## Trade Diversion Effects from Global Tensions—Higher Than We Think (wpiea2023234-print-pdf) ## Source details **Canonical URL:** [Trade Diversion Effects from Global Tensions—Higher Than We Think (wpiea2023234-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023234-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023234-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023234-print-pdf.pdf.json) --- ### Purpose and focus - Uses Mexico as an example to shed light on the likely impact of global trade tensions for countries with high trade exposure and supply linkages. - Studies two episodes, with a focus on the first: - (i) the 2018 U.S.-China trade tensions, - (ii) the 2014 U.S. sanctions on Russia. ### Contributions - Builds a unique industry-level dataset for Mexico exploiting input-output and supply chain linkages to quantify channels through which input-output linkages may play a role during global tensions. - Coverage: 258 industries (NAICS 4-digit) versus common cross-country sources coverage of 56 industries. - First attempt to capture both direct and indirect industry-level exposure across a large number of industries and to explore determinants of variation in trade diversion effects across Mexico’s industries. - Compares nationally sourced input-output tables (INEGI) with cross-country source (WIOD) and finds national tables show higher trade diversion effects. ### Data and methodology overview - Combined datasets: - (i) INEGI annual input-output table: 258 industries, 2003–18 (production structure); input-output table year used: 2015. - (ii) granular trade data (HS-6 digit) from UN Comtrade (annual and/or monthly). - (iii) WIOD (43 countries; 56 industries at ISIC 2-digit). - Empirical identification: - Difference-in-differences exploiting cross-industry variation in tariff exposure and timing (three tariff rounds: July 6, August 23, September 24 of 2018). - Treatment measures capture output (direct), upstream, downstream, total, and net tariffs constructed via Leontief-type inverse matrices and U.S. demand shares s_{k,US→MX} computed as 100 * (value of U.S. imports from Mexico in industry k in 2017) / (total U.S. imports from Mexico in 2017). - Regression frameworks: - Industry-month panel (2016/01–2019/05), baseline dynamic DiD: Y_{j,t} = α + β T_j · Post_{j,t} + ρ Y_{j,t−1} + η_j + ξ_t + ε_{j,t}. - Continuous treatment: Y_{j,t} = α + β Δτ_j · Post_{j,t} + ρ Y_{j,t−1} + η_j + ξ_t + ε_{j,t}. - Three-tariff decomposition: Y_{j,t} = α + β_1 Δτ^{out}_j · Post_{j,t} + β_2 Δτ^{down}_j · Post_{j,t} + β_3 Δτ^{up}_j · Post_{j,t} + ρ Y_{j,t−1} + η_j + ξ_t + ε_{j,t}. - Dynamic panel Arellano-Bond estimator used to address lagged dependent variable endogeneity; event-study (monthly β_τ) used to test pre-trends. ### Tariff exposure measures (definitions and construction) - Output tariff (direct): weighted average of reported U.S. import tariffs on Chinese HS 6-digit items aggregated to industry j by import value shares. - Upstream tariff τ_{j,up}: weighted average of tariffs on upstream industries k relative to j with weights containing d̃_{k→j} (Leontief inverse element) and s_{k,US→MX}. - Downstream tariff τ_{j,down}: weighted average of tariffs on downstream industries k relative to j with weights d̃_{j→k} and s_{k,US→MX}. - Total tariff: output + upstream + downstream. - Net tariff: output + downstream − upstream. - Continuous changes: - Δτ_j = Δτ_j^{output} + Δτ_j^{upstream} + Δτ_j^{downstream}. - Net change Δτ_j^{net} = Δτ_j^{out} − Δτ_j^{up} + Δτ_j^{down}. - Dummy constructions: - T^{out}_j = 1(Δτ^{out}_j ≥ 0), T^{up}_j = 1(Δτ^{up}_j ≥ 0), T^{down}_j = 1(Δτ^{down}_j ≥ 0), T_j = 1(Δτ^{total}_j ≥ 0), T^{net}_j = 1(Δτ^{net}_j ≥ 0). - Additional cutoff: T_j = 1 if Δτ_j ≥ median. ### Key empirical findings — aggregate effects - Overall positive trade diversion effect on Mexico’s exports to the U.S. during the 2018 U.S.-China tensions. - Main quantitative results: - A one-standard deviation increase in net tariff change (5.8 percentage points) on Chinese products increases Mexican exports to the U.S. by 6.4 percent (from continuous treatment, column 6, Table 3). - Output tariff one-standard deviation (5.6 percentage points) → exports to the U.S. up by 7.2 percentage points (column 1, Table 3). - Total tariff one-standard deviation (7.74) → exports increase by 6.2 percent (column 5, Table 3). - Dummy-based results (Arellano-Bond dynamic panel, dependent variable log imports, 258 industries × 41 months, Obs. 10,062): - Output tariff dummy: industries affected experienced on average 16.1 percent larger increase in U.S. imports after the trade tensions (Column 1, Table 2). - Net tariff dummy: industries exposed to a positive net tariff change increased their exports to the U.S. by 16.3 percent relative to industries with no net tariff increase (Column 6, Table 2). - Total tariff dummy: industries exposed to total tariff change had an 8.5 percent larger increase in exports (Column 5, Table 2). - Time pattern: - Event study shows β_t close to zero pre-2018 and positive after announcement; evidence of an anticipation effect with statistically significant diversion before formal implementation. ### Channels: output, upstream, downstream - Output (direct) tariffs play the most important and consistent role in driving positive trade