## wpiea2022040-print-pdf

## Source details

**Canonical URL:** [wpiea2022040-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022040-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022040-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022040-print-pdf.pdf.json)

---

### Context and motivation
- Starting April 2017, trade tensions escalated, particularly between the US and China.
- Until the signature of a “phase-one deal” in January 2020, average bilateral tariffs between the US and China increased to above 21%, from 8% and 3% before 2018, respectively on Chinese imports from the US and US imports from China (Brown, 2021).
- Among mostly OECD countries, average tariff levels had declined from roughly 4.5% in the mid-1990s to about 2% in 2015.
- Illustrative manufacturing value-added weighted tariff declines between 1995 and 2015:
  - India: from 32% to 11%
  - China: from 20% to 7%
  - Romania: from 14% to 0.7%
- Falling tariffs coincided with an increase in average global value chain (GVC) participation (forward + backward).

### Research question and empirical approach
- Core question: To what extent are countries and sectors not directly targeted by US and Chinese tariffs affected through lower demand for intermediate exports, higher costs for intermediate inputs, or lower competition in export markets?
- Empirical strategy:
  - Use a three-dimensional panel (country-sector-year) to calculate four tariff measures and estimate effects on value added, employment, labor productivity, and total factor productivity.
  - Four tariff measures (lagged by one year in estimation):
    - (a) domestic protection (average tariffs on competing imports);
    - (b) upstream tariffs (tariffs higher up in the value chain that increase input costs);
    - (c) downstream tariffs (tariffs further down the value chain that raise costs of a sector’s output for eventual consumers);
    - (d) diversion tariffs (ease of access to export markets compared to other international competitors).
  - Baseline panel regression:
    - ln(y_{c,s,t}) = α + β1 T^{up}_{c,s,t-1} + β2 T^{down}_{c,s,t-1} + β3 T^{dom}_{c,s,t-1} + β4 T^{div}_{c,s,t-1} + γ_{c,t} + δ_{c,s} + ε_{c,s,t}
  - Two types of fixed effects: country-sector (δ_{c,s}) and country-time (γ_{c,t}).
  - Estimation uses the linear estimator for high-dimensional fixed effects (Correia, 2017); residuals clustered at country-sector level.

### Data and empirical setup
- Key data sources:
  - Input-output matrices: 2016 OECD ICIO Tables, ISIC REV.3 sectoral classification (34 sectors, 65 countries).
  - Bilateral tariffs: World Integrated Trade Solution Platform; HS6 goods-level tariffs aggregated to sector-level using constant average import-shares; EU tariffs used for all member countries and tariffs among EU members assumed zero.
  - Real economic variables: sectoral real value added, employment, labor productivity, and total factor productivity from KLEMs databases (EU-KLEMs, World-KLEMs, Asia-KLEMs).
- Aggregation uses constant weights (import-, export- and input-output weights); missing observations at ends use last/earliest reported values.
- Sample availability:
  - Outcome variables: roughly half of countries in eventual sample for value added, employment, labor productivity.
  - 22 countries in TFP specifications: AUS, AUT, BEL, CAN, CZE, DEU, DNK, ESP, FIN, FRA, GBR, HUN, IRL, ITA, JPN, NLD, SVK, SVN, SWE and USA.
  - Employment and labor productivity regressions additionally include: CYP, EST, GRC, HRV, KOR, LTU, LUX, LVA, MLT, POL, PRT, and ROU.
  - Value added regressions further include China and India.
  - For Mexico and China, multiple observations per sector aggregated to one observation per sector.

### Construction of the four tariff measures
- Domestic protection:
  - Weighted average of bilateral tariffs τ_{c,i,s,t} applied by country c to imports from partner countries i that directly compete with sector s in year t.
  - Weights ω̄_{c,i,s,t} equal average share in country c’s imports of products from sector s coming from partner country i.
- Upstream tariffs:
  - Weighted average of tariffs applied on intermediate inputs imported directly and cumulative tariffs applied at previous production steps.
  - Notation: τ_{h,j} tariff by country-sector h on country-sector j; ā_{h,j} elements of input-output matrix Ā scaled by total output of h.
  - Scaled by share of imported inputs in a country-sector’s output; magnitudes substantially smaller than nominal tariff rates.
  - Matrix: τ_{h,·}^{up} = (e_{1×CS} B_{1×CS} (I − Ā)^{-1}), where B = Ā ∘ τ.
- Downstream tariffs:
  - Capture cumulative tariffs that a sector’s output faces downstream (direct and indirect exports).
  - Scaled by shares of total output that is exported; downstream tariffs tend to be larger than upstream tariffs.
  - Matrix: τ_{·,h}^{down} = ((I − Ā)^{-1} B^{down} e_{C×1}), where B^{down} = X̃ ∘ τ and X̃ is nominal exports scaled by GDP.
- Diversion tariffs:
  - Capture advantage sector i in country j enjoys when exporting to country p, relative to competitors.
  - Constructed by averaging tariffs partner p imposes on others weighted by partner p’s imports, then averaging across partners using home-country export shares ω̄_{c,s,p}^{X}.
  - Formula: τ_{c,s}^{div} = Σ_p ω̄_{c,s,p}^{X} [ Σ_i ω̄_{p,s,i}^{M} τ_{p,s,i} ].
  - Weights ω̄ are time-averaged import or export shares (1/T Σ ...).

