## 2.1 Industrial policies ## Source details **Canonical URL:** [2.1 Industrial policies](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025143.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025143.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025143.pdf.json) --- ### Context and research questions - After a period of decline following the liberalization wave of the 1990s, industrial policies (IPs) have been widely used in both advanced economies and emerging markets in recent years, particularly since 2017. - Mentions of IPs in the business press rose from under 1,000 in 1990 to over 18,000 in 2019 (Evenett et al., 2024). - Key questions: - How does the introduction of IPs relate to firm performance? - Which firms benefit more from IPs? - Do the effects of IPs spill over across sectors? - Dataset: novel industry-level database of IPs (Juhász et al., 2023) combined with firm-level ORBIS data, covering 2 million firms in 38 countries from 2011 to 2018. - Definition: IPs are “state actions aimed at transforming the structure of economic activity, typically by altering relative prices across sectors or directing resources toward specific industries or activities like exporting and R&D.” - Focus instruments: protectionist domestic subsidies, protectionist export incentives, liberalizing policies that reduce trade barriers. - Empirical approach: dynamic associations estimated using local projection methods (Jordà, 2005). Industry-targeted IP data imply estimates combine treated and untreated firms within industries and should be interpreted as lower bounds for firm-level costs and benefits. ### Main empirical findings by instrument - Protectionist domestic subsidies: - An additional protectionist domestic subsidy is associated with a 1 percent increase in value added (VA), payroll, and total factor productivity (TFP) after one to two years. - Effects are short-lived for VA, payroll, and TFP (decline over the medium term). - Capital: one additional subsidy associated with a gradual increase of more than 1 percent after three years; subsidies on average remain in place for 3 years. - Protectionist export incentives: - Mostly negative short-term association with firm-level outcomes. - Estimates point to a contraction of up to one percent in all firm-level variables after one to two years. - Negative associations tend to fade over time and, for TFP, even turn positive in the medium term. - Liberalizing trade barriers: - Robust positive association with value added and TFP, both increasing by 1 to 2 percent after two years. - Limited association with capital accumulation or payroll relative to protectionist policies. - Baseline-average-firm magnitudes (selected): - Liberalizing policy: 1.6 percent higher productivity (medium term); 1.2 percent higher value added (medium term); 0.8 percent more payroll (medium term); 0.4 percent more capital stock (medium term; not statistically significant). - Export incentive: short-term cost example — 0.5 percent lower productivity for the average firm in the first two years. ### Heterogeneity across firms - Dimensions explored: firm age and cash-to-assets ratio (proxy for credit constraints); also leverage ratio in robustness. - Protectionist domestic subsidies: - Younger firms: 2 percent increase in value added in one year. - Older firms: 0.5 percent increase in value added in one year. - More credit-constrained firms exhibit larger increases in capital after subsidy introduction (example: an additional subsidy associated with a 2 percent increase in the capital stock of firms with the largest cash flow to assets ratio; close to zero for firms with a low ratio). - Younger firms: temporary 1.5 percent increase in productivity; new subsidy linked to a 3.6 percent increase in capital stock of younger firms three years after announcement (negligible effects for older firms). - Protectionist export incentives: - Younger and more credit-constrained firms experience smaller short-term declines in value added and TFP, and faster and stronger recoveries. - Medium-term: one additional export incentive associated with 0.7 percent increase in productivity and value added of younger firms; for older firms the increase is close to zero. - Liberalizing policies: - Effects more homogeneous across firms for VA and productivity; capital accumulation and payroll show more robust associations for young firms. - Interpretation: - IPs more positively associated with firms facing larger frictions (younger, financially constrained). - Policies can cause reallocation across firms within an industry—winners and losers—implying ambiguous aggregate welfare effects ex ante. ### Industry-level distortions - Distortion gauge constructed from markups and external financial dependence (EFD) with country fixed effects removed; industries categorized into four dummies: D_HH, D_LL, D_HL, D_LH. - Main findings: - Positive relationship between IPs and firm-level value added is stronger in industries with higher levels of distortions. - Pattern stronger and more durable for factor accumulation (capital and payroll). - Industry-level distortions less central for TFP responses. - Example magnitude: an additional protectionist IP targeting a highly distorted industry is associated with a 1 percent increase in value added after 1 to 2 years; no increase for low-distortion industries. ### Spillovers along the supply chain - Exposure measures constructed from IO linkages (GTAP): Upstr_sct and Dwnstr_sct normalized so IO coefficients sum to 1. - Findings: - Upstream IPs: positive relationship with performance of downstream firms (productivity, VA, capital, payroll). - Mechanisms: alleviate capacity constraints, increase input productivity, potentially lower upstream product prices. - Downstream IPs: negative relationship with performance of upstream suppliers (reduced input demand). - Liberalizing IPs: positive spillovers regardless of stage; magnitudes 2 to 3 times larger than for protectionist IPs when targeting upstream sectors. - Implication: IPs targeting upstream sectors may benefit the economy more widely than downstream-targeted IPs, though some downstream interventions can be warranted (e.g., de-carbonization with network demand externalities). ### Tit-for-tat dynamics and international context - Construct Z_ict: activity of Protectionist (Red) IPs in industry i by the average country excluding country c, weighted by political distance (UN voting–based metric). - Main interaction finding: - The association between protectionist subsidies and firm value added in the targeted