeneity:
  - Vacancies demanding AI-related skills post higher wages but have not yet boosted overall employment in US local labor markets.
  - For occupations highly exposed to AI with limited scope for complementarity:
    - Employment levels are 6.3 percent lower than other regions for those occupations.
    - Given observed increase in new AI-skill postings (0.57 percentage point for high-exposure, low-complementarity occupations), the 6.3 percent lower employment translates into a predicted lower employment level of about 3.6 percent.
    - Another reported result: regions with greater demand for AI-related skills show employment levels 3.6 percent lower in the medium term (five years after skill appearance) for affected occupations.
  - Entry-level workers and young cohorts:
    - Early-career workers (ages 22–25) in the most AI-exposed occupations experienced a 13 percent relative decline in employment after ChatGPT release (cited study).
  - Vacancy-share dynamics:
    - 2.5 percent lower growth of high-exposure, low-complementarity vacancies relative to high-exposure, high-complementarity occupations (vacancy-share change analysis, 2019–23).
  - Effects on wages overall are generally insignificant for AI-related adoption, though modestly lower wages observed 2–3 years after AI skill entry for affected occupations.

### Demand-side drivers and firm-level patterns
- Aggregate and firm-level patterns:
  - Countries with large employment shares in professional, technical, and managerial occupations likely to experience higher demand for new skills.
  - Demand concentrated among young, innovative, and less financially constrained firms; larger, younger, more innovative, less financially constrained, and more productive firms post more new-skill vacancies.
  - Sectors with concentrated demand: technology, finance, and insurance.
- Talent-driven M&A (acqui-hire):
  - Acquirers that highlighted labor/talent issues prior to deals subsequently reduce external demand for new skills in job postings; associated with higher post-deal patenting and patent value (pattern consistent with substitution away from external hiring).
  - Policy concern: acquire-hire can raise market concentration and impede skill diffusion, warranting competition oversight.

### Supply-side characteristics, Skill Readiness Index, and Skill Imbalance Index
- Worker characteristics:
  - Among Lightcast US worker profiles (~18 million):
    - 85 percent of workers listing new skills have at least a bachelor’s degree (vs 60 percent among workers listing only non-new skills).
    - 84.5 percent of workers reporting new IT (AI) skills hold at least a bachelor’s degree.
    - AI-developer skills concentrated in ICT and STEM backgrounds (nearly 60 percent of AI-developer skills).
    - AI-user skills distributed more broadly across fields.
- Skill Readiness Index (Index: 0–1):
  - Components (equally weighted): share of graduates able to supply new IT and non-IT skills; adult participation in job-related learning; workforce literacy, numeracy, and adaptive problem-solving scores.
  - Findings: Ireland, Finland, and Denmark top the ranking; Chile, Italy, and Hungary at the lower end.
  - Correlation: sub-indicators highly correlated—countries producing graduates with new skills tend to invest more in lifelong learning and have higher adult skill levels.
- Skill Imbalance Index (Index: -1–1):
  - Construction: demand and supply ratios benchmarked to the United States; index = normalized (demand_ratio − supply_ratio) to range between −1 and 1.
  - Cross-country patterns (relative to US benchmark):
    - Potential future demand (using US benchmark) ranges from almost 5 percent of vacancies in Türkiye to 16 percent in Luxembourg.
    - Countries with high demand relative to domestic supply (example list includes Luxembourg, Sweden, the Netherlands, Brazil) should prioritize expanding worker training, integrating IT training across fields, strengthening STEM, and relying on outsourcing and foreign-born workers.
    - Countries with strong supply but modest demand (examples: Ireland, Poland, Australia) should focus on stimulating firm demand, innovation incentives, easier business creation, export promotion, and improved access to finance.
  - Correlations with AI Preparedness Index (AIPI):
    - Correlation between AI Preparedness Index and Skill Readiness Index: 0.61.
    - AI Preparedness Index vs Skill Imbalance Index: correlation coefficient 0.12 (uncorrelated in extended sample).
    - AI Preparedness Index strongly correlated with new skill demand ratio (0.86) and moderately with new skill supply ratio (0.61).

### Diffusion patterns — international and within-country
- International diffusion:
  - Primary originators among analyzed countries: United States, Germany, United Kingdom.
  - About half of new skills identified diffused across countries between 2021 and 2024.
  - Typical diffusion lags relative to the United States:
    - Other advanced economies (Denmark, United Kingdom): demand within two to four months.
    - Germany: wider range of lags.
    - Brazil and South Africa: average lags of about eight to nine months.
- Within-country (United States) diffusion:
  - Diffusion centered on large innovative hubs: California, New York, Texas.
  - Local adoption associated with higher education measures (literacy, numeracy, college attainment).
  - Commuting zones with higher union coverage display lower demand for new skills in US data.

### Occupational dynamics and concentration
- Vacancy trends (2021–2024 classification):
  - Occupations categorized as increasing, stable, or decreasing if share of vacancies rose/fell by more than 5 percent.
  - Increasing vacancies concentrated in professional occupations (notably health care), engineering, legal services, data analysis.
  - Decreasing vacancies concentrated in nonprofessional and routine-intensive occupations (freight handlers, drivers); some professional occupations (notably software developers) also in decline due to automation.
  - Cross-country heterogeneity: Brazil and the United Kingdom show larger composition changes; Denmark and the United States more stable; Germany and South Africa intermediate.
  - Aggregate shares across countries: declining occupations about 10 percent; expanding occupations about 10 to 20 percent; stable occupations between 30 and 60 percent.
- New skills distribution:
  - Skills universe: more than 30,000 distinct skills in Lightcast taxonomy.
  - US historical new-skill definition: a skill is “new” if less than 1 percent of job postings that list this skill from 2010 to 2024 are posted in 2010–11; must appear in at least 100 US postings to be included.
  - New skills more associated with computer and routine cognitive-related tasks than with interpersonal, routine manual tasks.

### Policy recommendations and adjustment considerations
- Five broad policy recommendations:
  1. Support worker mobility across occupations and regions—including active labor-market policies, collaboration with unions, affordable housing, and remote work—to accelerate diffusion and broaden access.
  2. Competition policies: restrict non-compete agreements and monitor mergers and acquisitions to prevent excessive concentration and preserve skill diffusion.
  3. Build effective lifelong learning systems—ensure retraining is widely accessible and aligned with emerging skill needs.
  4. Rethink education systems for the AI era—equip young people with cognitive, creative, and technical skills complementary to AI; reskilling for at-risk workers; improve career guidance; expand apprenticeships and internships; strengthen school-to-work transitions.
  5. Strengthen social protection and unemployment insurance systems to support longer job transitions and re-entry; examine AI’s broader societal role recognizing work provides income, dignity, and purpose.
- Additional policy priorities by Skill Imbalance position:
  - High-demand, constrained-supply countries: expand worker training opportunities; integrate IT training across fields; favor labor mobility; strengthen STEM education.
  - Strong-supply, lower-demand countries: stimulate innovation and improve access to finance so firms absorb and deploy skills.
  - Emerging market and low-income countries with both limited demand and supply: comprehensive human capital investments, adult learning, retraining programs, supported by international cooperation.
- Other policy-relevant findings:
  - Speed of adoption matters: rapid adoption can produce larger short-term disruptions; historical examples show uneven adjustment concentrated among middle-skill occupations.
  - AI may diffuse faster than prior technologies (example cited: around 100 million users in roughly two months for ChatGPT).

