## CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND

## Source details

**Canonical URL:** [CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND](https://www.imf.org/-/media/files/publications/weo/2025/october/english/ch3onlineannexes.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/weo/2025/october/english/ch3onlineannexes.pdf.md)
- [Structured JSON version](/-/media/files/publications/weo/2025/october/english/ch3onlineannexes.pdf.json)

---

### Classification of policy instruments and evaluation
- Direct support measures listed: “Financial grant”, “Production subsidy”, “Tax or social insurance relief”, “State aid, unspecified”, “Tax-based export incentive”, “Price stabilization”, “Export subsidy”, “Other export incentive”, “In-kind grant”, “Import incentive”, “State aid, nes”.
- Other measures listed: “Anti-dumping”, “Controls on commercial transaction and investment instruments”, “Controls on credit operations”, “Export ban”, “Export licensing requirement”, “Export quota”, “Export tariff quota”, “Export tax”, “Export-related non-tariff measure, nes”, “FDI: Entry and ownership rule”, “FDI: Financial incentive”, “FDI: Treatment and operations, nes”, “Import ban”, “Import licensing requirement”, “Import quota”, “Import tariff”, “Import tariff quota”, “Import-related non-tariff measure, nes”, “Internal taxation of imports”, “Labour market access”, “Local content incentive”, “Local content requirement”, “Local labour requirement”, “Local operations requirement”, “Local supply requirement for exports”, “Local value added incentive”, “Localisation, nes”, “Public procurement access”, “Public procurement localisation”, “Public procurement preference margin”, “Public procurement, nes”, “Repatriation & surrender requirements”, “Instrument unclear”.
- Only interventions identified as “industrial policies” by Juhász and others (2022, 2025) are considered.
- Each policy instrument is classified by GTA as liberalizing (green evaluation) or protectionist (red/amber evaluation).

### Data sources and construction
- Orbis (Bureau van Dijk) firm-level database:
  - Sector-level variables aggregated after cleaning steps described in Kalemli-Özcan and others (2015) and Gopinath and others (2017).
  - Variables used: labor compensation, age, total assets, operating revenue (gross output), tangible and intangible fixed assets, material costs, liabilities, earnings before interest and taxes, and cash flow.
  - All variables converted into constant 2010 US dollars.
  - Sectoral TFP and allocative efficiency calculated following Hsieh and Klenow (2009) and IMF (2024).
- GTAP Database:
  - Country-specific IO matrices constructed from GTAP.
  - Mapping: each NACE Rev. 2 4-digit industry code assigned to one or more of the 65 GTAP sectors.
  - GTAP sectors classified as energy: Coal; Oil; Gas; Petroleum & Coke; Electricity; Gas (manufacture, distribution).
  - Main GTAP variable used: “domestic purchases by firms at basic prices”, normalized to IO coefficients to construct upstream IP exposure measures as in Baquie and others (2025).
- OECD Inter-Country Input-Output Tables:
  - Granular IO tables for over 70 countries.
  - Raw data captures 45 narrowly defined industries; aggregated to 45 individual countries, the EU, and the rest of the world and to 20 industries.
- Additional models and data used for energy-sector analysis described in Online Annex 3.4 (extended infant industry model calibrated using clean technology data across four regions; GMMET model referenced).

### Stylized facts from Online Annex 3.2 (policy deployment and clean-technology diffusion)
- Subsidized financing and direct support (percent of total measures):
  - Advanced economies (AEs) supported firms with both trade finance and state loans.
  - Emerging market and developing economies (EMDEs) mainly used state loans.
  - Breakdown of direct support measures: advanced economies offered financial grants, while EMDEs relied more on tax incentives.
- Electricity demand and new needs:
  - Figures present Electricity Demand: Total by Country (2023) and New Needs (2030-estimates) in thousands of terawatt-hours.
  - Estimates for data centers (DCs) and electric vehicles (EVs) are for the world and come from OPEC and the IEA, respectively.
- Production and adoption volumes of clean technologies (2024; panels cover multiple series and horizons):
  - Cumulative solar production series cover 2004–2009 and 2004–2024.
  - Annual new installation and sales series cover 2015, 2019, 2023 and extend to 2025 and 2022–2025 for specific series.
  - Technologies and units shown include: EVs (Millions), Solar (GW), Wind (GW), EV/batteries (GWh), Solar polysilicon (Kilotonnes), Wind blades (GW), and cumulative solar panel production (Kilotonnes).
  - Regional series highlighted: China, United States, EU, Other Asia, ROW, and North America (for EV/batteries bars).
  - Sources: Bloomberg New Energy Finance; International Renewable Energy Agency; IMF staff calculations.

