## 3.1  Global Trade Alert Database

## Source details

**Canonical URL:** [3.1  Global Trade Alert Database](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025023-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025023-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025023-print-pdf.pdf.json)

---

### Description of the GTA database
- The Global Trade Alert (GTA) database consists of a collection of individual announcements of policies affecting trade relationships.
- GTA was originally launched at University of St. Gallen and is currently housed at the St.Gallen Endowment for Prosperity through Trade.
- The data compiles announcements and analyzes the text to extract information such as implementing jurisdiction, announcement date, type of intervention, and affected sector.
- In the language of this paper, a ‘targeted’ sector is a sector that is the direct ‘recipient’ of an IP measure.
- The data includes affected sector information using CPC (Rev. 2.1) 3-digit sector codes; some policies also contain more granular 6-digit HS2012 codes which are not used in this analysis.

### Data coverage and key counts
- Number of Countries: 195
- Number of Sectors: 329
- Number of Policy Tools: 29
- First Announcement Date: 6/20/2008
- Last Announcement Date: 5/31/2024
- Temporal coverage: between November 2008 and May 2024
- Implementing Jurisdictions: data covers policies from 195 countries, of which 31 are in Asia and Pacific.
- The data includes 329 CPC (Rev. 2.1) 3-digit sector codes.

### Policy tool composition (Top tools)
- The database includes 29 types of policies, including tariff and non-tariff measures.
- Globally most common policy types (counts and percents):
  - Subsidies: 2711746.8
  - Import Tariffs: 914415.8
  - Export Support: 637311.0
  - Antidumping: 26754.6
  - Export Restriction: 17773.1
  - Other: 1086418.7
- Total: 57950
- Note: Count is number of unique policy announcements that use given tool.
- Non-tariff measures are classified according to the MAST chapter from the UN Conference on Trade and Development.

### Regional, temporal, and income-group patterns (summary)
- The five top tools (subsidies, import tariffs, export support measures, antidumping measures, export restrictions) are frequently used in all regions.
- Subsidies are the most used tool in every year between 2009 and 2024.
- Import tariffs are the second most used tool in all years except 2012-2013.
- Tool use by income group: subsidies are the most frequently used tool in both advanced and emerging economies.
- Low income countries make more use of import tariffs and export restrictions; subsidies rank third in their most used tools.

### Data limitations and handling of magnitudes and sectors
- GTA metrics are based on counts of policies; there is no available data on the magnitude of IP.
- The analysis focuses on the extensive margin using binary variables rather than the count of policies: if a policy is implemented in a given country-sector-year, the binary variable takes the value one.
- A policy can target multiple sectors; without magnitude information, all listed sectors are treated equally.
- Sector information is missing for 19.3 percent of policies; observations without a CPC code are dropped for sector analysis.
- Table A1.4 breaks down missing sector information by intervention for the top tools.
- The highest share of missing sector information is in subsidies (15.3 percent) and export support measures (14.1 percent).

### Integration with input-output data and merged dataset
- GTA sector codes are converted from CPC Rev 2.1 to ISIC Rev 4 using a standard correspondence table from the UN Statistics Division.
- The merged dataset combines GTA and OECD ICIO data based on standard correspondence tables.
- Merged dataset coverage:
  - Number of Countries: 76
  - Number of Asian Countries: 19
  - Number of Sectors: 43
  - Year Available: 2009-2023
- Source tables for the merge: OECD (2023), Global Trade Alert (2024), Author calculation.
- Note: the OECD ICIO tables provide transaction matrices and measures of final demand, value added and total output.

### Limitations in merged data and temporal imputation
- The ICIO tables are available until 2020. To capture recent trends (2021–2023), the input-output structure from 2019 is imposed on 2021 to 2023.
- This approach abstracts from possible changes in input-output linkages post-covid; results for 2021 to 2023 should be interpreted as “holding constant the input-output structure”.

*Source: Global Trade Alert (2024), OECD (2023), author calculations.*

### 3.1  Global Trade Alert Database

### 3.1  Global Trade Alert Database

### Description of the GTA database
- The Global Trade Alert (GTA) database consists of a collection of individual announcements of policies affecting trade relationships.
- GTA was originally launched at University of St. Gallen and is currently housed at the St.Gallen Endowment for Prosperity through Trade.
- The data compiles announcements and analyzes the text to extract information such as implementing jurisdiction, announcement date, type of intervention, and affected sector.
- In the language of this paper, a ‘targeted’ sector is a sector that is the direct ‘recipient’ of an IP measure.
- The data includes affected sector information using CPC (Rev. 2.1) 3-digit sector codes; some policies also contain more granular 6-digit HS2012 codes which are not used in this analysis.

### Data coverage and key counts
- Number of Countries: 195
- Number of Sectors: 329
- Number of Policy Tools: 29
- First Announcement Date: 6/20/2008
- Last Announcement Date: 5/31/2024
- Temporal coverage: between November 2008 and May 2024
- Implementing Jurisdictions: data covers policies from 195 countries, of which 31 are in Asia and Pacific.
- The data includes 329 CPC (Rev. 2.1) 3-digit sector codes.