industry shrinks as geopolitically-distant countries introduce protective IPs targeting the same industry (interaction negative and statistically significant for ∆IP_Red,subsidies). - Similar but less robust negative finding for export incentives. - The association between liberalizing IPs and firm-level value added becomes stronger when other countries are also introducing IPs. - Interpretation: retaliatory or tit-for-tat IPs in other countries can dilute the domestic benefits of protectionist subsidies; liberalizing measures may yield amplified benefits when other countries pursue protectionist IPs. ### Data, sample, and key summary statistics - IPs data: Juhász et al. (2023) using machine learning on GTA policy text; aggregation to NACE Rev. 2; kept GTA policies announced and published within the same calendar year; stock of active IPs counted as policies announced but not yet removed; main shock = change in stock between years. - GTA evaluation categories: Red (protectionist), Amber (ambiguous), Green (liberalizing). - Firm data: ORBIS; cleaned following industry practices; winsorized at 1 and 99 percent; firms required to report at least four consecutive periods; nominal variables in 2015 USD. - IO matrices: GTAP; trade data: BACI (CEPII). - Final regression sample: - IPs introduced 2011–2018. - Over 2 million firms from 38 countries (11 EMDEs and 27 AEs) over 2011-2018. - Total observations: 8,515,018. - Selected summary statistics (Table 1, 2011-2018): - ln(VA): Mean 13.2; Std. dev. 1.71; P10 11.3; Median 13.1; P90 15.4 - ∆ ln(VA): Mean 0.03; Std. dev. 0.32; P10 -0.26; Median 0.02; P90 0.33 - ln(Capital): Mean 11.7; Std. dev. 2.38; P10 8.9; Median 11.6; P90 14.7 - ∆ ln(Capital): Mean 0.03; Std. dev. 0.61; P10 -0.40; Median -0.04; P90 0.55 - ln(Payroll): Mean 12.2; Std. dev. 1.81; P10 10.1; Median 12.1; P90 14.4 - ∆ ln(Payroll): Mean 0.05; Std. dev. 0.34; P10 -0.20; Median 0.03; P90 0.32 - ln(TFPQ): Mean 7.8; Std. dev. 1.26; P10 6.3; Median 7.7; P90 9.3 - ∆ ln(TFPQ): Mean 0.01; Std. dev. 0.48; P10 -0.45; Median 0.02; P90 0.43 - ∆IP_red: Mean 0.039; Std. dev. 0.304; P10 0; Median 0; P90 0 - ∆IP_red,subsidies: Mean 0.018; Std. dev. 0.180; P10 0; Median 0; P90 0 - ∆IP_red,expinc: Mean 0.020; Std. dev. 0.234; P10 0; Median 0; P90 0 - ∆IP_green: Mean 0.029; Std. dev. 0.170; P10 0; Median 0; P90 0 - ∆IP_green,tradebar: Mean 0.027; Std. dev. 0.165; P10 0; Median 0; P90 0 - ∆Upstr_red: Mean 0.171; Std. dev. 0.397; P10 -0.001; Median 0.008; P90 0.572 - ∆Dwnstr_red: Mean 0.181; Std. dev. 0.446; P10 -0.004; Median 0.023; P90 0.591 - Panel properties and composition: - Panel strongly balanced for outcomes of interest. - Average firm received 0.039 new protectionist (Red) IPs between two consecutive years: over 90% of this shock explained by protectionist subsidies (0.018) and protectionist export incentives (0.020). - Average firm experienced 0.029 more liberalizing (Green) IPs, mostly reductions in trade barriers (0.027). - Sharp increase in protectionist (Red) IPs after 2016; by 2018 protectionist domestic subsidies accounted for over a third of total protectionist IPs. - Compositional year-to-year changes motivated restricting sample to 2011-2018. ### Empirical specifications, identification, and robustness - Baseline local projections (h = 0,...,3) regressing lnY_ft+h − lnY_ft−1 on ΔIP by instrument k and GTA evaluation e, with two lags of dependent and independent variables, controls, and firm (α_f), country-year (α_ct), industry-year (α_it) fixed effects; standard errors clustered by country and industry. - Exposure along value chain measured via IO coefficients following Amiti and Konings (2007) style; upstream and downstream exposure normalized. - Heterogeneity specification allows β to vary by firm terciles (age, cashflow-to-assets, leverage). - Distortion heterogeneity interacts ΔIP with four distortion dummies to estimate differing βs across distortion categories. - Robustness checks: - LP DiD (Dube et al., 2024) accounting for staggered timing: broadly consistent with baseline; exception: export incentives where LP DiD finds larger and more significant medium-term associations with VA and TFP. - Instrumental variable (IV) strategy: instrument by actions taken by other countries (weighted by political distance or trade); IV results consistent with baseline. - Additional checks: address count nature of IP proxy, set of policies included, sample of countries, set of controls, lag structure; results robust across exercises. - Pre-trend/placebo tests: little evidence of pre-trends for domestic subsidies and export incentives in baseline; some pre-trends for export incentives (payroll) and a negative pre-trend for liberalizing IPs (interpreted as trend change consistent with local projections). - Limitations: - Approach compares relative performance of firms in targeted and non-targeted industries; does not assess aggregate welfare or fiscal costs. - Full welfare assessment would require structural general equilibrium analysis and information on size/fiscal cost of IPs, political economy, and retaliatory dynamics—beyond scope. - Potential undercoverage of IPs in some EMDEs (notably China) and lack of monetary policy-intensity measure in IP counts (correlation between IP counts and log value in 2023 is 0.52, statistically significant). ### Additional diagnostics and notes - Figure/appendix statistics (sample-average cumulative growth, Table A.4): ∆ ln VA_f t+h: h=0 = 3.19 ; h=1 = 5.95 ; h=2 = 7.01 ; h=3 = 8.22 - ∆ ln Capital_f t+h: h=0 = 3.29 ; h=1 = 5.60 ; h=2 = 6.66 ; h=3 = 5.83 - ∆ ln Payroll_f t+h: h=0 = 4.80 ; h=1 = 8.95 ; h=2 = 10.8 ; h=3 = 11.9 - ∆ ln TFPQ_f t+h: h=0 = 0.49 ; h=1 = 1.11 ; h=2 = 1.12 ; h=3 = 2.51 - Pre-trend placebo evidence (Table A.5): - VARed subsidies: Coefficient = 0.001 ; Std t-stat = 0.0037 ; p-value = 0.728 - VARed export incentives: Coefficient = -0.007 ; Std t-stat = 0.0036 ; p-value = 0.062 - VAGreen trade barriers: Coefficient = -0.025 ; Std t-stat = 0.0052 ; p-value = 0.000 - IV first-stage diagnostics: Kleibergen-Paap rk Wald F statistics truncated at 60 in figures reporting first stages for subsidies. _Italic: Source — IMF Working Paper (section 2.1; chapters 3.4 and 4.3, figures and appendix notes) — Industrial Policies and Firm Performance: A Nuanced Relationship._ ### 2.1 Industrial policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ### 2.1 Industrial policies ### Context and research questions - After a period of decline following the liberalization wave of the 1990s, industrial policies (IPs) have been widely used in both advanced