### Stylized learning-by-doing (LBD) model — structure and key equations
- Model overview:
  - Partial equilibrium model with two countries and two goods: a high-tech good (infant industry) and an outside good.
  - Consumers have quasilinear preferences with a CES aggregator over varieties of the high-tech good; notation and functional forms provided in the text.
  - Production exhibits country-level learning-by-doing: knowledge equals sum of past high-tech production quantities.
- Cost function and experience accumulation:
  - Domestic production costs in period t: c_{i,t}(Q^{hist}_{i,t}) = k_i Q^{hist}_{i,t}^{-β}
  - Experience accumulation: Q^{hist}_{i,t} = ∑_{s=0}^{t-1} Q_{i,s} for given Q^{hist}_{i,0}
  - Price of the high-tech good x_{i,j,t} consumed in country i and produced in country j: p_{i,j,t} = c_{j,t} (1+τ_i)(1−θ_j)
  - Household budget constraint: I_t + T_t = V_t + p_t′ X_t
  - Subsidies θ_j are non-discriminatory and uniform; revenues rebated to households; tariffs are ad-valorem τ_i.
- Equilibrium: consumers maximize utility subject to budget constraint taking prices as determined by the above.

### Model calibration (Online Annex Table 3.3.1)
- Parameter values:
  - Home Initial Cost: 1.0
  - Foreign Initial Cost: 0.7
  - Home Initial Experience: 2.0
  - Foreign Initial Experience: 10.0
  - Baseline Learning Rate (Percent): 19.0 (Home) and 19.0 (Foreign)
  - Trade Elasticity: 5.0 (Home) and 5.0 (Foreign)
  - Demand Elasticity: 2.0 (Home) and 2.0 (Foreign)

### Key simulation scenarios and quantitative policy parameterizations
- Baseline industrial policy (IP):
  - Production subsidy: 12 percent
  - Tariff: 10 percent
- Alternative starting-condition scenarios:
  - Starting closer to the frontier: foreign country cost advantage reduced to 15 percent (from baseline 30 percent).
    - Result: with IP introduced at t = 2, home country can leapfrog foreign technological leader within a few years as production volumes ramp up faster than in the baseline.
  - Starting farther from the frontier: foreign country cost advantage increased to 40 percent (vs baseline 30 percent).
    - Result: domestic learning-by-doing dynamics more muted, greater reliance on imports, domestic production increases slowly, consumer prices decline more slowly.
  - Lower domestic learning rate: home learning rate set to 9.5 percent (vs baseline 19 percent).
    - Result: lower domestic learning rate attenuates effectiveness of IP; outcomes compared with baseline where home learning rate is 19 percent.
  - No access to export market:
    - Implemented technically via a prohibitively high import tariff by the foreign country.
    - Result: effective market size shrinks considerably; domestic production grows only slowly under IP; domestic costs decline gradually with very limited catch-up to global frontier even after 20 years.
- Comparative diagnostics:
  - Squares in figures indicate period-20 outcomes in corresponding counterfactuals with no IP (0 percent subsidy and 0 percent tariff).

### Mechanisms and policy-relevant implications highlighted by the annex
- Conditions under which IP can enable sizable cost declines and catch-up are narrow:
  - Favorable starting conditions (smaller initial cost gap and higher initial experience) increase likelihood of successful catch-up and possible leapfrogging.
  - High domestic learning rates accelerate cost declines through cumulative production.
  - Access to larger effective markets, including exports, is critical to scale production and achieve meaningful learning-by-doing.
- Risks and limits:
  - IP is less effective when the home country starts far from the frontier, when sectors have low domestic learning rates, or when domestic markets are small / cut off from exports.
  - Mis-targeted IP (sector with low domestic learning) or absence of export opportunities can yield limited domestic production expansion and limited cost convergence.

### Quantitative 4-country Learning-by-Doing (LBD) model: setup and calibration
- Model scope: four country blocks — China, the EU, the US, and the rest of the world; simulated separately for renewable energy and electric vehicles.
- Trade modeling: Armington trade elasticity of 5.
- Renewable energy calibration (Online Annex Table 3.4.1):
  - Initial Cost (USD/kW): China 238; European Union 503; Rest of the World 73; United States 480.
  - Domestic Market Size (2023): China 293; European Union 65; Rest of the World 73; United States 31.
  - Learning Rate (Percent): China 21; European Union 21; Rest of the World 21; United States 21.
  - Sources: IEA; IRENA 2023; IMF staff compilation.
- Electric vehicle calibration (Online Annex Table 3.4.2):
  - Initial Cost (USD/kWh): China 94.0; European Union 139.0; Rest of the World 139.0; United States 123.0.
  - Domestic Market Size (Million vehicles, 2023): China 9.9; European Union 2.9; Rest of the World 2.1; United States 1.8.
  - Learning Rate (Percent): China 18.0; European Union 18.0; Rest of the World 18.0; United States 18.0.
  - Sources: Barwick and others 2024; BNEF 2024; IMF staff compilation.
- Tariff data note: Tariffs based on ITC Market Access Map for latest year available (2022 for most countries and 2021 in some instances); data retrieved on December 19th, 2024. Aggregations are simple averages.

