## sdnea2024002

## Source details

**Canonical URL:** [sdnea2024002](https://www.imf.org/-/media/files/publications/sdn/2024/english/sdnea2024002.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/sdn/2024/english/sdnea2024002.pdf.md)
- [Structured JSON version](/-/media/files/publications/sdn/2024/english/sdnea2024002.pdf.json)

---

### Executive Summary — key messages
- Gen AI can transform production and public services and enable governments to advance revenue mobilization and deliver more efficient public services (health care, education, public procurement, social transfers).
- Rapid diffusion of gen AI increases risks to labor markets relative to past general-purpose technologies.
- Major risks:
  - Further decline in the labor income share in national income.
  - Exacerbation of income and wealth inequality.
  - Reinforcement of market concentration and monopoly rents for dominant firms.
- Tax guidance:
  - Special taxes on AI to slow investment are not recommended.
  - Reconsider corporate tax incentives that encourage rapid labor displacement.
  - Strengthen general taxes on capital income to protect the tax base and offset rising wealth inequality.
- Social protection and labor-market policy priorities:
  - Broaden unemployment insurance coverage and generosity; improve portability; consider wage insurance.
  - Combine enhanced social insurance with active labor market policies (ALMPs) and lifelong learning.
  - Use digital technologies to expand social assistance coverage, especially in emerging market and developing economies (EMDEs).
- Public funding priorities:
  - Focus on fundamental research, infrastructure (especially in EMDEs), and AI applications in the public sector.
- Policy approach under uncertainty:
  - Prepare for both “business as usual” and highly disruptive scenarios; remain agile to steer innovation and cushion transition costs.

*Source: Executive Summary, STAFF DISCUSSION NOTES — Broadening the Gains from Generative AI: The Role of Fiscal Policies, INTERNATIONAL MONETARY FUND*

### Introduction — scope and framing
- Focus: the role of fiscal policies in broadening distributional gains from gen AI.
- Four specific questions addressed:
  - How social protection reduced labor market disruption in past automation episodes.
  - How to strengthen social spending during rapid technological transitions.
  - Whether to tax automation/AI to mitigate labor market disruptions and finance effects on workers.
  - How to design redistributive taxation given inequality and winner-take-all dynamics from gen AI.
- Caveats: high uncertainty on speed of AI capability improvement, adoption, substitution vs. complementarity, adaptation, policy responses, and productivity impacts.

### Rise of “cognitive” automation and labor-market risks
- Gen AI (large language models) can produce new content and expand tasks computers perform.
- Diffusion can be faster than steam engine, electricity, early computers, creating uncertain, rapid impacts.
- Labor-market implications:
  - Exposure extends beyond routine tasks to cognitive and high-skill nonroutine tasks.
  - AI could also automate blue-collar jobs via intelligent robots.
  - Risks include reduced labor share and exacerbated inequality via market power and rents.

### Empirical evidence and lessons from past automation
- Evidence on industrial robots:
  - At US commuting-zone level (2000–07), robot exposure measured via adjusted Bartik-type instrument.
  - UI generosity does not affect robotization’s impact on employment.
  - States with more generous UI saw wage declines from robotization about two-thirds smaller than other states.
  - Cushioning effect on wages is particularly pronounced for workers without a college degree; differences statistically significant at the 1 percent level.
- Poverty effect:
  - Robotization is estimated to increase poverty by "0.3 percentage point (3 percent increase)".
  - Most of the poverty increase is attenuated where social assistance is more generous.

### Model simulations and transition scenarios
- Model: HANK-DGSE extension with search-and-matching frictions; one sector subject to automation shock.
- Baseline calibration and shock:
  - Automated capital productivity increases by 300 percent by 2030, producing a "20 percent decline in employment in the new steady state (approached in about 15 years)."
  - New steady state shows wages increase "about 15 percent" as automation raises aggregate productivity.
  - McKinsey projection referenced: automation could replace "20 to 30 percent" of time spent on work activities by 2030.
- Calibration targets and selected parameter values:
  - Unemployment rate target: "7 percent".
  - Job-finding probability target: "45 percent per quarter".
  - Vacancy-filling probability target: "70 percent per quarter".
  - Replacement ratio set at "50 percent".
  - Elasticity of substitution in labor-automated capital (LAC) bundle: "2.5".
  - Share of labor in LAC bundle s_i: "0.99 (Labor income share of 61 percent)".
  - Other parameters: separation rate ρ = "0.06"; matching efficiency m̄ = "0.56"; matching elasticity μ = "0.5"; vacancy cost κ = "0.04"; worker’s wage bargaining power ψ̄ = "0.65"; discount factor β = "0.989"; depreciation rate δ = "0.025"; price adjustment cost κ_p = "30"; Taylor rule response to inflation γ_π = "2".
- Smaller-impact scenario:
  - Sectoral productivity shocks are about half the magnitude of baseline.
  - Output increases by about "17 percent".
  - Labor income share in affected sector falls by "10 percentage points".
  - Total labor income share in the economy is reduced by "5 percentage points".

### UI design trade-offs and illustrative policies
- Transition costs: unemployment rises temporarily due to sectoral reallocation and search frictions; consumption falls substantially for unemployed workers.
- Two illustrative UI designs in simulations:
  - (1) Permanent increase in the UI replacement ratio by "1 percentage point".
  - (2) Temporary asymmetric adjustment: increase replacement ratio in proportion to previous quarter’s unemployment gap by factor "0.6" once unemployment rises by more than "1 percentage point" relative to steady state.
- Findings:
  - Both mitigate consumption drops for unemployed workers.
  - Permanent increases can discourage job search, raise unemployment, and lower welfare.
  - Temporary UI adjustment aligned with unemployment yields higher welfare benefits in simulations.
- Suggested UI features for severe disruption:
  - Benefits depending on unemployment duration, linkage with training and re-skilling, portability, extended scope and flexibility, and consideration of wage insurance.

### Active labor market policies (ALMPs), training, and employer roles
- ALMPs (training, sectoral programs, apprenticeships) complement UI by shortening unemployment spells and improving matches.
- Evidence:
  - US sector-based training programs yielded earnings gains of "14–38 percent" in the year following training completion (with persistent gains).
- Policy implications:
  - Scale lifelong learning, preemptive skill acquisition, and assess re-skilling vs. alternatives (for example, early retirement) especially for older workers.
  - Employer-provided training may need public augmentation.

### Distinct challenges and policy timing in EMDEs
- EMDE characteristics:
  - Lower share of high-skill occupations, less prepared to adopt AI, large informal sectors, budget constraints, weaker institutions.
  - Larger shares of young people not in employment, education, or training.
  - Greater insurance and credit market imperfections and less personal wealth—larger welfare gains from UI.
- Model scenarios for EMDEs:
  - Faster deployment scenario: UI and ALMPs deployed "2.5 years after the initial shock".
  - Delayed deployment scenario: ALMPs "five years after the initial shock".
  - Outcomes averaged over "60 quarters"; bubble size proportional to time unemployment rate "1 percentage point" above initial steady state.

### Upgrading social assistance and delivery systems
- UI gaps:
  - Maximum benefit duration often less than "12 months"; most US states provide a maximum of "26 weeks".
  - Fragmented access, nonportable entitlements, minimum contribution durations exclude temporary workers.
  - Policy options: extend coverage, portability, flexibility, and consider wage insurance as a temporary subsidy for re-employed displaced workers.
- Social assistance:
  - Options range from enhanced means-tested guaranteed minimum income with integrated training to unconditional benefits (not generally recommended given fiscal costs and current evidence).
  - Integration of ALMPs with social assistance (conditional continued eligibility tied to job search or training) advised.
- Systems and infrastructure:
  - Require robust, universal information systems integrated across programs for beneficiary identification and efficient delivery.
  - Digital technologies can rapidly expand coverage in EMDEs but pose fiscal trade-offs.

### Tax systems, METRs, and investment incentives
- METR concept:
  - Marginal Effective Tax Rates (METRs) capture incentives for incremental investment; METR = 0 implies neutrality.
  - METR datasets used: ZEW (34 countries, 1998–2020) and OECD (74 countries, 2017–22).
  - Current METRs (equity finance, averaged across 74 countries) largely range between "12.5 to 35 percent".
- Asset tax treatment patterns:
  - Tax advantage for intellectual property is almost universal.
  - For the median country, tax treatment of software and hardware became more generous by "about 0.8 and 3 percentage points (relative to buildings)" over 2017–22.
  - Notable countries favoring software/hardware: Germany, the United States, The Netherlands, New Zealand, Singapore, Hong Kong SAR.
- Complementarity/substitution evidence:
  - Machinery and equipment tend to complement labor.
  - Computer hardware complements high-skilled labor but substitutes for low-skilled labor.
  - Software generally found to replace labor with estimated elasticity of substitution "1.7".

