## 1. AI Adoption Over Time and Across Countries

## Source details

**Canonical URL:** [1. AI Adoption Over Time and Across Countries](https://www.imf.org/-/media/files/publications/cr/2026/english/1gbrea2026002.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/cr/2026/english/1gbrea2026002.pdf.md)
- [Structured JSON version](/-/media/files/publications/cr/2026/english/1gbrea2026002.pdf.json)

---

### Overview and introduction
- AI technologies offer a significant opportunity to lift the UK's lackluster productivity growth, given the country's concentration in high-skill and knowledge-intensive services where AI shows greatest potential.
- Realizing AI gains depends on five enabling conditions: appropriate infrastructure (ICT and energy), financing for innovation, supportive regulation, skills, and resilience through trade openness.
- Economy-wide reforms (planning, energy markets, scale-up financing) are fundamental prerequisites and should be complemented by targeted AI-specific interventions in regulation, skills, and supply-chain resilience to address market failures.
- Model result summary:
  - Halving regulatory constraints and expanding the supply of AI-related skills jointly increase predicted output gains from AI by two thirds over the baseline.

### Potential of AI for UK productivity and growth
- Aggregate TFP gains from AI reported in the literature vary widely:
  - Medium-term impact on aggregate TFP level estimates range from below 1 to 35 percent over a decade.
- Two key determinants of aggregate AI gains in the model:
  - Share of tasks that can legally or technically be automated and performed with AI.
  - Average productivity gain and cost savings across AI-exposed tasks (including capital-labor substitution effects).
- Empirical and micro evidence cited:
  - AI could support the performance of between 23 to 80 percent of tasks.
  - Micro-level productivity improvements for AI-exposed tasks could range between 15 and 55 percent.
- UK positioning and comparative advantages:
  - Roughly 36 percent of the workforce hold a college degree.
  - Nearly two-thirds of the workforce are employed in AI-exposed occupations.
  - More than half of those workers are in high complementary occupations.
  - Misch and others (2025) estimate the UK’s TFP gains could range from 0.1 to 0.5 percent per year in the medium term.
  - UK sectors with comparative advantage: business services, finance, IT, pharma.
  - Labor cost measures place the UK above the OECD average, increasing gains from labor-saving AI adoption.

### Identified bottlenecks to AI development and diffusion
- Five enablers and associated bottlenecks:
  - Infrastructure (ICT and energy): barriers to construction, inadequate power density/energy integration in most datacenters, high electricity prices, grid connection delays up to 5 years in some sites.
  - Financing of innovation: shortages of long-term risk capital and scale-up financing; seed-stage attracts almost three-quarters of funding domestically while domestic share falls to less than one-third at scale-up; UK AI sector recipients raised only 40 percent of VC financing domestically in 2024.
  - Regulation: regulatory uncertainty (licensing, training, data privacy frameworks, copyright) constrains deployment; fragmentation across regulatory boundaries creates information frictions.
  - Skills: shortages of AI-relevant skills are the most important driver of cross-country adoption rates per variance decomposition; recruitment difficulties reported by more than 70 percent of companies (British Chamber of Commerce survey).
  - Trade openness and resilience: dependence on US-headquartered hyperscalers for compute, cloud infrastructure, and foundation models exposes the UK to supply decisions and policy changes elsewhere.
- Empirical and survey evidence:
  - Econometric analysis finds AI-relevant skills are the most important driver of cross-country variance in adoption rates, followed by infrastructure and openness.
  - Business surveys emphasize regulatory framework and access to financing as key factors shaping AI adoption and innovation.

### Model, calibration, and quantitative findings
- Modeling approach:
  - Endogenous technology adoption model calibrated to the UK; tasks within occupations have varying exposure to AI; firms adopt AI when AI capital costs compare favorably with wages and AI capability.
  - Model builds on task-based general equilibrium framework; occupational exposure measured using AI Occupational Exposure index mapped to UK household data.
- Baseline calibration assumptions:
  - AI adoption raises the aggregate capital share by approximately 10 percentage points in the long run.
  - AI reaches this long-run scale at twice the pace of routine automation, with steady state reached in 17 years rather than 34.
  - Cost-saving gains from AI per automated task set to 27 percent.
  - Under these assumptions, model produces cumulative TFP gains of 1.4 percent at the 10-year horizon.
- Commoditization assumption:
  - Baseline assumes rapid cost declines: cost of using AI models falling by roughly 90 percent per year; 99.7 percent decline in the cost efficiency frontier since March 2023; open-source models perform within 2 percent of frontier models on key benchmarks.
- Baseline quantitative projections:
  - Baseline cumulative gains relative to year 0:
    - TFP: Year 5 = 0.53, Year 10 = 1.40 (percent).
    - Output: Year 5 = 3.26, Year 10 = 7.96 (percent).
  - Baseline projected cumulative output gains: approximately 3 percent within the first 5 years; around 8 percent over a decade.
- Regulatory uncertainty and skill scenarios (modeled impacts over baseline):
  - Regulatory uncertainty modeled as holding back 20 percent of potential AI adoption on average.
  - Regulatory reform (removing 50 percent of estimated task-level regulatory constraints) modeled impacts:
    - Output: +1.24 percentage points (Year 5), +1.72 percentage points (Year 10) over baseline.
    - TFP: +0.22 percentage points (Year 5), +0.35 percentage points (Year 10) over baseline.
  - Skills expansion (halving AI wage premium via training, vocational pathways, immigration) modeled impacts:
    - Output: +0.83 percentage points (Year 5), +1.58 percentage points (Year 10) over baseline.
    - TFP: +0.15 percentage points (Year 5), +0.31 percentage points (Year 10) over baseline.
  - Combined effect:
    - Halving regulatory constraints and expanding AI-related skills jointly increase the output gains from AI by two thirds over the baseline.
- AI labor and wage dynamics:
  - AI labor share of adoption costs calibrated to 30 percent.
  - Current AI wage premium estimated at 14 percent in the UK.
  - Without policy intervention congestion markup grows from around 1.15× at year 5 to nearly 1.3× by year 10.
  - Modeled policy (halving the AI wage premium) implies increase in AI labor supply of approximately 27 percent at year 5 and 43 percent by year 10.

### Concentration risk, domestic alternatives, and insurance value
- Disruption modeling:
  - Restricted access to foreign frontier AI modeled as a markup on adoption costs with markup levels τ = 15 percent (moderate), τ = 25 percent (intermediate/central), τ = 40 percent (severe).
  - Domestic alternative assumed to deliver 90 percent of the benefits of the frontier model conditional on adoption.
- Central scenario (τ = 25 percent) impacts:
  - Baseline TFP gain at year 5 falls from 0.53pp to 0.08pp under restricted access.
  - Baseline output gain at year 5 falls from 3.26pp to 0.54pp—equivalent to approximately £82 billion per year in lost output relative to the baseline.
  - A domestic alternative recovers roughly half of this loss—approximately £40 billion per year.
- Severe disruption (τ = 40 percent):
  - Near-term productivity gain from AI is effectively lost, but domestic option still recovers approximately £57 billion per year.
- Policy implication:
  - Investment in domestic AI capacity (infrastructure, skills, incentives) serves as insurance against supply-side disruptions; the insurance value of modest domestic investments can significantly exceed current government commitments.
  - The government’s current AI commitment reported as over £2 billion is characterized as modest relative to potential benefits recovered under disruption scenarios.

### Identified constraints and UK-specific empirical facts
- AI adoption rates and patterns:
  - As of mid-2025, business surveys indicate the AI adoption rate in the UK equaled 34 percent (EU average 37 percent; US rate 36 percent).
  - UK adoption nearly doubled from 2023 to 2025.
  - Sectoral adoption: above-average in data processing, image processing, robotics; average in visual content creation and autonomous vehicles; below-average in text generation using LLMs.
- ICT and energy infrastructure:
  - UK datacenter capacity sizeable versus peers but many facilities lack necessary power density and energy integration for AI training/inference (DSIT and UKRI, 2025).
  - UK lags in AI supercomputer capacity and faces the highest electricity prices in the G7.
  - Real investment in AI-related physical and digital infrastructure has grown slower than peers; new orders stagnant since COVID-19 with some recent public sector activity.
  - Grid connection delays are a bottleneck; some sites face waits of up to 5 years.
- Financing for innovation:
  - Gross domestic expenditure on R&D = 2.75 percent of GDP in 2022 (OECD average 2.72 percent).
  - UK VC investment represented 0.17 percent of GDP in 2024, supporting more than 9,100 businesses.
  - Seed-stage companies attract almost three-quarters of funding domestically; at scale-up stage domestic share falls to less than one-third (TBI 2025).
  - In 2024, UK recipients of VC financing in the AI sector raised only 40 percent of such financing domestically, while 80 percent of UK’s VC financing was channeled abroad.
  - Restructuring costs estimated at 19 months of salary per employee in the UK, compared with 7.4 months in the United States and under 3 months in Denmark and Switzerland.
  - Three in four ICT founders identify access to growth capital as their primary obstacle (TechNation 2025).
- Regulation and uncertainty:
  - Since 2023 the UK relied on five non-statutory cross-sector principles implemented by existing regulators (FCA, ICO, CMA, Ofcom, and others).
  - Nearly 60 percent of government-commissioned survey respondents cited regulatory and policy obstacles as a barrier to AI adoption (McLean and Smith 2025).
  - Copyright rules and permissible use of data for training/deploying AI models are salient open issues.
- Skills and labor market:
  - UK scores highly on human capital dimension of IMF AI Preparedness Index, but ranks below average on perceived skillset of graduates.
  - Prevalence of workers with AI engineering skills lags the US substantially in financial services, professional services, and ICT sectors.
  - More than 70 percent of UK companies surveyed by the British Chamber of Commerce report recruitment difficulties (2025).
  - One in three UK founders of Tech firms says availability of talent is their biggest barrier (TechNation 2025).
  - Wage premium for AI-user skills is much higher in the UK than in the US (Jaumotte and others, 2026).
  - Net migration inflows of cross-country workers with AI skills have declined sharply in recent years.
- Trade and export strengths:
  - Digital trade represents more than half of the UK’s exports, twice the OECD and EU averages.
  - Digital trade exports have grown at nearly three times the rate of other UK exports.
  - UK sits in Tier 1 of the US AI Diffusion framework and faces no procurement restrictions on advanced GPUs.