### GMMET model description and integration with LBD outputs
- GMMET: large-scale, dynamic, non-linear, micro-founded, multicountry model to analyze short- and medium-term macroeconomic impacts of energy policies.
- Enhancements:
  - Detailed representation of trade in renewable energy capital goods and electric vehicles.
  - Electricity and transport sector calibration updated to reflect renewable generation and EVs.
- Integration approach:
  - Production cost changes from the infant-industry LBD model are implemented in GMMET as exogenous productivity changes applied to the transformation of final goods (clean tech goods modeled as a product of the final goods sector).
  - Equivalent to changing productivity of a clean tech sector combining capital, labor and intermediates, without a dedicated capital stock specific to clean tech production.
- Modeling assumptions:
  - Policy changes (tariffs and subsidies) are implemented in 2025.
  - All other policy settings are held constant at 2024 levels through the simulation period.
  - Announced or planned future GHG-emission-limiting policies are not modeled.
  - Governments’ budgets are balanced in each period through lump sum transfers in standard simulations.

### Industrial policy scenarios and policy parameters
- Three scenarios:
  - Baseline: status quo industrial policies (IPs) continue.
  - No IP: all industrial policies removed starting in 2025.
  - Reshoring: production subsidies introduced starting in 2025.
- Key subsidy parameters:
  - Reshoring subsidies: 15 percent electric vehicle production subsidy; 30 percent renewable production subsidy.
- Tariffs and subsidy data:
  - Tariffs drawn from ITC Market Access Map (2022/2021) supplemented with country-specific updates.
  - Baseline subsidy rates computed from NIPO and GTA following Rotunno and others (forthcoming); EVs and renewables subsidy rates set equal to NIPO-GTA implied subsidy rate on "electrical equipment" sector.

### Main quantitative findings and mechanisms
- Production costs and learning:
  - Production costs evolve from initial levels through learning-by-doing; speed of learning depends on cumulative domestic production and learning rates.
  - Circular relationship: production volumes ↔ production costs ↔ user prices.
- Sourcing and trade outcomes (EU in 2035, relative to 2024):
  - Reshoring scenario: larger subsidies lead to significant shifts to domestic production over time, initially driven by subsidies and later reinforced by domestic learning-by-doing.
  - No IP scenario: removal of tariffs and subsidies leads users to switch to cheaper Chinese products (imports increase), and domestic learning-by-doing is reduced.
- Production-cost dynamics (qualitative):
  - For renewables: due to large initial price differences, EU production share remains smaller without subsidies; even with reshoring subsidy, EU production costs do not reach current Chinese costs before 2035.
  - For EVs: with reshoring subsidy, EU production costs are around 7 years behind Chinese costs.
- Clean tech price changes in EU (2035 vs 2024):
  - Prices fall in all scenarios, primarily driven by learning-by-doing.
  - User prices are lowest in the Reshoring scenario for both technologies, driven by subsidies and more domestic learning-by-doing.
  - Decomposition of price change shows contributions from: domestic learning, foreign learning, import share, tariffs, and subsidies.
- Clean tech adoption in EU:
  - EV adoption increases more sharply than renewable generation because generators are longer-lived and require more planning and permitting; renewables face system-integration considerations; EV uptake is more sensitive to technology prices and less impacted by fuel prices than generation.
  - Adoption strongest under Reshoring scenario owing to lowest clean tech prices.

### Labor market, GDP, and macroeconomic impacts
- EU labor market (deviations from baseline, percent of labor market):
  - Reshoring scenario: increased domestic clean tech production raises labor in clean tech production but lowers labor in overall tradables sector due to exchange-rate appreciation and lower exports.
  - No IP scenario: clean tech production labor declines while tradables employment rises because of exchange-rate depreciation.
  - Aggregate labor broadly flat; sectoral shifts drive outcomes.
- GDP impacts by 2035 (relative to baseline):
  - Reshoring: GDP is 1.8% higher relative to the baseline in 2035. Drivers: sectoral productivity increases from additional learning-by-doing, shift to higher-productivity sector, and additional investment from cheaper renewables and electricity. Result is sensitive to the degree of learning.
  - No IP: GDP is 1.5% lower in 2035 relative to the baseline, driven by reduced learning-by-doing.
- Comparison note: contrasts with Wingender et al. (2024) who find tariffs on Chinese EVs only raise the GDP cost of the EV transition, including for countries heavily reliant on auto production.

### Short-term stimulatory effects of industrial policy via fiscal channels
- Fiscal financing scenarios compared:
  - (1) Reshoring subsidies financed by an increase in debt-to-GDP ratio in the first five years, repaid over the following five years.
  - (2) Reshoring subsidies financed by lump-sum taxes throughout (standard simulation).
- Fiscal expansion outcome:
  - Debt to GDP is 1.2 ppt higher after the first five years in the fiscal-expansion scenario.
  - The fiscal expansion leads to a temporary increase in consumption and GDP relative to the lump-sum-funded case:
    - Consumption increases in part due to hand-to-mouth households whose transfers are not reduced during the expansion.
    - GDP is initially boosted but remains only marginally higher, partly offset by increased imports and decreased exports and investment in non-targeted sectors.
    - After the expansionary period and during payback, GDP ends up lower, largely driven by the fiscal contraction required to repay the debt.