### Policy tool composition (Top tools)
- The database includes 29 types of policies, including tariff and non-tariff measures.
- Globally most common policy types (counts and percents):
  - Subsidies: 2711746.8
  - Import Tariffs: 914415.8
  - Export Support: 637311.0
  - Antidumping: 26754.6
  - Export Restriction: 17773.1
  - Other: 1086418.7
- Total: 57950
- Note: Count is number of unique policy announcements that use given tool.
- The frequency of all policy tools and more details on counts by policy tool and ’Other’ category are available in Appendix 1.
- Non-tariff measures are classified according to the MAST chapter from the UN Conference on Trade and Development.

### Regional, temporal, and income-group patterns (summary)
- Table A1.2 (Appendix 1) ranks the most used interventions by region: the five top tools (subsidies, import tariffs, export support measures, antidumping measures, export restrictions) are frequently used in all regions.
- Subsidies are the most used tool in every year between 2009 and 2024.
- Import tariffs are the second most used tool in all years except 2012-2013.
- Tool use by income group (Table A1.3): subsidies are the most frequently used tool in both advanced and emerging economies.
- Other commonly used tools across both income groups include import tariffs, export supports and export restrictions.
- Low income countries make more use of import tariffs and export restrictions; subsidies rank third in their most used tools.

### Data limitations and handling of magnitudes and sectors
- An important limitation: GTA metrics are based on counts of policies; there is no available data on the magnitude of IP.
- The analysis focuses on the extensive margin using binary variables rather than the count of policies: if a policy is implemented in a given country-sector-year, the binary variable takes the value one.
- A policy can target multiple sectors; without magnitude information, all listed sectors are treated equally.
- Sector information is missing for 19.3 percent of policies; observations without a CPC code are dropped for sector analysis.
- Table A1.4 (Appendix 1) breaks down missing sector information by intervention for the top tools.
- The highest share of missing sector information is in subsidies (15.3 percent) and export support measures (14.1 percent).

### Integration with input-output data and merged dataset
- GTA sector codes are converted from CPC Rev 2.1 to ISIC Rev 4 using a standard correspondence table from the UN Statistics Division.
- The merged dataset combines GTA and OECD ICIO data based on standard correspondence tables.
- Merged dataset coverage:
  - Number of Countries: 76
  - Number of Asian Countries: 19
  - Number of Sectors: 43
  - Year Available: 2009-2023
- Source tables for the merge: OECD (2023), Global Trade Alert (2024), Author calculation.
- Note: the OECD ICIO tables provide transaction matrices and measures of final demand, value added and total output.

### Limitations in merged data and temporal imputation
- The ICIO tables are available until 2020. To capture recent trends (2021–2023), the input-output structure from 2019 is imposed on 2021 to 2023.
- This approach abstracts from possible changes in input-output linkages post-covid; results for 2021 to 2023 should be interpreted as “holding constant the input-output structure”.

*Source: Global Trade Alert (2024), OECD (2023), author calculations.*

### 6.1  Results globally and by region

### 6.1 Results globally and by region

### Global and regional findings
- Global sample: small positive but statistically insignificant effect of domestic eigenvector centrality on IP (see Figure 6.1; full regressions in Table A2.1, Appendix 2).
- Asia Pacific: one standard deviation increase in centrality makes a sector 25 percent more likely to receive IP; result significant at the one percent level and robust to inclusion of control variables.
- Europe: one standard deviation increase in centrality makes a sector 10 percent less likely to receive IP; effect significant at the 5 percent level.
- Western Hemisphere, Africa, and the Middle East: weak positive relationship between IP and centrality.
- Data coverage is best in Asia and Europe, possibly reducing power for other regions.
- Notable contrast: Asia Pacific tends to target central sectors while Europe tends to target less central sectors.

### Control-variable patterns (summary of Appendix 2, Table A2.1)
- Global production centrality: not significant in the full sample; positive in Asia; negative in Europe.
- Sector upstreamness: generally not significant, except highly significant and positive in Middle East and Central Asia.
- Sector size and domestic final demand: IP tends to target smaller sectors (in terms of value added out of GDP) with lower domestic final demand.
- Exposure to imports: predictor of IP use in the full sample, as well as in Europe and in the Middle East and Central Asia—possibly reflecting attempts to lower the effective cost of imported inputs for domestic producers.
- Exports for intermediate use: small positive association with IP globally and in Europe.
- Combined implication: domestic and global production placement are important determinants of IP in Europe and Asia; Asia targets more central sectors while Europe targets less central sectors with respect to both domestic and global production. Subsidies (or IP more generally) to large sectors or those with high final demand are less prevalent. In the full sample, exposure to imported inputs is highly predictive of IP.

*Source: wpiea2025023-print-pdf, Section 6.1.*

### 1. IP tends to target smaller sectors, consistent with the fiscal implications of IP.

### 1. IP tends to target smaller sectors, consistent with the fiscal implications of IP.

### Key findings
- IP tends to target smaller sectors, consistent with the fiscal implications of IP.
- These patterns are consistent over the two time periods.
- Only in Asia is the targeting consistent: more central sectors are more likely to ‘receive’ industrial policy.
- The analysis presented are positive statements; further research is needed to draw policy recommendations.