### Policy implications and priorities
- Economy-wide reforms (fundamental prerequisites):
  - Translate planning reforms into materially shorter approval timelines for critical infrastructure.
  - Continue reforms to renewables permitting, grid connection processes, and electricity market design to narrow the UK's energy cost gap with peers.
  - Mobilize domestic capital for high-growth, innovation-intensive firms: consolidate and coordinate public support schemes; scale up public-finance institutions while preserving co-investment discipline; consider making R&D tax credits payable in cash for loss-making startups and reform capital gains tax treatment of capital losses and reinvestment.
- Targeted AI-specific interventions:
  - Provide regulatory clarity (e.g., on copyright, liability, conditions for AI-assisted decision-making) and operational guidance to reduce adoption uncertainty.
  - Expand supply of AI-relevant skills: streamline immigration pathways in the near term; expand postgraduate-level and employer co-investment training programs to address congestion and poaching externalities.
  - Improve access to scale-up financing and market-making interventions to foster a domestic AI ecosystem and capture insurance value against foreign model access disruptions.
  - Strengthen ICT and energy infrastructure and address energy pricing/infrastructure barriers relevant for data center deployment and continuous AI operation.
  - Preserve trade openness and manage resilience risks: diversify chip supply chains; ensure access to high-quality training data (e.g., National Data Library).
- Equity and redistribution:
  - Complement supply-side interventions with redistributive policies to address transitional labor-market disruptions, displacement risks, and rising wealth inequality from higher capital returns.
- Implementation emphasis:
  - Effective delivery requires scaling institutional capacity, coordination across authorities and investors, and systematic monitoring to translate reforms into shorter timelines and predictable outcomes that encourage private investment.

*Source: IMF staff chapter "1. AI Adoption Over Time and Across Countries" (June 24, 2026).*

### 1. AI Adoption Over Time and Across Countries ________________________________________ 9

### 1. AI Adoption Over Time and Across Countries

### Overview and introduction
- AI technologies offer a significant opportunity to lift the UK's lackluster productivity growth, given the country's concentration in high-skill and knowledge-intensive services where AI shows greatest potential.
- Realizing AI gains depends on five enabling conditions: appropriate infrastructure (ICT and energy), financing for innovation, supportive regulation, skills, and resilience through trade openness.
- The paper finds that economy-wide reforms (planning, energy markets, scale-up financing) are fundamental prerequisites, and should be complemented by targeted AI-specific interventions in regulation, skills, and supply-chain resilience to address market failures.
- Using a model of endogenous technology adoption calibrated to the UK, halving regulatory constraints and expanding the supply of AI-related skills could jointly increase the output gains from AI by two thirds over the baseline.

### Potential of AI for UK productivity and growth
- Aggregate TFP gains from AI vary widely in the literature; studies estimate a medium-term impact on aggregate TFP level ranging from below 1 to 35 percent over a decade.
- Two key model determinants of aggregate AI gains:
  - Share of tasks that can legally or technically be automated and performed with AI.
  - Average productivity gain and cost savings across AI-exposed tasks (including capital-labor substitution effects).
- Empirical and micro studies cited:
  - AI could support the performance of between 23 to 80 percent of tasks.
  - Micro-level productivity improvements for AI-exposed tasks could range between 15 and 55 percent.
- The UK’s positioning:
  - Roughly 36 percent of the workforce hold a college degree.
  - Nearly two-thirds of the workforce are employed in AI-exposed occupations.
  - More than half of those workers are in high complementary occupations (where AI improves productivity of existing workers rather than replaces them).
  - Misch and others (2025) estimate the UK’s TFP gains could range from 0.1 to 0.5 percent per year in the medium term, above the average of European countries and on par with estimated US gains.
- Comparative advantages for the UK:
  - Prevalence of occupations in business services, finance, IT, pharma—sectors where generative AI and LLMs show large micro-level performance gains.
  - Labor cost measures place the UK above the OECD average, increasing the gains from labor-saving AI adoption.

### Identified bottlenecks to AI development and diffusion
- The analysis focuses on five enablers and identifies bottlenecks in each area for the UK:
  - Infrastructure (ICT and energy): barriers to construction and energy costs can impede data center and ICT deployment.
  - Financing of innovation: constraints on business scale-up and access to scale-up financing limit domestic AI ecosystem growth.
  - Regulation: regulatory uncertainty (licensing, training, data privacy frameworks) can constrain deployment of AI across occupations and sectors.
  - Skills: shortages of AI-relevant skills are identified as a key constraint and the most important driver of cross-country adoption rates in variance decomposition.
  - Trade openness and resilience: risks from trade and geo-fragmentation that could disrupt sourcing of critical inputs and foreign model access.
- Econometric and survey evidence:
  - Econometric analysis of cross-country adoption rates shows AI-relevant skills are the most important driver, followed by infrastructure and openness.
  - Surveys of businesses emphasize regulatory framework and access to financing as key factors shaping AI adoption and innovation.

### Model and quantitative findings (selected)
- Model-calibrated findings to the UK:
  - Halving regulatory constraints and expanding the supply of AI-related skills jointly increase predicted output gains from AI by two thirds relative to the baseline.
- Additional qualitative quantitative points from the literature and micro evidence:
  - AI-exposed task share: between 23 to 80 percent.
  - Micro-level productivity gain per AI-exposed task: between 15 and 55 percent.
  - Literature range for medium-term aggregate TFP level impact: below 1 to 35 percent over a decade.
  - UK medium-term TFP gains estimate: 0.1 to 0.5 percent per year (Misch and others, 2025).

### Policy implications and priorities
- Economy-wide reforms (planning, energy markets, scale-up financing) are a fundamental prerequisite for AI development and adoption.
- Targeted AI-specific interventions should complement economy-wide reforms to address market failures:
  - Clarify and stabilize regulatory frameworks affecting AI deployment (licensing, training requirements, data privacy) to reduce uncertainty and constrain fewer AI-exposed tasks.
  - Expand the supply of AI-related skills—identified as the most critical driver of adoption—to realize complementarities and enable productivity improvements.
  - Improve access to scale-up financing to foster domestic AI ecosystem development and capture insurance value against foreign model access disruptions.
  - Strengthen ICT and energy infrastructure and address energy pricing/infrastructure barriers relevant for data center deployment and continuous AI operation.
  - Preserve trade openness and manage resilience risks to mitigate potential large output losses from disruptions to foreign model access.

*Source: IMF staff chapter "1. AI Adoption Over Time and Across Countries" (June 24, 2026).*

### 13.  We use this simple framework to identify possible bottlenecks impeding the

### 13.  We use this simple framework to identify possible bottlenecks impeding the realization of AI productivity gains.

### Framework overview
- The analysis examines trends, benchmarks the UK versus peers across indexes, studies responses from business surveys, and leverages independent assessments.
- The framework specifically considers five dimensions that could impede AI productivity gains:
  - (1) whether appropriate infrastructure is in place to support AI deployment at scale;
  - (2) whether financing is available for product and business-model innovation—especially for high-growth-potential firms;
  - (3) whether regulation provides an enabling and predictable environment facilitating investment in organizational and process transformations while safeguarding trust;
  - (4) whether a critical mass of skilled workers are available to drive adoption and innovation, ensuring effective use within firms; and
  - (5) whether openness to trade supports access to frontier inputs (on the import side) and expands market opportunities for UK exporters of AI-enabled goods and services, thereby increasing expected returns to effective adoption.

### Key analytical foci
- Infrastructure: assess readiness to support AI deployment at scale.
- Financing: evaluate availability of capital for product and business-model innovation, with emphasis on high-growth-potential firms.
- Regulation: gauge whether the regulatory environment is enabling, predictable, and trust-preserving for investments in organizational and process transformations.
- Skills: measure whether a critical mass of skilled workers exists to drive adoption, innovation, and effective firm-level use.
- Trade openness: determine whether import access to frontier inputs and export market opportunities for AI-enabled goods and services are sufficient to raise expected returns to adoption.

*Source: https://www.imf.org/-/media/files/publications/cr/2026/english/1gbrea2026002.pdf*

### Box 1. United Kingdom: AI Adoption Over Time and Across Countries

### Box 1. United Kingdom: AI Adoption Over Time and Across Countries

### AI adoption rates and patterns
- As of mid-2025, business surveys indicate the AI adoption rate in the UK equaled 34 percent, close to the EU average of 37 percent and the US rate of 36 percent.
- The UK’s AI adoption rates are growing fast, having almost doubled from 2023 to 2025.
- UK businesses report above-average adoption in:
  - data processing,
  - image processing,
  - robotics.
- UK businesses report average adoption in:
  - visual content creation,
  - autonomous vehicles.
- UK businesses report below-average adoption in:
  - text generation using LLMs.
- Business surveys and worker statistics consistently show UK adoption rates close to the US and the European average rates.

### ICT and energy infrastructure constraints
- UK datacenter capacity is sizeable relative to comparably sized European peers, but:
  - most existing facilities are designed for general-purpose enterprise workloads and lack the power density, energy integration, and technical specifications needed for AI training and inference (DSIT and UKRI, 2025).
- The UK lags in AI supercomputer capacity and faces the highest electricity prices in the G7.
- Real investment in AI-related physical and digital infrastructure has grown at a slower pace than peers.
- New orders for the construction of public and private infrastructure have been stagnant since the COVID-19 pandemic, with signs of increased activity more recently, especially in the public sector.
- Grid connection delays are a bottleneck for datacenters; some sites face waits of up to 5 years.
- Electricity prices are a significant determinant of AI adoption rates; elevated UK prices are a headwind for AI development (Haag 2025).
- Businesses can outsource computing/datacenter services to partly circumvent infrastructure constraints; recent planning reforms have begun to address acute bottlenecks.