### Empirical and methodological notes (selected)
- Wage regressions using job postings:
  - US postings: 54.3 million; UK postings: 5.6 million (2020–2024).
  - Share of job postings with any new skill: United States 9.2 percent; United Kingdom 8.8 percent.
  - Baseline regression: log(posted hourly wage) on indicator for any new skill with detailed fixed effects (4-digit ISCO × 6-digit NAICS × county/local authority × year) and controls for number of skills and salary type.
- Local exposure measure (US): shift-share variable using commuting-zone exposure based on 2000 occupational employment shares and nationwide expansion of new-skill demand; instruments: 2013 share of job postings requiring new skills in California and the United Kingdom, weighted by local occupational employment shares.
- Event-study and differences-in-differences specifications used for dynamic treatment timing (de Chaisemartin and d’Haultfoeuille 2023 methods referenced for staggered treatments).
- Macro-regression (country panel 2000–2023) includes productivity, task-creation proxy, task-destruction proxies (robot penetration, offshorability), product market institutions (HHI), and globalization controls; selected coefficient: Task creation 0.182*** (column 3) implying task creation predicts higher wages.

_International Monetary Fund — Staff Discussion Note: Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age (sdnea2026001)_

### Executive Summary ......................................................................................................

### Executive Summary

### A Dynamic Labor Market: scope and patterns
- New skills have grown substantially since the early 2010s; in Lightcast data, roughly 1 in 10 job postings requires at least one new skill in advanced economies, and the incidence is about half of that for emerging market economies.
- New skills appear first in the labor demand of advanced economies—particularly the United States—and then spread to other countries.
- Demand is concentrated in professional, technical, and managerial occupations.
- Information technology (IT) skills account for more than half of new skills, with a growing share linked to artificial intelligence (AI); sector-specific skills, such as in health care, are also increasing.
- AI-related new skills are distinguished from all new skills because they show important differences in labor-market impacts.
- Data sources and coverage:
  - Core demand measure: Lightcast vacancy data (online job postings).
  - Complementary sources: American Community Survey; German administrative labor market records (Sample of Integrated Labour Market Biographies); Compustat firm-level data; ILO and OECD statistics; millions of worker profiles from Lightcast.
- Limitations noted:
  - Lightcast overrepresents formal, professional, technical, and higher-skill occupations; bias is larger in emerging markets.
  - Country coverage varies (US postings from 2010; Brazil and South Africa from 2020), constraining historical perspective.
  - Vacancy data may understate new skill demand in economies relying on internal labor markets.

### Economywide return to new skills: wages and employment effects
- All new skills (aggregate):
  - At the job-posting level, new skills are associated with 3–3.4 percent higher wage offers in the United States and the United Kingdom.
  - In US local labor markets, an increase of 1 percentage point in the share of job postings with new skills is associated with an average wage gain of 2.3 percent and an employment gain of 1.3 percent.
  - In German local labor markets, employment impacts are statistically insignificant, but wages rise by 0.9 percent.
  - Both high-skilled workers and—through higher consumption of services—low-skilled workers capture the largest benefits; no significant benefits for middle-skilled workers (reinforcing job polarization).
  - Cross-country evidence: incorporation of new tasks (proxy for new skills) is associated with higher wages globally.
- AI-related new skills (subset):
  - Vacancies demanding new AI-related skills post higher wages but have not yet boosted overall employment in US local labor markets.
  - For occupations highly exposed to AI with limited scope for complementarity, employment levels are 3.6 percent lower in regions with greater demand for AI-related skills than in other regions five years after skill appearance.
  - Employment declines disproportionately affect white-collar middle-skilled jobs, young workers, and some categories of IT specialists.

### The demand and supply of skills: drivers and readiness
- Demand-side drivers:
  - Countries with large employment shares in professional, technical, and managerial occupations are likely to experience higher demand for new skills.
  - Young, innovative, and less financially constrained firms are important sources of demand.
  - Talent-driven mergers and acquisitions (“acquire-hire”) can secure scarce expertise but raise concerns about market concentration and skill diffusion.
- Supply-side characteristics:
  - Supply of new skills relies heavily on workers with tertiary education, particularly in STEM and IT fields, though IT skills are found across all fields of study.
  - A Skill Readiness Index is developed combining:
    - share of recent graduates able to supply new IT and non-IT skills,
    - indicators of retraining frequency,
    - workforce literacy and numeracy scores.
- Skill Imbalance Index:
  - Captures the relative gap between demand for new skills and domestic supply capacity across countries.
  - Policy priorities differ by position on this index:
    - High-demand, constrained-supply countries: expand worker training opportunities, integrate IT training across all fields of study, favor labor mobility across regions, and strengthen STEM education.
    - Strong-supply, lower-demand countries: stimulate innovation and improve access to finance so firms absorb and deploy skills.
  - IMF AI Preparedness Index can guide policymakers on priority policy areas.

### Adjustment, diffusion, and policy considerations
- Facilitating worker mobility across occupations and regions is critical for diffusion of new skills.
- Recommended policy actions:
  - Expand accessible and frequent retraining and upskilling programs for workers.
  - Integrate IT training across all fields of study and strengthen STEM education pipelines.
  - Promote active labor market policies and affordable housing to accelerate occupational and geographic mobility.
  - Encourage firm–union collaboration to facilitate worker adjustment to new technologies.
  - Limit non-compete agreements and other contractual restraints that may slow the spread of talent and skills.
  - Monitor and address competition concerns arising from talent-driven mergers and acquisitions to preserve skill diffusion.
  - Improve access to finance and policies that stimulate innovation in countries where supply outstrips current demand.
- Broader considerations:
  - The overall employment and distributional outcomes of new skills depend on productivity dynamics (including the potential for automation-driven labor replacement) and on consumer demand sensitivity to price changes (e.g., lower prices could expand usage and employment in some services).
  - Speed of adoption matters: rapid adoption can produce larger short-term disruptions; historical examples (industrial robots, IT revolution) show adjustment can be uneven and concentrated among middle-skill occupations.
  - AI may reach many occupations previously shielded from technological change and may diffuse faster than prior technologies (example: around 100 million users in roughly two months for ChatGPT).