### Should AI or automation be taxed?
- Theoretical rationales:
  - Production efficiency argues against special taxes on gen AI if capital taxes are neutral across assets.
  - Efficiency rationale for temporary automation tax: internalize social costs of excessive job displacement under high transition costs and frictions.
  - Equity rationale: mitigate rising top-income and capital shares.
- Quantitative estimate from literature:
  - Costinot and Werning (2023) estimate optimal tax rate on robots between "1 and 3.7 percent" of robot price, higher for more disruptive technologies.
- Model illustration:
  - Temporary automation tax financing UI can improve welfare when transition costs are high by discouraging substitution of labor.
  - Trade-off: automation tax reduces short-to-medium term wages (lower productivity) but can mitigate unemployment spikes; welfare may fall if transition costs are modest.
- Practical recommendation:
  - Specific tax on gen AI is not recommended due to identification, asset definition, relabeling, and international mobility/avoidance problems.
  - Alternative actions: reconsider preferential tax treatments that favor labor-displacing assets (adjust METR differentials); use AI regulation to include labor-market considerations.

### Broadening gains through taxation and redistribution
- Capital taxation importance:
  - Capital income is more concentrated among top income groups than labor income.
  - Decline in effective taxation of capital since the 1980s widened the gap between average tax rates on capital and labor to almost "10 percentage points" by 2018 in advanced economies.
- Policy options:
  - Strengthen corporate income tax and leverage global minimum tax.
  - Consider excess-profit taxes on monopoly rents.
  - Improve enforcement (for example, automatic exchange of information) and use gen AI to enhance enforcement.
  - Enhance taxation of capital gains.
  - Consider targeted labor tax relief via income tax credits or job credits (practical obstacles exist).
- Funding social protection:
  - Stronger capital income taxation will be more relevant as labor income share declines to finance social protection and education upgrades.

### Fiscal policies for AI innovation, deployment, and administrative capacity
- Public funding priorities now that AI is in commercial adoption:
  - Fundamental research.
  - Digital and electricity infrastructure, especially in EMDEs.
  - AI applications in public sector services and funding for upskilling/re-skilling.
- Administrative and governance capacity:
  - Invest in expertise to select and vet funding, update regulation, and develop procurement capacity.
  - Consider a dedicated agency model to coordinate private sector, academia, and stakeholders.
  - International cooperation (a “distributed CERN” model) could support global collaboration on AI innovation, use, and regulation.

### Conclusions and policy implications — actionable directions
- Prepare for alternative scenarios where AI either raises productivity and creates jobs or rapidly displaces jobs and increases inequality.
- Social protection:
  - Broaden UI eligibility (including self-employed and atypical contracts), ensure portability, and strengthen ALMPs.
  - Target social assistance to permanently displaced or indirectly affected workers; consider unconditional transfers only if justified by broader change in nature of work.
- Education:
  - Focus on quality, adaptability, and lifelong learning; foster higher-order skills to work alongside AI.
- Taxation:
  - Do not pursue narrowly targeted AI taxes; instead, rebalance corporate tax incentives and strengthen capital taxation and enforcement.
  - Consider carbon pricing for AI server energy consumption to reflect environmental externalities.
- Use gen AI to advance tax administration (tax enforcement and system redesign possibilities such as personalized VAT, lifetime-income tax, or real-time property tax contingent on information gains).
- Ensure innovation and deployment policies work in tandem with social protection, education, tax, and regulatory policies to broaden gains from gen AI.

*Source: sdnea2024002 — Broadening the Gains from Generative AI: The Role of Fiscal Policies (Staff Discussion Note)*

### Executive Summary _______________________________________________________________________________ 1

### Executive Summary

### Introduction
- Rapid advances in generative artificial intelligence (gen AI) hold immense potential to transform production processes and significantly accelerate productivity growth.
- Gen AI can revolutionize information availability and utilization, enabling governments to advance revenue mobilization and deliver more efficient public services across sectors, including health care, education, public procurement, and social transfers.
- A critical distinction between gen AI and past disruptive technologies (such as the steam engine, electricity, and early computers) lies in its potential for rapid diffusion, which increases risks to labor markets.

### Risks, distributional effects, and market structure
- Gen AI’s capabilities extend to more intelligent automation, potentially amplifying job losses in cognitive occupations beyond the low- and middle-skill, routine tasks affected by automation and robots to date.
- Consequences include:
  - Further decline in the labor income share in national income.
  - Exacerbation of income and wealth inequality.
  - Reinforcement of market concentration where dominant firms can enjoy monopoly rents.

### Tax policy guidance
- Special taxes on AI intended to slow AI investment are not recommended because they can be hard to operationalize and hamper productivity growth.
- Corporate tax incentives that encourage rapid labor displacement—prevalent in several advanced economies—should be reconsidered because they magnify the social costs of excessive labor market dislocation.
- Corporate tax distortions that hold up labor-saving investments—more prevalent in developing economies—can also be costly, especially in less-disruptive labor market scenarios.
- General taxes on capital income, which have systematically declined across the world during past decades, should be strengthened to:
  - Protect the tax base against a further decline in the labor-income share.
  - Offset rising wealth inequality.

### Social protection and labor-market policies
- Fiscal policies can cushion negative labor market and distributional effects of gen AI and help distribute gains more evenly.
- Recommended directions:
  - Broaden the coverage and generosity of unemployment insurance.
  - Improve portability of entitlements.
  - Consider forms of wage insurance.
  - Combine enhanced social insurance with active labor market policies to help workers manage transitions and adapt to changing skill requirements.
- Innovative approaches that harness digital technologies can facilitate expanded coverage of social assistance programs, particularly for those who suffer prolonged transition impacts or who work in the informal sector in emerging market and developing economies.
- Education and training policies must adapt to new realities to prepare workers for future jobs and offer lifelong learning. Sector-based training, apprenticeship, and upskilling and re-skilling programs could play a greater role in helping workers transition to new tasks and sectors.

### Public funding priorities for AI
- Now that AI has matured to the commercial adoption phase, public funding should focus on areas less likely to receive private sector investment, including:
  - Fundamental research.
  - Necessary infrastructure (particularly in emerging market and developing economies).
  - Applications in the public sector (education, health care, government administration).

### Policy approach under uncertainty
- Given uncertainty surrounding the transformative nature, impact, and pace of gen AI, policymakers must remain agile.
- Policy should aim to:
  - Create conditions that steer innovation and deployment to harness benefits of gen AI and serve collective human interests.
  - Be ready to cushion transition costs for workers and households.
  - Prevent rising inequality.
- Fiscal policies need to prepare for both “business as usual” and highly disruptive scenarios.

*Source: Executive Summary, STAFF DISCUSSION NOTES — Broadening the Gains from Generative AI: The Role of Fiscal Policies, INTERNATIONAL MONETARY FUND*

### Introduction

### Introduction

### Rise of “cognitive” automation
- Generative AI (gen AI) based on large language models can produce new content and is expanding the set of activities that computers can perform more efficiently than humans.  
- Gen AI can proliferate much faster than previous general purpose technologies (steam engine, electricity, early computers), creating uncertain and rapid impacts on production, firms, and public-sector operations.

### Untold benefits
- Potential gains for firms and industries: new revenue, cost savings, and improved products and processes.  
- Potential gains for governments: improved public service delivery (examples cited include fiscal operations, procurement, revenue collection through enhanced fraud detection and automated audit and assurance).  
- Potential gains for public services: personalized interactive learning, augmented reality, remote patient monitoring—potentially faster and more equitable reach of education and health care.  
- Gen AI–fueled advances in public-sector operations could inform more effective policy design and regulatory operations.

### Impact on labor markets and inequality
- Adoption of gen AI is likely to be uneven and rapid, risking disruption of labor markets.  
- Past automation waves displaced many routine tasks and drove down average wages and polarized wages and employment; earlier waves mostly displaced blue-collar (lower-skilled) workers, while evidence suggests white-collar (high-skilled) workers are most exposed to AI.  
- AI could also power more intelligent robots leading to further automation of blue-collar jobs—thus potentially amplifying job losses across low-skill and cognitive occupations.  
- Risks include a reduced labor share and wages relative to capital, and exacerbation of income and wealth inequality via rising market power and economic rents in concentrated, winner-take-all markets.