diversion effects. - Downstream tariffs show positive effects in many specifications but often statistically weaker when controlling jointly. - Upstream tariff effects are ambiguous: - Individual upstream coefficients often not statistically significant and can be mildly negative in joint specifications. - Conceptually, upstream channel can exert negative effects (limited short-run input availability for other domestic industries) or positive effects (increased domestic production of inputs to meet U.S. demand expands availability); aggregate sign depends on production response. - Wald tests: output, upstream, and downstream variables are jointly significant (reject null at 1 percent in joint specification Column 4, Table 2). ### Industry-level heterogeneity and correlates - Distributional facts: - Most industries’ exports to the U.S. in Mexico grew after trade tensions; average growth rate across industries: 6.25 percent. - Industry-level estimated trade diversion effects (U.S. import level, 258 obs.): Mean 0.0625, S.D. 0.4327, Min -1.5549, Max 5.6514. - Non-zero estimates (97 obs.): Mean 0.1663, S.D. 0.6955, Min -1.5549, Max 5.6514. - More than 75 percent of industries’ exports to the U.S. expanded during the trade tension period after controlling for macro variables. - Strong correlates (Table 5 correlations): - Change in U.S. imports from China: correlation coefficient -0.7181, p-value 0.0000. - Net tariff change: correlation coefficient 0.2500, p-value 0.0196. - Output tariff change: correlation coefficient 0.2742, p-value 0.0102. - Product substitutability (σ): correlation coefficient 0.2320, p-value 0.0002. - Weak or no correlation: - Export share to the U.S. in 2017: correlation 0.0149, p-value 0.8716. - Imported input value share in production (2016): correlation 0.0655, p-value 0.2956. - Export share in sales (2016): correlation 0.0624, p-value 0.3182. - GVC integration measures: positive but insignificant correlations (~0.06). - Interpretation: - Industry-level trade diversion magnitudes are most strongly associated with declines in U.S. imports from China, output and net tariff changes, and product substitutability with Chinese products. - Pre-existing trade exposure of Mexican industries to the U.S. market does not explain the positive diversion. ### Continuous-treatment magnitudes (selected coefficients and magnitudes preserved) - Table 3 reported coefficients (standard errors in parentheses): - Output tariff: 0.013*** (0.005) in column (1); 0.015** (0.007) in column (2). - Upstream tariff: 0.023 (0.015) in column (2); -0.034 (0.038) when with other tariffs. - Downstream tariff: 0.021 (0.014) in column (3); 0.009 (0.038) when joint. - Total tariff: 0.008*** (0.002) in column (5). - Net tariff: 0.011*** (0.002) in column (6). - Lagged ln(usimports): 0.158** (0.074) to 0.160** (0.074). - Magnitude interpretations: - Output one-standard deviation (5.6 percentage points) → exports to the U.S. ↑ 7.2 percentage points (column 1). - Upstream one-standard deviation (1.73) → exports ↑ 4.0 percent (insignificant, column 2). - Downstream one-standard deviation (2.24) → exports ↑ 4.7 percent (insignificant). - Total one-standard deviation (7.74) → exports ↑ 6.2 percent (column 5). - Net one-standard deviation (5.80) → exports ↑ 6.4 percent (column 6). ### Robustness and alternative specifications - Static panel with industry-clustered standard errors: - One standard deviation increase in total tariffs → Mexican exports to the U.S. of that industry go up by 4.6 percent after trade tensions (significant). - Alternative dependent variable (monthly growth rate): - One standard deviation increase in total tariffs → growth rate of Mexican exports to the U.S. goes up by 0.04 percentage points after trade tensions (significant). - Event-study monthly β_τ used to verify parallel trends; results show no pre-trend and positive post-announcement effects. - Multiple robustness checks reported in Appendix III (Tables A1–A4). ### Comparison: INEGI (national) vs WIOD (cross-country) - Granularity and magnitude differences: - INEGI dataset: 258 industries; WIOD matched dataset: 56 industries. - If an industry experienced a one standard deviation increase in net tariff change, exports to the U.S. increase by: - 6.4 percent using INEGI, - 1.4 percent using WIOD. - Table 7 examples (one S.D. increase results): Output: 7.2*** INEGI vs. 0.7 WIOD; Total: 6.2*** INEGI vs. 1.6 WIOD. - Industry-level means: - WIOD (56 obs.): industry-level mean trade diversion effect 4.24 percent; non-zeros (25 obs.) mean 9.49 percent. - INEGI (258 obs.): industry-level mean 6.25 percent; non-zeros (97 obs.) mean 16.63 percent. - Interpretation: qualitatively similar patterns but quantitatively larger estimated trade diversion effects using nationally sourced INEGI input-output linkages. ### 2014 U.S. sanctions on Russia — alternative episode evidence - Data: product-level U.S. import flows at HS 6-digit, monthly 2012/01–2016/12; product-level U.S. tariffs controlled. - Stylized aggregate facts (2012–16, Table 10): - Mexico: - Pre-sanction U.S. imports (billion $): 23.89; percent in 2012 GDP: 1.99% - Post-sanction U.S. imports (billion $): 24.74; percent in 2012 GDP: 2.06% - Change: 0.85 billion $, 0.07% - Russian Federation: - Pre-sanction U.S. imports (billion $): 2.29; percent