*International Monetary Fund — Staff Discussion Note: Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age*

### Section IV examines the new skills demand and supply and presents a new Skill Imbalance Index and Skill

### sdnea2026001 - Section IV examines the new skills demand and supply and presents a new Skill Imbalance Index and Skill Readiness Index for a wide range of countries

### II. A Dynamic Labor Market — Occupation-level Dynamics
- Vacancies from 2021 to 2024 are used to categorize occupations for each country as increasing, stable, or decreasing, based on whether the share of vacancies in a particular occupation rose or fell by more than 5 percent.
- Increasing vacancies are concentrated in professional occupations—most notably in health care—and in areas such as engineering, legal services, and data analysis.
- Decreasing vacancies are concentrated in nonprofessional and routine-intensive occupations, such as freight handlers and drivers; some professional occupations (most prominently software developers) are also in decline due to automation.
- Cross-country comparisons (six countries) show variation in labor market dynamism:
  - Brazil and the United Kingdom display larger changes in the composition of labor demand (higher shares of decreasing and increasing occupations).
  - Denmark and the United States exhibit greater stability.
  - Germany and South Africa are intermediate.
- Aggregate shares across countries:
  - Share of declining occupations: about 10 percent.
  - Share of expanding occupations: about 10 to 20 percent.
  - Share of stable occupations: between 30 and 60 percent.

### II.2 New Skills Demand — Definition and Measurement
- Skills universe: more than 30,000 distinct skills in the Lightcast taxonomy.
- New skills definition (United States historical series): a skill is “new” if less than 1 percent of job postings that list this skill from 2010 to 2024 in the United States are posted in 2010–11; the “year of emergence” is the first year it rises above the 1 percent threshold.
- New skills must appear in at least 100 job postings in the United States to be included in the analysis.
- Robustness: country-specific simpler definitions yield a similar distribution of new skills across occupations.

- Prevalence and patterns:
  - Roughly 1 in 10 vacancies lists a new skill in advanced economies.
  - In emerging market economies, the share is closer to 1 in 20.
  - New skills are closely tied to labor market dynamism:
    - A 1 percent increase in the number of new skills mentioned in postings within a given occupation is associated with about a 0.4 percentage point higher probability that the occupation is in the “increasing” group.
    - A 1 percent increase in new skills demanded is associated with roughly a 0.1 percentage point increase in the average annual growth rate of vacancies.
  - New skills appear more often in occupations that are expanding and rarely in those that are declining.

- Occupational and task concentration:
  - New skills are concentrated in managerial, technical, and professional occupations.
  - Most new skills are IT-related; IT shows greater “skill churn” because of rapid software/tool/platform evolution.
  - New skills are more closely associated with computer and routine cognitive-related tasks than with interpersonal, routine, or manual tasks.
  - New skills are also present across all economic sectors (US data): examples include telecare in health, social media management in sales and marketing, and climate engineering/climate resilience tied to the energy transition (many energy-transition skills predate 2010–11).

- Examples (top skills by occupation, United States, 2024): Software Developers, Dental Assistants and Therapists, Statistical/Mathematical Associate Professionals, Cashiers and Ticket Clerks. New skills in examples include Power BI*, Tableau*, Data Pipelines* (asterisk denotes new skills in the source table).

### AI-Related New Skills — Trends and Composition
- AI-related skills identification: AI-based text classification; taxonomy separates AI-user and AI-developer skills.
- Historical and recent prevalence (United States):
  - AI skills appeared in fewer than 1 percent of postings before 2015.
  - AI skills appeared in almost 5 percent of postings by 2025.
- Composition of AI-related postings in the United States (2024):
  - Roughly half of postings mention only AI-user skills.
  - One-fourth reference only AI-developer skills.
  - One-fourth reference both types of AI skills.
- Cross-country prevalence (2024):
  - Denmark and the United States display the highest prevalence of AI-related postings, with AI-user skills dominant.
  - Brazil and South Africa exhibit lower overall prevalence, below 2 percent.
- Examples of AI-developer skills: Python for machine learning, TensorFlow/PyTorch, model evaluation, ML-Ops.
- Distinction purpose: separate AI adoption/use (AI-user) from AI creation/build (AI-developer).

### II.3 New Skills Diffusion — International and Within-Country Patterns
- International diffusion:
  - New skills tend to emerge in advanced economies before spreading to emerging markets.
  - Germany, the United Kingdom, and the United States are identified as primary originators among the analyzed countries; many flows point toward Brazil and South Africa as recipients.
  - About half of the new skills identified diffused across countries between 2021 and 2024.
  - Typical diffusion timing relative to the United States (main originator):
    - Other advanced economies such as Denmark and the United Kingdom demand new skills within two to four months.
    - Germany shows a wider range of lags.
    - Brazil and South Africa experience average lags of about eight to nine months.
  - For skills that diffuse, speed is generally fast: diffusion occurs within a year for many skills.

- Within-country (United States) diffusion:
  - Diffusion across US states is highly uneven and centered on large innovative hubs.
  - California is a major point of origin for new skills, along with New York and Texas; these hubs diffuse skills outward to smaller states.
  - New skills are diffusing more rapidly in recent years.

- Local factors associated with adoption:
  - Areas with higher education levels (literacy, numeracy, college attainment) are significantly more likely to demand new skills that first appeared in California.
  - Commuting zones with higher union coverage and membership display lower demand for new skills, suggesting unions may be associated with lower adoption in the US context (the note cites contrasting evidence on unions’ role in worker transitions and technology adoption in other studies and country contexts).

*Source: STAFF DISCUSSION NOTES Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age — INTERNATIONAL MONETARY FUND*

### 1. Network of Skill Diffusion across Countries

### 1. Network of Skill Diffusion across Countries

### Global and US diffusion patterns
- Panel summaries (based on Lightcast job posting data; skill appearance = first recorded posting in each location):
  - Panel 1: global skill diffusion network across advanced and emerging market economies.
  - Panel 2: distribution of time lags in skill demand measured in months relative to the United States.
  - Panel 3: US state-level diffusion network.
  - Panel 4: regression results using California’s new skill adoption to predict adoption in other commuting zones (95 percent confidence intervals).
- Local policy and human-capital covariates measured prior to diffusion: Union membership and coverage; state non-compete index; college share (ACS 2010); land-use restrictiveness (2008 Wharton Index); literacy and numeracy (PIACC 2012); third grade (G3) math scores (Opportunity Lab).