### Alleviating costly transitions and broadening gains
- Social protection systems can help individuals adapt by offering financial support during unemployment, promoting new skills acquisition, and creating a safety net. Traditional mechanisms include payroll-based insurance (for example, unemployment benefits), lifelong education and training initiatives, and cash transfers and other noncontributory social assistance programs.  
- A key question raised: whether and how social policies must be reimagined in the face of disruptive AI-driven technological changes.

### Taxation of investment in AI and capital income
- Taxation and regulation could slow deployment of automation, mitigating disruptive labor market implications, but taxation can also distort productivity-enhancing investment and reduce economic growth. The net welfare balance is unclear and scenario-dependent.  
- Gen AI can increase top-income inequality; progressive income taxes, including taxation of capital income, are discussed as tools to address rising inequality while balancing efficiency trade-offs.  
- Labor substitution may reduce revenue if capital income is taxed less than labor income; developing economies exposed to “reshoring” and specializing in labor-intensive sectors are particularly at risk of losing tax revenue.

### Steering innovation in AI
- Fiscal policies may influence the direction of innovation and deployment—potentially favoring applications that expand rather than substitute human capabilities and generate new occupational tasks. The practical implications are unclear, but development opportunities in emerging markets could be significant.

### This note: scope and questions addressed
- Focus: the role of fiscal policies in supporting a more equal distribution of gains and opportunities from gen AI.  
- Four specific questions addressed:
  - How have social protection systems helped reduce labor market disruption during past episodes of automation?
  - Looking ahead, how can countries strengthen social spending during rapid technological transitions?
  - Have tax systems provided excessive incentives for automation? Should automation be taxed to mitigate labor market disruptions and pay for its effects on workers?
  - How should governments design redistributive taxation in the face of inequality and winner-take-all dynamics from gen AI—especially taxes on capital income?

### Previous work and contribution to literature
- Builds on recent studies of productivity and labor market impacts of AI and on pioneering fiscal-policy-focused work.  
- Contributions of this note:
  - New empirical analysis of the role of social protection systems during past automation waves.
  - Discussion of desirable characteristics of social spending in the face of disruptive technological developments.
  - Model simulations illustrating the impact of spending and tax policies on labor market outcomes and welfare.
  - Novel discussion of how current tax systems affect firms’ decisions to invest in labor-displacing capital assets, the case for taxing AI, and recommendations to enhance taxation of capital income.
  - Brief consideration of whether fiscal policies should promote innovation and deployment of gen AI.

### Caveats
- High uncertainty on how gen AI will evolve and affect economies, with plausible variation in:
  - the speed of AI capability improvement,
  - the degree of adoption across countries and firms and how technologies are used,
  - the extent AI replaces or complements different types of workers,
  - how people adapt to new work realities,
  - government policy responses,
  - implications for productivity growth and economic well-being.  
- Fiscal policies must adapt to changing conditions and prepare for both business-as-usual and highly disruptive scenarios.

### Upgrading social protection systems
- Objective: deliver stable employment and productivity growth (efficiency) while providing adequate worker protection (equity) in a world with gen AI.  
- Roles of key programs:
  - Unemployment insurance (UI): smooths consumption, enables better job search and matches, potentially improving job quality.
  - Active labor market policies (ALMPs): complementary to UI, can shorten unemployment spells via retraining and reducing information gaps.
  - Social assistance (cash transfers): provides financial support to low-income households during prolonged unemployment.

### Lessons from past automation waves
- Automation's labor-market impact hinges on whether technology substitutes for or complements worker tasks. Recent evidence shows displacement of routine tasks, falling average wages, and intensified job polarization.  
- Sectoral evidence: increased use of industrial robots in the United States reduced employment and wages—especially for manual and routine jobs—with displaced workers moving into lower-paying occupations. Similar displacement of lower-skilled workers observed in Europe, with worker adaptation over time in some contexts.  
- Conceptual and empirical approach used in this note: new evidence at the US commuting-zone level on how social protection (UI and social assistance) moderated long-term effects of industrial robots on employment and wages. Exposure to robots is measured via an adjusted Bartik-type measure combining industry-level robot use and baseline employment shares, adjusted for industry output expansion.

### Cushioning effect of unemployment insurance (empirical results)
- Empirical results (based on instrumental variable regressions, cross-sectional US commuting-zone data over 2000–07) indicate:
  - The impact of robotization on employment does not depend on UI generosity.
  - States with more generous UI benefits saw a smaller decline in wages as a result of robotization—about two-thirds smaller than other states.
  - This suggests more generous UI allows displaced workers to find jobs that better match their skills, improving labor allocation and potentially increasing worker productivity.
  - The cushioning effect on wages is particularly pronounced for workers without a college degree; differences between high and low UI states for workers with less than a college degree are statistically significant at the 1 percent level.
  - Regression specifications control for commuting zone demographics, the share of employment in manufacturing, exposure to Chinese imports, and the share of employment in routine jobs; UI generosity is measured as the product of the maximum legal benefit amount and its duration; sample period for dependent-variable changes is 2000–07.  

*Source: sdnea2024002 - Introduction*

### 0.3 percentage point (3 percent increase).

### sdnea2024002 - 0.3 percentage point (3 percent increase).

### Effect of robotization and generative AI on poverty and labor markets
- Robotization is estimated to increase poverty by "0.3 percentage point (3 percent increase)".  
- Most of the increase in poverty from robotization is attenuated in commuting zones where social assistance is relatively more generous.  
- Robotization historically displaced workers in routine and manual tasks; gen AI could potentially replace a broader spectrum of both routine and high-skill nonroutine tasks, implying more widespread impacts and a need for more fundamental changes in education, training, and policy frameworks.

### Lessons from past technological transitions
- The design of social protection systems played a role in ameliorating adverse labor market and poverty impacts in past automation episodes.  
- Technology-induced labor displacement often proceeds over a generation, with older workers leaving the workforce and fewer younger workers entering affected jobs.  
- If gen AI substitutes across a wider set of tasks, adjustment costs may be larger and require upgrades to existing systems.

### Strengthening social spending during rapid technological transitions
- Transition risks and model baseline:
  - A model-based HANK-DGSE analysis simulates a sizable acceleration in productivity of automated capital in one sector.
  - The increase in automation leads to a gradual reduction in labor demand in the affected sector, producing a "20 percent decline in employment in the new steady state (approached in about 15 years)."
  - The new steady state shows an increase in wages of "about 15 percent" as automation raises aggregate productivity.
  - McKinsey projection referenced: automation could replace "20 to 30 percent" of time spent on work activities by 2030 (model captures only the extensive margin).
- Transition costs and heterogeneity:
  - Unemployment rises temporarily because of costs of relocating workers across sectors and search/matching frictions.
  - Unemployment disproportionately hurts the most vulnerable groups, with substantial falls in consumption of unemployed workers.
  - Incomplete insurance creates precautionary saving motives that propagate shocks.
- UI design trade-offs and illustrative policies:
  - Two illustrative UI designs:
    - (1) Permanent increase in the UI replacement ratio by "1 percentage point".
    - (2) Temporary asymmetric adjustment: increase replacement ratio in proportion to the previous quarter’s unemployment gap by a factor of "0.6" once unemployment rises by more than "1 percentage point" relative to steady state.
  - Both options mitigate consumption drops for unemployed workers; permanent increases can discourage job search and raise unemployment, lowering overall welfare.
  - A temporary UI adjustment aligned with unemployment levels appears to yield the highest welfare benefits in the simulations.
  - Suggested UI design features for severe disruption scenarios: benefits depending on duration of unemployment spells, better linkage with training and re-skilling programs, portability, extended scope and flexibility, and consideration of wage insurance for displaced workers.
- Combining UI and ALMPs:
  - Active labor market policies (ALMPs) such as training and skill development improve employability and matching quality and, when combined with UI, can reduce transition costs and accelerate labor reallocation.
  - Policy efficacy depends on pace of automation and size of transition costs; skill specificity will require different ALMPs and broader access to effective training.