### Role of trade and intermediate goods
- In the global sample, high import reliance is a stronger predictor of IP than high exports.
- In Europe, both imports and exports of intermediate goods are positive predictors of IP.
- In advanced economies, IP targets sectors with high exports for intermediate use.
- In emerging markets, the overall share of global trade is more relevant; this could reflect a higher focus on commodities (or sectors early in the value chain) among emerging economies.

### Domestic demand and policy instruments
- IP tends to target sectors with less (domestic or foreign) final demand.
- Policy-makers may be using tools other than IP (for example, taxation and rebates) to subsidize domestic demand, which are unobserved in the analysis.

### Context, causality, and time horizon
- The context (the income and time-period) of the IP can inverse the targeting strategy completely.
- The analysis and extensions are informed by existing production network features, but governments might use IP to attempt to alter the production network; this would be a slow-moving process.
- From an econometric perspective, it is unlikely that such dynamic targeting would generate endogeneity between IP and centrality measures given the relatively short time horizon of 15 years.
- Nevertheless, governments may target sectors not because they are central but because they wish those sectors to become central; this cannot be captured by the data.

### Open questions and research agenda
- Better understanding is needed of the distinction between emerging and advanced economies, including how the level of development influences which sectors are more connected in production.
- Investigate other features of targeted sectors such as tradeable/non-tradeable status and country-sector comparative advantages to help interpret results.
- Compare multiple centrality measures to enrich the analysis.
- Develop a structural model to provide theoretical underpinning to the centrality measure and aid interpretation.

*Source: wpiea2025023-print-pdf - 1. IP tends to target smaller sectors, consistent with the fiscal implications of IP.*

### 8.1  Appendix 1: Additional Information on Data and Methodology

### 8.1  Appendix 1: Additional Information on Data and Methodology

### 8.1.1 Appendix 1a: Calculation of Centrality Measures
- Centrality measure used: Eigenvector centrality.
- Key properties:
  - Recursive structure: "a sector is central if it is connected to other central sectors."
  - Closely related to PageRank as a well-known application.
  - Does not require estimating a discounting parameter (unlike some diffusion measures).
- Mathematical definition provided:
  - λC = MC
  - (λI − M)C = 0
  - C is the centrality measure of interest and corresponds to the largest Eigenvalue (λ) of the adjacency (input-output) matrix (M).
- Implementation details:
  - Computation uses the Python package ‘networkx’.
  - Construction includes bi-directional input-output links weighted by transaction value.
- Relationship to other centrality measures (Bloch et al. (2023)):
  - Diffusion centrality and Katz-Bonacich centrality share the recursive structure where first-order connections weigh most and distant connections are discounted.
  - Diffusion centrality equals Katz-Bonacich centrality if the discount parameter is sufficiently small and all indirect connections are included (number of rounds = infinity).
  - If all indirect links are considered and the discount parameter is large (specifically, larger than the inverse of the largest eigenvalue of the adjacency matrix), diffusion centrality becomes Eigenvector centrality.
- Relevance to production networks:
  - Appropriate for networks with ‘cycles’ (production uses a sector’s output in its own production); main diagonal of the I-O matrix is non-zero.

### 8.1.2 Appendix 1b: Additional Descriptions of the Data
- General notes:
  - Tables A1.1 through A1.8c provide detailed dataset descriptions (tables enumerated in the source).
  - Regions follow official IMF categorization.
  - * denotes "Country is included in OECD ICIO table and merged dataset."
  - Countries in ’Other’ are included in regressions using global sample, but not in region or income sub-samples.
  - Sources cited in tables: Global Trade Alert (2024), OECD (2023), Author calculations.

- Table A1.1: Countries by Region
  - Regions listed: Africa; Asia-Pacific; Europe; Middle East, Central Asia; West. Hemisphere; Other.
  - Example country annotations include many entries with an asterisk (e.g., *Indonesia, *Japan, *United States of America).

- Table A1.2: Global Trade Alert, Top Tools by Region (Top 5 ranks shown)
  - Asia Pacific: Rank 1 Subsidy; Rank 2 Import Tariff; Rank 3 Export Support; Rank 4 Anti-dumping; Rank 5 Export Restriction.
  - Africa: Rank 1 Import Tariff; Rank 2 Subsidy; Rank 3 Export Restriction; Rank 4 Prohibition; Rank 5 Local Content Measure.
  - Europe: Rank 1 Subsidy; Rank 2 Export Support; Rank 3 Import Tariff; Rank 4 Anti-dumping; Rank 5 Export Restriction.
  - Middle East, Central Asia: Rank 1 Import Tariff; Rank 2 Subsidy; Rank 3 Export Restriction; Rank 4 Anti-dumping; Rank 5 Export Support.
  - West. Hemisphere: Rank 1 Subsidy; Rank 2 Import Tariff; Rank 3 Export Restriction; Rank 4 Anti-dumping; Rank 5 Gov. Local Content.
  - Source: Global Trade Alert (2024)