economies and emerging markets in recent years, particularly since 2017. - Mentions of IPs in the business press rose from under 1,000 in 1990 to over 18,000 in 2019 (Evenett et al., 2024). - Key questions the paper addresses: - How does the introduction of IPs relate to firm performance? - Which firms benefit more from IPs? - Do the effects of IPs spill over across sectors? - Dataset constructed by combining a novel industry-level database of industrial policies from Juhász et al. (2023) with firm-level data from ORBIS, covering 2 million firms in 38 countries from 2011 to 2018. - Industrial policies defined as “state actions aimed at transforming the structure of economic activity, typically by altering relative prices across sectors or directing resources toward specific industries or activities like exporting and R&D.” - Focus on the most prevalent instruments: protectionist domestic subsidies, protectionist export incentives, and liberalizing policies that reduce trade barriers. - Empirical approach: estimate dynamic associations using local projection methods (Jordà, 2005). Because IP data is industry-targeted (not firm-targeted), estimates combine effects on treated and untreated firms within industries and should be interpreted as lower bounds for firm-level costs and benefits. ### Main empirical findings by instrument - Protectionist domestic subsidies: - An additional protectionist domestic subsidy is associated with a 1 percent increase in value added (VA), payroll, and total factor productivity (TFP) after one to two years of implementation. - Effects are short-lived; VA, payroll, and TFP decline over the medium term. - More sustained effects on capital: one additional subsidy is associated with a gradual increase of more than 1 percent after three years. - Protectionist export incentives: - Mostly negative short-term association with firm-level outcomes. - Estimates point to a contraction of up to one percent in all firm-level variables after one to two years. - Negative associations tend to fade over time and, for TFP, even turn positive in the medium term. - Liberalizing trade barriers: - Robust positive association with value added and TFP, both increasing by 1 to 2 percent after two years. - Limited association with capital accumulation or payroll relative to protectionist policies. ### Heterogeneity across firms - Two firm dimensions explored: firm age and cash-to-assets ratio (proxy for credit constraints). - Protectionist domestic subsidies: - Younger firms: 2 percent increase in value added in one year. - Older firms: 0.5 percent increase in value added in one year. - More credit-constrained firms exhibit larger increases in capital after subsidy introduction. - Protectionist export incentives: - Younger and more credit-constrained firms experience smaller short-term declines in value added and TFP, and faster and stronger recoveries. - Liberalizing policies: - Effects appear more homogeneous across firms, consistent with broader, less targeted impact. ### Industry-level distortions - Constructed an industry-specific gauge combining external financial dependence (EFD) and markups (with country-specific factors stripped out) to capture industry-specific distortions. - Positive relationship between IPs and firm-level value added is stronger in industries with higher levels of distortions. - This pattern is stronger and more durable for factor accumulation (capital and payroll). - Industry-level distortions are less central for TFP responses. ### Spillovers along the supply chain - Cross-sectoral spillovers depend on the stage of the value chain targeted: - Upstream IPs (input-providing sectors): positive relationship with the performance of firms buying those inputs (downstream firms). - Downstream IPs (final-stage sectors): negative relationship with the performance of firms supplying inputs to those sectors (upstream firms). - Interpretation: IPs that temporarily raise productivity and capital in targeted industries can alleviate capacity constraints upstream but reduce input demand when targeted at downstream/final sectors. - Liberalizing IPs are associated with positive spillovers regardless of stage, with magnitudes typically larger than those for protectionist IPs. ### Tit-for-tat dynamics and international context - The estimated association between protectionist subsidies and firm value added in the targeted industry shrinks as other countries (specifically those that are distant from a geopolitical point of view) introduce protective IPs targeting the same industry. - A similar, less robust, finding exists for export incentives. - The association between liberalizing IPs and firm-level value added becomes stronger when other countries are also introducing IPs. ### Robustness, endogeneity, and methodological notes - Robustness exercises include: - Local projection difference-in-difference (LP DiD) methods to account for staggered treatment timing (Dube et al., 2024): LP DiD results broadly consistent with baseline; exception is export incentives, where LP DiD finds larger and more significant medium-term associations with VA and TFP. - Tests show little evidence of pre-trends, particularly for domestic subsidies and export incentives. - Instrumental variable (IV) strategy instruments instruments by actions taken by other countries; IV results are consistent with baseline findings. - Additional robustness checks address: count nature of IP proxy; set of policies included; sample of countries; set of controls and lag structure of local projections. Results remain robust across exercises. - Limitations: - Empirical approach compares relative performance of firms in targeted and non-targeted industries; does not assess aggregate welfare or absolute desirability of IPs. - Full welfare assessment would require structural general equilibrium analysis and information on the size/fiscal cost of IPs, political economy, and retaliatory dynamics—beyond the scope of this paper. ### Contribution to literature - Provides cross-country, cross-sector firm-level analysis of IPs using broad data coverage (2 million firms, 38 countries, 2011–2018). - Complements country-specific and GTA-based studies by exploring heterogeneity by firm characteristics and spillovers along supply chains. - Relates to studies leveraging detailed state-aid data in Europe (e.g., Criscuolo et al., 2019; Brandão‑Marques and Toprak, 2024) that also find positive links between subsidies and firm revenue/payroll and stronger effects for younger firms. *IMF Working Paper — Industrial Policies and Firm