### Distinct challenges for emerging market and developing economies (EMDEs)
- EMDEs tend to have:
  - Lower share of high-skill occupations (less exposed to some AI impacts) but are also less prepared to adopt AI.
  - More limited labor market policies and social protection due to large informal sectors, budget constraints, and less institutional capacity.
  - A larger share of young people not in employment, education, or training, raising adjustment concerns.
  - Larger insurance and credit market imperfections and less personal wealth, increasing the potential welfare gains from UI.
- Model illustration for EMDEs:
  - Simulations calibrate smaller and more gradual exposure to automation but more constrained policy space.
  - Two policy scenarios for EMDEs: one with larger, faster deployment of UI and ALMPs (deployment "2.5 years after the initial shock") and one with smaller, delayed implementation (ALMPs "five years after the initial shock").
  - Outcomes averaged over "60 quarters"; bubble size in the figure proportional to time unemployment rate is "1 percentage point" above initial steady state.

### Upgrading unemployment insurance and social assistance systems
- UI features and gaps:
  - Coverage, generosity, and design matter for cushioning AI effects.
  - Maximum benefit duration is usually less than "12 months" in many countries; most US states provide a maximum of "26 weeks".
  - Access to UI is fragmented, not portable, and minimum contribution durations often exclude temporary workers and (re-)entrants; maximum benefit durations often exclude the long-term unemployed.
  - Policy options: extend scope, portability, flexibility, and consider wage insurance as a temporary subsidy for displaced re-employed workers.
- Social assistance enhancements:
  - Options range from enhanced means-tested guaranteed minimum income programs with integrated training/job transition services to unconditional benefits for all (the latter likely fiscal costly and not desirable given current limited widespread labor market impacts).
  - Integration of ALMPs with social assistance is advised (for example, conditional continued eligibility tied to participation in job search support or skills training).
- Systems and infrastructure:
  - Robust, universal information systems for beneficiary identification and verification are required, integrated across programs with efficient delivery and strong institutional frameworks.
  - For EMDEs, digital technologies can enable rapid expansion of coverage, but trade-offs between coverage and fiscal cost exist.

### Upskilling, training, and employer roles
- Lifelong learning and preemptive skill acquisition are important; examples include unconditional adult training grants (Singapore).
- Evidence:
  - Sector-based training programs in the United States yielded earnings gains of "14–38 percent" in the year following training completion (with persistent gains).
  - Upskilling can outperform on-the-job training for workers displaced by offshoring.
- Policy implications:
  - Assess viability of re-skilling versus alternatives (for example, early retirement) especially for older workers.
  - Employer-provided training may need substitution or augmentation by public ALMPs.

### Tax systems, AI, and investment incentives
- Current tax differentiation:
  - Tax systems differentiate asset categories (equipment, structures, inventory, intellectual property) via measures such as accelerated depreciation and investment tax credits, summarized by marginal effective tax rate (METR).
  - Neutral tax systems equalize METRs across assets; differences create incentives/disincentives.
- Complementarity with labor:
  - Evidence suggests:
    - Machinery and equipment tend to be labor complements.
    - Computer hardware complements high-skilled labor but substitutes for low-skilled labor.
    - Software is generally found to replace labor, with an estimated elasticity of substitution of "1.7".
  - Gen AI investments can include both labor-saving and labor-complementing assets (purchased software, data infrastructure, high-performance computing, AI researchers, employee training).
- Existing METR patterns:
  - Tax advantage for intellectual property (acquired patents, utility models, trademarks) is almost universal.
  - Most economies impose a higher tax on acquired software and computer hardware than on buildings, though notable exceptions (countries favoring software/hardware) include Germany, the United States, The Netherlands, New Zealand, Singapore, and Hong Kong SAR.
  - For the median country, tax treatment of software and hardware became more generous by "about 0.8 and 3 percentage points (relative to buildings)" over 2017–22.
- Changes over time:
  - Policy examples: US TCJA (effective 2018) allowed full expensing of acquired software and computer hardware (full expensing will expire by 2026); Germany implemented similar approach in 2021; India reduced statutory CIT in 2020; Hungary, Slovak Republic, and Czech Republic introduced accelerated depreciation for acquired intellectual property in the early 2000s.

### Should AI or automation be taxed?
- Theoretical considerations:
  - Production efficiency principle argues against special taxes on gen AI if capital income taxes should be neutral across assets.
  - Efficiency rationale for a temporary automation tax: internalize social costs of excessive job displacement under high transition costs and labor market frictions.
  - Equity rationale: mitigate wage inequality if automation increases capital or top-income shares; predistribution arguments can support taxation of automation.
- Quantitative estimates:
  - Costinot and Werning (2023) estimate an optimal tax rate on robots between "1 and 3.7 percent" of robot price, with higher optimal rates for more disruptive technologies.
- Model-based illustration:
  - HANK-DGSE simulations show a temporary automation tax financing UI can improve welfare when transition costs are high by discouraging substitution of labor and internalizing labor market/credit frictions.
  - Trade-offs: automation tax reduces short-to-medium term wages (due to lower productivity) but can mitigate unemployment spikes and yield short-term welfare gains; welfare likely falls if transition costs are modest.
- Practical considerations and recommendation:
  - Implementing taxes targeted to AI/automation is practically difficult: identifying tax base, asset definitions, relabeling risks, and international mobility/avoidance.
  - A specific tax on gen AI is therefore not recommended.
  - Alternative: reconsider preferential tax treatments that effectively favor labor-displacing assets (adjust METR differentials), and use AI regulation (for example, EU AI Act requiring employer notification before deploying “high-risk AI systems”) to include labor-market considerations.

### Broadening gains through taxation and redistribution
- Importance of capital income taxation:
  - Capital income is considerably more concentrated among top income groups than labor income; rising capital shares can amplify inequality.
  - Declining effective taxation of capital since the 1980s has widened the gap between average tax rates on capital and labor to almost "10 percentage points" by 2018 in advanced economies.
  - In EMDEs average tax rates on both labor and capital are generally much lower; capital income may be taxed more than labor due to corporate tax importance.
- Policy measures to strengthen capital taxation and revenue:
  - Strengthen corporate income tax (CIT) and leverage global minimum tax to reduce tax competition.
  - Consider supplemental taxes on excess profits from monopoly rents (not necessarily AI-specific).
  - Improve enforcement (for example, automatic exchange of information) and employ gen AI to enhance enforcement against tax evasion.
  - Enhance taxation of capital gains, which are often preferentially treated and concentrated among top earners.
  - Consider targeted labor tax relief via income tax credits or job credits for employers (examples: EITC) but recognize potential practical obstacles (relabeling, targeting challenges).
- Funding social protection:
  - Effective taxation of capital income becomes more relevant as labor income share declines and needs to finance upgraded social protection systems.

### Fiscal policies for AI innovation, deployment, and administrative capacity
- Fiscal roles in innovation:
  - Past AI advances benefited from decades of public funding and subsidies; as AI enters commercial adoption phase, fiscal support should prioritize areas with high social returns:
    - Fundamental research, digital and electricity infrastructure (especially in EMDEs), AI applications in public sector services, and funding for upskilling/re-skilling and social protection upgrades.
- Scope for AI in EMDE development:
  - AI can improve education, targeted human capital investments, financial access, and risk management (disease prevention, disaster management).
  - Constraints: weak digital infrastructure, limited entrepreneurial ecosystems, and scarce local AI expertise—areas for government intervention.
- Need to upgrade administrative and governance capacity:
  - Governments must invest in expertise to select and vet funding, update regulation as AI evolves, and develop procurement capacity.
  - A dedicated agency (inspired by models like the US National Institutes of Health) could mobilize private sector, academia, and stakeholders to track AI developments and coordinate accountability and data governance.
  - International cooperation is important; a “distributed CERN” model for AI could inform global collaboration on innovation, use, and regulation.

*Italic: Source — sdnea2024002 - 0.3 percentage point (3 percent increase).*

### Conclusions and Policy Implications

### Conclusions and Policy Implications

### Forward-thinking policies
- Uncertainty about the speed of AI progress and its economic impact in the short, medium, and long term requires preparedness for alternative scenarios (Korinek and Suh 2024).
- Policy responses should differ depending on whether AI:
  - Raises labor productivity and creates new jobs, or
  - Rapidly displaces jobs, reduces wages, and increases inequality.
- Countries must assess whether social protection, education, and tax systems are fit for purpose and flexible enough to cope with a wide range of potential scenarios.