### Descriptive patterns of tariff measures
- Aggregate trends:
  - Upstream and downstream tariffs have declined alongside domestic protection; sizes differ due to scaling.
  - Upstream and downstream tariffs (scaled) are substantially smaller than diversion and import tariffs (not scaled).
  - Downstream tariffs tend to be larger than upstream tariffs because downstream weights use shares of total output that is exported, whereas upstream weights use shares of imported inputs in total output.
- Cross-sectional correlations (reported coefficients):
  - Upstream vs downstream: 0.85 across sectors (0.62 when the textile sector is dropped) and 0.68 across countries.
  - Upstream vs domestic protection: 0.77 across countries and 0.73 across sectors.
  - Downstream vs domestic protection: 0.35 across countries and 0.56 across sectors; highly insignificant if either the textile sector or Korea is excluded.
- Historical changes (1995–2010 correlations):
  - Upstream vs downstream changes: 0.46 across countries and 0.80 across sectors.
  - Change in upstream vs change in domestic protection: 0.83 across countries and 0.25 across sectors (insignificant across sectors).
  - Change in downstream vs change in domestic protection: 0.20 across countries and 0.02 across sectors (both highly insignificant).
- Sectoral and country variation:
  - Downstream tariffs relatively more important for basic and fabricated metals or chemicals.
  - Downstream tariffs relatively less important for sectors relying on heavily taxed intermediate inputs, e.g., food beverage and tobacco.
  - Upstream tariffs relatively more important for assembly economies (e.g., Korea and China) than for raw-material exporters (e.g., Australia and Finland).
- Illustrative example: US tariff increase on semiconductor imports from China raises upstream tariffs for chip-using industries, raises diversion tariffs for competitors (e.g., Taiwan Province of China, South Korea), increases downstream tariffs for suppliers to microchip users, while domestic protection supports US producers.

### Baseline empirical results (Table 1: effects of a 1 ppt increase; dependent variables in natural log; tariff measures lagged one year)
- Coefficient estimates (standard errors in parentheses); residuals clustered at country-sector level; * p<0.10 ** p<0.05 *** p<0.01.
- T upstream:
  - VA: -19.41*** (6.810)
  - Empl.: -10.24*** (3.853)
  - L-Prod.: -6.53* (3.847)
  - TFP: -11.52** (4.696)
- T dom. protection:
  - VA: 0.02 (0.646)
  - Empl.: -0.88* (0.487)
  - L-Prod.: 0.92 (0.885)
  - TFP: 0.37 (0.383)
- T downstream:
  - VA: -14.47** (6.027)
  - Empl.: -0.50 (3.790)
  - L-Prod.: -12.61*** (4.672)
  - TFP: -13.19** (6.394)
- T diversion:
  - VA: 5.14* (2.792)
  - Empl.: 3.50* (1.836)
  - L-Prod.: 1.29 (2.100)
  - TFP: -2.70 (4.188)
- Fixed effects: Country-Year FE Yes; Country-Ind. FE Yes.
- Sample sizes and fit:
  - N: 6774 (VA), 6776 (Empl.), 6144 (L-Prod.), 4112 (TFP).
  - R2: 0.733 (VA), 0.995 (Empl.), 0.734 (L-Prod.), 0.693 (TFP).
- Key takeaways:
  - Upstream and downstream tariffs: large, negative, statistically significant effects on value added, labor productivity, and TFP; upstream tariffs also significantly reduce employment.
  - Domestic protection: generally small and insignificant effects; borderline-significant negative effect on employment (-0.88*).
  - Diversion tariffs: positive and weakly significant effects on value added (5.14*) and employment (3.50*); no significant productivity gains.

### Baseline estimates: detailed interpretation (Section 3.1)
- Real value added:
  - Upstream and downstream tariffs are drags on activity; diversion tariffs tend to support value added; domestic protection shows no effect.
- Employment:
  - Employment declines with higher upstream tariffs and increases with higher diversion tariffs.
  - Employment appears unaffected by downstream tariffs.
  - Country-sector fixed effects explain 96% of variation for employment; adding country-year FE brings R2 to 0.9945, implying the four tariff coefficients are identified on only 0.05% of the variation used for employment estimates.
- Labor productivity and TFP:
  - Labor productivity strongly and negatively affected by upstream and downstream tariffs.
  - Domestic protection and diversion tariffs far from statistically different from zero for labor productivity.
  - TFP results closely follow labor productivity results.

### Magnitude of effects and standardization (Section 3.2, Table 2: effects of a 1 standard-deviation increase)
- Standardized coefficients (one-standard-deviation increase; standard errors in parentheses):
  - T upstream:
    - VA: -16.18*** (5.68)
    - Empl.: -8.55*** (3.21)
    - L-Prod.: -5.45* (3.21)
    - TFP: -9.60** (3.91)
  - T dom. protection:
    - VA: 0.10 (3.80)
    - Empl.: -5.17* (2.86)
    - L-Prod.: 5.42 (5.20)
    - TFP: 2.17 (2.25)
  - T downstream:
    - VA: -11.93** (4.97)
    - Empl.: -0.41 (3.12)
    - L-Prod.: -10.40*** (3.85)
    - TFP: -10.87** (5.27)
  - T diversion:
    - VA: 10.98* (5.96)
    - Empl.: 7.45* (3.92)
    - L-Prod.: 2.75 (4.48)
    - TFP: -5.76 (8.94)
- Interpretation: one-standard deviation changes in upstream or downstream tariffs have roughly comparable effects on value added to an equivalent-sized change in diversion tariffs.