### Financing for innovation and the “missing middle”
- Gross domestic expenditure on R&D stood at 2.75 percent of GDP in 2022, marginally above the OECD average of 2.72 percent.
- UK VC investment represented 0.17 percent of GDP in 2024, supporting more than 9,100 businesses.
- Most recent VC funding has been channeled into AI businesses; the number of newly funded AI startups and total private investment for AI startups has increased rapidly.
- Structural weakness: high-growth domestic firms struggle to scale domestically due to a shortage of long-term risk capital:
  - Seed-stage companies attract almost three-quarters of funding domestically; at scale-up stage domestic share falls to less than one-third (TBI 2025).
  - In 2024 in the AI sector, UK recipients of VC financing raised only 40 percent of such financing domestically, while 80 percent of UK’s VC financing was channeled abroad, mostly to US startups.
  - US and EU startups’ VC financing is more likely to come from domestic sources (70 and 50 percent respectively); US and EU VC funds are less likely to be invested abroad (15 and 55 percent respectively).
- UK firms face restructuring costs estimated at 19 months of salary per employee, compared with 7.4 months in the United States and under 3 months in Denmark and Switzerland—this structural rigidity compounds investor reluctance at the growth stage.
- Financing constraints are frequently cited as a top barrier to digital technology adoption, especially for SMEs and young ICT firms; three in four ICT founders identify access to growth capital as their primary obstacle (TechNation 2025).

### Regulatory framework and uncertainty
- Since 2023 the UK has relied on existing laws and regulatory guidance based on five non-statutory cross-sector principles: safety, transparency, fairness, accountability, and contestability—implemented by existing regulators (FCA, ICO, CMA, Ofcom, and others).
- The principles-based, sector-led approach has avoided the compliance costs associated with the EU AI Act.
- Nevertheless, regulatory uncertainty can dampen effective AI adoption:
  - Regulation never ranks as the main barrier overall but features prominently, especially for larger firms and in regulated sectors.
  - Nearly 60 percent of a government-commissioned survey respondents cited regulatory and policy obstacles as a barrier to AI adoption (McLean and Smith 2025).
  - Copyright rules and unclear permissible use of data for training/deploying AI models are salient issues.
  - Fragmentation across existing regulatory boundaries creates an information friction—firms cannot easily determine what compliance requires when multiple regimes apply.

### Skills and labor market constraints
- The UK scores highly on the human capital dimension of the IMF AI Preparedness Index, reflecting high average years of schooling and good digital skills among the active population.
- The UK ranks below average on the perceived skillset of graduates.
- Prevalence of workers with AI engineering skills lags the US substantially in financial services, professional services, and ICT sectors.
- As of 2025, more than 70 percent of UK companies surveyed by the British Chamber of Commerce report recruitment difficulties; ICT is among the most affected sectors.
- Recruitment difficulties are most pronounced for professional and managerial positions (engineers, lawyers, finance professionals, HR managers).
- One in three UK founders of Tech firms says availability of talent is their biggest barrier to growth (TechNation 2025).
- Wage premium for AI-user skills is much higher in the UK than in the US, signaling shortages (Jaumotte and others, 2026).
- High-frequency indicators show skill shortages have increased over the past decade and are at historically elevated levels.
- Net migration inflows of cross-country workers with AI skills exist but have declined sharply in recent years.

### Trade openness, concentration risk, and export strengths
- Digital trade represents more than half of the UK’s exports, twice the OECD and EU averages, underpinned by comparative advantage in digitally deliverable services (financial and professional services).
- Digital trade exports, including computer services, have grown at nearly three times the rate of other UK exports.
- The UK sits in Tier 1 of the US AI Diffusion framework and faces no procurement restrictions on advanced GPUs.
- The UK’s AI ecosystem depends heavily on US-headquartered hyperscalers for compute, cloud infrastructure, and foundation models, exposing it to supply decisions and policy changes elsewhere.

### Policy responses: economy-wide and targeted measures
- Both economy-wide reforms and AI-specific interventions are needed to alleviate bottlenecks.
- Economy-wide policies (examples and objectives):
  - Infrastructure / Planning reform:
    - 2025 Planning and Infrastructure Act aims to speed delivery of nationally significant infrastructure, including digital and energy infrastructure.
    - Datacenters designated as critical national infrastructure in September 2024; can be opted into the Nationally Significant Infrastructure Projects regime.
    - AI Growth Zones under the AI Opportunities Action Plan provide fast-track pathways with pre-cleared planning approval and access to clean power.
    - Key implementation challenges: scale institutional capacity (local planning authorities), strengthen coordination between planning authorities, network operators, and investors, and systematically monitor approval times and project completions.
  - Energy and grid reforms:
    - Ofgem reformed the grid connection regime in April 2025 to a “first ready and needed, first connected” process; connections queue has been reduced substantially.
    - Clean Power 2030 Action Plan and establishment of Great British Energy on a statutory footing aim to expand domestic clean generation and improve long-term price predictability.
    - Government advanced nuclear capacity plans including final investment decision for Sizewell C.
    - Near-term challenge: reforms take time to lower prices while AI electricity demand expands rapidly; immediate focus needed on accelerating permitting of new renewables, connections reform, network buildout, and electricity market design.
  - Innovation financing:
    - Pension fund consolidation under the Mansion House Accord to build scale for domestic investment.
    - Listing and market reforms to lower frictions and improve exit opportunities.
    - British Business Bank’s expanded Growth Guarantee Scheme and the National Wealth Fund aim to crowd in private capital.
    - AI-specific financings include the £500m Sovereign AI Unit and the £100m advance market commitment for UK AI hardware startups.
    - Observed gaps: the Growth Guarantee Scheme is several times smaller, relative to GDP, than comparable schemes; the National Wealth Fund's £27.8 billion capital over nine years is modest relative to peer policy banks.
    - Potential further reforms proposed:
      - consolidate fragmented public support schemes and improve coordination so they cover larger financing rounds;
      - scale up public-finance institutions (BBB and NWF) over time to peer footprints while preserving co-investment discipline;
      - make R&D tax credits payable in cash for loss-making startups and reform capital gains tax treatment of capital losses and reinvestment.
- Targeted AI-specific interventions:
  - Address market failures specific to AI (regulatory uncertainty on permissible use, AI-specific skills shortages, access to frontier models).
  - Near-term temporary interventions may be warranted given competitor economies’ rapid investments and a narrow window to shape competitive position.
  - Government’s AI strategy commits significant resources toward these challenges.

### Key policy implications and priorities
- Economy-wide reforms are essential pre-requisites to address bottlenecks (planning, energy costs, financing) and will deliver broad productivity gains beyond AI.
- Targeted AI interventions complement economy-wide reforms where market failures are AI-specific and where speed matters (skills, regulatory clarity, access to frontier infrastructure and models).
- Effective delivery, scaling institutional capacity, coordination across authorities and investors, and systematic monitoring are critical to translate reforms into materially shorter timelines and predictable outcomes that encourage private investment.

*Source: Box 1. United Kingdom: AI Adoption Over Time and Across Countries (IMF).*

### Box 2. United Kingdom: The Government’s AI Strategy

### Box 2. United Kingdom: The Government’s AI Strategy

### Strategic orientation and progress
- AI is a strategic priority in the UK's 10-year industrial strategy (DBT and DSIT, 2025) and was designated one of the three "Big Choices" by the Chancellor in her March 2026 Mais lecture.
- Anchoring documents and programs: January 2025 AI Opportunities Action Plan; UK Compute Roadmap; Digital & Technologies Sector Plan; AI Growth zones program; AI for Science Strategy.
- The AI strategy complements the authorities' Growth Mission, which focuses on higher public investment, planning reform, mobilizing patient capital, and expanding labor supply (IMF UK 2026 Staff report).
- As of early 2026, the government reports had met 38 of the 50 actions set out in the Action Plan, with progress tracked through an online dashboard (DSIT 2026).

### Specific commitments (enabling conditions)
- Skills
  - Committed £200m to upskill the workforce through AI courses.
  - Expand the Growth and Skills Levy for employer-sponsored short AI and digital courses.
  - Develop new AI scholarship programs, including the £17.6m Spärck AI 100 scholarships.
  - Launch the £160m TechFirst program to support over 4,000 graduates and researchers.
  - Establish a Global Talent Taskforce to attract top researchers.
- Infrastructure and compute
  - Plans to invest £1.75bn to build public compute capacity through the AI Research Resource program, in partnership with the University of Edinburgh.
  - Develop AI Growth Zones (AIGZs) for accelerated data center buildout with streamlined planning approval and access to clean power.
  - Additional investment of £100m for a National Data Library to house high-value public sector datasets and make them AI-ready.
- Innovation financing
  - A £500m Sovereign AI Unit has been set up to allocate access to public compute, provide early-stage financing to startups alongside the British Business Bank, and finance the buildup of AI infrastructure.
  - An "advance market commitment" of £100m to act as a first customer for UK startups building AI hardware products.
- Regulation and safety
  - Text and data mining regime reformed.
  - The Data Use and Access Act (DUAA) was passed in 2025.
  - Increased funding for regulators to scale up their AI capabilities.
  - Government provided £240m for the AI Security Institute to expand frontier model testing, foundational safety, and societal resilience.
- Public sector adoption
  - Government committed to make the public sector a leader in AI adoption, supporting diffusion of AI tools across central government and public services.