- Table A1.3: Global Trade Alert, Top Tools by Income Group
  - Advanced: Rank 1 Subsidy; Rank 2 Export Support; Rank 3 Import Tariff; Rank 4 Export Measures, Other; Rank 5 Gov. Local Content Requirement.
  - Emerging: Rank 1 Subsidy; Rank 2 Import Tariff; Rank 3 Anti-dumping; Rank 4 Export Price Controls; Rank 5 Export Licenses.
  - Low-Income: Rank 1 Import Tariff; Rank 2 Export Restriction; Rank 3 Subsidy; Rank 4 Internal taxes/charges on Imports; Rank 5 Tariff Rate Quotas.
  - Source: Global Trade Alert (2024)

- Table A1.4: Global Trade Alert, Sector Codes for Top 5 Tools
  - Intervention Type / Known Sector / Missing Sector / Total / Percent Missing
  - Overall: 57950 known, 13895 missing, 71845 total, 19.3 percent missing.
  - Subsidies: 27117 known, 4899 missing, 32016 total, 15.3 percent missing.
  - Import Tariffs: 9144 known, 841 missing, 9985 total, 8.4 percent missing.
  - Export Support: 6373 known, 1049 missing, 7422 total, 14.1 percent missing.
  - Antidumping: 2675 known, 261 missing, 2936 total, 8.9 percent missing.
  - Export Restriction: 1777 known, 100 missing, 1877 total, 5.3 percent missing.
  - Count is number of unique policy announcements that use given tool. Source: Global Trade Alert (2024)

- Table A1.5: Count of Interventions by Policy Tool
  - Policy Tool: MAST Chapter / Unique Policies
  - L Subsidies (excluding export subsidies): 27117
  - Tariff measures: 9144
  - P6 Export-support measures: 6373
  - D1 Antidumping: 2675
  - P3 Export licences, quotas, prohibition and other restrictions: 1777
  - P9 Export measures, n.e.s.: 1556
  - P4 Export price-control measures, including additional taxes and charges: 1341
  - M3 Government Procurement Local Content Requirement: 1335
  - E6 Tariff-rate quotas (TRQ): 1084
  - I1 Local content measures: 766
  - FDI measures: 759
  - E1 Non-automatic import-licensing procedures: 724
  - F7 Internal taxes and charges levied on imports: 721
  - E3 Prohibitions other than for SPS and TBT reasons: 537
  - D2 Countervailing measure: 436
  - Instrument unclear: 425
  - D31 General (multilateral) safeguard: 332
  - E2 Quotas: 330
  - Capital control measures: 109
  - Migration measures: 93
  - M5 Government Procurement Tendering Process: 86
  - M1 Government Procurement Market Access Restrictions: 66
  - M2 Government Procurement Domestic Price Preference: 65
  - G Finance measures: 53
  - D32 Agricultural special safeguard: 20
  - C4 Import monitoring, surveillance and automatic licensing measures: 17
  - B Technical barriers to trade: 4
  - I2 Trade-balancing measures: 3
  - N Intellectual Property: 2
  - Non-tariff measures are classified according to the MAST chapter from the UN Conference on Trade and Development.
  - Source: Global Trade Alert (2024)

- Table A1.6: Count of Interventions by Sector (selected entries; "Num. in Figures" and Unique Interventions)
  - Agriculture, hunting, forestry A01_02: Num. in Figures 0, Unique Interventions 20442.0
  - Fishing and aquaculture A03: Num. in Figures 1199, Unique Interventions 08.0
  - Chemical products C20: Num. in Figures 1015, Unique Interventions 309.0
  - Computer, electronic and optical equipment C26: Num. in Figures 1610, Unique Interventions 912.0
  - Basic metals C24: Num. in Figures 1497, Unique Interventions 01.0
  - Food products, beverages and tobacco C10T12: Num. in Figures 592, Unique Interventions 32.0
  - Other entries listed across many sectors; note: "A policy intervention can target multiple sectors, hence the count in this table exceeds the number of policies."
  - Note: "*Mining support service activities" and "*Wholesale and retail trade; repair of motor vehicles" not included after data merge. Certain sectors (ISIC 45, 46, 47) are not yet included in the conversion tables.
  - Source: Global Trade Alert (2024), OECD (2023), Author calculations

- Table A1.7a: Average Sector Centrality, Advanced Economies by Time Period (Mean value of the normalized domestic eigenvector centrality)
  - Construction: All years 2.686, 2009-2016 2.573, 2017-2023 2.815
  - Professional, scientific and technical activities: All years 2.025, 2009-2016 2.034, 2017-2023 2.016
  - Financial and insurance activities: All years 1.682, 2009-2016 1.921, 2017-2023 1.409
  - Real estate activities: All years 0.795, 2009-2016 0.802, 2017-2023 0.787
  - Food products, beverages and tobacco: All years 0.687, 2009-2016 0.718, 2017-2023 0.65
  - ... (table continues with sectors showing positive and negative normalized eigenvector centrality values, ordered by average for all years)
  - Observations ordered by average for all years. Source: OECD (2023), Author calculations