### The future of social protection
- AI-induced labor market transformations may redefine employment and the skills demanded by employers, requiring comprehensive reassessment of labor policies and social protection mechanisms.
- If AI behaves like a general-purpose technology, a range of work tasks will be automated but new opportunities will also arise; in this case:
  - Current social protection systems provide a solid foundation, particularly in advanced economies.
  - Unemployment insurance eligibility rules should be robust to radical uncertainty and coverage broadened to encompass self-employed workers and those with atypical employment contracts.
  - Active labor market policies (ALMPs) should aim to improve skill acquisition for workers capable of adapting to new market requirements.
  - Social assistance benefits should target those permanently displaced or indirectly affected by labor market disruptions.
  - Sector-based training, apprenticeships, and upskilling and re-skilling programs could play a greater role in helping workers transition to new tasks and sectors.
- If the nature of work changes dramatically (for example, if tasks became increasingly unnecessary), policy responses might include:
  - Broader sharing of gains via unconditional transfers.
  - Consideration of the design and infrastructure required for unconditional transfer policies.
- AI could be leveraged to improve efficiency and quality of social protection systems, while being used in conjunction with traditional systems to reduce data privacy risks.

### Educational systems
- Education and training policies should be geared to:
  - Upskilling workers to cope with structural changes in the workplace.
  - Matching skill and task demands of new technologies (OECD 2023c).
- Spending matters, but quality and adaptability of education are decisive in preparing workers for change.
- Given high uncertainty about which skills will be needed, educational systems must be flexible in responding to market demands while keeping equity and access in mind.
- Educational systems could take advantage of gen AI to foster higher-level skills such as critical thinking, analysis, and strategy.
- Develop skills to work alongside AI systems, not only with existing technologies.

### Taxing AI
- A special tax on gen AI to slow adoption and prevent labor displacement would be hard to design and implement and risks hampering productivity growth, including where AI augments labor.
- Countries should reconsider the design of corporate tax systems and how they incentivize investments in automation:
  - Reconsideration of tax incentives, such as capital allowances, in countries where they are more generously applied to labor-displacing software or intangibles than to other assets.
  - Where corporate tax systems impose much higher tax burdens on AI, deployment may be held up and productivity growth reduced.
- Consider income tax credits and job credits to mitigate excessive labor displacement from automation, even if not targetable to particular occupations.
- Given large energy consumption by AI servers, taxing associated carbon emissions is recommended to reflect external environmental costs in the price of the technology.

### Capital taxation
- The average tax on capital income has declined in advanced economies over past decades; reversing this trend is important, especially under a disruptive AI scenario.
- Rationales for enhancing capital income taxes:
  - Low taxation of capital relative to labor can contribute to excessive labor displacement and exacerbate labor market frictions.
  - Capital income taxes help address increasing inequality from rising market power and economic rents in winner-take-all markets.
  - Large labor displacements that reduce the labor income share will erode the tax base and reduce public revenue.
  - Enhancing capital income taxes will boost revenue mobilization needed to finance higher education and social spending in the arrival of automation.
- Policy actions include:
  - Restoration of the corporate income tax.
  - Well-designed excess profit taxes.
  - Higher personal income taxes on capital through better enforcement of automatic information exchange between countries.
  - Enhanced taxation of capital gains.

### Advancing tax systems through gen AI
- Gen AI has significant potential to further advance tax administration practices and improve tax enforcement, building on digital transformations that have already visibly reduced tax evasion (Amaglobeli and others 2023).
- The AI-associated information revolution can enable tax system redesign by transforming information systems and management.
- Potential innovations in tax design enabled by improved information:
  - A personalized progressive value-added tax.
  - An income tax based on lifetime income.
  - A real-time market-value-based property tax.

### AI that serves people
- Fiscal policies can promote AI innovation and deployment in applications with greater social benefits (education, health care, government).
- Finding the right policy response requires upgrading administrative and analytical capacity to monitor and evaluate trends in technological advances.
- Ensuring AI deployment for the common good and equitable distribution of benefits requires governmental and intergovernmental action and innovative engagement of the private sector, academia, and civil society.
- Innovation and deployment policies should work in tandem with social protection, education, tax, and regulatory policies to broaden the gains of gen AI for all.

### Box 1 — Taxation and the Decline of the Labor Income Share: Key findings
- The labor income share has fallen steadily since the 1980s in most advanced economies.
- Lower corporate income tax (CIT) rates can affect the labor income share through several channels, including:
  - Boosting investment in labor-substituting capital and reducing employment and wages.
  - Inducing self-employed entrepreneurs to report income as profits instead of wages, artificially increasing the capital share.
  - Offsetting effects where capital and labor are complementary, raising labor productivity and wages.
- Empirical panel regression (42 advanced and emerging market economies) results:
  - For each percentage point reduction in the statutory CIT rate, the labor share falls 0.1 percent.
  - For each percentage point reduction in the top statutory PIT rate, the labor share increases 0.11 percent.
  - The reduction in the average statutory CIT rate from 27.7 to 23.9 percent during 2005–18, combined with a slight increase in the average top PIT rate from 41.9 to 43.7 percent, is estimated to have reduced the labor share by 0.58 percentage point in advanced economies over this period.
- Contextual projection:
  - Using a CES function with equal shares of the two sectors and an elasticity of 0.9 results in similar findings.

### Annex 1 — Model framework and calibrated scenarios
- Model extension:
  - Extends a tractable HANK-DGSE model with labor market frictions (Ravn and Sterk 2021) to incorporate automation.
  - Two intermediate sectors employ labor, traditional capital, and automated capital; automated capital can substitute for labor (Berg, Buffie, and Zanna 2018).
  - Only one sector is subject to an automation shock to capture cross-sector labor flows and study policies supporting sectoral mobility.
  - Final goods produced with a Cobb-Douglas function: y_t = A_t (y_t^1)^ψ (y_t^2)^(1−ψ).
- Labor market features:
  - Search and matching frictions follow Diamond-Mortensen-Pissarides tradition.
  - Three household types: firm owners, employed workers, unemployed workers.
  - Job separation probability and search frictions create idiosyncratic income risk; employed workers self-insure via precautionary saving.
  - Policy tools modeled: unemployment insurance and active labor market policies facilitating sectoral mobility, funded with labor income taxes and ensuring budget neutrality each period.
  - Nominal rigidities and a Taylor-rule-based interest rate included.
- Sectoral production specifics:
  - CES production with elasticity parameters and automated capital productivity A_t^iii.
  - Labor–automated capital bundle v_t^i defined with elasticity and labor share s_i.
- Baseline simulation and shocks:
  - A series of shocks increase the productivity of automated capital by 300 percent by 2030, resulting in a decrease of labor in the affected sector by 20 percent.
  - This is consistent with McKinsey (2023) projection that automation could replace the time spent on work activities by 20 to 30 percent by 2030; the model captures only the extensive margin while the projection includes both extensive and intensive margins.

### Calibration — selected parameters and targets
- Targets used to calibrate labor market parameters:
  - Unemployment rate target: 7 percent (ranges between 5 percent in the US and 10 percent in euro area countries).
  - Job-finding probability target: 45 percent per quarter.
  - Vacancy-filling probability target: 70 percent per quarter.
  - Replacement ratio set at 50 percent.
  - Other parameters calibrated to match a 20–30 percent consumption loss upon unemployment.
- Key calibrated parameter values (selected):
  - Elasticity of substitution in labor-automated capital (LAC) bundle: 2.5.
  - Elasticity between traditional capital and the LAC: 0.50.
  - Share of labor in LAC bundle s_i: 0.99 (Labor income share of 61 percent).
  - Share of traditional capital in sectoral production α_i: 0.54 (Traditional capital income share of 35 percent).
  - Share of sectoral input in final goods production ψ: 0.5 (Assumption of symmetric sectors).
  - Separation rate ρ: 0.06.
  - Matching efficiency m̄: 0.56.
  - Matching elasticity μ: 0.5.
  - Vacancy cost κ: 0.04 (About 15 percent of average wage as in standard value).
  - Worker’s wage bargaining power ψ̄: 0.65.
  - Replacement rate initial value r̄: 0.5.
  - Other benefit of unemployed worker r: 0.26.
  - Response of unemployment income support to unemployment rate λ: 0.4.
  - Persistence of hiring cost ρ_tttt: 0.5.
  - Hiring cost parameter γ: 0.15.
  - Discount factor β: 0.989.
  - Firm owners’ risk aversion σ_o: 0.283.
  - Depreciation rate δ: 0.025.
  - Price elasticity ε_p: 6.
  - Price adjustment cost κ_p: 30 (matches average frequency of price changes every three quarters).
  - Taylor rule smoothing ρ: 0.5.
  - Taylor rule response to inflation γ_π: 2.