---

### II.4 Individual Return to New Skills

### Empirical approach
- Data: Lightcast job postings with posted wages, skills, 6-digit NAICS industry, 4-digit ISCO occupation, pay period, county, year of posting (2020–2024).
- Regression design:
  - OLS with detailed fixed effects: number of skills in vacancy; fixed effect for every unique combination of year × 6-digit industry × county × 4-digit ISCO occupation; posted pay period categories interacted with year.
  - Purpose: capture marginal gain of a new skill over an existing skill within very similar vacancies.

### Key findings on wage returns
- Presence of a new skill in a vacancy is associated with about 3 to 3.4 percent higher wages.
- Wage premium increases with multiple new skills:
  - Four or more new skills: 15.1 percent in the United Kingdom.
  - Four or more new skills: 8.5 percent in the United States.
- Occupational coverage:
  - Wage gains found for new skills across all occupation types in both the United Kingdom and the United States, except new skills in low-skill occupations in the United Kingdom (no wage gain detected).
- By skill category, highest wage premiums linked to:
  - Information technology
  - Business and data analysis
  - Engineering

### AI-specific wage returns
- United Kingdom:
  - AI-developer and AI-user skills associated with posted wage premiums of about 7.5–8 percent within occupations.
- United States:
  - AI-developer skills: high wage premiums of above 8 percent.
  - AI-user skills: smaller premium close to 2 percent.
  - New non-AI skills: premium of about 2.5 percent.

---

### III. Economywide Return to New Skills

### III.1 United States — Local labor market effects
- Exposure measure: shift-share variable (commuting-zone exposure based on 2000 occupational employment shares and nationwide expansion of new-skill demand).
- Instrumental variables: 2013 share of job postings requiring new skills in California and the United Kingdom, weighted by local occupational employment shares.
- Wage effects:
  - A 1 percentage point increase in job postings requiring new skills raises average hourly wages by 2.3 percent.
  - Observed increase in new skill postings over the period: 3.3 percentage points → predicted wage gain about 7.6 percent.
- Employment and population effects:
  - Same 1 percentage point increase linked to 1.3 percent higher employment → predicted rise about 4.3 percent given observed change.
  - Estimated effect on population: 5.3 percent.
- Occupational distribution of gains:
  - High-skill occupations (managers, engineers) benefit the most.
  - Low-skill occupations also experience gains, likely via aggregate income effects increasing demand for services.
  - White-collar and blue-collar effects smaller and statistically insignificant.
- Heterogeneity by local policies and initial conditions (interactions measured prior to adoption, in 2013 or earlier):
  - Larger wage and employment effects where:
    - Third-grade math scores (proxy for neighborhood quality) are stronger.
    - Land-use policies are less restrictive.
    - Labor markets are more flexible (weaker non-compete agreements).
  - Stricter non-compete agreements dampen pass-through into wages.
  - Stronger prior union presence associated with slightly smaller employment gains (interaction not statistically significant).
  - Higher unemployment insurance generosity associated with smaller employment gains.
- Interpretation: local institutions and policies shape magnitude of wage and employment translation from new-skill adoption.

### III.2 Germany — Local labor market effects
- Data and period: Lightcast data covering 2019–23; 96 German commuting zones; instruments: early US adoption (2014) and change in US adoption (2014–18).
- Average effects:
  - A 1 percentage point increase in vacancies requiring new skills raises average wages in local labor markets by 0.9 percent.
  - Employment effects are statistically insignificant.
- Heterogeneity by institutions:
  - Regions with union coverage below the 25th percentile:
    - New-skill adoption associated with higher wages but lower employment (both statistically significant).
  - Regions with stronger union presence:
    - Wage effect slightly decreases (marginally significant).
    - Employment effect increases significantly → suggests unions mitigate displacement pressures, possibly via job protection coupled with some wage moderation.
  - Retraining intensity:
    - Regions below the 25th percentile of retraining share: new-skill adoption exerts negative impact on employment, positive but insignificant wage effect.
    - With higher retraining levels, employment effect improves (insignificant).

### III.3 Cross-Country impact of new tasks on wages
- Country-level panel (2000–2023) using macro task-creation measure (tasks mapped across countries).
- Findings:
  - Task creation robustly predicts stronger wage growth, conditional on productivity and institutional controls.
  - Task-destruction indicators (routine task intensity, offshorability, robot penetration) are insignificant once conditioning on productivity.
- Implication: new tasks help explain deviations between wage growth and productivity growth.

### III.4 Effect of AI skills on employment
- Exposure and risk:
  - Young workers’ employment concentrated in occupations with high exposure and low complementarity to AI → higher displacement risk.
  - Pattern more accentuated for college-educated young workers due to higher exposure.
- Evidence on entry-level employment:
  - Brynjolfsson, Chandar, and Chen (2025): since release of ChatGPT, early-career workers (ages 22-25) in the most AI-exposed occupations experienced a 13 percent relative decline in employment.
  - Vacancy evidence (US, 2019–23): local AI adoption disproportionately reduces vacancies in high-exposure, low-complementarity occupations.
  - Pizzinelli and others (2023): a one-standard-deviation higher commuting-zone adoption in 2019 (about 0.18 percentage point, comparable to the gap between Boston, MA, and Portland, OR) is associated with a 0.4 percentage point lower vacancy share after controls.
- Interpretation: generative-AI adoption can reduce entry-level hiring particularly where tasks are automatable rather than complementary.

---

*Source: sdnea2026001 - 1. Network of Skill Diffusion across Countries (IMF staff discussion notes; data and figures based on Lightcast, ACS, CPS, Opportunity Lab, PIACC, Reinmuth and Rockall 2023, Wharton Residential Land Use Regulatory Index, and IMF staff calculations).*

### 2.5 percent lower growth of high-exposure, low-complementarity vacancies, relative to high-exposure, high-

### sdnea2026001 - 2.5 percent lower growth of high-exposure, low-complementarity vacancies, relative to high-exposure, high-

### AI exposure, vacancy changes, and occupational categories
- 2.5 percent lower growth of high-exposure, low-complementarity vacancies, relative to high-exposure, high-complementarity occupations.
- Vacancy-share change analysis (2019–23) compares HEHC (high-exposure, high-complementarity) and HELC (high-exposure, low-complementarity) occupations at the commuting-zone level against an industry-based measure of AI adoption.
- HEHC = high-exposure, high-complementarity; HELC = high-exposure, low-complementarity; LE = low-exposure.
- AI adoption at the commuting-zone (CZ) level is estimated using the US Census Bureau’s Annual Business Survey module on firms’ AI use during 2016–18; industry-level adoption shares are employment-weighted to construct a CZ indicator of AI presence in local production activities.
- Country comparisons in figures include BRA, GBR, USA and age-group breakdowns: Youth (20–24), Prime (25–54), Older workers (55–64).