Performance: A Nuanced Relationship* ### 2.1 Industrial policies ### 2.1 Industrial policies ### Data on industrial policies (IPs) - Source: Juhász et al. (2023) construct a global database of industrial policies from 2009–2022 using machine learning on policy text descriptions from the GTA database. - Definition of IPs: “state actions aimed at transforming the structure of economic activity, typically by altering relative prices across sectors or directing resources toward specific industries or activities like exporting and R&D”. - Aggregation and coding: - IPs aggregated from HS product code and CPC industry codes to NACE Rev. 2 industry codes using correspondences from the UN Statistics Division. - Applied recommended reporting-lag adjustment and only kept GTA policies announced and published within the same calendar year. - For each country and industry in a given year, the stock of active IPs is counted as policies announced but not yet removed. - Main outcome: the change in the stock of IPs in a given country and industry between two years (the “IP shock”). - GTA evaluation categories: - Red: protectionist (almost certainly discriminate against foreign commercial interests). - Amber: ambiguous (“likely involve discrimination against foreign commercial interests”); not the focus of this paper. - Green: liberalizing towards foreign commercial interests. - Policy instruments: 66 GTA instruments aggregated into 5 broad groups (per UN MAST classification): trade barriers (export and import), domestic subsidies, export incentives, local content requirements, and other instruments (FDI/public procurement measures, among others). - Caveats and data limitations: - Analysis does not identify firms specifically targeted by IPs; results capture direct and second-round impacts within industries. - Potential undercoverage: IPs in some key emerging markets, notably China, could be missing because (i) database starts in 2009 and (ii) the database focuses on national-level activities while implementation in some countries is decentralized. - IP counts do not capture monetary value (policy intensity). Correlation between IP counts and the log of their value in 2023 is 0.52 and statistically significant (NIPO/Evenett et al., 2024). - Methodology may undercount IPs that employ export and import barriers, which are more frequently used by EMDEs. - GTA underlying database does not capture subsidies in some EMDEs, potentially causing undercounting. ### Firm-level data - Source: ORBIS (Bureau van Dijk), covering around 300 million companies globally; harmonized cross-country financial information up to 2021. - Cleaning and sample construction: - Follow Kalemli-Ozcan et al. (2015), Gopinath et al. (2017), Gal (2013); ensures data accounts on average for at least 40% of total output reported in official sources. - Brazil and the US included despite lower coverage in some years. - Drop firms in: Financial Activities (NACE Rev 2 2-digit 64-66), Public Administration and Education (NACE 2-digits 84-85), Utilities (NACE 2-digits 35-39), Activities of Households as Employers and Extraterritorial Organizations (NACE 2-digits above 97). - Winsorize all firm-level variables at the 1 and 99 percent. - Restrict to firms reporting at least four consecutive periods. - Restrict analysis to the 2009-2021 period for firm-level data; final regression sample restricted to IPs introduced in 2011–2018 to form a balanced panel. - Main firm-level variables: - Value added: difference between operating turnover and material costs. - Capital stock: tangible fixed assets. - Total wage bill (payroll). - Productivity: TFPQ, following Hsieh and Klenow (2009) and IMF (2024). - Additional: firms’ age, cashflow to assets ratio, leverage ratio. - All nominal variables converted to U.S. dollars of 2015. ### Other data - Input-output (IO) matrices: calculated with GTAP database; NACE Rev. 2 4-digit industry codes assigned to one or more of the 65 GTAP sectors. - Main GTAP variable used: “domestic purchases by firms at basic prices” between GTAP sectors; normalized and used as IO coefficients. - Trade data: BACI database (CEPII), bilateral trade flows at HS 6-digit level across more than 200 countries; used to construct the IP trade intensity index. ### Final sample and summary statistics - Final regression sample: - IPs introduced in 2011–2018. - After merging IP, firm-level, and IO databases: over 2 million firms from 38 countries (11 EMDEs and 27 AEs) over 2011-2018. - Total observations: 8,515,018. - Key summary statistics (Table 1: Summary statistics of the main regression sample, 2011-2018): - ln(VA): Mean 13.2; Std. dev. 1.71; P10 11.3; Median 13.1; P90 15.4 - ∆ ln(VA): Mean 0.03; Std. dev. 0.32; P10 -0.26; Median 0.02; P90 0.33 - ln(Capital): Mean 11.7; Std. dev. 2.38; P10 8.9; Median 11.6; P90 14.7 - ∆ ln(Capital): Mean 0.03; Std. dev. 0.61; P10 -0.40; Median -0.04; P90 0.55 - ln(Payroll): Mean 12.2; Std. dev. 1.81; P10 10.1; Median 12.1; P90 14.4 - ∆ ln(Payroll): Mean 0.05; Std. dev. 0.34; P10 -0.20; Median 0.03; P90 0.32 - ln(TFPQ): Mean 7.8; Std. dev. 1.26; P10 6.3; Median 7.7; P90 9.3 - ∆ ln(TFPQ): Mean 0.01; Std. dev. 0.48; P10 -0.45; Median 0.02; P90 0.43 - ∆IP_red (change in stock of Protectionist (Red) IPs): Mean 0.039; Std. dev. 0.304; P10 0; Median 0; P90 0 - ∆IP_red,subsidies: Mean 0.018; Std. dev. 0.180; P10 0; Median 0; P90 0 - ∆IP_red,expinc: Mean 0.020; Std. dev. 0.234; P10 0; Median 0; P90 0 - ∆IP_green: Mean 0.029; Std. dev. 0.170; P10 0; Median 0; P90 0 - ∆IP_green,tradebar: Mean 0.027; Std. dev. 0.165; P10 0; Median 0; P90 0 - ∆Upstr_red: Mean 0.171; Std. dev. 0.397; P10 -0.001; Median 0.008; P90 0.572 - ∆Dwnstr_red: Mean 0.181; Std. dev. 0.446; P10 -0.004; Median 0.023; P90 0.591 - Summary observations drawn from Table 1 and Figures: - The panel is strongly balanced with no missing observations for outcomes of interest. - The average firm received 0.039 new protectionist (red) IPs in its industry between two consecutive years. - Over 90% of this shock is explained by introduction of protectionist subsidies (0.018) and protectionist export incentives (0.020). - The average firm experienced 0.029 more liberalizing (green) IPs, mostly reductions in trade barriers (0.027). - The average firm saw 0.171 (upstream) and 0.181 (downstream) more protectionist IPs in related sectors. - Composition over time (Figure 1 insights): - Sharp increase in total number of protectionist (Red) IPs after 2016. - About a third of IPs implemented in 2016 were protectionist export incentives. - Compositional change by 2018: protectionist domestic subsidies accounted for over a third of total protectionist IPs. - Substantial year-to-year changes in composition motivated restricting sample to 2011-2018 for a balanced panel. ### Empirical specifications and identification - Baseline local projections (Equation (1)): - Regress lnY_ft+h − lnY_ft−1 on changes in the stock of IPs by instrument k (subsidies, export incentives, trade barriers, local-content requirements, other) and GTA evaluation e (Red, Amber, Green), with h = 0,...,3. - Include two lags of dependent and independent variables, controls, and fixed effects: firm (α_f), country-year (α_ct), industry-year (α_it). - Standard errors clustered by country and industry. - Key coefficients: - β_ke_h: dynamic effect of one additional IP (by instrument and GTA evaluation) on firm outcomes. - θ_up_h, θ_dwn_h: indirect impacts through upstream and downstream exposure (value-chain spillovers). - Measures of exposure along the value chain (Amiti and Konings (2007) style): - Upstr_sct = sum_{s'≠s} io_cs'→s · IP_s'ct, normalized so sum of IO coefficients equals 1. - Dwnstr_sct = sum_{s'≠s} io_cs→s' · IP_s'ct, normalized. - OtherInd_ict = sum_{i'∈s, i'≠i} io_cs→s · IP_i'ct, scaling IPs in other industries within same sector by IO coefficients. - Controls X_ict include change and 2 lags of other GTA policies not classified as IPs but targeting the industry, and IPs in other industries within same sector. - Econometric choices motivated by Chudik and Pesaran (2015) rule of thumb on lags. ### Firm-level heterogeneity - Extended specification (Equation (5)): - Allows effects β_qke_h to vary across firm terciles q based on firm characteristic Z_ft−1 (age, cash flow-to-assets ratio, leverage ratio). - Q_Z_qct−1 represents tercile q of distribution of Z among firms in country c and year t−1. - Coefficients β_qke_h capture dynamic relationships for different firm types. ### Industry-level heterogeneity: role of distortions - Distortion measures: - Markups: calculated following Duval et al. (2024). - External Financial Dependence (EFD): calculated following Rajan and Zingales (1998). - Construction: - For each country-industry pair, calculate median markup and EFD; regress on industry and country fixed effects; use industry fixed effects as industry-level gauge of distortions. - Construct four dummies: - D_HH: high markup and high EFD. - D_LL: low markup and low EFD. - D_HL: high markup and low EFD. - D_LH: low markup and high EFD. - Appendix Table A.2 provides examples of industries in each distortion category. - Extended specification (Equation (6)): - Interacts ΔIP_e_ict with the four distortion dummies to estimate β_HH,e_h, β_HL,e_h, β_LH,e_h, β_LL,e_h. - Focus for comparison: β_HH,e_h versus β_LL,e_h (relationship between IPs and firm-level outcomes in industries with high distortions vs low distortions). *Source: IMF Working Paper (chapter 2.1), equations, tables, and text as provided in the source content.* ### 3.4 Tit-for-tat industrial policies and firm performance ### 3.4 Tit-for-tat industrial policies and firm performance ### Tit-for-tat construct and empirical specification - Constructed measure Z_ict for the number of Protectionist (Red) IPs implemented in industry i, between t−1 and t, by the average country in the world excluding country c: - Z_ict = Σ_{c′≠c} ω_{c′t} · ΔIP^Red_{ic′t}, ω_{c′t} = IPD_{cc′t−1} / Σ_{c′′} IPD_{cc′′t−1} - Countries are weighted by political distance IPD_{cc′t−1} (UN voting–based metric from Bailey et al. (2017)). - Main regression (dynamic firm-level horizon h = 0,...,3): - ln Y_{f,t+h} − ln Y_{f,t−1} = Σ_{instr_k} Σ_{GTA_eval_e} φ^h_{ke} · ΔIP_{ke,ict} · Z_ict + η Z_ict + Σ_{instr_k} Σ_{GTA_eval_e} β^h_{ke} ΔIP_{ke,ict} + ... + 2 Σ_{j=1} Σ_{instr_k} Σ_{GTA_eval_e} λ_{ke,t−j} IP_{ke,ict−j} + controls + firm, country-year, industry-year fixed effects + ε_{ft} - The main coefficient of interest φ^h_{ke} tests whether the relationship between IP instrument k and economic performance at horizon h depends on the amount of IPs implemented by less geopolitically aligned countries (tit-for-tat dynamics). ### Caveats and robustness methodologies - Baseline specification controls extensively: - Industry-year FEs, country-year FEs, firm FEs, lags of dependent and independent variables to address omitted variables and dynamics; pre-trends checked (3 years prior). - Robustness 1 — IP intensity index: - IP_Trade_{ict} = Σ_{policy_p} Σ_{product_q∈ic} (Trade_{qic} / Trade_{ic}) · 1{policy_p affects product_q} - Trade_{qic} and Trade_{ic} are averaged between 2012 and 2022. - Robustness 2 — LP difference-in-differences (LP DiD) following Dube et al. (2024), Cugat and Manera (2024), Ahn et al. (2024): - Clean treatment firms: treated for the first time in 3 years (stabilization lag). - Clean control firms: never directly exposed to IPs over 2011-2018. - Sample restricted to over 330,00 observations in the clean treatment group and over 5 million observations in the clean control group. - Robustness 3 — 2SLS IV strategy (Jordà and Taylor, 2016): - Instrument change in IPs of type k that country c implements in industry i with IPs implemented by other countries c′≠c. - For subsidies and export incentives, instrument uses protective subsidies introduced by other countries; for trade-liberalizing policies, uses liberalizing policies by other countries. - Aggregation weights vary by IP type: for subsidies, weight by political distance between c and c′; for export incentives and trade barriers, weight by trade flows between c and c′. ### Baseline results by instrument (average firm in treated industries) - Protectionist domestic subsidies: - One additional subsidy associated with a sustained increase of over 1 percent in the capital stock for the average firm. - Value added increases by 1 percent in the aftermath of a domestic subsidy, but this is fully reverted 3 years down the road. - Subsidies: positive short-term link to value added, TFP, and payroll but turn negative in the medium term. - Average protectionist subsidy policy remains in place for 3 years (on average). - Protectionist export incentives: - Short-term costs: an additional export incentive associated with 0.5 percent lower productivity for the average firm in the first two years after implementation. - Medium-term: mild positive association between export incentives and productivity, but with little statistical significance. - Value added and capital also experience short-term declines followed by medium-term recoveries that offset initial losses, but recoveries insufficient to improve these