### Historical case studies (summary from Online Annex 3.5)
- Korea (1973–1979):
  - Export-oriented industrial policies under President Park Chung Hee targeting six strategic sectors: steel, nonferrous metals, transportation, machinery, electronics, and petrochemicals.
  - Objectives: military and industrial modernization and long-term development amid political/security concerns and potential U.S. troop withdrawal.
  - Implementation vehicle: large business conglomerates (chaebol).
- Brazil (broadly 1930s–1980s; focus on 1970s):
  - Import-substitution industrialization strategy guided by two national development plans.
  - 1972–74 plan: emphasized infrastructure development and expansion of state-owned enterprises.
  - 1975–79 plan (post-1973 oil crisis): focused on manufacturing, energy, transportation, and communications; aimed to reduce dependence on oil imports via domestic oil production and alternative energy (including ethanol and nuclear).
- Comparative assessment:
  - Korea’s experience is often considered more successful than Brazil’s, reflected in the literature.
- Annex Figure 3.5.1 key finding:
  - Korea experienced higher growth rates of manufacturing value added and real GDP during the period 1973–79 compared to Brazil.
  - Period covered: 1973–79.
  - Manufacturing value added (annual growth rate) axis tick marks: 0 4 8 12 16 20 24 28 32.
  - Real GDP (annual growth rate; right scale) axis tick marks: 0 2 4 6 8 10 12 14 16.
  - Manufacturing definition: ISIC divisions 15–37.
  - Graphical conventions: boxes = interquartile range; squares = median; whiskers = range of maximum and minimum values.
  - Sources: World Bank, World Development Indicators; and IMF staff calculations.

### Data and calibration for multicountry model and policy counterfactuals
- Model baseline calibrated to a pre-2025 steady state.
- Input-output linkages: Data from the OECD’s inter-country input-output tables capturing flows of intermediate goods and breakdown of sectoral output between intermediate use and final demand.
- Country coverage:
  - 46 individual countries (12 AEs with the EU counted as one country block and 34 EMDEs).
  - All remaining countries subsumed as the rest of the world (ROW), producing a 47th country block in results aggregation.
- Tariff calibration: pre-2025 tariff data from ITC MacMap and TRAINS.
- Subsidy calibration: country-sector-level subsidy rates from NIPO using the estimation strategy from Rotunno and others (2025).

### Policy counterfactuals and clean tech (electrical equipment) results
- Three counterfactual subsidy scenarios:
  - Optimal targeted subsidies in all goods-producing sectors (optimal level proportional to sectoral increasing returns to scale; methodology from Ju and others (2024)).
  - Targeted industrial policy applied only to the clean tech sector (electrical equipment): subsidies in the electrical equipment sector raised to the optimal level while subsidies in all other sectors remain at baseline.
  - Uniform subsidies across all sectors (AEs set subsidies uniformly across all sectors); fiscal envelope set approximately equal to the optimal-subsidies-all-goods counterfactual.
- Baseline subsidies computed from NIPO and Global Trade Alert (GTA) following Rotunno and others (forthcoming).
- Clean tech counterfactual outcomes (AEs average, weighted by shares in total output):
  - Sectoral output in the clean tech sector rises by 65 percent.
  - TFP and employment in the clean tech sector also grow sizably (exact sectoral TFP and employment changes presented in model figures as percentage changes; aggregate TFP changes reported in percentage points).
  - Aggregate TFP slightly increases due to the large boost to the clean tech sector despite the targeted sector being relatively small.
  - Subsidy costs reported as change relative to the status quo baseline (size of subsidies shown as percent of GDP in figures).
- Model caveat:
  - The model features labor as the only factor of production; employment effects should be interpreted as changes in factor demand because calibration uses data on sectoral value added (including payments to factors other than labor).

*International Monetary Fund | October 2025 — Online annex material for CHAPTER 3*

### CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND

### CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND RESILIENCE?

### Classification of policy instruments and evaluation
- Direct support measures listed: “Financial grant”, “Production subsidy”, “Tax or social insurance relief”, “State aid, unspecified”, “Tax-based export incentive”, “Price stabilization”, “Export subsidy”, “Other export incentive”, “In-kind grant”, “Import incentive”, “State aid, nes”.
- Other measures listed: “Anti-dumping”, “Controls on commercial transaction and investment instruments”, “Controls on credit operations”, “Export ban”, “Export licensing requirement”, “Export quota”, “Export tariff quota”, “Export tax”, “Export-related non-tariff measure, nes”, “FDI: Entry and ownership rule”, “FDI: Financial incentive”, “FDI: Treatment and operations, nes”, “Import ban”, “Import licensing requirement”, “Import quota”, “Import tariff”, “Import tariff quota”, “Import-related non-tariff measure, nes”, “Internal taxation of imports”, “Labour market access”, “Local content incentive”, “Local content requirement”, “Local labour requirement”, “Local operations requirement”, “Local supply requirement for exports”, “Local value added incentive”, “Localisation, nes”, “Public procurement access”, “Public procurement localisation”, “Public procurement preference margin”, “Public procurement, nes”, “Repatriation & surrender requirements”, “Instrument unclear”.
- Only interventions identified as “industrial policies” by Juhász and others (2022, 2025) are considered.
- Each policy instrument is classified by GTA as liberalizing (green evaluation) or protectionist (red/amber evaluation).