*International Monetary Fund — Conclusions and Policy Implications (sdnea2024002)*

### 0.125 Standard value

### 0.125 Standard value

### Smaller Impact Scenario
- Scenario setup:
  - Sectoral productivity shocks to automated capital are about half the magnitude of those in the baseline scenario.
- Key quantitative outcomes:
  - Output increases by about 17 percent.
  - The labor income share in the affected sector falls by 10 percentage points.
  - Total labor income share in the economy is reduced by 5 percentage points.
- Dynamics and interpretation:
  - The transition remains costly due to sectoral mobility frictions, but it is less severe than in the baseline.
- Sources and notes:
  - Results and figures are from IMF staff calculations; the alternative scenario is similar to Cazzaniga and others (2024).

### Annex 2. Effective Tax Rates

H3: Marginal Effective Tax Rates (METRs)
- Concept and measurement:
  - METRs capture incentives for incremental investment decisions at the breakeven point where the marginal product of capital equals the cost of capital.
  - METRs are calculated using tax code parameters: statutory tax rate, tax depreciation, deductions for financing costs, and other capital allowances or tax credits.
  - METR = 0 implies tax neutrality for investment; positive and equivalent METRs across asset types distort overall investment but not allocation between assets.
- Data sources and coverage:
  - Two internationally comparable METR datasets used: OECD and ZEW Leibniz Centre for European Economic Research.
  - ZEW data: three main assets (buildings, machinery, acquired patents) for 34 countries over 1998–2020.
  - OECD data: eight different assets (including acquired software and computer hardware) for 74 countries over 2017–22.
- Empirical pattern:
  - Current METRs (based on equity finance and averaged across 74 countries) are well above zero.
  - The bulk of observations are in the range of 12.5 to 35 percent.
- Additional note:
  - Annex Figure 2.2 reports corporate tax bias for labor-saving assets by economy (Bottom 10), where bias is measured as the METR for each asset type relative to the METR for buildings; negative (positive) values denote lower (higher) METR relative to buildings.

H3: Average Tax Rates (ATRs)
- Construction and attribution:
  - ATRs relate historical tax revenues to the tax base and follow Bachas and others (2022).
  - Tax categories (OECD Revenue Statistics methodology): category 1200 (CIT) fully attributed to capital; category 2000 (SSC) fully attributed to labor; category 1100 (PIT) attributed partially to capital and partially to labor (PIT split varies by country and year).
  - The note applies a different definition of ATR on capital income by excluding property and wealth taxes to better reflect taxes affecting firms’ automation decisions.
  - Consumption taxes are excluded.
  - Totals are divided by capital and labor income shares computed from national accounts data.
- Historical trends:
  - ATR on corporate income and on capital income (at the personal level) have both decreased since 1980.
  - Capital income ATR saw significant reductions in the early 1980s and the early 2000s.
  - The CIT component is more affected by business cycles and has trended downward since the late 1970s.
  - Since 2016, automatic exchange of bank information may have contributed to a slight recovery of personal tax components.
- Visualization:
  - Annex Figure 2.3 shows trends in the ATRs for labor and capital in individual countries (sources: OECD; IMF staff calculations).

### Annex 3. Corporate Taxes and Labor Income Share

H3: Data
- Labor income share:
  - Source: Bachas and others (2022) provide a long time series for 155 countries since 1965 using national accounts (SNA).
  - Labor income share includes compensation of employees plus a share of mixed income (operating surplus of private unincorporated enterprises).
  - Benchmark assumption: labor share of mixed income = 75 percent (i.e., 25 percent of mixed income treated as capital income); this share is time- and country-invariant in the series.
- Country-level tax rates and macro variables:
  - Statutory corporate and personal income tax rates: IMF Fiscal Affairs Department Tax Rate Database.
  - Nontax macro determinants: average hourly wage, average price of capital relative to consumption, trade openness, population, financial development index, inflation from the IMF World Economic Outlook database.

H3: Specification
- Baseline regression (sample: 42 advanced and emerging market economies during 1990–2008):
  - ALA_it = β1 CIT_it + β2 PIT_it + Λ′ X_it + α_i + η_t + ε_it
  - ALA_it: labor income share in country i, year t.
  - X_it: vector of structural and institutional characteristics including:
    - labor costs = log of average labor income per hour worked (US dollars, constant prices);
    - capital costs = average price of capital relative to consumption;
    - trade openness = (exports + imports) / GDP;
    - level of financial development.
  - CIT_it: statutory corporate income tax rate.
  - PIT_it: top statutory personal income tax rate.
  - α_i: country fixed effects; η_t: year fixed effects; ε_it: error term.

H3: Results
- Summary of regression findings (Annex Table 3.1):
  - Column (1) to (4) correspond to specifications for full sample, full sample with PIT, advanced economies, and emerging market & developing economies, respectively.
  - Statutory CIT rate coefficients:
    - Column (1): 0.01 (standard error (0.02))
    - Column (2): 0.02 (standard error (0.02))
    - Column (3) [advanced economies]: 0.10* (standard error (0.06)) — significant at the 10 percent level
    - Column (4) [emerging market & developing]: -0.01 (standard error (0.03))
  - Statutory PIT rate coefficients (where included):
    - Column (2): -0.04** (standard error (0.02))
    - Column (3): -0.11*** (standard error (0.04))
    - Column (4): -0.06** (standard error (0.03))
  - Controls: included in all specifications (Y).
  - Year FE: Y in all columns.
  - Country FE: Y in all columns.
  - Observations:
    - Column (1): 1086
    - Column (2): 1086
    - Column (3): 399
    - Column (4): 687
  - R squared:
    - Column (1): 0.89
    - Column (2): 0.89
    - Column (3): 0.93
    - Column (4): 0.91
  - Significance notation:
    - *p < .10; **p < .05; ***p < .01.
- Interpretation of coefficients (from specification (3), advanced economies):
  - A 1 percentage point reduction in the statutory CIT rate, all else equal, reduces the labor income share in advanced economies by 0.1 percentage point.
  - A 1 percentage point reduction in the top statutory PIT rate increases the labor income share by 0.11 percentage point.
- Historical context and implied effect:
  - The average statutory CIT rate decreased from 27.7 to 23.9 percent in the sample of 32 advanced countries during 2005–18, a reduction of 3.8 percentage points.
  - The average top statutory PIT rate increased from 41.9 to 43.7 percent during the same period, an increase of 1.8 percentage points.
  - Together, holding other factors constant, these changes would imply a reduction in labor share of 0.58 percent (Box Figure 1.1).

*Source: IMF staff compilation and calculations as presented in sdnea2024002 - 0.125 Standard value.*