variables within the considered horizon. - Liberalizing trade-barrier policies: - An additional liberalizing policy associated with: - 1.6 percent higher productivity (medium term), - 1.2 percent higher value added (medium term), - 0.8 percent more payroll (proxy for wages and employment) (medium term), - 0.4 percent more capital stock (medium term; not statistically significant). - Trade liberalization linked with improved medium-term performance consistent with heterogeneous-firm trade models (import tariff reductions fostering productivity and growth). - Aggregate interpretation: - Protectionist subsidies: short-term improvements only (capital more sustained). - Export incentives: short-term costs, weak average-firm benefits. - Liberalizing IPs: medium-term gains in productivity and value added. ### Heterogeneity by firm characteristics - By firm age: - Subsidies: - Stronger link for younger firms; negligible or negative for older firms. - Temporary 2 percent improvement in value added for younger firms. - Temporary 1.5 percent increase in productivity for younger firms. - New subsidy linked to a 3.6 percent increase in the capital stock of younger firms three years after announcement, with negligible effects for older firms. - Export incentives: - Adjustment period less pronounced for younger firms; initial changes often non-significant or mildly positive for young firms. - Medium-term: one additional export incentive associated with 0.7 percent increase in productivity and value added of younger firms; for older firms the increase is close to zero. - Liberalizing measures: - Differences across firms by age less pronounced for value added and productivity. - Capital accumulation and payroll show more robust associations for young firms. - By financial constraints (cash flow to assets ratio as reliance on internal funds): - Subsidies: - Larger capital accumulation for financially constrained firms. - An additional subsidy associated with a 2 percent increase in the capital stock of firms with the largest cash flow to assets ratio; close to zero for firms with a low ratio. - Export incentives and liberalizing measures: - Some evidence of stronger medium-term gains for younger and more financially constrained firms, though differences are not always statistically significant. - Interpretation: - IPs have stronger positive associations for firms expected to face larger frictions (younger, financially constrained). - Policies can cause reallocation across firms within an industry—potential winners and losers—implying ambiguous aggregate welfare effects ex ante. - Appendix evidence (Figure A.1) indicates reallocation from older to younger firms after export incentive IPs can improve allocative efficiency (Hsieh and Klenow (2009) methodology). *Source: wpiea2025143 - 3.4 Tit-for-tat industrial policies and firm performance* ### 4.3 The relevance of industry-level distortions and position in the supply chain ### 4.3 The relevance of industry-level distortions and position in the supply chain ### Measure and empirical approach - Industry-level distortions are proxied by combining two industry characteristics: markups and external financial dependence (see Section 3). - For simplicity, the analysis focuses on the overall count of protectionist IPs. - Equation (6) is estimated allowing the response of firm-level variables to an additional industrial policy to vary with the industry’s degree of distortions. - Industrial policies are split according to their GTA evaluation; discussion focuses on protective (“Red”) measures and the total stock of protective measures (no instrument-level breakdown for this part). ### Main findings: targeting distorted industries - IPs targeting industries with higher levels of distortions are associated with stronger performance by the average firm in the industry. - An additional protectionist IP targeting a highly distorted industry is associated with a 1 percent increase in the value added of firms operating in that industry after 1 to 2 years, while there is no increase in the value added of firms in low distortion industries when these are targeted. - Capital and payroll show more pronounced and durable effects when IPs target highly distorted industries. - For productivity, the difference in the response of firms in high and low distortion industries is less pronounced; domestic subsidies’ prevalence and the finding that domestic subsidies do not appear to have sustained effects on productivity are consistent with this outcome. - Results are consistent with prior model-based evidence that poorly targeted IPs can lead to economic losses because adverse terms-of-trade effects may outweigh gains from economies of scale, while well-targeted policies can yield economic dividends to the implementing country (with potential negative spillovers on other countries). ### Cross-sector spillovers and supply-chain position - IPs propagate across sectors through input-output (IO) linkages; effects depend on whether targeted sectors are upstream or downstream relative to a firm. - Empirical focus: coefficients associated with IPs in upstream and downstream sectors relative to a firm (results presented for the last horizon of analysis). - Findings: - IPs targeting upstream sectors are linked to medium-term increases in the productivity, value added, capital stock, and payroll of firms in downstream sectors. - Mechanisms: temporary lifting of productivity and capital stock in upstream sectors can (i) alleviate capacity constraints, (ii) increase the productivity of inputs flowing downstream, and (iii) potentially lower upstream product prices, reducing intermediate input costs for downstream producers. - IPs targeting downstream sectors are negatively associated with firm performance in upstream sectors. - Intuition: by increasing productivity and lowering input demand in downstream sectors, IPs can push down demand for inputs from upstream sectors. - Overall implication: IPs in upstream sectors may benefit the economy more widely than IPs targeting downstream sectors, though some downstream sectors with distortions (e.g., de-carbonization of iron and steel production where network externalities on the demand side are present) may warrant downstream interventions. - Liberalizing IPs (those fostering trade) that target either upstream or downstream sectors are positively associated with firm performance in the medium term. - For IPs targeting upstream sectors, magnitudes associated with liberalizing IPs are 2 to 3 times as large as those for protectionist IPs. ### Interaction