### Simulation of interconnected tariff changes (Figure 5 example)
- Simulated reciprocal uniform 1 ppt tariff increase between the US and China:
  - Most negative effects on value added occur for China and the US themselves, driven by upstream and downstream tariff changes.
  - Domestic protection increases but shows little positive effect due to near-zero coefficient.
  - Other countries generally benefit from trade diversion, but downstream—and to a lesser extent upstream—tariff increases partially offset gains (notably for Korea, Japan, Canada).

### Comparison with general-equilibrium models
- Estimated effects here are larger than typical theoretical general-equilibrium model predictions.
  - Example extrapolation: a reciprocal and uniform increase in bilateral tariff by 25 percentage points between the US and China would reduce Chinese value added by roughly 5½ % in the long run and US value added by roughly 3% according to the coefficients.
  - IMF (2019) model simulations suggest likely GDP effects may be only a fraction of this.
- Possible explanations for larger estimates:
  - Country-year fixed effects may absorb general equilibrium adjustments (exchange rates, factor cost changes).
  - Sample focuses on manufacturing sectors, which are particularly sensitive to tariffs.

### Robustness highlights
- Alternative specifications and robustness checks:
  - Individual vs jointly estimated effects (Table 3): jointly estimated upstream -19.41*** (6.810); individually estimated upstream -22.30*** (7.464).
  - Alternative weighting schemes for tariff aggregation (Table 4: 1995-2010, 1995, 2010 weights) — upstream and downstream coefficients broadly robust:
    - T upstream: -19.41*** (6.81), -20.63*** (7.11), -17.18** (8.01) across columns.
    - T downstream: -14.47** (6.03), -15.84** (7.53), -11.99*** (4.63) across columns.
    - T diversion: 5.14* (2.79), 2.19 (2.67), 4.46** (2.01) across columns.
  - Lag sensitivity (Table 5): effects decline with increasing lag length but main conclusions broadly unchanged.
    - Contemporaneous (LX.): T upstream -20.85*** (7.04); LX. T downstream -15.37** (6.37); LX. T diversion 5.87** (2.81).
    - L1: T upstream -19.41*** (6.81); T downstream -14.47** (6.03); T diversion 5.14* (2.79).
    - L2: T upstream -18.53*** (6.62); T downstream -12.75** (5.64); T diversion 4.37 (2.68).
    - L3: T upstream -17.18*** (6.18); T downstream -11.43** (5.25); T diversion 3.73 (2.54).

### Conclusions and policy implications
- Main findings:
  - Indirect tariffs (upstream and downstream) impose economically and statistically significant negative effects on sectoral value added, labor productivity, and TFP; upstream tariffs also significantly reduce employment.
  - Diversion tariffs can generate positive effects on value added and employment through trade diversion.
  - Direct domestic protection generally does not improve economic prospects in the protectionist country; effects on protected sector are negligible or negative, and higher input costs harm the home country.
  - Bilateral tariff wars hurt the two parties most; third countries can benefit net from trade diversion, though downstream/upstream effects may offset some gains.
- Policy-relevant implications:
  - Policymakers should account for indirect tariff exposure through GVCs when assessing costs of tariff policy; burdens from increased input costs and lost downstream demand can outweigh limited domestic protection gains.
  - Trade policy analysis should incorporate input-output linkages to capture upstream and downstream effects and potential trade diversion impacts.
- Limitations and open questions:
  - Focus on tariffs excludes non-tariff barriers and other trade costs; findings may understate total effects of trade tensions where non-tariff measures and uncertainty rise.
  - Historical tariff declines were gradual; unclear whether tariff increases have symmetric effects or nonlinearities.
  - Global value chain adaptation over time may generate dynamics not fully captured; further research needed.

*IMF Working Paper: The Effect of Tariffs in Global Value Chains — extract from wpiea2022040-print-pdf (Introduction, Baseline Estimates, Magnitude of Effects, Conclusion).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Context and motivation
- Starting April 2017, trade tensions escalated, particularly between the US and China.
- Until the signature of a “phase-one deal” in January 2020, average bilateral tariffs between the US and China increased to above 21%, from 8% and 3% before 2018, respectively on Chinese imports from the US and US imports from China (Brown, 2021).
- This episode reversed a decades-long trend in trade policy: among mostly OECD countries, average tariff levels had declined from roughly 4.5% in the mid-1990s to about 2% in 2015.
- Examples of manufacturing value-added weighted tariff declines between 1995 and 2015 cited in the source:
  - India: from 32% to 11%
  - China: from 20% to 7%
  - Romania: from 14% to 0.7%
- Falling tariffs, together with improved transportation and communication technologies, coincided with an increase in average global value chain (GVC) participation (forward + backward).

### Research question and approach
- Core question: To what extent are countries and sectors not directly targeted by US and Chinese tariffs affected through lower demand for intermediate exports, higher costs for intermediate inputs, or lower competition in export markets?
- Empirical strategy: use a three-dimensional panel (country-sector-year) to calculate four tariff measures and estimate their effects on value added, employment, labor productivity, and total factor productivity.
- Four tariff measures constructed for each country-sector-year:
  - (a) average tariffs on competing imports (degree of domestic protection from foreign competition);
  - (b) tariffs higher up in the value chain (which increase the cost of inputs);
  - (c) tariffs further down in the value chain (which increase the costs of the own output for the eventual consumer);
  - (d) the ease of access to export markets compared to other international competitors.