### Regulatory uncertainty: constraints and implications
- The 2025 Data Use and Access Act (DUAA) simplifies compliance and enables responsible data sharing, but important questions remain open:
  - Application of UK copyright law to AI model training is under ongoing review; DUAA required the Secretary of State to publish an impact assessment and report by March 2026 (DUAA, ss135–137).
  - The government indicated it will not introduce copyright reforms until confident they meet objectives (DSIT, 2026).
  - Interim legal uncertainty may discourage firms from training AI models using UK-held data.
  - DUAA introduces changes to the UK's data protection framework that diverge from EU norms; the EU renewed the UK's data adequacy status to 2031, but longer-term implications depend on divergence between regimes.
- Industry demand for greater regulatory clarity is significant (DSIT, 2024); firms in regulated sectors report regulators may lack resources to keep pace with AI developments (CBI 2025), possibly slowing deployment even where no explicit prohibition exists.
- Policy options to enhance clarity and adoption:
  - Provide cross-cutting and operational depth to the regulatory framework with assurance tools demonstrating compliance with AI governance principles and concrete guidance developed iteratively with industry and stakeholders (model cited: Singapore).
  - Accelerate resolution of outstanding legal questions on copyright, liability for AI-generated outputs, and conditions for AI-assisted decision-making in healthcare, financial services, and education.
  - Improve coordination and accountability across the decentralized regulatory model by building on existing requests for regulators to report on strategic approaches to AI (DSIT, 2024).
  - Provide additional resources and technical capacity to sectoral regulators.
  - Consider maintaining compatibility with EU and US regulatory frameworks where possible to avoid trade frictions and preserve data adequacy status.

### Skills bottlenecks and policy responses
- Market features two reinforcing externalities:
  - Congestion: broader AI adoption raises demand for specialists, bidding up wages and raising adoption costs, which holds back further adoption.
  - Poaching: trained workers can be hired away by competitors, reducing firms' incentives to invest in training and slowing aggregate supply growth of specialists.
- Current policy targeting and gaps
  - Shortages most acute for professional and managerial roles—the AI engineers, data scientists, and implementation specialists needed for deployment.
  - Government focus on short introductory courses is welcome but may be insufficient to build deep specialist capabilities at scale.
  - Programs to attract world-class researchers target a narrow cohort.
  - Growth and Skills Levy reform expands employer-sponsored training but has defunded Master's-level apprenticeships for learners aged 22 and over, which may work against acquisition of AI-specific skills.
  - Need for wider access to postgraduate-level programs aimed at deepening specialist skills.
- Recommended complementary approaches
  - Near term: expand access to global AI talent via immigration; streamline visa pathways for AI-relevant occupations in response to the sharp recent decline in inflows of workers with AI-relevant skills.
  - Medium term: structured employer co-investment in AI training to address poaching externalities by narrowing the gap between private returns to training firms and broader social returns.

### Resilience: concentration risk and coordination failures
- Concentration of frontier AI development in a small number of foreign firms creates vulnerability; current access is broadly competitive but could be disrupted by export restrictions, supply chain bottlenecks, or market power.
- Coordination failure risk: no individual firm has incentive to invest in domestic alternatives that are commercially unprofitable today but collectively valuable under disruption.
- Government responses already initiated:
  - Investment in sovereign AI capacity: commitments include the £1.75bn AI Research Resource for public compute and AI Growth Zones for data center buildout.
- Policy focus recommended
  - Build conditions for a domestic AI ecosystem that could scale rapidly if needed (rather than replicating frontier capabilities directly): invest in compute infrastructure and energy capacity, diversify chip supply chains, ensure access to high-quality training data via the National Data Library, and maintain regulatory compatibility with like-minded partners.
  - Modeling in Section E indicates modest investments in domestic resilience could recover a substantial share of output lost under disruption scenarios and that the insurance value of such investments significantly exceeds the government's current AI commitment of over £2 billion.

### Quantifying benefits: model setup and baseline assumptions
- Modeling approach
  - Uses a model of endogenous technology adoption calibrated to the UK economy (Rockall and others, 2025); occupations consist of tasks with varying degrees of exposure to AI.
  - Firms decide whether to adopt AI for tasks by comparing cost of AI capital with wages and AI capability; tasks with largest cost savings relative to labor are automated first.
  - Model builds on task-based general equilibrium framework of Moll and others (2022), extended to study AI diffusion; occupational exposure measured with AI Occupational Exposure index of Felten and others (2021), mapped to UK household data from the Wealth and Assets Survey.
- Baseline calibration assumptions
  - AI adoption raises the aggregate capital share by approximately 10 percentage points in the long run (anchored to increase observed in UK routine automation between 1980 and 2014).
  - AI reaches this long-run scale at twice the pace of routine automation, with steady state reached in 17 years rather than 34.
  - Cost-saving gains from AI per automated task set to 27 percent (average of micro-empirical estimates).
  - Under these assumptions, model produces cumulative TFP gains of 1.4 percent at the 10-year horizon.
- Commoditization trajectory assumption
  - Assumes cost of AI technologies continues to fall rapidly; evidence cited: cost of using AI models falling by roughly 90 percent per year, with a 99.7 percent decline in the cost efficiency frontier since March 2023 (Appenzeller 2024).
  - Open-source models now perform within 2 percent of frontier models on key benchmarks (Stanford HAI 2025).
  - Baseline assumes competitive dynamics lead to continued partial transmission of cost declines to users; final scenario relaxes this assumption.

### Modeling results and scenario findings
- Baseline productivity impacts (under commoditization and calibrated assumptions)
  - Model projects cumulative output gains of approximately 3 percent within the first 5 years, rising to around 8 percent over a decade—substantially larger than the 1.4 percent TFP gain at the 10-year horizon (difference reflects capital deepening).
- Regulatory uncertainty scenario
  - Regulatory uncertainty modeled as holding back 20 percent of potential AI adoption on average, calibrated to executives reporting significant regulatory limitations (PwC Global Compliance Survey 2025).
  - Regulatory exposure index constructed from O*NET work context measures (responsibility for outcomes, others’ health and safety, consequences of error, and freedom and frequency of decisions).
  - Many occupations with greatest AI potential (managers, healthcare and education providers, legal practitioners, financial service providers) also face greatest regulatory uncertainty.
  - Regulatory reform modeled as resolving half of this constraint, representing measures that clarify permissible uses and strengthen legal protections; the remaining half reflects substantive safety and accountability requirements that should remain.
  - Under this scenario, providing regulatory clarity would boost output by 1.24 percentage points (TFP impact reported in source beyond this excerpt).

### Key numeric figures (as reported)
- 10-year industrial strategy (DBT and DSIT, 2025); Mais lecture March 2026.
- Government met 38 of the 50 actions set out in the Action Plan (early 2026).
- Skills commitments: £200m; £17.6m Spärck AI 100 scholarships; £160m TechFirst program to support over 4,000 graduates and researchers.
- Infrastructure and data commitments: £1.75bn AI Research Resource; £100m National Data Library.
- Innovation financing: £500m Sovereign AI Unit; £100m advance market commitment.
- Regulation and safety: DUAA passed in 2025; £240m for the AI Security Institute.
- Modeling calibration: aggregate capital share increase of approximately 10 percentage points; steady state reached in 17 years; cost-saving gains from AI set to 27 percent; cumulative TFP gains of 1.4 percent at the 10-year horizon.
- Modeled constraint: regulatory uncertainty holds back 20 percent of potential AI adoption on average.
- Reported historical cost decline: cost of using AI models falling by roughly 90 percent per year; 99.7 percent decline in the cost efficiency frontier since March 2023.
- Open-source models perform within 2 percent of frontier models on key benchmarks.
- Baseline projected cumulative output gains: approximately 3 percent within first 5 years; around 8 percent over a decade.
- Government's current AI commitment reported as over £2 billion.

*Source: Box 2. United Kingdom: The Government’s AI Strategy (extracted from the source content provided).*

### 0.22 percentage points within 5 years, growing to 1.72 and 0.35 percentage points respectively by

### 1gbrea2026002 - 0.22 percentage points within 5 years, growing to 1.72 and 0.35 percentage points respectively by

### Productivity impacts of regulatory reform
- Regulatory reform that removes 50 percent of estimated task-level regulatory constraints on AI use is modeled to:
  - Add 1.24 percentage points to output at year 5 and 1.72 percentage points at year 10 (gains over baseline).
  - Add 0.22 percentage points to TFP at year 5 and 0.35 percentage points at year 10 (gains over baseline).
- Rationale:
  - As technology costs fall and adoption expands, regulatory uncertainty becomes the dominant barrier to adoption, so resolving regulatory uncertainty yields larger productivity gains over the medium and long run.
  - Cost-based interventions (adoption subsidies or tax breaks) that lower adoption costs deliver comparable near-term productivity gains but substantially smaller gains at longer horizons because their value is eroded as AI adoption costs decline.

### Skill shortages, labor externalities, and impacts of expanding AI talent supply
- Decomposition of adoption costs:
  - Adoption costs = technology component (falls over time) + labor component (rises over time).
  - AI labor share of adoption costs calibrated to 30 percent, consistent with lower bound estimates that 29–49 percent of frontier AI costs are attributable to labor.
- AI wage premium and congestion:
  - Current AI wage premium estimated at 14 percent in the UK (PwC UK AI Jobs Barometer, 2024).
  - Without policy intervention AI labor supply does not expand to meet rising demand; congestion markup grows from around 1.15× at year 5 to nearly 1.3× by year 10.
  - AI labor's share of total costs rises from around 39 percent to 46 percent between year 5 and year 10.
- Modeled policy: halving the AI wage premium through expanded training, vocational pathways, and immigration.
  - Output gains from halving the premium: 0.83 percentage points within 5 years, growing to 1.58 percentage points by year 10 (gain over baseline).
  - TFP gains from halving the premium: 0.15 percentage points at year 5, rising to 0.31 percentage points at year 10.
  - Implied increase in AI labor supply: approximately 27 percent at year 5, rising to 43 percent by year 10.
- Distributional and transitional considerations:
  - The analysis does not capture transitional labor market disruptions; other work using the same model framework suggests higher output leads to higher wages overall but many workers may see lower wages due to displacement.
  - Wealth inequality is likely to increase through higher capital returns driving up risky asset prices, highlighting the need for complementary redistributive policies.

### Complementarity of regulatory reform and skills investment
- Regulatory reform and skills investment deliver similar headline gains but operate through different channels and are complementary:
  - Regulatory reform expands the set of tasks for which firms are willing to deploy AI, unlocking productivity gains beyond what cheaper adoption alone can achieve.
  - Skills investment alleviates a self-reinforcing bottleneck that grows over time.
- Combined effect:
  - Pursued together, the two policies could jointly increase the output gains from AI by two thirds over the baseline (relative complementarity shown in modeled scenarios).
- Shared assumption:
  - Both policy scenarios assume UK firms continue to enjoy access to frontier AI models on competitive terms; the next subsection examines risk if that assumption fails.