### References

### References (sdnea2024002)

### AI economics, automation, and technological transitions
- Acemoglu, Daron. 2021. “ Harms of AI.”  NBER Working Paper 29247, National Bureau of Economic Research, Cambridge, MA
- Acemoglu, Daron. 2024. “The Simple Macroeconomics of AI.” NBER Working Paper 32487, National Bureau of Economic Research, Cambridge, MA.
- Acemoglu, Daron, David Autor and Simon Johnson. 2023. “Policy Insight 123: Can We Have Pro-Worker AI? Choosing a Path of Machines in Service of Minds”, CEPR Policy Insight No 123, CEPR Press, Paris & London.
- Acemoglu, Daron, and Simon Johnson. 2023. Power and Progress: Our 1000-Year Struggle over Technology and Prosperity. New York: Public Affairs.
- Acemoglu, Daron, and Pascual Restrepo. 2018. “ The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment.” American Economic Review 108 (6): 1488–542.
- Acemoglu, Daron, and Pascual Restrepo. 2020. “ Robots and Jobs: Evidence from US Labor Markets.” Journal of Political Economy 128 (6): 2188–244.
- Acemoglu, Daron, Claire Lelarge, and Pascual Restrepo. 2020. "Competing with Robots: Firm-Level Evidence from France." AEA Papers and Proceedings, 110: 383-88.
- Adão, Rodrigo, Martin Beraja, and Nitya Pandalai-Nayar. 2024. “ Fast and Slow Technological Transitions.” https://economics.mit.edu/sites/default/files/inline-files/transitions_JPE_Macro_revised_final.pdf.
- Aghion, Philippe, Celine Antonin, Simon Bunel, and Xavier Jaravel.  2022. “The Effects of Automation on Labor Demand: A Survey of the Recent Literature.”  CEPR Discussion Paper DP16868, Centre for Economic Policy Research, London. https://ssrn.com/abstract=4026751
- Autor, D. 2022. “The Labor Market Impacts of Technological Change: From Unbridled Enthusiasm to Qualified Optimism to Vast Uncertainty.” NBER Working Paper 30074, National Bureau of Economic Research, Cambridge, MA
- Autor, David. 2024. “Applying AI to Rebuild Middle Class Jobs.”   NBER Working Paper 32140, National Bureau of Economic Research, Cambridge, MA.
- Brynjolfsson, Erik, Danielle Li, and Lindsey R. Raymond. 2023. “Generative AI at Work.”   NBER Working Paper 31161, National Bureau of Economic Research, Cambridge, MA.
- Eloundou, Tyna, Sam Manning, Pamela Mishkin, and Daniel Rock. 2023. “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.” https://doi.org/10.48550/arXiv.2303.10130.
- Felten, Edward W., Manav Raj, and Robert Seamans. 2023. “ Occupational Heterogeneity in Exposure to Generative AI.”  https://dx.doi.org/10.2139/ssrn.4414065.
- Noy, Shakked, and Whitney Zhang. 2023. “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence.” Science 381 (6654):  187–92.
- Korinek, Anton. 2023a. “Generative AI for Economic Research: Use Cases and Implications for Economists.” Journal of Economic Literature 61 (4):1281–317.
- Korinek, Anton. 2023b. “Scenario Planning for an A(G)I Future.” Finance and Development 60 (4): 30–33.
- Korinek, Anton, and Donghyun Suh. 2024. “  Scenarios for the Transition to AGI.” NBER Working Paper 32255, National Bureau of Economic Research, Cambridge, MA.
- Restrepo, Pascual.  2023. “Automation: Theory, Evidence, and Outlook.”  NBER Working Paper 31910, National Bureau of Economic Research, Cambridge, MA.
- Prettner, Klaus, and Holger Strulik.  2020. “Innovation, Automation, and Inequality: Policy Challenges in the Race Against the Machine.” Journal of Monetary Economics 116: 249–65.
- Frey, Carl Benedikt (2019). The Technology Trap: Capital, Labor, and Power in the Age of Automation. Princeton: Princeton University Press.
- Baldwin, Richard. The Globotics Upheaval: Globalization, Robotics, and the Future of Work. 2019. Oxford, UK: Oxford University Press.
- Berg, Andrew, Edward F. Buffie, and Luis-Felipe Zanna. 2018. Should We Fear the Robot Revolution? (The correct answer is yes). Journal of Monetary Economics 97: 117–48.
- Dauth Wolfgang, Sebastian Findeisen, Jens Suedekum, and Nicole Woessner. 2021. “The Adjustment of Labor Markets to Robots.” Journal of the European Economic Association 19 (6): 3104–153.
- Graetz, Georg, and Guy Michaels. 2018 “Robots at Work.” Review of Economics and Statistics 100 (5): 753–68.
- Thuemmel, Uwe. 2023.  “Optimal Taxation of Robots.” Journal of the European Economic Association 21 (3): 1154–90.
- ZeW. 2021. “Effective Tax Levels Using the Devereux/Griffith Methodology—Update 2020, Final Report.” Project for the EU Commission TAXUD/2020/DE/308, Leibniz Centre for European Economic Research, Mannheim.
- Webb, Michael. 2020. “The Impact of Artificial Intelligence on the Labor Market.” Unpublished.

### Taxation, corporate tax policy, and public finance
- Acemoglu, Daron, Andrea Manera, and Pascual Restrepo. 2020. “Does the U.S. Tax Code Favor Automation?” Brookings Papers on Economic Activity ( Spring): 231–300.
- Amaglobeli, David, Ruud A. de Mooij,  Andualem Mengistu, Mariano Moszoro, Manabu Nose, Soheib Nunhuck,  Sailendra Pattanayak,  Lorena Rivero del Paso, Frankosiligi Solomon, Rebecca Sparkman,  Hervé Tourpe, Gerardo Uña. 2023. Transforming Public Finance Through GovTech, IMF Staff Discussion Note 2023/4, International Monetary Fund, Washington, DC.
- Banks, James, and Peter Diamond. 2011. “The Base for Direct Taxation.” In Dimensions of Tax Design, edited by Stuart Adam, Tim Besley, Richard Blundell, Stephen Bond, Robert Chote, Malcolm Gammie, Paul Johnson, Gareth Myles, and James Poterba, 548–674. Oxford, UK: Oxford University Press.
- Bond, Stephen,  and Jing Xing. 2015. “ Corporate Taxation and Capital Accumulation: Evidence from Sectoral Panel Data for 14 OECD Countries.” Journal of Public Economics 130: 15–31.
- Devereux, Michael P., Li Liu, and Simon Loretz. 2014. “ The Elasticity of Corporate Taxable Income: New Evidence from UK Tax Records.” American Economic Journal: Economic Policy 6 (2): 19–-53.
- de Mooij, Raud A., and Gaëtan Nicodème. 2008. “ Corporate Tax Policy and Incorporation in the EU.” International Tax and Public Finance 15:  478–98.
- Hebous, Shafik,  and Michael Keen. 2023. “Pareto-Improving Minimum Corporate Taxation.” Journal of Public Economics 225: 104952.
- Hebous, Shafik, Alexander D.  Klemm, Geerten Michielse, and Carolina Osorio Buitron. 2024. “How to Tax Wealth.” IMF How-to Note 2024/001, International Monetary Fund, Washington, DC.
- Hebous, Shafik, Dinar Prihardini, and Nate Vernon. 2022. "Excess Profit Taxes: Historical Perspective and Contemporary Relevance." IMF Working Paper 2022/187, International Monetary Fund, Washington, DC.
- Kaymak, Barış,  and Immo Schott. 2023. “Corporate Tax Cuts and the Decline of the Manufacturing Labor Share.”   International Finance Discussion Paper 1379, Board of Governors of the Federal Reserve System, Washington, DC.
- Kaymak, Barış, and Immo Schott. 2023. “Corporate Tax Cuts and the Decline of the Manufacturing Labor Share.” International Finance Discussion Paper 1379, Board of Governors of the Federal Reserve System, Washington, DC.
- Kazzaniga, Mauro, Florence Jaumotte, Longji Li, Giovanni Melina, Augustus J. Panton, Carlo Pizzinelli, Emma J. Rockall, and Marina Mendes Tavares. 2024. “Gen-AI: Artificial Intelligence and the Future of Work.” IMF Staff Discussion Note SDN2024/001, International Monetary Fund, Washington, DC.
- OECD. 2023d. “Corporate Tax Statistics 2023.” Paris.  https://doi.org/10.1787/f1f07219-en
- Zwick, Eric, and James Mahon. 2017. “Tax Policy and Heterogeneous Investment Behavior.” American Economic Review 107 (1): 217–48.
- Liu, Li.  2011. “Do Taxes Distort Corporations’  Investment Choices? Evidence from Industry-Level Data.” University of Oxford Centre for Business Taxation Working Paper 2010/104, Oxford, UK.
- Maffini, Giorgia, Jing Xing, and Michael P. Devereux. 2019. “The Impact of    Investment Incentives: Evidence from UK Corporation Tax Returns.” American Economic Journal: Economic Policy 11 (3): 361–89.
- Ohrn, Eric. 2019. “ The Effect of Tax Incentives on U.S. Manufacturing: Evidence from State Accelerated Depreciation Policies.” Journal of Public Economics 180:  104084.
- Fuest, Clemens, Andreas Peichl, and Sebastian Siegloch. 2018. "Do Higher Corporate Taxes Reduce Wages? Micro Evidence from Germany." American Economic Review 108 (2): 393–418.
- Garrett, Daniel G., Eric Ohrn, and Juan Carlos Suárez Serrato. 2020. “ Tax Policy and Local Labor Market Behavior.” American Economic Review: Insights 2 (1): 83–100.
- Hebous, Shafik, Alexander D. Klemm, Geerten Michielse, and Carolina Osorio Buitron. 2024. “How to Tax Wealth.” IMF How-to Note 2024/001, International Monetary Fund, Washington, DC.