### Main empirical findings (overview)
- Upstream and downstream tariffs generally hurt economic activity:
  - Coefficients are negative and significant for value added, labor productivity, and total factor productivity.
- Higher domestic protection (output tariffs) is generally found to be insignificant, except for a borderline-significant negative effect on employment.
- Higher tariffs on competitors are associated with higher value added and employment, but not improved productivity.
- Illustrative simulations provide a sense of the relative magnitudes of the effects.

### Theoretical grounding and contribution to literature
- Builds on theoretical work emphasizing tariff accumulation and magnification when goods cross borders at several stages of production (Yi 2003, 2010; Koopman, Wang, and Wei 2014; Rouzet and Miroudot 2013).
- Incorporates models that highlight the role of intermediate inputs in accumulation of tariff effects (Arkolakis, Costinot and Rodriguez-Clare 2012; Eaton et al. 2016; Hsieh and Ossa 2016; Shikher 2011; Caliendo and Parro 2015; Vandenbussche, Connell and Simons 2019).
- Extension relative to prior empirical literature:
  - Examines a broader set of dependent variables including value added and employment (in addition to productivity).
  - Jointly estimates effects not only of domestic protection and input tariffs, but also tariffs further down the value chain and tariffs on competitors, highlighting potential omitted-variable bias in prior work.

### Scope and limitations
- Focuses on tariffs and abstracts from other trade costs (e.g., non-tariff barriers, transportation costs) for practical and conceptual reasons.
- Uses bilateral, goods-specific, ad-valorem tariffs which combine with input-output logic; non-tariff barriers are less amenable to the same treatment.
- Leaves investigation of whether conclusions extend to other types of trade costs for further research.

### Data and empirical setup (summary)
- Three key data types:
  - Input-output matrices: 2016 version of the OECD Inter-Country Input-Output (ICIO) Tables, ISIC REV.3 sectoral classification.
  - Bilateral tariffs: from the “World Integrated Trade Solution Platform”; HS6 goods-level tariffs aggregated to sector-level using constant average import-shares as weight; linear interpolation when goods-level tariffs not reported every year; EU tariffs used for all member countries and tariffs among EU members assumed zero.
  - Real economic variables: sectoral real value added, employment, labor productivity, and total factor productivity from KLEMs databases (EU-KLEMs, World-KLEMs, Asia-KLEMs). Baseline labor productivity proxy: gross value added per hour worked; alternative: gross value added divided by employment. Estimation uses natural logarithms of outcome variables.
- Aggregation uses constant weights (import-, export- and input-output weights) so tariff measure dynamics reflect trade policy decisions rather than composition changes.
- Empirical model estimates the effect of the respective tariff measures on value added, employment, labor productivity, and total factor productivity; robustness checks and illustrative simulations are reported in subsequent sections.

*Source: wpiea2022040-print-pdf (1. Introduction)*

### conclusions do not change significantly when different weights are chosen.

### wpiea2022040-print-pdf - conclusions do not change significantly when different weights are chosen.

### Data and sample
- Input-output tables include 34 sectors for 65 countries, and tariff data is nearly universal.
- Data availability for outcome variables is more limited:
  - Only roughly half of the countries are part of the eventual sample for value added, employment and labor productivity.
  - 22 countries are in the specifications using TFP.
- For TFP, the eventual sample includes: AUS, AUT, BEL, CAN, CZE, DEU, DNK, ESP, FIN, FRA, GBR, HUN, IRL, ITA, JPN, NLD, SVK, SVN, SWE and USA.
- Regressions of employment and labor productivity additionally include: CYP, EST, GRC, HRV, KOR, LTU, LUX, LVA, MLT, POL, PRT, and ROU.
- Regression of value added further include China and India.
- For Mexico and China, multiple observations per sector were aggregated to one observation per sector to keep the exercise tractable.
- Missing observations at the end (beginning) of the sample use the last (earliest) reported value.

### Construction of tariff measures (four measures)
- The paper constructs four tariff measures: (i) domestic protection, (ii) upstream tariffs, (iii) downstream tariffs, and (iv) diversion tariffs.

- Domestic protection:
  - Captures the degree of protection producers in country c and sector s enjoy from international competition.
  - Calculated as the weighted average of bilateral tariffs τ_{c,i,s,t} applied by country c to imports from partner countries i that directly compete with sector s in year t.
  - Weights ω̄_{c,i,s,t} equal the average share in country c’s imports of products from sector s that comes from partner country i.
  - A higher degree of domestic protection is expected to support domestic demand.

- Upstream tariffs:
  - Also called input tariffs in other studies; measure the weighted average of tariffs applied on intermediate inputs imported directly and, following Rouzet and Miroudot (2013), include cumulative effects of tariffs on inputs irrespective of which border crossing imposed them.
  - Includes (i) the tariff imposed by country c on inputs imported directly by sector s and (ii) tariffs applied at previous production steps to inputs sourced domestically or from abroad.
  - A higher upstream tariff makes inputs more expensive (negative supply shock).
  - Notation: τ_{h,j} is the tariff imposed by country-sector h on country-sector j; ā_{h,j} are elements of the input-output matrix Ā scaled by total output of h.
  - Upstream tariffs are scaled by the share of imported inputs in a country-sector’s output, making their magnitudes substantially smaller than nominal tariff rates.
  - Matrix expression: τ_{h,·}^{up} = (e_{1×CS} B_{1×CS} (I − Ā)^{-1}), where B = Ā ∘ τ (element-by-element multiplication).