### Data sources and construction
- Orbis (Bureau van Dijk) firm-level database:
  - Sector-level variables aggregated after cleaning steps described in Kalemli-Özcan and others (2015) and Gopinath and others (2017).
  - Variables used: labor compensation, age, total assets, operating revenue (gross output), tangible and intangible fixed assets, material costs, liabilities, earnings before interest and taxes, and cash flow.
  - All variables converted into constant 2010 US dollars.
  - Sectoral TFP and allocative efficiency calculated following Hsieh and Klenow (2009) and IMF (2024).
- GTAP Database:
  - Country-specific IO matrices constructed from GTAP.
  - Mapping: each NACE Rev. 2 4-digit industry code assigned to one or more of the 65 GTAP sectors.
  - GTAP sectors classified as energy: Coal; Oil; Gas; Petroleum & Coke; Electricity; Gas (manufacture, distribution).
  - Main GTAP variable used: “domestic purchases by firms at basic prices”, normalized to IO coefficients to construct upstream IP exposure measures as in Baquie and others (2025).
- OECD Inter-Country Input-Output Tables:
  - Granular IO tables for over 70 countries.
  - Raw data captures 45 narrowly defined industries; aggregated to 45 individual countries, the EU, and the rest of the world and to 20 industries.
- Additional models and data used for energy-sector analysis described in Online Annex 3.4 (extended infant industry model calibrated using clean technology data across four regions; GMMET model referenced).

### Stylized facts from Online Annex 3.2 (policy deployment and clean-technology diffusion)
- Subsidized financing and direct support (percent of total measures):
  - Advanced economies (AEs) supported firms with both trade finance and state loans.
  - Emerging market and developing economies (EMDEs) mainly used state loans.
  - Breakdown of direct support measures: advanced economies offered financial grants, while EMDEs relied more on tax incentives.
- Electricity demand and new needs:
  - Figures present Electricity Demand: Total by Country (2023) and New Needs (2030-estimates) in thousands of terawatt-hours.
  - Note: Estimates for data centers (DCs) and electric vehicles (EVs) are for the world and come from OPEC and the IEA, respectively.
- Production and adoption volumes of clean technologies (2024; panels cover multiple series and horizons):
  - Panel references and time spans: cumulative solar production 2004–2009 and 2004–2024; annual new installation and sales across 2015, 2019, 2023 and series extending to 2025 and 2022–2025 for specific series.
  - Technologies and units shown include: EVs (Millions), Solar (GW), Wind (GW), EV/batteries (GWh), Solar polysilicon (Kilotonnes), Wind blades (GW), and cumulative solar panel production (Kilotonnes).
  - Regional series highlighted: China, United States, EU, Other Asia, ROW (rest of the world), and North America (for EV/batteries bars in panel note).
  - Sources: Bloomberg New Energy Finance; International Renewable Energy Agency; IMF staff calculations.

### Stylized learning-by-doing (LBD) model — structure and key equations
- Model overview:
  - Partial equilibrium model with two countries and two goods: a high-tech good (infant industry) and an outside good.
  - Consumers have quasilinear preferences with a CES aggregator over varieties of the high-tech good; notation and functional forms provided in the text.
  - Production exhibits country-level learning-by-doing: knowledge equals sum of past high-tech production quantities.
- Cost function and experience accumulation (equations reproduced from source):
  - Domestic production costs in period t: c_{i,t}(Q^{hist}_{i,t}) = k_i Q^{hist}_{i,t}^{-β}
  - Experience accumulation: Q^{hist}_{i,t} = ∑_{s=0}^{t-1} Q_{i,s} for given Q^{hist}_{i,0}
  - Price of the high-tech good x_{i,j,t} consumed in country i and produced in country j:
    - p_{i,j,t} = c_{j,t} (1+τ_i)(1−θ_j)
  - Household budget constraint:
    - I_t + T_t = V_t + p_t′ X_t
  - Subsidies θ_j are non-discriminatory and uniform; revenues rebated to households; tariffs are ad-valorem τ_i.
- Equilibrium: consumers maximize utility subject to budget constraint taking prices as determined by the above.

### Model calibration (Online Annex Table 3.3.1)
- Parameter values preserved exactly as in source:
  - Home Initial Cost: 1.0
  - Foreign Initial Cost: 0.7
  - Home Initial Experience: 2.0
  - Foreign Initial Experience: 10.0
  - Baseline Learning Rate (Percent): 19.0 (Home) and 19.0 (Foreign)
  - Trade Elasticity: 5.0 (Home) and 5.0 (Foreign)
  - Demand Elasticity: 2.0 (Home) and 2.0 (Foreign)
- Note: calibration aims to broadly mimic countries that lag behind the global frontier; elasticities calibrated to standard literature values; learning rates mimic historic learning rates documented in chapter.