- Table A1.7b: Average Sector Centrality, Emerging Markets by Time Period
  - Food products, beverages and tobacco: All years 1.625, 2009-2016 1.683, 2017-2023 1.558
  - Construction: All years 1.494, 2009-2016 1.385, 2017-2023 1.618
  - Agriculture, hunting, forestry: All years 0.828, 2009-2016 0.846, 2017-2023 0.808
  - ... (continues with sectors and corresponding normalized eigenvector centrality values)
  - Observations ordered by average for all years. Source: OECD (2023), Author calculations

- Table A1.7c: Average Sector Centrality, Low-Income Countries by Time Period
  - Food products, beverages and tobacco: All years 1.765, 2009-2016 1.806, 2017-2023 1.719
  - Textiles, leather and footwear: All years 0.969, 2009-2016 0.983, 2017-2023 0.954
  - Agriculture, hunting, forestry: All years 0.894, 2009-2016 0.883, 2017-2023 0.906
  - ... (continues with sectors and corresponding normalized eigenvector centrality values)
  - Observations ordered by average for all years. Source: OECD (2023), Author calculations

- Table A1.8a: Count of Affected Country-Periods, Advanced Economies by Time Period (Count of country-sector-year that have at least one intervention)
  - Agriculture, hunting, forestry: All years 498.0, 2009-2016 269.0, 2017-2023 229.0
  - Fishing and aquaculture: All years 485.0, 2009-2016 254.0, 2017-2023 231.0
  - Chemical and chemical products: All years 480.0, 2009-2016 253.0, 2017-2023 227.0
  - Food products, beverages and tobacco: All years 478.0, 2009-2016 254.0, 2017-2023 224.0
  - Textiles, leather and footwear: All years 468.0, 2009-2016 247.0, 2017-2023 221.0
  - ... (table continues; sectors ordered by total count across all years)
  - Source: GTA (2024), Author calculations

- Table A1.8b: Count of Affected Country-Periods, Emerging Markets by Time Period
  - Agriculture, hunting, forestry: All years 443.0, 2009-2016 238.0, 2017-2023 205.0
  - Food products, beverages and tobacco: All years 409.0, 2009-2016 211.0, 2017-2023 198.0
  - Fishing and aquaculture: All years 408.0, 2009-2016 216.0, 2017-2023 192.0
  - Chemical and chemical products: All years 406.0, 2009-2016 221.0, 2017-2023 185.0
  - Basic metals: All years 383.0, 2009-2016 209.0, 2017-2023 174.0
  - ... (table continues)
  - Source: GTA (2024), Author calculations

- Table A1.8c: Count of Affected Country-Periods, Low-Income Countries by Time Period
  - Agriculture, hunting, forestry: All years 44.0, 2009-2016 22.0, 2017-2023 22.0
  - Food products, beverages and tobacco: All years 30.0, 2009-2016 14.0, 2017-2023 16.0
  - Fishing and aquaculture: All years 29.0, 2009-2016 11.0, 2017-2023 18.0
  - Chemical and chemical products: All years 29.0, 2009-2016 11.0, 2017-2023 18.0
  - Textiles, leather and footwear: All years 27.0, 2009-2016 11.0, 2017-2023 16.0
  - ... (table continues down to sectors with counts of 1.00)
  - Source: GTA (2024), Author calculations

Italic: Source: wpiea2025023-print-pdf - 8.1  Appendix 1: Additional Information on Data and Methodology