- Downstream tariffs:
  - Mirror image of upstream tariffs; capture cumulative tariffs that the output of a given sector faces downstream in the value chain, including direct exports and indirect exports after further processing.
  - A higher downstream tariff makes a country-sector’s output more expensive and is expected to reduce demand.
  - Downstream tariffs are scaled by shares of total output that is exported; therefore downstream tariffs tend to be larger than upstream tariffs.
  - Matrix expression: τ_{·,h}^{down} = ((I − Ā)^{-1} B^{down} e_{C×1}), where B^{down} = X̃ ∘ τ and X̃ is nominal exports scaled by GDP.

- Diversion tariffs:
  - Capture the advantage sector i in country j enjoys when exporting to country p, relative to competitors.
  - Constructed in two steps:
    1. For each partner country p and sector i, calculate the average tariff imposed on all other countries than home country c, weighted by the imports of country p.
    2. For home country c, average the tariffs the partner countries p impose on others using export shares ω̄_{c,s,p}^{X} as weights.
  - An increase in the diversion tariff should support demand for the country-sector in question.
  - Formula: τ_{c,s}^{div} = Σ_p ω̄_{c,s,p}^{X} [ Σ_i ω̄_{p,s,i}^{M} τ_{p,s,i} ].
  - Weights ω̄ are time-averaged import or export shares (1/T Σ ...).

### Description and illustrative patterns of tariff measures
- Three broad empirical patterns:
  1. Aggregate trends:
     - Upstream and downstream tariffs have come down in line with domestic protection (nominal tariff rates).
     - Dynamics are similar, but sizes differ due to scaling: upstream and downstream tariffs (right-hand scale) are substantially smaller than diversion and import tariffs (not scaled by output or exports).
     - Downstream tariffs tend to be larger than upstream tariffs because downstream weights use shares of total output that is exported, whereas upstream weights use shares of imported inputs in total output.
  2. Cross-sectional correlations:
     - Significant cross-sectional correlation between upstream and downstream tariffs and with domestic protection.
     - Countries/sectors with high upstream tariffs also tend to have high downstream tariffs.
     - Correlation sensitivity: correlation with domestic protection is slightly lower and more sensitive to outliers (particularly downstream tariffs).
     - Reported correlation coefficients:
       - Upstream vs downstream: 0.85 across sectors (0.62 when the textile sector is dropped) and 0.68 across countries.
       - Upstream vs domestic protection: 0.77 across countries and 0.73 across sectors.
       - Downstream vs domestic protection: 0.35 across countries and 0.56 across sectors; highly insignificant if either the textile sector or Korea is excluded.
  3. Substantial variation by sector and country position in GVCs:
     - Downstream tariffs relatively more important for sectors such as basic and fabricated metals or chemicals.
     - Downstream tariffs relatively less important for sectors relying on heavily taxed intermediate inputs, such as food beverage and tobacco.
     - Upstream tariffs relatively more important for countries that assemble intermediate inputs (e.g., Korea and China) than for countries with significant raw material exports (e.g., Australia and Finland).

- Historical declines and exogenous variation:
  - Countries have seen upstream and downstream tariffs decline, often independently of their own domestic protection.
  - Change between 1995 and 2010 shows pronounced correlation between declines in downstream and upstream tariffs, while link with domestic protection is weak and sensitive to outliers.
  - Correlation coefficients for changes:
    - Upstream vs downstream changes: 0.46 across countries and 0.80 across sectors.
    - Change in upstream vs change in domestic protection: 0.83 across countries and 0.25 across sectors (insignificant across sectors).
    - Change in downstream vs change in domestic protection: 0.20 across countries and 0.02 across sectors (both highly insignificant).
  - This provides a source of largely exogenous variation: indirect tariff exposure was reduced largely independently of domestic trade policy.

- Concrete illustrative example (semiconductor tariffs):
  - A US tariff increase on semiconductor imports from China:
    - Domestic protection: makes US production more lucrative; supports US domestic producers.
    - Upstream tariff: industries relying on microchips face higher input prices (upstream tariff increases).
    - Diversion tariff: competitors (e.g., Taiwan Province of China or South Korea) become more competitive; their diversion tariff increases.
    - Downstream tariff: suppliers to microchip users see their downstream tariffs increase.
  - The example references a June-2018 US list proposing a 25% tariff on certain Chinese imports as context for plausibility.

### Empirical strategy
- Baseline panel regression at the country-sector-year level:
  - ln(y_{c,s,t}) = α + β1 T^{up}_{c,s,t-1} + β2 T^{down}_{c,s,t-1} + β3 T^{dom}_{c,s,t-1} + β4 T^{div}_{c,s,t-1} + γ_{c,t} + δ_{c,s} + ε_{c,s,t}
  - Dependent variables ln(y_{c,s,t}) are logs of real value added, employment, labor productivity, or total factor productivity.
  - Two types of fixed effects:
    - Country-sector fixed effects (δ_{c,s}) absorb structural and time-invariant aspects of the country-sector.
    - Country-time fixed effects (γ_{c,t}) control for time-varying macro drivers.
  - Tariff measures are lagged by one year to avoid front running.
  - Estimation uses the linear estimator for high-dimensional fixed effects developed by Correia (2017).
  - Baseline estimation generally includes 13 manufacturing sectors for up to 35 countries.
  - The fixed effects identification implies coefficients are estimated from within-country industry variation over time; aggregate or general equilibrium effects (including exchange rate effects) cannot be identified.