### Concentration risk, domestic alternatives, and insurance value of domestic AI capacity
- Modeled disruption scenarios:
  - Restricted access to foreign frontier AI modeled as a markup on adoption costs with three markup levels: τ = 15 percent (moderate), τ = 25 percent (intermediate/central), τ = 40 percent (severe).
  - Domestic alternative is assumed to deliver 90 percent of the benefits of the frontier model conditional on adoption.
- Impact under central scenario (τ = 25 percent):
  - Baseline TFP gain at year 5 falls from 0.53pp to 0.08pp under restricted access.
  - Baseline output gain at year 5 falls from 3.26pp to 0.54pp under restricted access—equivalent to approximately £82 billion per year in lost output relative to the baseline.
  - A domestic alternative recovers roughly half of this loss—approximately £40 billion per year.
- Impact under severe disruption (τ = 40 percent):
  - Near-term productivity gain from AI is effectively lost, but domestic option still recovers approximately £57 billion per year.
- Policy implications:
  - Investment in the UK’s domestic AI capacity (infrastructure, skills, incentives) serves as insurance against supply-side disruptions and can substantially reduce exposure.
  - Relevant support policies include investment in data center capacity and energy infrastructure, diversified chip supply chains, subsidized compute for model development, and facilitated access to training data.
  - The government’s current AI investment commitment of over £2 billion is characterized as modest relative to potential benefits recovered under disruption scenarios.

### Key quantitative findings and calibration notes
- Baseline cumulative gains relative to year 0:
  - TFP: Year 5 = 0.53, Year 10 = 1.40 (percent).
  - Output: Year 5 = 3.26, Year 10 = 7.96 (percent).
- Policy gains over baseline (percentage points):
  - Regulatory reform: TFP +0.22 (Year 5), +0.35 (Year 10); Output +1.24 (Year 5), +1.72 (Year 10).
  - Skills investment: TFP +0.15 (Year 5), +0.31 (Year 10); Output +0.83 (Year 5), +1.58 (Year 10).
- Policy gains over baseline (percent of baseline):
  - Regulatory reform: TFP +41% (Year 5), +25% (Year 10); Output +38% (Year 5), +22% (Year 10).
  - Skills investment: TFP +28% (Year 5), +22% (Year 10); Output +25% (Year 5), +20% (Year 10).
- Calibration and technical details:
  - AI labor share of adoption costs calibrated to 30 percent.
  - AI wage premium currently estimated at 14 percent in the UK.
  - Assumption in regulatory scenario: regulation holds back 20 percent of potential AI adoption; regulatory reform removes 50 percent of estimated task-level regulatory constraints.
  - Domestic alternative effectiveness assumed at 90 percent of frontier benefits conditional on adoption.

### Policy recommendations and priorities from the analysis
- Economy-wide reforms (Growth Mission priorities) to enable AI gains:
  - Ensure recent planning reforms translate into materially shorter approval timelines for critical infrastructure.
  - Continue reforms to renewables permitting, grid connection processes, and electricity market design to narrow the UK's energy cost gap with peers.
  - Mobilize domestic capital for high-growth, innovation-intensive firms—better coordination and consolidation of public support schemes and fiscal incentives tailored to financially constrained startups.
- Targeted AI-specific interventions:
  - Expand supply of AI-relevant skills: streamline immigration pathways in the near term; ramp up education and training programs; expand university programs, sponsored vocational training, and immigration pathways.
  - Provide regulatory clarity (e.g., on copyright issues) and facilitate use of AI in regulated sectors to reduce adoption uncertainty.
  - Invest in enabling conditions for a resilient domestic AI ecosystem rather than attempting to replicate frontier capabilities directly.
- Equity and redistribution:
  - Complement supply-side interventions with redistributive policies to address displacement risks and rising wealth inequality from higher capital returns.

*Source: IMF country chapter content (United Kingdom).*

### Annex II. Drivers of Business Adoption of AI Across Countries

### Annex II. Drivers of Business Adoption of AI Across Countries

### Regression framework and data
- Cross-section regression estimated:
  - y_i = x_i β + z_i γ + e_i
  - y_i is the AI adoption rate mid-2025 in country i.
  - x_i includes ICT infrastructure, human capital stock (both from the AI preparedness index in Cazzaniga and others, 2024), trade openness (exports + imports / GDP from World Bank WDI), and electricity prices (UK Department for Energy Security & Net Zero; derived from IEA Energy Prices and Taxes). Human capital and infrastructure index are based on year 2023 or before.
  - z_i are controls in some specifications: log GDP per capita and the country-level World Uncertainty Index (Ahir, Bloom, and Furceri, 2022).
- To avoid feedback from AI adoption to predictors, 2019 values are used for electricity variables, GDP per capita, openness, and the uncertainty index.
- Focus: fundamental, structural drivers across countries rather than temporary deviations.

### Main empirical findings
- Enabling conditions: ICT infrastructure, human capital, and trade openness are meaningful drivers of AI adoption.
- Energy prices and uncertainty also matter: higher industrial electricity prices and greater economic/policy uncertainty are associated with lower AI adoption.
- Variance decomposition (squared semi-partial correlations) using the specification in Column (1) indicates that AI-relevant skills are the most important unique driver of cross-country variance in adoption rates, followed by infrastructure and openness. The remainder of R-squared is jointly explained by correlated predictors.

### Regression estimates (Table II.1: Long-Run Determinants of AI Adoption Across Countries)
- Notes: All variables standardized except GDP per capita. Robust standard errors in brackets. Significance: * p<0.10, ** p<0.05, *** p<0.01.
- Observations and R-squared:
  - Observations: 30 in Columns (1)-(3); 26 in Columns (4)-(5).
  - R-squared: 0.759 (1); 0.764 (2); 0.788 (3); 0.869 (4); 0.869 (5).

- Coefficients and standard errors by column:
  - Human capital:
    - (1): 5.69*** (1.54)
    - (2): 4.80** (1.95)
    - (3): 5.25*** (1.49)
    - (4): 5.15*** (1.78)
    - (5): 5.16** (1.83)
  - ICT infrastructure:
    - (1): 7.30*** (2.35)
    - (2): 6.59** (2.54)
    - (3): 7.00*** (2.25)
    - (4): 8.69*** (2.46)
    - (5): 8.64*** (2.56)
  - Openness:
    - (1): 1.86** (0.86)
    - (2): 1.49 (0.99)
    - (3): 2.05** (0.83)
    - (4): 2.33** (1.03)
    - (5): 2.31** (1.08)
  - Log GDP per capita:
    - (2): 2.33 (3.07)
    - (3): 0.98 (3.37)
    - (5): 1.08 (3.58)
  - Log electricity price:
    - (3): -2.77* (1.49)
    - (5): 0.19 (1.73)
  - Uncertainty:
    - (4): -2.01** (0.85)
    - (5): -2.06* (0.99)
  - Constant:
    - (1): 32.64*** (1.23)
    - (2): 9.16 (30.98)
    - (3): 32.67*** (1.18)
    - (4): 24.15 (33.83)
    - (5): 23.16 (35.85)

### Interpretation and robustness
- Column (1) establishes robust positive associations of AI adoption with human capital, ICT infrastructure, and trade openness.
- These associations remain largely unchanged after controlling for GDP per capita (Column (2)).
- Including log electricity price (Column (3)) shows significantly lower AI adoption when electricity prices are higher.
- Including the World Uncertainty Index (Column (4)) finds higher uncertainty significantly associated with lower adoption, consistent with uncertainty deterring investment in new technology.
- Column (5) includes all variables and confirms robustness of results.

*Source: Annex II. Drivers of Business Adoption of AI Across Countries*

### 31161. Published as Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work."

### 1gbrea2026002 - 31161. Published as Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work."

### Major cited research and working papers
- Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work." Quarterly Journal of Economics, 140(2), 889–942. https://www.nber.org/papers/w31161
- Carella, A., Deb, P. and Haider, N., 2024. “Construction Planning Reforms for Growth and Investment” IMF Selected Issues Paper (SIP/2024/031). Washington, D.C.: International Monetary Fund https://www.imf.org/-/media/files/publications/selected-issues-papers/2024/english/sipea2024031.pdf
- Cazzaniga, M., Jaumotte, F., Li, L., Melina, G., Panton, A.J., Pizzinelli, C., Rockall, E.J. and Tavares, M.M., 2024. "Gen-AI: Artificial Intelligence and the Future of Work." IMF Staff Discussion Note SDN/2024/001. https://doi.org/10.5089/9798400262548.006
- Cerutti, E.M., Pascual, A.I.G., Kido, Y., Li, L., Melina, G., Tavares, M.M. and Wingender, P., 2025. "The Global Impact of AI: Mind the Gap." IMF Working Paper 2025/076. https://doi.org/10.5089/9798229008570.001
- Comin, D. and Hobijn, B., 2004. "Cross-Country Technology Adoption: Making the Theories Face the Facts." Journal of Monetary Economics, 51(1), 39–83. https://doi.org/10.1016/j.jmoneco.2003.07.003
- Comin, D. and Hobijn, B., 2010. "An Exploration of Technology Diffusion." American Economic Review, 100(5), 2031–2059. https://doi.org/10.1257/aer.100.5.2031

### United Kingdom–focused sources and policy documents
- Confederation of British Industry (CBI), 2025. "AI Adoption among UK Firms." Unpublished CBI presentation.
- Cottier, B., Rahman, R., Fattorini, L., Maslej, N. and Owen, D., 2024. "How Much Does It Cost to Train Frontier AI Models?" Epoch AI, June 2024 (updated January 2025). https://epoch.ai/blog/how-much-does-it-cost-to-train-frontier-ai-models
- Council of Economic Advisers, 2025. "AI Talent Report." The White House, January 14, 2025. https://bidenwhitehouse.archives.gov/cea/written-materials/2025/01/14/ai-talent-report/
- Department for Business and Trade and Department for Science, Innovation and Technology (DBT and DSIT), 2025. Digital and Technologies Sector Plan. Policy Paper. https://www.gov.uk/government/publications/digital-and-technologies-sector-plan
- Department for Education (DfE), 2025. Skills England: Sector Evidence on the Growth and Skills Offer. https://assets.publishing.service.gov.uk/media/6863f18b3464d9c0ad609ddf/Skills_England_-_sector_evidence_on_the_growth_and_skills_offer.pdf https://www.gov.uk/government/publications/skills-england-sector-evidence-on-the-growth-and-skills-offer
- Department for Science, Innovation and Technology (DSIT), 2025. Artificial Intelligence Sector Study

### Observations from the citation set (document-level)
- The unit draws on academic research, IMF staff notes and working papers, policy papers, and government reports spanning 2004–2025.
- Multiple entries explicitly address AI, generative AI, technology diffusion, AI costs, and workforce/talent considerations, indicating a cross-cutting focus on technological adoption, economic impacts, and policy responses.
- Several entries are IMF outputs (SIP/2024/031; SDN/2024/001; IMF Working Paper 2025/076), reflecting institutional engagement with AI's economic implications.