### 8.2  Appendix 2: Main Empirical Results

### 8.2 Appendix 2: Main Empirical Results

### Global and Regional Specifications (Table A2.1)
- Dependent variable: Interventions Binary.
- Eigenvector Centrality, Domestic:
  - Global (1): 0.025 (0.035)
  - Global (2): 0.022 (0.034)
  - Western Hemisphere: 0.198 (0.154)
  - Asia Pacific: 0.222 (0.067) ∗∗∗
  - Europe: -0.102 (0.049) ∗∗
  - Africa: 0.170 (0.197)
  - Middle East, Central Asia: 0.200 (0.116) ∗
- Eigenvector Centrality, Global:
  - Global (1): 0.064 (0.044)
  - Global (2): 0.050 (0.047)
  - Western Hemisphere: 5.829 (2.835) ∗∗
  - Asia Pacific: -9.797 (2.390) ∗∗∗
  - Europe: 40.833 (75.323)
  - Africa: 10.944 (34.504)
- Upstreamness:
  - Global (1): 0.085 (0.083)
  - Global (2): -0.114 (0.250)
  - Western Hemisphere: 0.205 (0.146)
  - Asia Pacific: -0.223 (0.143)
  - Europe: -0.286 (0.341)
  - Africa: 0.484 (0.179) ∗∗∗
- Final Demand, Domestic, 2nd Lag:
  - Global (1): -1.204 (0.243) ∗∗∗
  - Global (2): -2.124 (0.714) ∗∗∗
  - Western Hemisphere: -1.102 (0.426) ∗∗∗
  - Asia Pacific: -2.217 (0.440) ∗∗∗
  - Europe: -0.986 (0.820)
  - Africa: 0.324 (0.498)
- Export Intermediates, 2nd Lag:
  - Global (1): 0.565 (0.334) ∗
  - Western Hemisphere: -0.691 (1.071)
  - Asia Pacific: -0.708 (0.569)
  - Europe: 1.173 (0.573) ∗∗
  - Europe (alternative): -0.405 (1.078)
  - Africa: -0.328 (1.006)
- Import Intermediates, Domestic, 2nd Lag:
  - Global (1): 3.779 (0.591) ∗∗∗
  - Global (2): 2.149 (2.340)
  - Western Hemisphere: 0.572 (0.870)
  - Asia Pacific: 5.116 (1.150) ∗∗∗
  - Europe: -0.573 (1.493)
  - Africa: 3.514 (1.437) ∗∗
- Sector Size, Value Added, 2nd Lag:
  - Global (1): -2.596 (0.279) ∗∗∗
  - Global (2): -3.459 (0.802) ∗∗∗
  - Western Hemisphere: -2.266 (0.473) ∗∗∗
  - Asia Pacific: -3.462 (0.496) ∗∗∗
  - Europe: -1.328 (1.096)
  - Africa: -0.866 (0.690)
- Share of Global Exports, 2nd Lag:
  - Global (1): 0.763 (1.589)
  - Global (2): -2.846 (3.777)
  - Western Hemisphere: 4.618 (2.635) ∗
  - Asia Pacific: 1.724 (2.810)
  - Europe: -16.014 (14.843)
  - Africa: 3.918 (5.165)
- Fixed effects: Country FE and Year FE included in all specifications.
- Observations by column: 42345; 39996; 4341; 8700; 21075; 2280; 3600.
- Intra-class variance, ρ: 0.594; 0.492; 0.494; 0.359; 0.582; 0.415; 0.393.
- Panel-level variance, ln(σ2u): 1.571 (0.036) ∗∗∗; 1.157 (0.043) ∗∗∗; 1.165 (0.135) ∗∗∗; 0.610 (0.095) ∗∗∗; 1.522 (0.064) ∗∗∗; 0.848 (0.197) ∗∗∗; 0.758 (0.146) ∗∗∗.

### By Income Group (Table A2.2)
- Samples: Advanced and Emerging (column 1); Advanced (2); Emerging (3).
- Eigenvector Centrality, Domestic:
  - Advanced and Emerging: -0.113 (0.044) ∗∗
  - Advanced: -0.116 (0.049) ∗∗
  - Emerging: 0.176 (0.059) ∗∗∗
- Emerging indicator: -0.891 (0.653).
- Emerging × Eigenvector Centrality, Domestic: 0.322 (0.075) ∗∗∗.
- Eigenvector Centrality, Global:
  - Advanced and Emerging: 0.081 (0.045) ∗
  - Advanced: 0.089 (0.048) ∗
  - Emerging: 10.066 (4.825) ∗∗
- Upstreamness:
  - Advanced and Emerging: 0.082 (0.087)
  - Advanced: -0.223 (0.145)
  - Emerging: 0.279 (0.107) ∗∗∗
- Final Demand, Domestic, 2nd Lag:
  - All: -1.288 (0.259) ∗∗∗; Advanced: -1.863 (0.434) ∗∗∗; Emerging: -1.041 (0.314) ∗∗∗
- Final Demand, Foreign, 2nd Lag:
  - All: -0.913 (0.512) ∗; Advanced: -1.598 (0.810) ∗∗; Emerging: -0.648 (0.655)
- Export Intermediates, 2nd Lag:
  - All: 0.549 (0.357)
  - Advanced: 1.485 (0.571) ∗∗∗
  - Emerging: -0.769 (0.467) ∗
- Import Intermediates, Domestic, 2nd Lag:
  - All: 4.299 (0.653) ∗∗∗; Advanced: 4.747 (1.128) ∗∗∗; Emerging: 3.101 (0.782) ∗∗∗
- Sector Size, Value Added, 2nd Lag:
  - All: -2.622 (0.296) ∗∗∗; Advanced: -3.553 (0.506) ∗∗∗; Emerging: -2.074 (0.359) ∗∗∗
- Share of Global Exports, 2nd Lag:
  - All: 0.995 (1.627)
  - Advanced: -0.453 (2.071)
  - Emerging: 6.567 (3.196) ∗∗
- Observations: 37431; 20145; 17286.
- Intra-class variance, ρ: 0.503; 0.570; 0.440.
- Panel-level variance, ln(σ2u): 1.202 (0.045) ∗∗∗; 1.471 (0.061) ∗∗∗; 0.950 (0.068) ∗∗∗.