### Main results (baseline estimates)
- Table 1: The Effect of Tariffs on Economic Outcome Variables
  - Estimates are effects of a 1 ppt increase in the respective tariff measures. Dependent variables are expressed in natural logarithm. Tariff measures are lagged by one year. Residuals clustered at country-sector level.
  - Column layout: (1) VA; (2) Empl.; (3) L-Prod.; (4) TFP.
  - Coefficient estimates (standard errors in parentheses):
    - T upstream:
      - VA: -19.41*** (6.810)
      - Empl.: -10.24*** (3.853)
      - L-Prod.: -6.53* (3.847)
      - TFP: -11.52** (4.696)
    - T dom. protection:
      - VA: 0.02 (0.646)
      - Empl.: -0.88* (0.487)
      - L-Prod.: 0.92 (0.885)
      - TFP: 0.37 (0.383)
    - T downstream:
      - VA: -14.47** (6.027)
      - Empl.: -0.50 (3.790)
      - L-Prod.: -12.61*** (4.672)
      - TFP: -13.19** (6.394)
    - T diversion:
      - VA: 5.14* (2.792)
      - Empl.: 3.50* (1.836)
      - L-Prod.: 1.29 (2.100)
      - TFP: -2.70 (4.188)
  - Fixed effects: Country-Year FE Yes; Country-Ind. FE Yes.
  - Sample sizes and fit:
    - N: 6774 (VA), 6776 (Empl.), 6144 (L-Prod.), 4112 (TFP).
    - R2: 0.733 (VA), 0.995 (Empl.), 0.734 (L-Prod.), 0.693 (TFP).
  - Significance legend: * p<0.10 ** p<0.05 *** p<0.01.

- Key empirical takeaways:
  - Upstream tariffs: large, negative, and statistically significant effects on value added, employment, labor productivity (weakly), and TFP.
    - Example: a 1 ppt increase in upstream tariffs associated with -19.41 on VA (ln scale) and -10.24 on employment.
  - Downstream tariffs: large, negative, and statistically significant effects on value added, labor productivity, and TFP; effect on employment is not significant.
    - Example: a 1 ppt increase in downstream tariffs associated with -14.47 on VA and -12.61 on labor productivity.
  - Domestic protection: generally small coefficients; only employment shows a weakly significant negative coefficient (-0.88*).
  - Diversion tariffs: positive and weakly significant coefficients for value added (5.14*) and employment (3.50*); effects on labor productivity and TFP are not significant.

### Interpretation and implications
- Indirect tariffs (upstream and downstream), which capture cumulative tariff exposure through global value chains and are scaled by input or export shares, exert larger negative effects on sectoral economic outcomes than the direct domestic protection measure.
- Diversion tariffs, reflecting protection of competitors, can confer relative demand advantages and are associated with positive effects on value added and employment in the home sector.
- The identification strategy—using within-country, across-sector variation and country-time fixed effects—helps isolate tariff effects from macroeconomic confounders but does not capture general equilibrium effects (including exchange rate adjustments).
- Historical declines in indirect tariffs across countries and sectors provide plausibly exogenous variation that contributes to identification and suggests indirect tariff exposure moved independently of domestic trade policy.

*IMF Working Paper: The Effect of Tariffs in Global Value Chains (extract).*

### 3.1 Baseline Estimates

### 3.1 Baseline Estimates

### Effects on Real Value Added
- Dependent variables: real value added, employment, labor productivity, total factor productivity (푦
௖,௦,௧).
- Column 1 of Table 1: a one percentage-point increase in tariff measures.
- Upstream and Downstream tariffs: negative and significant coefficients on value added (tariffs higher up as well as further down in the value chain are drags on activity).
- Diversion tariff: higher diversion tariff (lower tariff barriers compared to competitors) tends to support value added.
- Domestic protection: no effect on value added.

### Employment
- Employment suffers under higher upstream tariffs and benefits from higher diversion tariffs (column 2).
- Coefficients for employment are slightly smaller than for value added, consistent with employment often not changing one-for-one with value added.
- Employment seems unaffected by downstream tariffs.
- Domestic protection: coefficients on domestic protection are negative and borderline statistically significant for employment.
- Variation and identification concerns:
  - Country-sector fixed effects alone explain 96% of the variation for employment.
  - Addition of country-year fixed effects brings R2 to 0.9945.
  - This implies the four tariff coefficients are identified on only 0.05% of the variation (the difference with the R2 in column (2)).
  - Cross-sectoral variation (the type of variation estimation relies upon) is significantly lower for employment than for value added or productivity.

### Labor Productivity and Total Factor Productivity (TFP)
- Labor productivity most strongly affected by upstream and downstream tariffs: coefficients negative and statistically significant (column 3).
  - Upstream tariff point estimate somewhat smaller than for value added.
  - Downstream tariff point estimate very similar in size to value added effect (consistent with negligible effect on employment).
- Domestic protection and diversion tariffs: effects far from being statistically different from zero for labor productivity.
- TFP results follow labor productivity results closely.