### Impact of AI skills on employment (key empirical findings)
- For workers in jobs with high exposure and low complementarity to AI—about 30 percent of total employment (Cazzaniga and others 2024)—a clear and sustained decline in employment growth is observed.
- AI-related skills show employment levels that are 6.3 percent lower than other regions for occupations that are highly exposed to AI with limited scope for complementarity.
- Given the observed increase in new skill postings (which amounts to 0.57 percentage point for workers in high-exposure, low-complementarity occupations over the period considered), the 6.3 percent lower employment translates into a predicted lower employment level of about 3.6 percent.
- Effects are insignificant for workers in high-exposure, high-complementarity occupations.
- For low-exposure occupations, employment levels are lower—by about 4 to 5 percent—in years one and three relative to other regions, but insignificant otherwise.
- The impact on wages is overall insignificant; however, workers in high-exposure, low-complementarity occupations experience modestly lower wages 2–3 years after the AI skill entry.
- In the medium term (after five years), regions with a 1 percentage point higher demand for AI adoption at the commuting-zone level are estimated to experience the noted employment impacts (see accompanying figures and methods).
- Methodology note: dynamic event study estimates of log employment following the entry of AI skills use the difference-in-differences estimator developed by de Chaisemartin and d’Haultfoeuille (2023) to accommodate staggered, recurrent, and nonbinary treatments.

### Firm-level demand for new skills and patterns of hiring
- Analysis uses US job vacancy data from Lightcast matched to firm-level information from Compustat (covering mainly larger, publicly listed companies); the match covers 83 percent of the total market capitalization of Compustat firms.
- Demand for new skills is concentrated in the technology, finance, and insurance sectors (examples cited include Salesforce, Cognizant, and JPMorgan Chase).
- Firms posting vacancies with new skills are:
  - larger,
  - younger,
  - more innovative,
  - less financially constrained,
  - more productive.
- Corporate evidence indicates competition for scarce, high-demand skills has intensified, fueling “acquire-hire” mergers and acquisitions; such transactions can slow external hiring for new skills by acquirers while increasing patenting activity and value-added, but may raise competition and market-power concerns warranting oversight.

### Worker characteristics and distribution of new skills
- Lightcast data covering nearly 18 million US worker profiles is used to assess worker characteristics associated with new skills.
- Education and skill incidence:
  - 85 percent of workers listing new skills on their profiles have at least a bachelor’s degree, compared with 60 percent among workers who list only non-new skills.
  - 84.5 percent of workers reporting new IT (AI) skills hold at least a bachelor’s degree.
  - Among new non-IT skills holders, 86 percent have at least a bachelor’s degree.
- Field-of-study patterns:
  - Workers with ICT degrees report the highest number of new skills, followed by natural sciences, mathematics, statistics, or security and transport.
  - AI-developer skills are concentrated among workers with ICT and STEM backgrounds—accounting for nearly 60 percent of all AI-developer skills.
  - AI-user skills are distributed far more broadly across fields, reflecting AI’s general-purpose nature.
- Non-STEM workers who list IT skills are often more likely to report additional new non-IT skills, suggesting bundled adoption outside STEM.

### Cross-country demand, supply, and indices of imbalance and readiness
- Potential future demand for new skills (using US as benchmark) varies across countries: it ranges from almost 5 percent of vacancies in Türkiye to 16 percent in Luxembourg.
- Across countries, IT skills account for the largest share of new skills sought by employers, followed by business and data analysis skills and social and administrative competencies.
- Potential supply of new skills (using OECD graduates by field of study and US field-specific prevalence) also varies markedly: Ireland and Poland have a large share of recent STEM graduates with potential to supply new skills, while Mexico and Brazil have about one-third of that share and Luxembourg about one-fifth.
- Regional patterns:
  - Northern and western European countries often have relatively higher potential demand and relatively strong domestic supply.
  - Many emerging economies show both weaker potential demand and more limited supply.
  - Southern and eastern European economies lie between these extremes.
- Policy implications from a new Skill Imbalance Index (relative to the United States, normalized to 1):
  - Countries with high demand relative to domestic supply (e.g., Luxembourg, Sweden, the Netherlands, Brazil) should focus on expanding worker training (including IT training across all fields), strengthening STEM education, and potentially relying on outsourcing and foreign-born workers.
  - Countries with strong supply capacity but modest demand (e.g., Ireland, Poland, Australia) should prioritize stimulating firms’ demand and helping companies absorb and deploy skills via innovation incentives, easier business creation, export promotion, and reduced financial constraints.
  - Countries with smaller imbalances (e.g., Chile, Estonia) face less immediate urgency but still need policies on both sides of the skills market to remain competitive.
- Skill Readiness Index (24 countries) combines:
  - share of graduates able to supply new IT and non-IT skills,
  - share of adults aged 25–65 who participated in job-related learning,
  - workforce proficiency measured by the OECD’s Programme for the International Assessment of Adult Competencies literacy and numeracy scores.
  - Components are normalized and equally weighted to produce a single index where higher values reflect a larger pool of graduates, more frequent retraining, and stronger foundational skills.

_Staff Discussion Notes: Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age — INTERNATIONAL MONETARY FUND_

### 1. Demand: Share of Employment Requiring New Skills

### 1. Demand: Share of Employment Requiring New Skills

### Measurement and Panels
- Panel 1: Share of employment requiring new skills (Percent). Assumes the US occupational distribution of new skill requirements. Job postings listing multiple new skills are classified according to the most common skill.
- Panel 2: Share of graduates with new skills (Percent of population aged 20–24). Shows 2013–2021 actuals; 2022 and 2023 assume the same graduation rates as 2021, then apply the US distribution of graduates with new skills by field of study. The annual flow numerator is scaled by 5 to match the five-year cohort denominator.
- Panel 3: Supply versus demand relative to the United States (Share of graduates with new skills (Ratio relative to USA); New skill demand (Ratio relative to USA)).
- Panel 4: Skill Imbalance Index (Index:-1–1). Calculated as the difference between supply and demand (Panel 3 axes), normalized by the maximum absolute value of these differences to range between –1 and 1.
- Country labels use International Organization for Standardization (ISO) country codes.

### Skill Readiness Index: findings
- Index specification: Skill Readiness Index (Index: 0–1) is the average of four normalized components:
  - 1) graduates with new IT skills;
  - 2) graduates with new non-IT skills;
  - 3) average adult skills in literacy, numeracy, and adaptive problem-solving (OECD);
  - 4) adult participation in job-related learning (OECD).
- Key cross-country observations:
  - Northern European economies appear well positioned to supply new skills, reflecting continuous learning and adult training.
  - Ireland, Finland, and Denmark top the ranking with high shares of graduates linked to both new IT and non-IT skills, strong adult literacy and numeracy performance, and robust retraining systems.
  - Chile, Italy, and Hungary rank at the lower end, reflecting weaker tertiary specialization in IT, fewer retraining opportunities, and lower adult skill levels.
- Correlation: Sub-indicators are highly correlated—countries that produce graduates with new skills also tend to invest more in lifelong learning and achieve higher adult skill levels.