*Source: 1gbrea2026002 - 31161. Published as Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work."*

### 2024. https://www.gov.uk/government/publications/artificial-intelligence-sector-study-2024

### UNITED KINGDOM: Why Have UK Gilt Yields Moved Ahead of G7 Peers?

### Key stylized facts on yields and the term premium
- Since 2022, UK gilt yields—particularly at long maturities—have risen and moved above G7 peers; the shift began in 2022 and coincided with the gilt market turmoil following the September 2022 “mini budget” announcement.
- The level of UK yields reflects both elevated expected short rates and the term premium; this paper focuses on the latter (the 10-year sovereign bond term premium).
- Over 2022–26, both expected short rates and the term premium contributed to increases in long-term gilt yields, but increases in the UK term premium have more than offset declines in expected short rates over the past two years, bringing long-term yields to their highest level since 2008. (Latest data: 5/31/2026.)
- In March 2026, UK CPI inflation was highest among G7 peers and had been on the higher end of the range since mid-2023; commensurately, the UK policy rate was also on the upper end of the G7-range in recent years.
- The contribution of expected short-term rates to long-term yields jumped at the start of the 2026 conflict in the Middle East, illustrating short-rate sensitivity to exogenous shocks.

### Gilt market structure and long-term trends
- Debt maturity and composition
  - UK debt’s weighted average maturity is just under 14 years, compared to 6 to 9 years for other G7 countries (reference date: 2/12/2026).
  - The UK issues a large share of long-dated bonds and a relatively high proportion of inflation-linked securities and has a limited reliance on short-term bills.
- Investor base shifts
  - The UK has a large defined benefit pension fund sector that historically supported long-term demand for gilts.
  - Pension funds and insurance corporations have reduced their share of gilt holdings over the past two decades; the Bank of England partly offset this through asset purchases, but quantitative tightening is reducing BoE holdings.
  - Office for Budget Responsibility (2025) estimate: the decline in the pensions sector’s gilt holdings could push up interest rates on government debt by around 0.8 percentage points, assuming the stock of debt remains close to 100 per cent of GDP.
  - Hedge funds account for 60 percent of the daily gilt market trading volume, versus 55 percent in the euro area and 30 percent in the US.
  - An index (Holders Vulnerability Index = Share of gilts held by foreign investors / Share of gilts held by Central Bank, Insurance Corporations and Pension Funds) shows a positive correlation with the term premium (2025Q4 observation cited), consistent with a shift toward more price-sensitive marginal investors.
- Market size, liquidity, and sovereign credit
  - The UK gilt market is less deep and liquid than the largest benchmark markets (market size reference date: 3/6/2026).
  - Government debt: debt-to-GDP rose from around 80 percent before the pandemic to just below 100 percent in recent years.
  - UK sovereign credit rating declined from AAA to AA/AA- over the past two decades (rating series up to 2026).
- Market fragility since September 2022
  - Economic Policy Uncertainty (EPU) spiked around the Brexit referendum and the September 2022 turmoil.
  - Market feedback suggests the September 2022 episode marked a structural shift, after which domestic risk factors have played a larger role and gilt markets have become more prone to volatility spikes.

### Empirical approaches and main analytical findings
- Methods employed
  - Factor-model (PCA) on observed yield changes (daily and weekly) to assess the time-varying importance of global versus domestic factors.
  - Structural Bayesian VAR (BVAR) to decompose the UK term premium into contributions from identified global and UK-specific shocks (UK term premium as variable of interest).
  - G7 panel regression of term premia (monthly, first differences) to relate term-premium changes to macro/credit fundamentals, risk, and supply–demand factors.
  - UK time-series regressions to assess changes in sensitivities after September 2022 (term premium and gilt yield volatility analyses).
- Principal empirical findings
  - Global factors account for the majority of UK yield variation over time, but their explanatory share has varied between 60 and 90 percent for gilt yields over 2020–26 (12-month rolling PCA).
  - After September 23, 2022, the share of variation explained by global factors declined markedly—particularly at the long end (30-year)—indicating a greater role for domestic factors; the global factor share rose again through 2024 and then declined in 2025 as domestic factors regained prominence.
  - The PCA results (12-month window) show that domestic factors became more important after the 2022 turmoil and again toward 2025; this coincided with a widening UK term premium.
  - BVAR decomposition (daily data, sign restrictions) identifies separate shocks: Global Risk, Global Term Premium, UK Risk, US Risk, and Structural demand/supply. Structural demand/supply shocks are identified by a rise in the UK term premium combined with a decline in the UK swap spread (interpreted as weaker demand for gilts relative to swaps and potential supply–demand imbalances).
  - BVAR results: around September 2022, UK-specific risk shocks dominated and drove a sharp increase in the UK term premium. From early 2024 to end-2025—when the term premium created a wedge between long-term yields and expected short rates—contributions came from a mix of global and UK-specific factors; structural demand/supply shocks became more prominent from 2024 onward (moving from negative to positive).
  - The shift in investor base toward more price-sensitive investors, increased net supply (issuance or QT), and constraints in market intermediation/liquidity are plausible mechanisms behind the structural demand/supply contribution to term premia; the model cannot separately identify these mechanisms.

### Policy implications and recommended focus areas
- Policy should focus on three areas:
  - Managing the quantity and composition of bond supply.
  - Reinforcing policy credibility (fiscal policy, debt management, predictable policies).
  - Strengthening market resilience (liquidity, market functioning, intermediation capacity).
- Rationale:
  - The term premium captures time-varying risk compensation, uncertainty, and bond market structure factors; credible and predictable fiscal and debt-management policies can help contain the term premium by reducing uncertainty and improving the market’s capacity to absorb shocks.
  - Structural shifts—declining stable domestic demand, growing foreign participation, and a larger share of hedge funds in trading—can increase required compensation for holding gilts and make the market more vulnerable to stress.
  - Given model dependence of term premium estimates, policy should aim to address underlying supply–demand and market-structure vulnerabilities identified across multiple empirical approaches.

*Source: IMF Selected Issues Paper chapter: “Why Have UK Gilt Yields Moved Ahead of G7 Peers?” (United Kingdom chapter), using data and analyses referenced through May 31, 2026.*

### Annex Figure 22 shows the impulse response functions for the model.

### Annex Figure 22 shows the impulse response functions for the model.

### Cross-Country Perspective on Supply–Demand Factors
- A cross-country panel regression across G7 countries examines the relationship between term premia and macroeconomic, risk, and structural supply–demand factors.
- Variables are grouped into three categories:
  - Macro-credit factors: Purchase Managers Index (lagged); Debt-to-trend-GDP ratio.
  - Sources of uncertainty: Equity Implied Volatility (VIX); Inflation uncertainty (average 6-month deviation of y/y CPI from survey expectation, lagged); Economic Policy Uncertainty (Baker-Bloom–Davis).
  - Structural demand/supply: Net issuance / percentage change in nominal amount of government debt outstanding; Central bank assets: ratio of central bank assets over the amount of government debt outstanding; Pension fund assets: ratio of pension fund assets over the amount of government debt outstanding.
- Quarterly and annual data are interpolated to a monthly frequency where needed. The dataset spans 2000–2025 at a monthly frequency.
- The analysis is subject to limitations: co-movement of drivers, differing data availability and frequency across jurisdictions, and a relatively small G7 sample limiting statistical power. Results identify correlations rather than causal relationships.

### Panel Regression Findings (G7)
- Key empirical findings:
  - In a levels regression: VIX and net issuance are positively associated with term premia; larger central bank balance sheets are associated with lower term premia.
  - In a first-differences specification: robust positive relationship between net issuance and term premia; continued role for the VIX.
  - Stronger growth expectations (proxied by lagged PMIs) show some evidence of association with a higher term premium.
  - Increases in pension fund assets are associated with lower term premia.
  - Effect of central bank flows is not statistically significant at conventional levels in first-differences.
- Economic significance (Annex Table 7 cited):
  - A percent increase (over 3 months) of the outstanding amount of debt through net issuance is associated with a 3.8 basis points increase of the term premium.
  - A one percentage point increase in the pension fund holdings of gilts is associated with a 5.4 basis points decrease of the term premium.

### Role of Domestic Uncertainty (UK Time-Series Evidence)
- A UK-specific time-series regression includes net issuance and a UK-specific Economic Policy Uncertainty (EPU) index, and tests changes in sensitivities using a post-2022-turmoil dummy.
- Key quantified result:
  - A one standard deviation change in the monthly EPU (80 index points) is associated with a 4.6 bps increase in the term premium, after the September 2022 gilt market turmoil.
- The net issuance coefficient in the time-series analysis is noted as similar in magnitude to the panel regression’s net issuance coefficient.