### Consistency with Literature and Economic Intuition
- Findings align with many papers that upstream/input tariffs are more relevant for productivity than domestic protection/output tariffs (citations in source: Amiti and Konings (2007), Topolova & Khandelwal (2011), Ahn et al. (2018)).
  - Ahn et al. (2018) finds output tariffs statistically insignificant in almost all specifications, consistent with this study.
- Empirical evidence on tariffs' effects on value added and employment is limited and diverse; this study is the first, to the authors' knowledge, to empirically study effects on value added in a cross-country setup.
- Micro studies report negative, neutral, or positive labor market effects depending on context and labor market rigidities (references in source).
- Comparable results:
  - Barattieri and Cacciatore (2020): temporary trade barriers have insignificant effects on protected sector but hurt industries further down the value chain—consistent with insignificant effects of domestic protection and negative effects of upstream tariffs.
  - Feenstra and Sasahara (2018): partly disagree; suggest tariffs can increase employment in the protected sector—results not fully consistent with this study or Barattieri and Cacciatore (2020).
- Trade diversion:
  - Few empirical studies investigate trade diversion explicitly.
  - Inclusion of multilateral tariff in Feenstra and Sasahara (2018) produces results consistent with this study's estimated employment effects of diversion tariffs.
  - Studies of trade diversion in free-trade agreement contexts (Magee (2004), Carter and Steinbach (2018), Mattoo, Mulabdic and Ruta (2017)) find some but generally weak trade diversion—broadly consistent with this study's results for value added.

---

### 3.2 The Magnitude of the Effects

### Standardization Approach (One-Standard-Deviation Effects)
- Problem: tariff measures differ considerably in size, complicating direct comparisons of one percentage-point effects.
- First approach: standardize all tariff measures so coefficients show effect of a one-standard deviation change.
- Table 2 findings (effect of a 1 standard-deviation increase):
  - For Value Added (VA), Employment (Empl.), Labor Productivity (L-Prod.), Total Factor Productivity (TFP):
    - T upstream: -16.18*** (VA), -8.55*** (Empl.), -5.45* (L-Prod.), -9.60** (TFP). SEs: (5.68), (3.21), (3.21), (3.91).
    - T dom. protection: 0.10 (VA), -5.17* (Empl.), 5.42 (L-Prod.), 2.17 (TFP). SEs: (3.80), (2.86), (5.20), (2.25).
    - T downstream: -11.93** (VA), -0.41 (Empl.), -10.40*** (L-Prod.), -10.87** (TFP). SEs: (4.97), (3.12), (3.85), (5.27).
    - T diversion: 10.98* (VA), 7.45* (Empl.), 2.75 (L-Prod.), -5.76 (TFP). SEs: (5.96), (3.92), (4.48), (8.94).
  - Country-Year FE: Yes for all columns.
  - Country-Ind. FE: Yes for all columns.
  - N: 6774 (VA), 6768 (Empl.), 6144 (L-Prod.), 4112 (TFP).
  - R2: 0.733 (VA), 0.995 (Empl.), 0.734 (L-Prod.), 0.693 (TFP).
  - Note: Dependent variables are expressed in natural logarithm. Tariff measures are lagged by one year. Residuals clustered at country-sector level. SE in parentheses; * p<0.10 ** p<0.05 *** p<0.01.
- Relative changes: coefficients of upstream and downstream tariffs decline somewhat relative to Table 1; coefficients of domestic protection and diversion tariff increase.
- Interpretation example: for value added, one-standard deviation changes in upstream or downstream tariffs have roughly comparable effects to trading partners reducing tariffs imposed on competitors by the same extent.

### Simulation Approach (Interconnected Tariff Changes)
- Rationale: tariff increases by one country can affect multiple tariff measures for many countries to differing degrees; standardization ignores these interconnections.
- Use simulations to illustrate how a given tariff change likely affects respective tariffs and economic prospects of concerned countries.
- Figure 5: simulated effect of a reciprocal uniform 1 ppt tariff increase between the US and China on value added of US, China and main trading partners.
  - Effects most negative for China and the US themselves, driven by significant changes in upstream and downstream tariffs and large coefficients associated with them.
  - Domestic protection increases but barely shows up due to coefficient very close to zero.
  - Other countries generally benefit from significant trade diversion, but downstream tariffs—and to a lesser extent upstream tariffs—partly offset gains in value added for countries such as Korea, Japan and Canada.

### Comparison with General-Equilibrium Models
- Estimated effects here are significantly larger than those from theoretical general-equilibrium models.
  - Example: reciprocal and uniform increase in bilateral tariff by 25 percentage points between the US and China would reduce Chinese value added by roughly 5½ % in the long run and US value added by roughly 3% according to the coefficients.
  - Model simulations in IMF (2019) suggest likely effects on GDP may be only a fraction of this.
- Possible reasons for larger estimated effects:
  - General equilibrium effects may be absorbed by country-year fixed effects in this framework (e.g., exchange rate changes, factor cost changes that attenuate macro impact of tariffs).
  - Sample limited to manufacturing sectors, which are likely particularly sensitive to tariffs.

### Robustness (overview leading into Section 4)
- Baseline estimates shown in Table 1 are supplemented by robustness checks:
  - Standardization of tariffs (Table 2).
  - Simulations for interconnected tariff changes (Figure 5).
  - Robustness to alternative weights for tariff aggregation (1995 and 2010 weights compared in Table 4).
  - Sensitivity to lag order (Table 5): effects decline with increasing lag length but conclusions broadly unchanged; some marginally significant coefficients become indistinguishable from zero with longer lags.