### By Time Period (Table A2.3)
- Samples: 2009-2023 (1); 2009-2016 (2); 2017-2023 (3).
- Eigenvector Centrality, Domestic:
  - 2009-2023: -0.043 (0.040)
  - 2009-2016: -0.071 (0.048)
  - 2017-2023: 0.115 (0.040) ∗∗∗
- 2017-23 indicator: 2.172 (0.095) ∗∗∗.
- 2017-23 × Eigenvector Centrality, Domestic: 0.163 (0.040) ∗∗∗.
- Eigenvector Centrality, Global:
  - 2009-2023: 0.063 (0.045)
  - 2009-2016: 0.094 (0.055) ∗
  - 2017-2023: 0.044 (0.041)
- Final Demand, Domestic, 2nd Lag:
  - 2009-2023: -1.161 (0.242) ∗∗∗
  - 2009-2016: -2.178 (0.333) ∗∗∗
  - 2017-2023: -0.905 (0.258) ∗∗∗
- Import Intermediates, Domestic, 2nd Lag:
  - 2009-2023: 3.769 (0.588) ∗∗∗
  - 2009-2016: 5.842 (0.799) ∗∗∗
  - 2017-2023: 3.194 (0.608) ∗∗∗
- Sector Size, Value Added, 2nd Lag:
  - 2009-2023: -2.600 (0.278) ∗∗∗
  - 2009-2016: -4.033 (0.389) ∗∗∗
  - 2017-2023: -3.120 (0.285) ∗∗∗
- Observations: 39996; 21334; 18662.
- Intra-class variance, ρ: 0.492; 0.627; 0.378.
- Panel-level variance, ln(σ2u): 1.158 (0.043) ∗∗∗; 1.710 (0.054) ∗∗∗; 0.693 (0.059) ∗∗∗.

### By Policy Tool — Global (Tables A2.4a and A2.4b)
- Subsidies (Table A2.4a, col 1):
  - Eigenvector Centrality, Domestic: 0.078 (0.028) ∗∗∗
  - Eigenvector Centrality, Global: 0.001 (0.029)
  - Upstreamness: 0.082 (0.063)
  - Final Demand, Domestic, 2nd Lag: -0.722 (0.195) ∗∗∗
  - Import Intermediates, Domestic, 2nd Lag: 1.222 (0.433) ∗∗∗
  - Sector Size, Value Added, 2nd Lag: -1.457 (0.238) ∗∗∗
  - Share of Global Exports, 2nd Lag: 2.603 (1.378) ∗
  - Observations: 37851; ln(σ2u): 0.201 (0.060) ∗∗∗; ρ: 0.271.
- Export Restrictions (Table A2.4a, col 3):
  - Eigenvector Centrality, Domestic: 0.169 (0.061) ∗∗∗
  - Eigenvector Centrality, Global: -0.382 (0.148) ∗∗∗
  - Upstreamness: -0.604 (0.139) ∗∗∗
  - Final Demand, Domestic, 2nd Lag: -2.059 (0.450) ∗∗∗
  - Export Intermediates, 2nd Lag: 2.053 (0.470) ∗∗∗
  - Import Intermediates, Domestic, 2nd Lag: 3.827 (0.796) ∗∗∗
  - Sector Size, Value Added, 2nd Lag: -3.245 (0.562) ∗∗∗
  - Observations: 37200; ln(σ2u): 1.352 (0.077) ∗∗∗; ρ: 0.540.
- Export Support (Table A2.4a, col 5):
  - Eigenvector Centrality, Domestic: 0.143 (0.038) ∗∗∗
  - Eigenvector Centrality, Global: -0.013 (0.027)
  - Upstreamness: -0.244 (0.087) ∗∗∗
  - Final Demand, Domestic, 2nd Lag: -0.990 (0.273) ∗∗∗
  - Import Intermediates, Domestic, 2nd Lag: 1.397 (0.567) ∗∗
  - Sector Size, Value Added, 2nd Lag: -2.304 (0.312) ∗∗∗
  - Share of Global Exports, 2nd Lag: -0.247 (1.609)
  - Observations: 35055; ln(σ2u): 0.644 (0.069) ∗∗∗; ρ: 0.367.
- Import Tariff and Antidumping (Table A2.4b):
  - Import Tariff (col 1): Eigenvector Centrality, Domestic: -0.070 (0.057)
    - Export Intermediates, 2nd Lag: 1.904 (0.535) ∗∗∗
    - Sector Size, Value Added, 2nd Lag: -5.029 (0.436) ∗∗∗
    - Share of Global Exports, 2nd Lag: -8.423 (2.639) ∗∗∗
  - Import Tariff, Only (col 2): Eigenvector Centrality, Domestic: -0.168 (0.049) ∗∗∗
    - Eigenvector Centrality, Global: 0.070 (0.036) ∗∗
    - Sector Size, Value Added, 2nd Lag: -3.215 (0.330) ∗∗∗
    - Observations: 38286; ln(σ2u): 0.888 (0.055) ∗∗∗; ρ: 0.425.
  - Antidumping, main (col 4): Eigenvector Centrality, Domestic: -0.265 (0.077) ∗∗∗
    - Export Intermediates, 2nd Lag: 3.020 (0.546) ∗∗∗
    - Import Intermediates, Domestic, 2nd Lag: 2.380 (0.719) ∗∗∗
    - Sector Size, Value Added, 2nd Lag: -4.950 (0.562) ∗∗∗
    - Observations: 32781; ln(σ2u): 1.906 (0.069) ∗∗∗; ρ: 0.672.
  - Antidumping, Only (col 5): Eigenvector Centrality, Domestic: -0.310 (0.112) ∗∗∗
    - Observations: 24786; ln(σ2u): 0.172 (0.183).