### Demand and supply patterns
- Incidence:
  - In advanced economies, roughly 1 in 10 job postings now requires at least one new skill.
  - The incidence in emerging market economies is about half of that in advanced economies.
- Concentration:
  - New skills are concentrated in professional, technical, and managerial occupations.
  - IT-related competencies—particularly AI—are at the forefront.
- Diffusion:
  - New skills demand originates mainly in advanced economies, especially the United States, and then spreads to other economies.
  - Within countries, diffusion is shaped by local educational attainment and labor market institutions that support technological change.
- Supply composition:
  - Development of new skills relies substantially on individuals with tertiary education, especially those with backgrounds in information technology.
  - IT competencies are present across a broad range of academic disciplines.
  - Much demand for IT skills relates to AI-user skills demanded across many non-IT roles.

### Economic impacts and heterogeneity
- Wage and employment impacts:
  - Many new skills are associated with higher wage offers and local employment and wage gains, particularly benefiting high- and low-skill workers.
  - Despite higher wages for AI-related vacancies, the rise in demand for new AI skills has not generated an increase in economywide employment so far.
  - Regions with greater demand for AI-related skills show lower employment levels than other regions for occupations highly exposed to AI with limited scope for complementarity, posing challenges especially for young workers.
  - New non-AI skills do boost overall employment.
- Risks:
  - Dynamics may reinforce job polarization and the shrinking of the middle class, and risk amplifying inequalities if adoption remains concentrated in certain sectors, firms, or regions.
  - Many firms will adopt AI through ready-made tools or outsourcing rather than by expanding in-house technical teams.

### Skill Imbalance Index: policy implications
- The Skill Imbalance Index compares supply and demand of new skills and complements the Skill Readiness Index and the IMF AI Preparedness Index (AIPI).
- Use-case guidance from index patterns:
  - Countries with high demand and constrained domestic supply (examples: Sweden, the Netherlands, Brazil): prioritize expanding lifelong training opportunities, integrating IT skills into all fields of study, and strengthening STEM education.
  - Countries with strong supply but modest demand (examples: Ireland, Poland, Australia): prioritize stimulating firm demand through innovation incentives, easier business creation, and improved access to finance.
  - Emerging market economies and low-income countries with both limited demand and supply: need both sets of policies and fundamental investment in human capital, including enhanced education systems, adult learning, and retraining programs, supported by international cooperation.
- Relationship with AI Preparedness Index:
  - The AIPI focuses on AI across human capital and labor market policies, digital infrastructure, innovation and economic integration, and regulation and ethics.
  - The AIPI correlates with the demand and supply measures of new skills but not with the Skill Imbalance Index itself.

### Five broader policy recommendations (as stated)
- First: Support worker mobility across occupations and regions—including active labor-market policies, potentially organized in collaboration with unions, affordable housing, and remote work—to accelerate diffusion and broaden access to opportunities.
- Second: Competition policies should safeguard against excessive concentration by restricting the use of non-compete agreements and monitoring mergers and acquisitions, targeting cases likely to lead to excessive concentration and weakened skill diffusion across firms.
- Third: Build effective lifelong learning systems—ensure retraining is widely accessible and aligned with emerging skill needs to enable worker adaptation.
- Fourth: Rethink education systems to prepare the workforce for the AI era—equip young people with cognitive, creative, and technical skills complementary to AI; provide reskilling for workers whose tasks are most at risk of automation; improve career guidance; expand apprenticeships and internships; and strengthen school-to-work transitions.
- Fifth: Strengthen social protection and unemployment insurance systems to support those facing longer job transitions or difficulties re-entering the labor market; examine the broader role of AI in society recognizing that work provides income, dignity, and purpose.

### Box: The Impact of Mergers and Acquisitions on New Skills Demand (summary)
- Context:
  - Acute shortages of specialized skills in US technology-intensive sectors have coincided with a rise in mergers and acquisitions (M&A) aimed at internalizing tacit know-how (acqui-hiring).
  - Motivations for deals can include talent acquisition, intellectual property, reinforcing market positions, or limiting antitrust scrutiny.
- Empirical approach:
  - Analysis merges Compustat firm-level data with Lightcast job postings, Capital IQ transcripts, and a database of completed M&A.
  - Acquirers are classified by whether labor/talent keywords appear in the three years prior to a deal.
- Findings:
  - Acquirers whose pre-deal calls highlighted labor/talent issues subsequently reduce their demand for new skills in job postings; acquirers without such mentions show no material change.
  - On average, post-deal patenting levels and patent value are higher among the labor-mentioning group, though differences are not statistically significant at conventional levels and the group was already more innovation-active pre-deal.
  - Pattern is consistent with substitution away from external hiring once specialized teams are internalized; talent-driven M&A may concentrate market power and impede skill diffusion.

### Data and coverage notes
- Key indices and ranges explicitly used:
  - Skill Readiness Index (Index: 0–1).
  - Skill Imbalance Index (Index:-1–1).
- Data sources include Lightcast job postings and profile data (published June 2025), OECD, ILO, PIACC, Compustat, Capital IQ, and other listed sources for specific figures and annexes.
- Country and figure coverage details appear in Annex I (Figures and country lists).

*Sources: International Labour Organization; Lightcast; Organisation for Economic Co-operation and Development;; and IMF staff calculations.*

### Annex Table 1.2. Country Sample Coverage

### Annex Table 1.2. Country Sample Coverage

### Country sample (ISO3 — Country — Income Group)
- AUS — Australia — AE
- AUT — Austria — AE
- BEL — Belgium — AE
- BGR — Bulgaria — EM
- BRA — Brazil — EM
- CHE — Switzerland — AE
- CHL — Chile — EM
- COL — Colombia — EM
- CRI — Costa Rica — EM
- CZE — Czech Republic — AE
- DEU — Germany — AE
- DNK — Denmark — AE
- ESP — Spain — AE
- EST — Estonia — AE
- FIN — Finland — AE
- FRA — France — AE
- GBR — United Kingdom — AE
- GRC — Greece — AE
- HRV — Croatia — AE
- HUN — Hungary — EM
- IRL — Ireland — AE
- ISR — Israel — AE
- ITA — Italy — AE
- LTU — Lithuania — AE
- LUX — Luxembourg — AE
- LVA — Latvia — AE
- MEX — Mexico — EM
- NLD — The Netherlands — AE
- NOR — Norway — AE
- POL — Poland — EM
- PRT — Portugal — AE
- ROU — Romania — EM
- SVK — Slovak Republic — AE
- SVN — Slovenia — AE
- SWE — Sweden — AE
- TUR — Türkiye — EM
- USA — United States — AE
- ZAF — South Africa — EM

### Note on labels
- Country labels use International Organization for Standardization (ISO) country codes.
- AE = advanced economies; EM = emerging market economy.