### Impact of the September 2022 Episode on Gilt Yield Volatility
- Gilt yield volatility (six-month moving standard deviation of yield changes) regressed on US Treasury yield volatility and VIX, with interactions for post-September 2022.
- Post-September 2022 findings:
  - US yield volatility is more strongly associated with gilt volatility.
  - Gilt yields have become more responsive to global risk conditions (VIX).
- Interpretation: the gilt market became more fragile, with volatility transmitted more readily across markets; increased propensity to enter high-volatility regimes can amplify and prolong instability via forced deleveraging.

### Summary of Empirical Results
- Two coherent drivers behind the rise in the UK term premium:
  1. Gradual long-term structural change in demand–supply conditions in the gilt market (net issuance increases; declining structural demand from long-term institutional investors).
  2. A regime shift after the September 2022 gilt market turmoil: greater role for domestic risk in driving the term premium and higher propensity for volatility spikes.
- Cross-method consistency:
  - Factor model (PCA): Global factors explain most yield variation, but their importance declines after September 2022.
  - Structural: BVAR: Structural supply–demand shocks contribute notably to recent term premium increases.
  - Panel regression (G7): Term premia positively associated with global risk (VIX) and net issuance; pension funds reduce premia.
  - UK time-series regression: Sensitivity of term premia to policy uncertainty increases after 2022.
  - Volatility regression: Gilt volatility becomes more sensitive to US volatility and global risk after 2022.
- Additional interpretive points:
  - The footprint of long-term institutional investors—such as pension funds and central banks—has become much smaller.
  - Evidence for other factors (e.g., central bank assets) is less robust or limited by data; more detailed data on outright government bond holdings could be more informative.

### Discussion and Policy Recommendations
- Policy responses should address both long-term structural supply–demand imbalances and the post-September 2022 regime shift.
- Recommendation cluster 1 — Reduce volume and adjust composition of gilt supply:
  - Stay the course on planned fiscal consolidation to limit net gilt issuance.
  - Prioritize fiscal measures: avoid raising fiscal deficit targets; develop contingency plans on tax and revenue sides if consolidation is pushed off course.
  - Calibrate issuance composition to avoid concentrating supply in segments with limited absorption capacity.
    - Actions already noted: shortening maturity of issuance; reduced share of index-linked gilts; consultation concluded on bill issuance; consideration of switch auctions.
  - Manage balance sheet normalization (quantitative tightening) to avoid unnecessarily amplifying bond-supply pressures:
    - Implement QT in a predictable and pre-determined manner to preserve market stability and the Bank rate as the active tool of monetary policy.
    - Given reduced structural demand for long-term gilts, skewing sales away from longer maturity gilts could be considered if market conditions warrant.
- Recommendation cluster 2 — Strengthen policy credibility and predictability:
  - Clear communication and well-signaled policy changes to anchor expectations and reduce sharp repricing in sovereign yields.
  - Recent reforms noted: authorities did not announce new measures in the spring, making the Fall budget the sole fiscal event; the Office for Budget Responsibility now assesses rule compliance at the time of the Fall budget; fiscal buffers increased.
  - Continue implementation of the structural reform agenda to build sustained credibility.
- Recommendation cluster 3 — Enhance resilience in the core funding market ecosystem:
  - Address intermediation constraints: repo market functioning, dealer balance sheet capacity, and the broader institutional framework.
  - System Wide Exploratory Scenario (SWES) evidence: individual actions can amplify shocks via procyclical asset sales when intermediation is limited.
  - Post-2022 LDI reforms improved resilience but further steps may be needed.
  - The FPC has received feedback on gilt repo market resilience; the BoE intends to publish a comprehensive update, including potential policy proposals, in early 2027.
- Note: Anchoring inflation expectations and maintaining credible monetary policy remain essential because expectations of future short rates have also contributed to the rise in long-term gilt yields.

*Source: Bloomberg, IMF staff calculations. Last data: 3/6/2026.*

### Annex I. Supplementary Data and Empirical Results

### Annex I. Supplementary Data and Empirical Results

### A. Data Set
- Table I.1: United Kingdom: Data Set — Variables, start/end dates, frequency, and sources
  - Term premium: Feb-99 to Nov-25, Daily, Bloomberg, IMF staff calculations
  - Debt/GDP: Mar-00 to Dec-25, Quarterly, Refinitiv
  - Central bank assets: Jan-00 to Nov-25, Monthly, Haver
  - Pension fund assets: Dec-06 to Dec-24, Annual, OECD
  - Inflation uncertainty: Dec-03 to Dec-25, Monthly, Bloomberg, IMF staff calculations
  - GDP (nominal): Mar-00 to Dec-25, Quarterly, Haver
  - Economic policy uncertainty: Jan-00 to Dec-25, Monthly, www.policyuncertainty.com
  - Government bonds outstanding: Mar-00 to Sep-25, Quarterly, Haver
  - PMI: Aug-01 to Dec-25, Monthly, Bloomberg
  - VIX: Jan-00 to Dec-25, Monthly, Bloomberg

### B. Additional Stylized Facts
- Inflation
  - UK inflation has been higher than in G7 peers in recent years.
  - Figure labeling: "CPI Inflation March 2026 (Percent y/y)"; timeframe visible through Jan-17 to Mar-26.
- Policy rates
  - The UK’s policy rate has been on the upper end of the G7 range over the past years.
  - Figure labeling: "Policy Rates & CPI Inflation March 2026 (Percent, percent)".
- Real and inflation components of gilt yields
  - The real component of gilt yields has increased while the inflation component has been relatively stable since 2022.
  - Figure labeling: "UK 10-Year Gilt Yield Real and Inflation Components Change Since 2022 (Percentage points)"; change from 1/1/2022 until 5/31/2026.
- Swap spread
  - The swap spread has turned negative in recent years.
  - Figure labeling: "UK 10-Year Swap Spread (Basis points)". Note: swap spread computed as difference between the 10-year SONIA swap rate and the 10-year gilt yield.
- Projected pensions sector gilt holdings
  - Pensions sector gilt holdings are expected to decline as defined benefit schemes shrink.
  - Figure labeling: "Projected Pensions Sector Gilt Holdings (Percent of GDP)"; projection horizon shown from 2024-25 through 2048-49.
  - Source: Office for Budget Responsibility (2025).

### C. Background on G7 Panel Regression
- Regression setup
  - Panel regression with dependent variable term premium. G7, monthly data.
  - Country fixed effects included. Standard errors clustered at country level.
- Estimated coefficients (columns (1) Macro; (2) Macro+Uncertainty; (3) Macro+Uncertainty+Structural)
  - ∆PMI_{t−1}: 0.001 (0.001), 0.001 (0.001), 0.002* (0.001)
  - ∆(Debt/GDP)_t trend: 0.000 (0.000), -0.000 (0.000), -0.000 (0.000)
  - ∆VIX_t: 0.002* (0.001), 0.002* (0.001)
  - ∆U_t inflation: 0.040 (0.069), 0.053 (0.063)
  - ∆U_t policy: -0.000 (0.000), -0.000 (0.000)
  - Net issuance: 3.763** (1.142) — appears in column (3)
  - ∆[CBAssets_t / Gov.Bonds_t]: -0.108 (0.128)
  - ∆[PFAssets_t / Gov.Bonds_t]: -5.410*** (0.995)
  - Constant: -0.003*** (0.000), -0.002*** (0.000), -0.003*** (0.000)
- Model statistics
  - Observations: 2138.000 (col 1), 1947.000 (col 2), 1802.000 (col 3)
  - R-squared: 0.002 (col 1), 0.006 (col 2), 0.036 (col 3)
- Significance notation
  - * p < 0.10, ** p < 0.05, *** p < 0.01

### D. Background on UK Time Series Regression
- Regression setup
  - Time series regression, UK, Monthly Data, Dependent Variable Term Premium.
  - Newey-West standard errors (lag 4). Includes issuance (3-month change) and policy uncertainty interaction with a dummy that takes the value of 1 during and after September 2022.
- Estimated coefficients (column (1))
  - Net issuance: 4.068*** (1.001)
  - ∆U_t policy: -0.006 (0.018)
  - post Sep.2022: 1.111 (1.794)
  - post Sep.2022 x ∆U_t policy: 0.064** (0.026)
  - Constant: 0.090 (0.788)
- Observations: 304.000
- Significance notation
  - * p < 0.10, ** p < 0.05, *** p < 0.01

### E. Background on UK BVAR Analysis
- Sample and methodology
  - BVAR sample consists of daily data spanning 2004-2026.
  - Sign-restrictions are imposed on the first two periods of the impulse response functions (0-2).
  - Uses a standard Minnesota prior, and a 4-period lag-structure.
- Figure I.6: BVAR Impulse Response Functions — computed using the BEAR Toolbox.