### Key simulation scenarios and quantitative policy parameterizations
- Baseline industrial policy (IP) in simulations:
  - Production subsidy: 12 percent
  - Tariff: 10 percent
- Alternative starting-condition scenarios examined:
  - Starting closer to the frontier: foreign country cost advantage reduced to 15 percent (from baseline 30 percent).
    - Result: with IP introduced at t = 2, home country can leapfrog foreign technological leader within a few years as production volumes ramp up faster than in the baseline.
  - Starting farther from the frontier: foreign country cost advantage increased to 40 percent (vs baseline 30 percent).
    - Result: domestic learning-by-doing dynamics more muted, greater reliance on imports, domestic production increases slowly, consumer prices decline more slowly.
  - Lower domestic learning rate: home learning rate set to 9.5 percent (vs baseline 19 percent).
    - Result: lower domestic learning rate attenuates effectiveness of IP; outcomes compared with baseline where home learning rate is 19 percent.
  - No access to export market:
    - Implemented technically via a prohibitively high import tariff by the foreign country.
    - Result: effective market size shrinks considerably; domestic production grows only slowly under IP; domestic costs decline gradually with very limited catch-up to global frontier even after 20 years.
- Comparative notes on dynamics:
  - The dynamic paths of production costs, production volumes, and consumer prices critically depend on starting conditions, domestic learning rates, and market size (exports).
  - Squares in figures indicate period-20 outcomes in corresponding counterfactuals with no IP (0 percent subsidy and 0 percent tariff).

### Mechanisms and policy-relevant implications highlighted by the annex
- Conditions under which IP can enable sizable cost declines and catch-up are narrow:
  - Favorable starting conditions (smaller initial cost gap and higher initial experience) increase likelihood of successful catch-up and possible leapfrogging.
  - High domestic learning rates (higher β in text interpretation) accelerate cost declines through cumulative production.
  - Access to larger effective markets, including exports, is critical to scale production and achieve meaningful learning-by-doing.
- Risks / limits illustrated:
  - IP is less effective when the home country starts far from the frontier, when sectors have low domestic learning rates, or when domestic markets are small / cut off from exports.
  - Policy design matters: mis-targeted IP (sector with low domestic learning) or absence of export opportunities can yield limited domestic production expansion and limited cost convergence.

*International Monetary Fund | October 2025 — Online annex material for CHAPTER 3*

### CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND

### CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND RESILIENCE?

### Quantitative 4-country Learning-by-Doing (LBD) model: setup and calibration
- Model scope: four country blocks — China, the EU, the US, and the rest of the world; simulated separately for renewable energy and electric vehicles.
- Consumer preferences and production structure follow the stylized LBD in Online Annex 3.3, extended to four countries.
- Trade modeling: Armington trade elasticity of 5.
- Renewable energy calibration (Online Annex Table 3.4.1):
  - Initial Cost (USD/kW): China 238; European Union 503; Rest of the World 503; United States 480.
  - Domestic Market Size (2023): China 293; European Union 65; Rest of the World 73; United States 31.
  - Learning Rate (Percent): China 21; European Union 21; Rest of the World 21; United States 21.
  - Sources: IEA; IRENA 2023; IMF staff compilation.
- Electric vehicle calibration (Online Annex Table 3.4.2):
  - Initial Cost (USD/kWh): China 94.0; European Union 139.0; Rest of the World 139.0; United States 123.0.
  - Domestic Market Size (Million vehicles, 2023): China 9.9; European Union 2.9; Rest of the World 2.1; United States 1.8.
  - Learning Rate (Percent): China 18.0; European Union 18.0; Rest of the World 18.0; United States 18.0.
  - Sources: Barwick and others 2024; BNEF 2024; IMF staff compilation.
- Tariff data note: Tariffs based on ITC Market Access Map for latest year available (2022 for most countries and 2021 in some instances); data retrieved on December 19th, 2024. Aggregations are simple averages.

### GMMET model description and integration with LBD outputs
- GMMET: large-scale, dynamic, non-linear, micro-founded, multicountry model to analyze short- and medium-term macroeconomic impacts of energy policies.
- Enhancements for this analysis:
  - Detailed representation of trade in renewable energy capital goods and electric vehicles.
  - Electricity and transport sector calibration updated to reflect renewable generation and EVs.
  - Broader calibration as in Carton et al (2023).
- Integration approach:
  - GMMET does not include endogenous LBD; production cost changes from the infant-industry LBD model are implemented in GMMET as exogenous productivity changes applied to the transformation of final goods (clean tech goods are modeled as a product of the final goods sector).
  - This implementation is equivalent to changing productivity of a clean tech sector combining capital, labor and intermediates, without a dedicated capital stock specific to clean tech production.
- Modeling assumptions:
  - Policy changes (tariffs and subsidies) are implemented in 2025.
  - All other policy settings are held constant at 2024 levels through the simulation period.
  - Announced or planned future GHG-emission-limiting policies (for example, planned increases in EU Fit for 55 stringency) are not modeled.
  - Governments’ budgets are balanced in each period through lump sum transfers in standard simulations.