*Source: Annex Table 1.2. Country Sample Coverage (sdnea2026001).*

### Annex Figure 2.1. Skill Appearance Time Lag

### sdnea2026001 - Annex Figure 2.1. Skill Appearance Time Lag

### Skill diffusion timing across U.S. states (Annex Figure 2.1)
- The figure reports distributions of time lags in new skill appearance across U.S. states measured in months relative to California.
- States are ordered by their diffusion lag in the 2020–24 period.
- Positive values indicate slower appearance relative to California.
- Periods shown: 2010–14, 2015–19, 2020–24.

### Key implication
- The visualization documents cross-state heterogeneity in the speed of new-skill diffusion, using California as the reference (zero months).

### Data sources and notes
- Sources: Lightcast; and IMF staff calculations.
- Notes: Time lags are measured in months relative to California; states ordered by diffusion lag in 2020–24.

---

### Macro-regression framework and results (Annex Table 2.1)
- Sample: unbalanced country–year panel for 2000–23 covering real hourly wages, labor productivity (GDP per hour), task content measures, product market institutions, and globalization variables.
- Task destruction proxies: robot penetration (International Federation of Robotics) and offshorability indices mapped from occupations to national employment.
- Task creation proxy: O*NET “emerging tasks,” aggregated to country–year employment shares.
- Institutional controls: product market concentration (HHI) and IMF Structural Reforms Database measures.
- Estimation: country and year fixed effects, employment weights, one-year lags for task variables, standard errors clustered by country.

- Announced macro-regression findings:
  - "Task creation consistently predicts higher wages across countries."
  - "The estimated coefficients suggest that a one standard deviation increase in task creation correlates with approximately 18 percent higher wages."

- Selected coefficient estimates from Annex Table 2.1 (Dep: ln(wage)):
  - ln(Productivity): 0.668 (column 1), 1.642** (column 2), 1.851** (column 3), 1.857** (column 4).
  - Task creation: 0.182*** (column 3), 0.178** (column 4).
  - HHI, Business regulation, Robot exposure, Offshoring included in columns 3–4 with reported point estimates:
    - HHI: 0.799 (col 3), 0.818 (col 4).
    - Business regulation: -0.071 (col 3), -0.073 (col 4).
    - Robot exposure: -0.181 (col 3), -0.182 (col 4).
    - Offshoring: -0.016 (col 3), -0.026 (col 4).
  - Import/GDP included in column 4: -0.003.
  - Financial integration included in column 4: 0.000.
  - Observations: 812 (col 1), 426 (col 2), 426 (col 3), 416 (col 4).
  - R-squared: 0.607 (col 1), 0.657 (col 2), 0.661 (col 3), 0.657 (col 4).

- Data sources cited for macro-regression: Brookings Institution, IMF Structural Reforms Database, International Federation of Robotics, OECD, O*NET, World Bank World Development Indicators, and IMF staff calculations.
- Note: HHI = Herfindahl-Hirschman Index.

---

### Wage regression using job postings (Annex III)

### Job postings data (Annex III.1)
- Coverage: United Kingdom and the United States, 2020–2024.
- Job-posting sample sizes:
  - United States: 54.3 million postings.
  - United Kingdom: 5.6 million postings.
- Data contain: 4-digit ISCO occupations, 6-digit NAICS industry, county/local authority, hourly wage, salary type.
- Wage construction: if posted wage specified as interval, midpoint taken; all salary frequencies harmonized to hourly units; outcome = log of posted hourly wages.
- Share of job postings with any new skill:
  - United States: 9.2 percent.
  - United Kingdom: 8.8 percent.

### Baseline empirical specification (Annex III.2)
- Regression: log(wage)j = β * I(any new skill)j + δoccupation,industry,country,year + δsalarytype,year + δ#skills + εj.
- Identification: within 4-digit ISCO × 6-digit industry × county × year cells; controls for salary type by year; δ#skills controls for number of skills listed.
- Robust standard errors clustered at the occupational level.
- Heterogeneity analyses:
  - Wage premium by number of new skills (1, 2, 3, 4+).
  - Wage premium by skill category (IT, business and data analysis, etc.).
  - Wage premium by occupation type (high-skill, white-collar, blue-collar, low-skill).
  - Classification into AI Developer skills, AI User skills, and non-AI skills.

### Key interpretation
- Coefficient β represents the percentage difference in posted wages between job postings that list a new skill versus those that list only existing skills, conditional on detailed fixed effects.

---

### Demand and supply of new skills (Annex IV)

### Demand estimation (IV.1)
- Assumption: U.S. occupational distribution of new skill requirements applies to all countries.
- Demand of new skill s in country c computed as a weighted sum across occupations using country employment shares and U.S. Lightcast job postings by occupation:
  - Demand formula (symbolic): Demand_of_new_skill_c,s = Σ_emp(c,o) × Postings_with_new_skill_USA,o,s / Σ_emp_USA,o × Σ_emp(c,o) × Postings_USA,o / Σ_emp_USA,o
  - Notation: c = country, s = skill category, o = occupation.
- Data source for employment by occupation: International Labour Organization.

### Supply estimation (IV.2)
- Share of graduates with new skills among population aged 20–24 estimated assuming U.S. distribution of graduates with new skills by field of study applies to all countries.
- Supply formula (symbolic): Share_of_graduates_with_new_skills_c,y = (Σ_share_USA,f,y × Graduates_c,f,y) × 5 / Pop_20–24,c,y
  - c = country, f = field of study, y = year.
- Construction details:
  - 푆ℎ푎푟푒 표푓 푔푟푎푑푢푎푡푒푠 푤푖푡ℎ 푛푒푤 푠푘푖푙푙푠_U푆퐴,푓,푦 estimated using degree information from Lightcast US profiles graduating in 2022–24 and ChatGPT-4o-mini model to infer field of study distribution; earlier years adjusted by ratios relative to 2023.
  - Graduates_c,f,y from OECD data.
  - Multiplied by five to match annual flow numerator with five-year age cohort denominator (ages 20–24).

### Skill Imbalance Index construction (IV.3)
- Both demand and supply are benchmarked to the United States (U.S. = frontier of new-skill adoption).
- Steps:
  - Express demand and supply as ratios relative to U.S. values.
  - Compute difference: demand_ratio − supply_ratio.
  - Normalize by maximum absolute value of these differences so index ranges between −1 and 1.
- Annex Figure 4.1 extends the Skill Imbalance Index to IT and non-IT skills separately.