*Source: Annex I. Supplementary Data and Empirical Results (1gbrea2026002).*

### References

### References

### Selected literature cited
- Adrian, Tobias; Richard K. Crump; and Emanuel Moench (2013). Pricing the Term Structure with Linear Regressions. Journal of Financial Economics, 110(1), 110–138.
- Baker, Scott R.; Nicholas Bloom; Steven J. Davis (2015). Measuring Economic Policy Uncertainty. CEP Discussion Papers dp1379, Centre for Economic Performance, LSE.
- Brandt, Lennart; Saint Guilhem, Arthur; Schröder, Maximilian; Van Robays, Ine (2021). What drives euro area financial markets? The role of US spillovers and global risk. Working Paper Series 2560, European Central Bank.
- Cochrane, John H.; Monika Piazzesi (2005). Bond Risk Premia. American Economic Review, 95(1), 138–160.
- Dieppe, Alistair; van Roye, Björn; Legrand, Romain (2016). The BEAR toolbox. Working Paper Series 1934, European Central Bank.
- Don H. Kim; Jonathan H. Wright (2005). An arbitrage-free three-factor term structure model and the recent behavior of long-term yields and distant-horizon forward rates. Finance and Economics Discussion Series 2005-33, Board of Governors of the Federal Reserve System (U.S.).
- Giese, Julia; Michael Joyce; Jack Meaning; Jack Worlidge (2023). Preferred habitat investors in the UK government bond market. Bank of England Staff Working Paper No. 939.
- Greenwood, R.; Vissing-Jorgensen, A. (2018). The Impact of Pensions and Insurance on Global Yield Curves. Harvard Business School, Working Paper 18-109.
- Hanson, Samuel G.; Stein, Jeremy C. (2015). Monetary policy and long-term real rates. Journal of Financial Economics, vol. 115(3), pages 429-448.
- International Monetary Fund (2025). October 2025 Global Financial Stability Report, Annex 1.7.
- Kadiric S. (2022). The determinants of sovereign risk premiums in the UK and the European government bond market: the impact of Brexit. Int Econ Econ Policy. 2022;19(2):267-298.
- Kaminska, Iryna; Haroon Mumtaz (2022). Monetary policy transmission during QE times: role of expectations and term premia channels. Bank of England Staff Working Paper No. 978.
- Krishnamurthy, A.; Vissing-Jorgensen, A.; Gilchrist, S.; Philippon, T. (2011). The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy. Brookings Papers on Economic Activity, 215–287.
- Leippold, Markus; Felix Matthys (2022). Economic Policy Uncertainty and the Yield Curve. Review of Finance, Volume 26, Issue 4, Pages 751–797.
- Lengyel, Andras (2026). Government bond issuance surprises and the term structure of interest rates in the UK. Journal of Banking & Finance, Volume 189, 2026, 107715.
- McCauley, Robert; Eli Remolona (2000). Size and liquidity of government bond markets. BIS Quarterly Review, November.
- Office for Budget Responsibility (2025). Fiscal risks and sustainability. CP 1343, July 2025.
- Organisation for Economic Co-operation and Development (2026). Global Debt Report 2026. 4 March 2026.
- Panigrahi, Lisa; Amarjot Sidhu (2026). What were the drivers of UK long-term interest rates in 2025? Bank of England.
- Greenwood, Robin; Vayanos, Dimitri (2014). Bond Supply and Excess Bond Returns. The Review of Financial Studies, Volume 27, Issue 3, Pages 663–713.
- Miranda-Agrippino, Silvia; Hélène Rey (2020). U.S. Monetary Policy and the Global Financial Cycle. The Review of Economic Studies, Volume 87, Issue 6, Pages 2754–2776.
- Stehn, Sven Jari (2025). How Vulnerable are the UK’s Public Finances? Goldman Sachs Research, 26 September 2025.
- Vayanos, D.; Vila, J.-L. (2021). A Preferred-Habitat Model of the Term Structure of Interest Rates. Econometrica, 89: 77-112.

### Annex: Labor Taxation in the UK — Executive points (A–F)
H3 Motivation (A)
- Labor taxes (income tax and national insurance contributions) account for just under half of all tax revenues in the UK, or around 16 percent of GDP.
- The UK’s tax-to-GDP ratio has increased by about 5½ ppts since the mid-1960s and now stands at around 35 percent.
- Labor-tax revenues have risen by almost 4 percentage points of GDP since the mid-2010s and are projected to increase by a further 2 percentage points of GDP by the end of the decade under current fiscal plans (due to income tax threshold freezes beyond 2028).
- Key policy question: how much additional revenue can be raised from labor taxes without materially weakening labor supply and growth?

H3 Labor Taxes in the UK (B) — structure and thresholds
- Average tax wedge for an average earner in the UK is lower than the OECD and G7 averages; however, marginal tax rates for high earners in the UK are significantly above those in other G7 countries.
- Important thresholds and “tax cliffs”:
  - Basic income tax rate begins above the £12,570 personal allowance (PA).
  - Withdrawal of Child Benefit (CB) at earnings above £60,000.
  - Tapering of the personal allowance at earnings over £100,000.
- Universal Credit (UC) withdrawal creates high participation tax rates for low earners.
- Marginal effective tax rates (METRs) vary substantially across earnings and produce “bunching” just below high METR thresholds.

H3 Model and Data (C)
- Microsimulation model based on the Mirrlees framework, combining UK Family Resources Survey (FRS) 2023–24 microdata with UKMOD to map taxes and transfers into disposable incomes, PTRs, and METRs.
- Behavioral responses captured on two margins: participation (extensive) and hours/earnings conditional on working (intensive).
- Baseline elasticities:
  - Average intensive-margin elasticity = 0.15.
  - Average extensive-margin elasticity with respect to in-work income = 0.25 (declining profile in income; lowest participation responses for high-income workers).
- Participation elasticities by income quintile (Table 1, baseline calibration):
  - Lowest 20 percent: 0.49
  - Next 20 percent: 0.37
  - Middle 20 percent: 0.21
  - Next 20 percent: 0.13
  - Highest 20 percent: 0.08
- Social marginal welfare weights inferred via inverse-tax approach conditional on assumed elasticities.

H3 Simulation Results (D) — scenarios and key quantitative outcomes
- Three reform scenarios:
  - Scenario 1: Uniform tax increase (upward shift of METR across distribution).
  - Scenario 2: Segmented (targeted) increases applied to (i) bottom 25 percent; (ii) middle 65 percent; (iii) top 10 percent.
  - Scenario 3: Non-linear (Mirrlees optimal) reform where marginal tax schedule is endogenously determined given revenue targets, elasticities and welfare parameters.
- Scenario 1 results (uniform increases):
  - Model suggests uniform marginal tax increases could raise labor tax revenues by up to 6.5 percent—equivalent to around 1 percent of GDP—but could reduce labor supply by as much as 13.5 percent.
- Scenario 2 results (targeted 5 percentage-point METR increases; Table 2, percent relative to baseline):
  - Bottom 25%: Change in tax revenue = 3.0; Change in labor earnings = -1.6; Change in average tax rate = 3.9
  - Middle 65%: Change in tax revenue = 1.9; Change in labor earnings = -1.9; Change in average tax rate = 1.5
  - Top 10%: Change in tax revenue = -0.5; Change in labor earnings = -0.9; Change in average tax rate = 1.1
  - All (uniform 5 ppts): Change in tax revenue = 4.3; Change in labor earnings = -4.4; Change in average tax rate = 3.9
- Scenario 3 results (non-linear optimal reforms):
  - An “optimal” revenue-neutral reform could raise aggregate labor supply by 3 percent.
  - Non-linear reforms that raise up to 4.5 percent in additional revenues could result in still higher aggregate labor supply.
  - Optimal schedules tend to reduce marginal tax rates for earners above £100,000 while increasing marginal rates for low and middle-income earners in many optimal calibrations.

H3 Robustness (E)
- Higher behavioral elasticities reduce revenue potential but increase efficiency gains from reform:
  - 25 percent increase in elasticities: intensive = 0.19; extensive = 0.31.
  - 50 percent increase in elasticities: intensive = 0.23; extensive = 0.38.
  - With higher elasticities, smaller revenue increases may still improve aggregate labor supply; the green line in Figure 4.A indicates labor income taxes could be increased by up to 5 percent while still raising aggregate labor supply under sufficiently strong behavioral responses.
- Inequality aversion calibration:
  - Baseline inequality aversion parameter = 1.0.
  - Alternative values considered: 0.5 (lower redistributive motive) and 1.5 (higher redistributive motive).
  - Higher inequality aversion shifts the earnings-revenue frontier (Figure 4.B) such that for lower revenue increases (below 5 percent) labor supply responds more negatively to the optimal schedule, while for larger revenue increases labor supply losses are smaller than baseline.
- Effects are highly non-linear: small changes in elasticities or inequality aversion can produce disproportionately large changes in optimal schedules and revenue/earnings outcomes, especially at the top of the distribution.

H3 Policy Conclusions (F)
- Limited scope for significant additional revenue from uniform labor tax increases without materially weakening labor supply; uniform increases amplify existing distortions (high PTRs at the bottom and high METRs at the top).
- Targeted increases toward the bottom of the earnings distribution, coupled with more generous in-work transfers, can raise revenue with smaller efficiency costs but involve equity trade-offs.
- More fundamental, non-linear reforms that strengthen incentives to work at the bottom (e.g., smoother benefit withdrawal, conditional in-work support) while reducing abrupt cliff edges at higher incomes could deliver higher revenues with a better efficiency-equity trade-off.
- Policy focus should move beyond headline rate increases to redesigning the tax-benefit schedule to preserve work incentives and support medium-term growth while addressing fiscal pressures.

*Source: 1gbrea2026002 - References.*

### References

### References

### Works cited

- Adam, S. and Phillips, D. (2013). “An ex-ante analysis of the effects of the UK Government's welfare reforms on labour supply in Wales” IFS Report No. R75.
- Bourguignon, F., and Spadaro, E. (2012). “Tax-benefit revealed social preferences”Journal of Economic Inequality, 10: 75–108.
- Brewer, M., Saez, E. and Shephard, A. (2010). "Means-testing and tax rates on earnings" in Dimensions of Tax Design: the Mirrlees Review, 1(4): 90-201.
- Mirrlees, J. ed., 2011.Tax by design: The Mirrlees review. Oxford University Press (OUP) Oxford.
- Piketty, T., and Saez, E. (2013). “Optimal labor income taxation” InHandbook of Public Economics, Volume 5, edited by Alan J. Auerbach, Raj Chetty, Martin Feldstein, and Emmanuel Saez, 391–474. Amsterdam: Elsevier.
- Richiardi, M., Collado, D. and Popova, D., 2021.UKMOD–A new tax-benefit model for the four nations of the UK (No. CeMPA WP 7/21).CeMPA Working Paper Series.
- Jacobs, B., Jongen, E.L.,  and Zoutman, F.T. (2017). “Revealed social preferences of Dutch political parties”Journal of Public Economics, 156, 81–100.
- Jacquet, L, Lehmann, E. and Van der Linden, B. (2013) “Optimal redistributive taxation with both extensive and intensive responses”Journal of Economic Theory, 148 (5): 1770-1805.
- Zoutman, F.T., Jacobs, B. and Jongen, E.L. (2013). “Optimal redistributive taxes and redistributive preferences in the Netherlands”Erasmus University Rotterdam.

*Source: 1gbrea2026002 - References*

---


_Source: https://www.imf.org/-/media/files/publications/cr/2026/english/1gbrea2026002.pdf_