### Industrial policy scenarios and policy parameters
- Three scenarios modeled:
  - Baseline: status quo industrial policies (IPs) continue.
  - No IP: all industrial policies removed starting in 2025.
  - Reshoring: production subsidies introduced starting in 2025.
- Key subsidy parameters (Online Annex Tables 3.4.3 and 3.4.4; scenario notes):
  - Reshoring subsidies: 15 percent electric vehicle production subsidy; 30 percent renewable production subsidy.
- Tariffs and subsidy data:
  - Tariffs drawn from ITC Market Access Map (2022/2021) supplemented with country-specific updates.
  - Baseline subsidy rates computed from NIPO and GTA following Rotunno and others (forthcoming); EVs and renewables subsidy rates set equal to NIPO-GTA implied subsidy rate on "electrical equipment" sector.

### Main quantitative findings and mechanisms
- Production costs and learning:
  - Production costs evolve from initial levels through learning-by-doing; speed of learning depends on cumulative domestic production and learning rates.
  - Circular relationship: production volumes ↔ production costs ↔ user prices.
- Sourcing and trade outcomes (EU in 2035, relative to 2024):
  - Reshoring scenario: larger subsidies lead to significant shifts to domestic production over time, initially driven by subsidies and later reinforced by domestic learning-by-doing.
  - No IP scenario: removal of tariffs and subsidies leads users to switch to cheaper Chinese products (imports increase), and domestic learning-by-doing is reduced.
- Production-cost dynamics (qualitative):
  - For renewables: due to large initial price differences, EU production share remains smaller without subsidies, limiting EU learning-by-doing; even with reshoring subsidy, EU production costs do not reach current Chinese costs before 2035.
  - For EVs: with reshoring subsidy, EU production costs are around 7 years behind Chinese costs.
- Clean tech price declines in EU (2035 vs 2024):
  - Prices fall in all scenarios, primarily driven by learning-by-doing.
  - User prices are lowest in the Reshoring scenario for both technologies, driven by subsidies and more domestic learning-by-doing.
  - Decomposition of price change shows contributions from: domestic learning, foreign learning, import share, tariffs, and subsidies.
- Clean tech adoption in EU:
  - EV adoption increases more sharply than renewable generation due to:
    - Generators being longer-lived assets and requiring more planning and permitting.
    - Renewables having system-integration considerations (intermittency, transition needs).
    - EV uptake being more sensitive to technology prices and less impacted by fuel prices than generation.
  - Adoption strongest under Reshoring scenario owing to lowest clean tech prices.

### Labor market, GDP, and macroeconomic impacts
- EU labor market (deviations from baseline, percent of labor market):
  - Reshoring scenario: increased domestic clean tech production raises labor in clean tech production but lowers labor in overall tradables sector due to exchange-rate appreciation and lower exports.
  - No IP scenario: clean tech production labor declines while tradables employment rises because of exchange-rate depreciation.
  - Aggregate labor broadly flat; sectoral shifts drive outcomes.
- GDP impacts by 2035 (relative to baseline):
  - Reshoring: GDP is 1.8% higher relative to the baseline in 2035. Drivers: sectoral productivity increases from additional learning-by-doing, shift to higher-productivity sector, and additional investment from cheaper renewables and electricity. Result is sensitive to the degree of learning.
  - No IP: GDP is 1.5% lower in 2035 relative to the baseline, driven by reduced learning-by-doing.
- Comparison note: This contrasts with Wingender et al. (2024) who find tariffs on Chinese EVs only raise the GDP cost of the EV transition, including for countries heavily reliant on auto production.

### Short-term stimulatory effects of industrial policy via fiscal channels
- Fiscal financing scenarios compared:
  - (1) Reshoring subsidies financed by an increase in debt-to-GDP ratio in the first five years, repaid over the following five years.
  - (2) Reshoring subsidies financed by lump-sum taxes throughout (standard simulation).
- Fiscal expansion outcome:
  - Debt to GDP is 1.2 ppt higher after the first five years in the fiscal-expansion scenario.
  - The fiscal expansion leads to a temporary increase in consumption and GDP relative to the lump-sum-funded case:
    - Consumption increases in part due to hand-to-mouth households whose transfers are not reduced during the expansion.
    - GDP is initially boosted but remains only marginally higher, partly offset by increased imports and decreased exports and investment in non-targeted sectors.
    - After the expansionary period and during payback, GDP ends up lower, largely driven by the fiscal contraction required to repay the debt.

### Historical case studies (summary from Online Annex 3.5)
- Korea (1973–1979):
  - Export-oriented industrial policies under President Park Chung Hee targeting six strategic sectors: steel, nonferrous metals, transportation, machinery, electronics, and petrochemicals.
  - Objectives: military and industrial modernization and long-term development amid political/security concerns and potential U.S. troop withdrawal.
  - Implementation vehicle: large business conglomerates (chaebol).
- Brazil (broadly 1930s–1980s; focus on 1970s):
  - Import-substitution industrialization strategy guided by two national development plans.
  - 1972–74 plan: emphasized infrastructure development and expansion of state-owned enterprises.
  - 1975–79 plan (post-1973 oil crisis): focused on manufacturing, energy, transportation, and communications; aimed to reduce dependence on oil imports via domestic oil production and alternative energy (including ethanol and nuclear).
- Comparative assessment:
  - Korea’s experience is often considered more successful than Brazil’s, as reflected in the literature.