with international tit-for-tat dynamics - Countries often introduce IPs when other countries are conducting IPs (tit-for-tat dynamic); benefits from IPs can depend on other countries’ retaliatory actions. - Empirical strategy: interact the change in IPs targeting industry i with a variable gauging how active other countries are in targeting the same industry, with more weight on geopolitically-distant countries. - A larger value of this variable means geopolitically-distant countries are more active in protectionist IPs. - Findings: - The interaction between ∆IP_Red,subsidies and the variable capturing IP activity in other countries is negative and statistically significant. - Interpretation: the potential benefits of subsidies for the value added of firms in the targeted industry are diluted when other countries are actively introducing IPs. - For export incentives the interaction term is negative but not statistically significant. - By contrast, the positive association between new liberalizing trade policies and firm-level value added is amplified when other countries are conducting protectionist IPs. - Reinforces benefits of policies reducing trade barriers, especially during periods of high tit-for-tat IPs in other countries. ### Robustness checks and alternative specifications - LP DiD (Dube et al., 2024) constructing clean control (never treated) and clean treatment (first time treated) groups: - Confirms short-term positive association between subsidies and both value added and TFP. - Shows less marked improvements in capital after subsidies compared to baseline. - For export incentives, LP DiD suggests negative or non-significant relationships with value-added and capital, and a medium-term improvement in productivity that appears more quickly than in baseline. - Positive associations for liberalizing IPs appear at shorter horizons in LP DiD. - Heterogeneity: - Younger and capital-constrained firms have stronger performance links to IPs; LP DiD points to larger differences by firm age and financial constraint. - Pre-trends: - Tests using regressions of lagged firm-level outcomes on contemporaneous IPs (only for lags greater than those included in baseline) find: - No evidence of pre-trends for domestic subsidies. - Some pre-trends for export incentives for certain outcomes (most notably payroll). - A statistically significant negative pre-trend for liberalizing IPs; interpreted as a change in trend indicative that liberalizing IPs affect outcomes in the direction shown by local projections. - Policy intensity alternative: - Using the share of trade in a given country and industry affected by IPs (trade coverage, Equation (9)) yields similar association patterns; short-term positive (subsidies) and medium-term positive (export incentives) associations are less precise when using trade exposure. - Endogeneity concerns and IV exercise: - Constructed weighted count of IPs introduced by other countries as an instrument (for subsidies weighted by geopolitical distance; for trade-related IPs weighted by bilateral trade). - IV results are consistent with baseline: value added and TFP increase with subsidies in the short term and fall thereafter; capital has a more sustained increase. For export incentives, VA and TFP improve in the medium term after an adjustment period. Liberalizing trade barriers link to medium-term improvement in VA and TFP. - First-stage coefficients indicate instrument is strongly correlated with the IP instrument of interest. - Additional robustness battery (summary of checks): - Consider all policies of a given GTA evaluation; consider all policies GTA published within a year of announcement; exclude China, Brazil, and the US; add firm-level controls; drop firms with abnormal growth; remove over-represented countries and re-weight regressions; focus on countries with good ORBIS coverage; control for 3 lags of dependent and independent variables. - Most general patterns from baseline hold across these exercises. Exception: for all GTA liberalizing policies, increase in value added observed in the short term (rather than medium-term recoveries) and change in capital is not significant. - Aggregation and alternative measures: - Aggregating firm-level information at the industry level produces similar links between IPs and performance. - Categorizing industries by mean markups (rather than median markups) preserves findings. - Using the leverage ratio as an alternative proxy for firm financial constraint confirms stronger links for younger and more financially constrained firms. - Results robust to using TFP measures from Ackerberg et al. (2015) rather than TFPQ measures from Hsieh and Klenow (2009). - Remaining endogeneity caveats: - Firm-specific policies that substitute for industry-level IPs (attributing firm-specific policies to all firms in an industry) could bias estimates (likely bias downwards positive coefficients and bias upwards negative coefficients). - Cannot control for degree of political connections a firm has; however, findings that younger and more financially constrained firms benefit more from IPs, together with evidence that politically connected firms tend to be older, larger, and less financially constrained, mitigate this concern. ### Implications highlighted in this section - IPs targeted at industries with larger distortions produce stronger positive firm-level outcomes (value added, capital, payroll) than IPs targeting less distorted industries. - Position in supply chain matters: - Upstream-targeted IPs have positive cross-sectoral spillovers on downstream firms’ productivity, value added, capital, and payroll. - Downstream-targeted IPs are negatively associated with upstream firm performance. - Liberalizing IPs generate uniformly positive spillovers and larger magnitudes than protectionist interventions. - The effectiveness of subsidies can be diluted when other countries are actively introducing IPs (tit-for-tat dynamics), while liberalizing policies may be especially beneficial when other countries pursue protectionist IPs. _Italic: Source — IMF Working Paper section "4.3 The relevance of industry-level distortions and position in the supply chain"._ ### 1. VA by Age ### 1. VA by Age ### Description of empirical design and measures - The figure plots the percent change in firm-level value added (VA) 0, 1, 2 and 3 years after the implementation of a policy. - Firms are grouped into the 1st (red) and 3rd (blue) terciles of the distribution of firm age. - Estimated transformation: 100×(exp(β_qke_h)−1). - Standard errors are