### Labor markets, social protection, and policy interventions
- Autor, David H., Lawrence F. Katz, and Alan B. Krueger. 1998. Computing Inequality: Have Computers Changed the Labor Market? Quarterly Journal of Economics 113 (4):  1169–213.
- Autor, David H., William R. Kerr, and Adriana D. Kugler. 2007. “Does Employment Protection Reduce Productivity? Evidence from US States.” Economic Journal 117 (521): 189–217.
- Berman, Eli, John Bound, and Zvi Griliches. 1994. “Changes in the Demand for Skilled Labor within U.S. Manufacturing: Evidence from the Annual Survey of Manufacturers.” Quarterly Journal of Economics 109 (2): 367–97.
- Braxton, J. Carter, and Bledi Taska. 2023. “Technological Change and the Consequences of Job Loss.” American Economic Review 113 (2): 279–316.
- Brollo,  Fernanda. 2024. “Strengthening Social Protection to Pave the Way for Technological Innovation: Evidence from the U.S.” IMF Working Paper 2024/095, International Monetary Fund, Washington, DC.
- Brollo Fernanda, David Coady, Samir Jahan, and Riki Matsumoto. 2024a. “Challenges Facing SSNs in Emerging and Developing Economies.” IMF Working Paper 2024/096, International Monetary Fund, Washington, DC.
- Brollo Fernanda, Gabriel Lara Ibarra, and Ricardo Campante Vale. 2024b. “Strengthening Income Stabilization through Social Protection in Emerging and Developing Economies: The Brazilian Experience.” IMF Working Paper 2024/052, International Monetary Fund, Washington, DC.
- Brussevich, Mariya, Era Dabla-Norris, and Salma Khalid. Forthcoming. “Is Technology Widening the Gender Gap? Automation and the Future of Female Employment.” International Monetary Fund, Washington, DC.
- Hyman, Benjamin G., Brian K. Kovak, Adam Leive, and Theodore Naff. 2021. “ Wage Insurance and Labor Market Trajectories.” AEA Papers and Proceedings 111: 491–95.
- Humlum, Anders, Jakob Munch, and Mette Rasmussen. 2023. “What Works for the Unemployed? Evidence from Quasi-Random Caseworker Assignments.” University of Chicago Becker Friedman Institute for Economics Working Paper 2023-43, Chicago, IL.
- Katz, Lawrence F., Jonathan Roth, Richard Hendra, and Kelsey Schaberg. 2022. “Why Do Sectoral Employment Programs Work? Lessons from WorkAdvance.” Journal of Labor Economics 40 (S1): S249–S291.
- Nichols, Austin, and Jesse Rothstein. 2015. “The Earned Income Tax Credit (EITC).” NBER Working Paper 21211, National Bureau of Economic Research, Cambridge, MA.
- International Labour Organization (ILO). 2022. “Youth Not in Employment, Education or Training in Asia and the Pacific: Trends and Policy Considerations.” Geneva.
- IMF. 2022. “IMF Engagement on Social Safety Net Issues for Surveillance and Program Work.” IMF Technical Notes and Manuals 2022/000, Washington, DC.
- IMF. 2017a. Fiscal Monitor: Achieving More with Less. Washington, DC, April.
- IMF. 2017b. Fiscal Monitor: Tackling Inequality. Washington, DC, October.
- IMF. 2024. “Expanding Frontiers: Fiscal Policies for Innovation and Technology Diffusion.” Fiscal Monitor, Chapter 2, Washington, DC, April.
- Organisation for Economic Co-operation and Development (OECD). 2023b. “Income Support for Jobseekers: Trade-offs and Current Reforms.” Paris.
- Organisation for Economic Co-operation and Development (OECD). 2023c. “OECD Skills Outlook 2023: Skills for a Resilient Green and Digital Transition.” Paris. https://doi.org/10.1787/27452f29-en
- Pizzinelli, Carlo, Augustus J. Panton, Marina Mendes Tavares, Mauro Cazzaniga, and Longji Li. 2023. “Labor Market Exposure to AI: Cross-Country Differences and Distributional Implications.” IMF Working Paper 2023/216, International Monetary Fund, Washington, DC.
- Chamley, Christophe. 1986. “  Optimal Taxation of Capital Income in General Equilibrium with Infinite Lives.” Econometrica 54 (3): 607–22.
- Judd, Kenneth L. 1985. “Redistributive Taxation in a Simple Perfect Foresight Model. Journal of Public Economics 28 (1): 59–83.
- Marimon, Ramon, and Fabrizio Zilibotti. 1999. “Unemployment vs. Mismatch of    Talents: Reconsidering Unemployment Benefits.” Economic Journal 109 (455): 266–91.

### Macro, monetary policy, and modeling contributions
- Blanchard, Olivier J., and Jean Tirole. 2008. “The Joint Design of Unemployment Insurance and Employment Protection: A First Pass.” Journal of the European Economic Association 6 (1): 45–77.
- Branch, William A., Nicolas Petrosky-Nadeau, and Guillaume Rocheteau. 2016. “Financial Frictions, the Housing Market, and Unemployment.” Journal of Economic Theory 164: 101–35.
- Challe, Edouard. 2020. “Uninsured Unemployment Risk and Optimal Monetary Policy in a Zero-Liquidity Economy.” American Economic Journal: Macroeconomics 12 (2): 241–83.
- Challe, Edouard, Julien Matheron, Xavier Ragot, and Juan F. Rubio-Ramirez. 2017. “Precautionary Saving and Aggregate Demand.” Quantitative Economics 8 (2): 435–78.
- Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans.  2005. “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy.” Journal of Political Economy 113 (1): 1–  45.
- Christoffel, Kai, Keith Kuester, and Tobias Linzert. 2009. “ The Role of Labor Markets for Euro Area Monetary Policy.” European Economic Review  53 (8): 908–36.
- Gertler, Mark, Luca Sala, and Antonella Trigari.  2008. “An Estimated Monetary DSGE Model with Unemployment and Staggered Nominal Wage Bargaining.” Journal of Money, Credit and Banking 40 (8):1713–64.
- Gomes, Sandra, Pascal Jacquinot, and Massimiliano Pisani. 2012. “The EAGLE: A Model for Policy Analysis of Macroeconomic Interdependence in the Euro Area.” Economic Modelling 29 (5):  1686–714.
- Ravn, Morten O.,  and Vincent Sterk.  2021. “Macroeconomic Fluctuations with HANK & SAM: An Analytical Approach.” Journal of the European Economic Association 19 (2): 1162–202.
- Smets, Frank, and Rafael Wouters.  2007. “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach.” American Economic Review 97 (3): 586–606.
- Walsh, Carl E. 2012. Monetary policy and resource mobility. Technical report.

### Measurement, capital, and sectoral investment
- Berndt, Ernst R.,  and Catherine J. Morrison. 1995. “High-Tech Capital Formation and Economic Performance in U.S. Manufacturing Industries; An Exploratory Analysis,” Journal of Econometrics 65 (1): 9-43.
- Schaller, Huntley. 2006. “Estimating the Long-Run User Cost Elasticity.” Journal of Monetary Economics 53 (4): 725–36.
- Fatica, Serena. 2017. “Measurement and Allocation of Capital Inputs with Taxes: A Sensitivity Analysis for OECD Countries.” Review of Income and Wealth 63 (1): 1–29.
- Beraja, Martin, and Nathan Zorzi. 2024 “Inefficient Automation.” The Review of Economic Studies.
- Baily, Martin Neil, Erik Brynjolfsson, and Anton Korinek.  2023. “Machines of Mind: The Case for an AI-Powered Productivity Boom.”  Brookings Institution, Washington, DC.
- House, Christopher L., and Matthew D. Shapiro. 2008. “Temporary Investment Tax Incentives: Theory with Evidence from Bonus Depreciation.” American Economic Review 98 (3): 737–68.
- Schaller, Huntley. 2006. “Estimating the Long-Run User Cost Elasticity.” Journal of Monetary Economics 53 (4):  725–36.
- ZEW. 2021. “Effective Tax Levels Using the Devereux/Griffith Methodology—Update 2020, Final Report.” Project for the EU Commission TAXUD/2020/DE/308, Leibniz Centre for European Economic Research, Mannheim.

### Energy, environment, and cross-cutting analyses
- de Vries, Alex. 2023. “The Growing Energy Footprint of Artificial Intelligence.” Joule 7 (10): 2191–94.
- OECD. 2023a. “OECD Employment Outlook 2023: Artificial Intelligence and the Labour Market.” Paris.
- United Nations.  2023. “Interim Report: Governing AI for Humanity.” New York.

*Broadening the Gains from Generative AI: The Role of Fiscal Policies — Staff Discussion Note No. SDN/2024/002*

---


_Source: https://www.imf.org/-/media/files/publications/sdn/2024/english/sdnea2024002.pdf_