*Source: CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND RESILIENCE?, Online annexes (International Monetary Fund | October 2025).*

### Annex Figure 3.5.1, which shows that Korea

### Annex Figure 3.5.1, which shows that Korea

### Key finding
- Korea experienced higher growth rates of manufacturing value added and real GDP during the period 1973–79 compared to Brazil.

### Figure details and measures
- Period covered: 1973–79.
- Outcomes plotted:
  - Manufacturing value added (annual growth rate).
    - Axis tick marks shown: 0 4 8 12 16 20 24 28 32.
  - Real GDP (annual growth rate; right scale).
    - Axis tick marks shown: 0 2 4 6 8 10 12 14 16.
- Manufacturing definition: includes industries classified under ISIC divisions 15–37.
- Graphical conventions:
  - Boxes denote the interquartile range.
  - Squares denote the median.
  - Whiskers denote the range of maximum and minimum values.

### Data sources and calculations
- Sources: World Bank, World Development Indicators; and IMF staff calculations.

### Interpretation note from the annex
- Contrasting Brazil and Korea during 1973–79 offers broad yet valuable insight into the importance of policy design, implementation, and complementary measures for contemporary industrial policy strategies.

*Source: Online Annex Figure 3.5.1, ch3onlineannexes (World Economic Outlook, International Monetary Fund | October 2025).*

### CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND

### CHAPTER 3: INDUSTRIAL POLICY: MANAGING TRADE-OFFS TO PROMOTE GROWTH AND RESILIENCE?

### Data and calibration
- Model baseline calibrated to a pre-2025 steady state.
- Input-output linkages: Data from the OECD’s inter-country input-output tables capturing flows of intermediate goods and breakdown of sectoral output between intermediate use and final demand.
- Country coverage in the model:
  - 46 individual countries (12 AEs with the EU counted as one country block and 34 EMDEs).
  - All remaining countries subsumed as the rest of the world (ROW), producing a 47th country block in results aggregation.
- Tariff calibration: pre-2025 tariff data from ITC MacMap and TRAINS.
- Subsidy calibration: country-sector-level subsidy rates from NIPO using the estimation strategy from Rotunno and others (2025).

### Policy counterfactuals and scenarios
- Three counterfactual subsidy scenarios considered:
  - Optimal targeted subsidies in all goods-producing sectors (optimal level proportional to sectoral increasing returns to scale; methodology from Ju and others (2024)).
  - Targeted industrial policy applied only to the clean tech sector (identified as the “electrical equipment” sector). In this counterfactual, subsidies in the electrical equipment sector are raised to the optimal level while subsidies in all other sectors remain at baseline.
  - Uniform subsidies across all sectors (AEs set subsidies uniformly across all sectors, regardless of external economies of scale). Fiscal envelope set to approximately equal the fiscal envelope in the optimal-subsidies-all-goods counterfactual.
- Baseline subsidies are computed from NIPO and Global Trade Alert (GTA) following the methodology in Rotunno and others (forthcoming).
- Figures reported average subsidies by sector under baseline and counterfactuals, with sector codes defined as in the Development of the OECD Inter-Country Input-Output Database 2023, table A D.2.

### Results — clean tech (electrical equipment) counterfactual
- Scope: IPs introduced for the AEs in the sample (Australia, Canada, European Union, Iceland, Israel, Japan, Korea, New Zealand, Norway, Switzerland, United Kingdom, United States). Results reported as weighted average effect across all AEs (weights are shares in total output by AEs).
- Key quantitative outcomes:
  - Sectoral output in the clean tech sector rises by 65 percent.
  - TFP and employment in the clean tech sector also grow sizably (exact sectoral TFP and employment changes presented in model figures as percentage changes; aggregate TFP changes reported in percentage points).
- Aggregate effects:
  - Aggregate TFP slightly increases due to the large boost to the clean tech sector despite the targeted sector being relatively small.
  - Subsidy costs reported as change relative to the status quo baseline (size of subsidies shown as percent of GDP in figures).
- Interpretation notes:
  - The model features labor as the only factor of production; employment effects should be interpreted as changes in factor demand because calibration uses data on sectoral value added (including payments to factors other than labor).

### Figures and diagnostics (as described)
- Online Annex Figures referenced:
  - Online Annex Figures 3.7.1–3.7.3: average subsidy rates for AEs across industries; comparisons of baseline versus optimal and baseline versus uniform subsidies by sector (percent of sectoral output).
  - Online Annex Figure 3.7.4: sectoral and aggregate effects of industrial policy in the clean tech sector (Panel 1: percentage change in sectoral output, TFP, and employment for targeted and untargeted sectors; Panel 2: percentage changes in aggregate TFP, energy import share, and size of subsidies as percent of GDP).
- Notes on figure construction:
  - Baseline and counterfactual subsidy values plotted on dual scales where appropriate (baseline on left scale, counterfactual on right scale).
  - Sector codes on x-axes use OECD activity codes (e.g., A01_02, A03, B05_06, …, C31T33, S).

*International Monetary Fund | October 2025 — CHAPTER 3 online annexes*

---


_Source: https://www.imf.org/-/media/files/publications/weo/2025/october/english/ch3onlineannexes.pdf_