clustered by country and NACE Rev. 2 4-digit industry. - Shaded areas represent 90 percent confidence intervals. - Sources: Juhász et al. (2023), GTA, ORBIS, GTAP. ### Context within related figures - Analogous figures accompany VA by CF, Productivity by Age, Productivity by CF, Capital by Age, Capital by CF, Payroll by Age, and Payroll by CF for the same policy shocks. - Comparable notes and estimation approaches are used for protectionist domestic subsidies, protectionist export incentives, and liberalizing trade barrier policies in subsequent figures. ### Key statistics and reported sample averages (from Table A.4) - ∆ ln VA_f t+h: h=0 = 3.19 ; h=1 = 5.95 ; h=2 = 7.01 ; h=3 = 8.22 - ∆ ln Capital_f t+h: h=0 = 3.29 ; h=1 = 5.60 ; h=2 = 6.66 ; h=3 = 5.83 - ∆ ln Payroll_f t+h: h=0 = 4.80 ; h=1 = 8.95 ; h=2 = 10.8 ; h=3 = 11.9 - ∆ ln TFPQ_f t+h: h=0 = 0.49 ; h=1 = 1.11 ; h=2 = 1.12 ; h=3 = 2.51 - Notes: Values are the sample average of cumulative growth of each outcome over different horizons relative to h=−1. Values are multiplied by 100 and expressed in percent. Sources: ORBIS. ### Pre-trend (placebo) evidence relevant to VA and related outcomes (from Table A.5) - VARed subsidies: Coefficient = 0.001 ; Std t-stat = 0.0037 ; p-value = 0.728 - VARed export incentives: Coefficient = -0.007 ; Std t-stat = 0.0036 ; p-value = 0.062 - VAGreen trade barriers: Coefficient = -0.025 ; Std t-stat = 0.0052 ; p-value = 0.000 - Interpretation: Table A.5 reports placebo regression estimates of change in log outcome between t−3 and t−1 on changes in IPs between t−1 and t to test for pre-trends; standard errors clustered by country and industry. Sources: Juhász et al. (2023), GTA, ORBIS, GTAP. ### Estimation and inference details emphasized in the figure note - Local projections are used (see transformation 100×(exp(β_qke_h)−1)). - Standard errors clustered by country and NACE Rev. 2 4-digit industry. - Confidence regions shown are 90 percent confidence intervals. *Source: IMF Working Paper chapter/figure notes and Appendix (Juhász et al. (2023), GTA, ORBIS, GTAP).* ### 1. Protectionist Domestic Subsidies ### 1. Protectionist Domestic Subsidies ### Overview of empirical exercises - Figures and tables focus on the relationship between protectionist domestic subsidies and firm- and industry-level performance, using firm-level outcomes: VA, TFP (productivity), payroll, and capital stock. - Time horizons analyzed: 0, 1, 2 and 3 years after policy implementation (often reported as "0,1,2 and 3 years"). - Main estimation expressions reported in the source: - Aggregate industry-level effects: 100×(exp(βke h)−1). - Firm-level heterogeneous effects: 100×(exp(βqke h)−1). - Standard errors are clustered at the country level and NACE Rev. 2 4-digit industry for industry-level estimates; clustered by country and NACE Rev. 2 4-digit industry for firm-level estimates. - Shaded areas in figures represent 90 percent confidence intervals. ### Firm-level heterogeneous effects (Figures A.3, A.14) - Outcomes analyzed by firm characteristics: - VA by Age; VA by CF. - Productivity by Age; Productivity by CF. - Capital by Age; Capital by CF. - Payroll by Age; Payroll by CF (Figure A.14 includes payroll). - Heterogeneity is frequently shown by terciles of the firm-characteristic distribution (e.g., firms in the 1st and 3rd terciles are colored red and blue respectively in Figure A.14). - Figure A.14 explicitly states: "This figure plots the percent change in each firm-level outcome (VA, TFP, payroll and capital stock), 0,1,2 and 3 years after the implementation of a protectionist domestic subsidy, for firms in the 1st (red) and 3rd (blue) terciles of the distribution of firm characteristics, where age and cash flow to assets ratio are included in the same specification. These are estimated following Equation (5): 100×(exp(βqke h)−1). Standard errors are clustered by country and NACE Rev. 2 4-digit industry. Shaded areas represent 90 percent confidence intervals." ### Industry-level effects (Figure A.9) - Figure A.9 description: "This figure plots the percent change in each industry-level outcome (VA, TFP, payroll and capital stock), 0,1,2 and 3 years after the implementation of a protectionist domestic subsidy, estimated in an aggregate version of Equation (1): 100×(exp(βke h)−1). Standard errors are clustered at the country level country and NACE Rev. 2 4-digit industry. Shaded areas represent 90 percent confidence intervals." ### Instrumental variables and identification (Figure A.7 notes relevant to subsidies) - Instrument for protectionist subsidies: protectionist IPs in other countries and the same industry, weighted by political distance. - First stage Kleibergen-Paap rk Wald F statistics truncated at 60. - Figure A.7 contains a "First stage" panel and a "Second stage: Subsidies" panel among others. ### Robustness and additional checks (Figure A.8) - A range of robustness exercises overlayed on the link between IPs and the average firm: 1) considering all GTA policies for each evaluation instead of focusing on IPs; 2) including GTA policies published within a year from announcement; 3) including 3 lags of dependent and independent variables; 4) adding firm controls; 5) dropping extreme growth outliers in each horizon; 6) dropping firms from China, and from the US and Brazil; 7) dropping firms from Spain; 8) dropping firms from France; 9) dropping firms from Italy; 10) restricting attention to countries in Cravino and Levchenko (2016); 11) weighting regressions by the inverse of the number of firms in each country; 12) weighting regressions by lag firm value added. - Exercise 1) is highlighted in blue. - Results for TFP closely follow results for VA and are omitted for brevity in the source. Results by exercise are available upon request. ### Contextual analyses and related diagnostics - Figures compare subsidies against other IP dimensions (protectionist export incentives and liberalizing trade barriers) across the same outcomes and heterogeneities (age, cash flow, leverage, trade intensity). - Trade-intensity analysis excludes firms and policies in services, which are not assigned a HS product code. - Additional figures categorize industries by mean markups and examine how IPs vary by industry distortions and firms’ leverage ratio. ### Data sources and provenance - Sources cited consistently in figures and notes: Juhász et al. (2023), GTA, ORBIS, GTAP. *Industrial Policies and Firm Performance: A Nuanced Relationship — Working Paper No. WP/2025/143* --- _Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025143.pdf_