### Extension using UNESCO UIS (IV.3 notes)
- To increase country coverage, OECD graduate-field data are complemented with UNESCO UIS data where OECD data are unavailable.
- Trade-offs: UNESCO covers fewer years, broader field categories (16 → 11 mapping), and covers bachelor’s and master’s degrees (ISCE 6 and 7) only.
- Inclusion criteria for UNESCO-based countries:
  1. Not covered by OECD data.
  2. ILO provides occupational employment distributions at 2-digit ISCO level.
  3. UNESCO UIS covers at least four years between 2013–24 and latest year is 2022 or later.
  4. Country is not low-income.
- This adds 18 additional countries to the Skill Imbalance Index.

### Skill Readiness Index (IV.4)
- Constructed as the simple average of four normalized components:
  1. Graduates with new IT skills.
  2. Graduates with new non-IT skills.
  3. Adult participation in job-related learning (OECD 2023).
  4. Average adult skills in literacy, numeracy, and adaptive problem-solving score (OECD 2023).
- Normalization procedure per component: convert values into standard scores, add 3, divide by 6, rescaling so values between the 0.1st and 99.9th percentiles correspond to range 0–1.

### Relationship to AI Preparedness Index (IV.5)
- The Skill Readiness Index and Skill Imbalance Index complement the AI Preparedness Index (AIPI) of Cazzaniga and others (2024).
- Key comparisons:
  - The AI Preparedness Index focuses on preparedness to adopt AI technologies across four areas (human capital and labor market policies, digital infrastructure, innovation and economic integration, regulation and ethics).
  - The Skill Readiness Index is human-capital focused and based on granular new skills (AI-related and non-AI new IT and non-IT skills).
  - The Skill Imbalance Index compares projected supply and demand relative to the United States.
- Correlations reported:
  - Correlation coefficient between AI Preparedness Index and Skill Readiness Index: 0.61.
  - Demand and supply measures underlying the Skill Imbalance Index are also correlated with the AI Preparedness Index (Annex Figure 4.3, panels 3 and 4).
  - The Skill Imbalance Index itself (extended sample) is uncorrelated with the AI Preparedness Index (Annex Figure 4.3, panel 2).

---

*Source: sdnea2026001 - Annex Figure 2.1. Skill Appearance Time Lag (sdnea2026001 - Annex Figure 2.1. Skill Appearance Time Lag, IMF staff; sources within: Lightcast; IMF staff calculations; O*NET; International Federation of Robotics; OECD; World Bank World Development Indicators; Brookings; ILO; UNESCO UIS).*

### Annex V. Mergers and Acquisitions Analysis

### Annex V. Mergers and Acquisitions Analysis

### V.1. Data
- Time period covered: 2010–24.
- Four US firm-level sources combined:
  - (1) Capital IQ earnings and M&A call transcripts to identify talent-motivated deals.
  - (2) Compustat North America for fundamentals and identifiers.
  - (3) Lightcast job-posting microdata to measure firms’ demand for “new skills”.
  - (4) A database of completed M&As with effective completion year and acquirer identifiers.
- Transcript processing:
  - Transcripts are tokenized and searched for a curated dictionary of labor/talent terms (for example, “talent acquisition,” “hiring freeze,” “labor shortage,” “skills gap,” “acqui-hire”).
  - An indicator is set to one if any acquirer mentions these terms in the three years preceding deal completion.
- Lightcast postings:
  - De-duplicated at firm–occupation–location–date and mapped to SOC codes.
  - The new-skills index follows the definition discussed in Annex III, and is aggregated to firm-year shares.
- Firm linkage:
  - Firms are linked across data sets via gvkey/ticker and name-matching with manual resolution of large mergers.
- Measured outcomes:
  - (1) The change in the firm’s new-skills share in postings (percentage points).
  - (2) Innovation outcomes using patent value and forward citations (Kogan and others 2017, log values).

### V.2. Methodology
- Empirical framework: standard event-study approach.
- Specification (as presented):
  - y_it = ∑_{r=−3}^{4} β_r · rel_year_{r,i t} + α_i + γ_{n,t} + ε_it
- Notation and components:
  - y_it denotes the outcome variable for firm i in year t (can be the percentage of postings requiring new skills, the log real value of patents, or the log forward citations of patents).
  - rel_year_{r,i t} represents the number of years before or after the effective completion of the M&A deal.
  - α_i captures firm fixed effects.
  - γ_{n,t} captures industry-by-year fixed effects.
- Event window indicated: r from −3 to 4 around deal completion.
- Controls: firm fixed effects and industry-by-year fixed effects to account for time-invariant firm heterogeneity and industry trends.

### Annex Figure 4.3 — Relation with AI Preparedness Index (scatter-plot correlations)
- Panels and comparisons (each dot = country):
  - Panel 1: AI Preparedness Index vs Skill Readiness Index.
    - Correlation coefficient: 0.61
  - Panel 2: AI Preparedness Index vs Skill Imbalance Index (overall).
    - Correlation coefficient: 0.12
    - Note: country coverage in panels 2, 3, and 4 is the same as in Annex Figure 4.2.
  - Panel 3: AI Preparedness Index vs New Skill Demand (Ratio relative to USA).
    - Correlation coefficient: 0.86
  - Panel 4: AI Preparedness Index vs New Skill Supply (Ratio relative to USA).
    - Correlation coefficient: 0.61
- Data sources cited for figure: Lightcast, OECD, UNESCO UIS, ILO, PIACC, Cazzaniga and others (2024), and IMF staff calculations.
- Visual axis ticks and ranges shown in figure (preserved as presented):
  - AI preparedness index axis ticks include: 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85.
  - Skill readiness index axis ticks include: 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8.
  - New skill demand and supply panels present ratios relative to USA with axis tick labels including: 0, 0.5, 1, 1.5, 2 (as visually represented).

### Outcomes, interpretation, and analytic focus
- Primary firm-level outcome measures:
  - Percentage-point change in firm’s new-skills share in job postings.
  - Log real value of patents (patent value, log).
  - Log forward citations of patents (log).
- Identification strategy:
  - Variation in relative-year indicators around completed M&A deals identifies changes in outcomes attributable to acquisitions, conditional on firm and industry-year controls.
- Talent-motivated deals:
  - Identified via textual hits on labor/talent dictionary terms in acquirer call transcripts within the three years before deal completion, enabling classification of “acqui-hire” or talent-driven M&A activity.

*Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age — Annex V. Mergers and Acquisitions Analysis (Staff Discussion Note No. SDN/2026/001)*

---


_Source: https://www.imf.org/-/media/files/publications/sdn/2026/english/sdnea2026001.pdf_