### Asia Pacific — Time Periods, Income Groups, and Policy Tools (Tables A2.5a and A2.5b)
- Asia Pacific, by income/time (Table A2.5a):
  - Eigenvector Centrality, Domestic:
    - Advanced Asia: 0.063 (0.080)
    - Emerging Asia: 0.488 (0.158) ∗∗∗
    - 2009-2016: 0.156 (0.104)
    - 2017-2023: 0.376 (0.106) ∗∗∗
  - Eigenvector Centrality, Global:
    - Advanced Asia: 1.538 (6.780)
    - Emerging Asia: 12.726 (5.695) ∗∗
    - 2009-2016: 11.810 (4.230) ∗∗∗
    - 2017-2023: -1.003 (4.537)
  - Final Demand, Domestic, 2nd Lag: generally negative and significant for most subsamples (e.g., 2009-2016: -1.252 (0.556) ∗∗; 2017-2023: -1.451 (0.459) ∗∗∗).
  - Sector Size, Value Added, 2nd Lag: consistently negative and significant (e.g., Advanced Asia: -3.305 (0.974) ∗∗∗).
- Asia Pacific, by policy tool (Table A2.5b):
  - Subsidies: Eigenvector Centrality, Domestic: 0.084 (0.101)
  - Export Restrictions: Eigenvector Centrality, Global: 8.620 (2.792) ∗∗∗; Final Demand, Domestic, 2nd Lag: -0.602 (0.445)
  - Export Support: Eigenvector Centrality, Domestic: 0.235 (0.111) ∗∗; Eigenvector Centrality, Global: 5.024 (1.631) ∗∗∗
  - Import Tariff: Eigenvector Centrality, Domestic: 0.192 (0.143)
  - Antidumping: Eigenvector Centrality, Domestic: -0.118 (0.145)
  - Observations per tool: Subsidies 6735; Export Restrictions 6720; Export Support 6540; Import Tariff 5400; Antidumping 6045.

### Additional Empirical Results and Robustness (Appendix 3)
- Including Low-Income Countries (Table A3.1):
  - Eigenvector Centrality, Domestic:
    - All: -0.114 (0.043) ∗∗∗
    - Advanced: -0.116 (0.049) ∗∗
    - Emerging: 0.176 (0.059) ∗∗∗
    - Low Income: 0.377 (0.126) ∗∗∗
  - Low Income indicator: -4.839 (0.558) ∗∗∗
  - Low Income × Eigenvector Centrality, Domestic: 0.500 (0.144) ∗∗∗
  - Eigenvector Centrality, Global for Low Income: -93.904 (223.989)
  - Observations: 39996; 20145; 17286; 2565.
  - Panel-level variance, ln(σ2u) for All: 1.135 (0.043) ∗∗∗.
- Time-split robustness (Table A3.2):
  - Splits considered: 2018 split; 2019 split; 2020 split.
  - Eigenvector Centrality, Domestic (pre/post splits):
    - Pre-period: approximately -0.041 to -0.039 (standard errors ~0.039-0.040)
    - Interaction (post × Eigenvector Centrality, Domestic): 0.191 (0.041) ∗∗∗; 0.219 (0.044) ∗∗∗; 0.293 (0.057) ∗∗∗ for successive splits.
  - Import Intermediates, Domestic, 2nd Lag stable at ~3.77 (0.588) ∗∗∗ across splits.
- Policy-tool dynamics over time (Table A3.3):
  - 2017-23 indicator and interactions with Eigenvector Centrality, Domestic vary by tool:
    - Subsidies: 2017-23 = 3.001 (0.090) ∗∗∗; 2017-23 × Eigenvector Centrality, Domestic = 0.165 (0.040) ∗∗∗
    - Export Restrictions: 2017-23 = 4.566 (0.257) ∗∗∗; 2017-23 × Eigenvector Centrality, Domestic = -0.433 (0.095) ∗∗∗
    - Export Support: 2017-23 = 1.903 (0.124) ∗∗∗; 2017-23 × Eigenvector Centrality, Domestic = -0.143 (0.054) ∗∗∗
    - Import Tariff: 2017-23 = -0.037 (0.120)
    - Antidumping: 2017-23 = -2.045 (0.152) ∗∗∗; 2017-23 × Eigenvector Centrality, Domestic = -0.157 (0.088) ∗
  - Export Intermediates, 2nd Lag and Import Intermediates, Domestic, 2nd Lag show significant positive coefficients for several tools (e.g., Export Restrictions: Export Intermediates = 1.994 (0.467) ∗∗∗; Import Tariff: Export Intermediates = 1.896 (0.536) ∗∗∗).
- Excluding China (Table A3.4 and A3.5):
  - Overall patterns remain similar when China is excluded.
  - Eigenvector Centrality, Domestic:
    - All (excluding China): -0.110 (0.044) ∗∗
    - Advanced: -0.116 (0.049) ∗∗
    - Emerging: 0.179 (0.060) ∗∗∗
  - Asia Pacific excluding China:
    - Emerging Asia: 0.548 (0.172) ∗∗∗
    - 2017-2023: 0.380 (0.117) ∗∗∗
  - Observations and variance measures reported in respective tables; panel-level variance remains significant in many subsamples.

*Production Network Features of Industrial Policy — Working Paper No. WP/2025/023*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025023-print-pdf.pdf_