### Selected Robustness Table Highlights
- Table 3 (Individual vs Jointly Estimated Effects on Value Added):
  - Jointly estimated T upstream: -19.41*** (SE 6.810).
  - Individually estimated T upstream (column 2): -22.30*** (SE 7.464).
  - T dom. protection jointly: 0.02 (SE 0.646); individually reported -0.58 (SE 0.699) in one specification.
  - T downstream jointly: -14.47** (SE 6.027); individually -14.20** (SE 6.724).
  - T diversion jointly: 5.14* (SE 2.792); individually 2.33 (SE 2.171).
  - Country-Year FE and Country-Ind. FE: Yes for all columns. N = 6774. R2: 0.733 (joint), 0.730, 0.722, 0.725, 0.722.
  - Note: Effects reported are for a 1 ppt increase in respective tariff measures; dependent variable is natural logarithm of value added. Tariff measures lagged by one year. Residuals clustered at country-sector level. SE in parentheses; * p<0.10 ** p<0.05 *** p<0.01.
- Table 4 (Effect on Value Added with Alternative Weights):
  - Weights based on 1995-2010 (column 1), 1995 (column 2), 2010 (column 3).
  - T upstream: -19.41*** (6.81), -20.63*** (7.11), -17.18** (8.01).
  - T dom. protection: 0.02 (0.65), 0.31 (0.48), 0.06 (0.67).
  - T downstream: -14.47** (6.03), -15.84** (7.53), -11.99*** (4.63).
  - T diversion: 5.14* (2.79), 2.19 (2.67), 4.46** (2.01).
  - Country-Year FE and Country-Ind. FE: Yes for all columns. N = 6774. R2: 0.733, 0.733, 0.731.
  - Note: Effects of a 1 ppt increase in respective tariffs when 1995 and 2010 weights used. Dependent variable is natural logarithm of value added. Tariff measures lagged by one year. Residuals clustered at country-sector level. SE in parentheses; * p<0.10 ** p<0.05 *** p<0.01.

*IMF Working Paper — The Effect of Tariffs in Global Value Chains (section 3.1 Baseline Estimates and accompanying material provided).*

### 5.  Conclusion

### 5.  Conclusion

### Main findings
- Tariffs have wide-ranging effects in global value chains, affecting countries and sectors not directly targeted.
- Negative effects on economic outcomes are economically and statistically significant from tariffs imposed higher up as well as further down in the value chain.
- Positive effects can arise from tariffs imposed on competitors (trade diversion effects).
- Domestic protection generally does not improve economic prospects in the protectionist country because:
  - Effects on the protected sector are negligible or even negative.
  - The home country tends to suffer most from more expensive intermediate inputs.
- Bilateral tariff wars hurt the two parties involved most; third countries can benefit on net from trade diversion.
- Even unilateral tariffs often do not benefit the imposing country because imports from the targeted country are likely replaced by imports from elsewhere.
- Empirical results align with recent theoretical gravity models emphasizing trade in value added and with higher-frequency studies of recent US-China trade tensions, which show negative effects from tariffs encountered further down the value chain outweighing potential domestic protection gains.

### Key statistics (Table 5: Effect on Value Added with Increasing Lag Length)
- Effect of a 1 ppt increase in the respective tariff measures (dependent variable: value added in natural logarithm). Residuals clustered at the country-sector level. SE in parentheses; * p<0.10 ** p<0.05 *** p<0.01.
- Contemp.:
  - LX. T upstream: -20.85*** (7.04)
  - LX. T dom. protection: 0.05 (0.66)
  - LX. T downstream: -15.37** (6.37)
  - LX. T diversion: 5.87** (2.81)
  - Country-Year FE: Yes; Country-Ind. FE: Yes; N: 7187; R2: 0.744
- L1:
  - LX. T upstream: -19.41*** (6.81)
  - LX. T dom. protection: 0.02 (0.65)
  - LX. T downstream: -14.47** (6.03)
  - LX. T diversion: 5.14* (2.79)
  - Country-Year FE: Yes; Country-Ind. FE: Yes; N: 6774; R2: 0.733
- L2:
  - LX. T upstream: -18.53*** (6.62)
  - LX. T dom. protection: -0.02 (0.64)
  - LX. T downstream: -12.75** (5.64)
  - LX. T diversion: 4.37 (2.68)
  - Country-Year FE: Yes; Country-Ind. FE: Yes; N: 6361; R2: 0.719
- L3:
  - LX. T upstream: -17.18*** (6.18)
  - LX. T dom. protection: -0.00 (0.58)
  - LX. T downstream: -11.43** (5.25)
  - LX. T diversion: 3.73 (2.54)
  - Country-Year FE: Yes; Country-Ind. FE: Yes; N: 5948; R2: 0.709

### Contribution to the literature
- Constructs four tariff measures using global input-output matrices to capture richer spillovers than prior empirical work.
- First cross-country empirical investigation of downstream tariffs and tariff-induced diversion of value added.
- Complements theoretical gravity models that focus on sectoral trade linkages and trade in value added (examples cited in the source).

### Limitations and open questions
- Focus on tariffs likely underestimates overall effects of trade tensions because non-tariff barriers and increased uncertainty have risen in tandem; the paper cannot support or refute whether effects generalize to non-tariff barriers.
- Historical tariff data limitations:
  - Tariffs historically declined rather gradually; unclear if tariff increases mirror declines or if asymmetries exist.
  - Unclear whether effects are linear (proportional to tariff changes).
  - Global value chains adapt over time, potentially causing non-negligible consequences not fully captured.
- These questions are left for further research.

### Robustness and dynamics
- Effects tend to decline as lag length is increased (tables in the appendix compare contemporaneous and lagged effects across value added, employment, labor productivity, and total factor productivity).
- Additional appendix results show individual versus jointly estimated tariff effects and alternative weighting schemes (1995-2010 averages, 1995, 2010) broadly robust the main findings.

*IMF Working Paper No. WP/22/40 — "The Effect of Tariffs in Global Value Chains" (Conclusion section).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022040-print-pdf.pdf_