in 2012 GDP: 0.10% - Post-sanction U.S. imports (billion $): 1.36; percent in 2012 GDP: 0.06% - Change: -0.93 billion $, -0.04% - Dynamic regressions and local projections: - Estimated time-dummy coefficients increase in months following March 2014 and December 2014. - Local projection impulse response: nominal exports from Mexico increase by about 10 percent four months after the sanction (horizon k = 4). ### Stylized aggregate and sectoral facts on Mexico (Appendix II) - Export and import shares in production (2003–18): - Export share increased from 14 percent in 2003 to 22 percent in 2018. - Import share in production increased from 26 percent in 2003 to 42 percent in 2018. - U.S. share in Mexico’s trade (2003 → 2020): - U.S. share in Mexico’s exports: 88 percent in 2003 → 79 percent in 2020. - U.S. share in Mexico’s imports: 62 percent in 2003 → 44 percent in 2020. - Top trading partners (2020): exports — U.S., Canada, China, Germany, Japan; imports — U.S., China, South Korea, Japan, Germany. - Sectoral notes (2018): - Motor vehicle manufacturing and related industries together account for 32 percent of exports in 2018. - Semiconductor and other electronic component manufacturing and computer and peripheral equipment manufacturing together comprise 11 percent of exports. - Semiconductor industry is the largest importing industry (about 9 percent of total imports in 2018). ### Policy-relevant messages and implications - Differential short-term impacts: - Some countries and industries can be short-term beneficiaries of policy-driven geoeconomic fragmentation (GEF) via trade diversion, but higher tariffs and geopolitical tensions reduce global welfare. - Importance of input-output linkages: - Proper accounting of supply linkages matters; including upstream and downstream channels changes estimated trade diversion magnitudes. - Data source matters: - Using nationally sourced richer input-output tables (INEGI) yields larger estimated trade diversion effects than cross-country sources (WIOD); policy analysis should prefer country-sourced IO data where available. - Determinants for winners/losers: - Industries with higher product substitutability with China and those facing larger declines in U.S. imports from China tend to gain more. - Pre-existing exposure to the U.S. market is not a strong predictor of gaining from diversion. *IMF Working Paper — "Trade Diversion Effects from Global Tensions—Higher Than We Think", wpiea2023234-print-pdf.* ### 1. Introduction ........................................................................................................ ### 1. Introduction ### Purpose and focus - Uses Mexico as an example to shed light on the likely impact of global trade tensions for countries with high trade exposure and supply linkages. - Studies two episodes, with a focus on the first: - (i) the 2018 U.S.-China trade tensions, - (ii) the 2014 U.S. sanctions on Russia. ### Contributions (threefold) - Builds a unique industry-level dataset for Mexico exploiting input-output and supply chain linkages to quantify channels through which input-output linkages may play a role during global tensions. - Nationally sourced input-output tables show higher trade diversion effects than results from typically used cross-country sources (e.g., WIOD). - Coverage: 258 industries versus cross-country coverage of 56 industries. - Uses difference-in-differences method and input/output tariff literature to estimate the impact on Mexico’s exports during U.S.-China trade tensions. - First attempt to capture both direct and indirect industry-level exposure across a large number of industries. - First attempt to explore determinants of variation in trade diversion effects across Mexico’s industries. - Examines the impact on a third country from sanctions by studying the 2014 U.S. sanctions on Russia and their effect on Mexico. ### Data and methodology overview - Combines three datasets: - (i) input-output table of the Mexican economy from INEGI, - (ii) granular trade data (HS-6 digit) from UN Comtrade, - (iii) cross-country input-output table from World Input-Output Database (WIOD). - Allows: - granular input-output linkages matched with trade data, - detailed information on sources of imported inputs for specific industries, - comprehensive usage of each industry's products in all industries across countries. - Empirical strategy: - Estimates impact on Mexico’s exports to the U.S. from the first three rounds of U.S. tariffs on China imposed on July 6, August 23, and September 24 of 2018. - Considers direct trade diversion effect from output tariffs and indirect effects through upstream and downstream tariffs. - Uses difference-in-differences exploiting variation of tariff exposure across industries: - Cross-section dimension: industries with higher U.S. import tariff imposed on Chinese products versus those less affected. - Time dimension: before and after the trade tensions. ### Key empirical findings (high-level) - Overall positive trade diversion effect on Mexico’s exports to the U.S. - A one-standard deviation increase in net tariff change (5.8 percentage points) on Chinese products increases Mexican exports to the U.S. by 6.4 percent. - Channels: - Output tariffs play a more important role. - Some evidence of a positive impact through downstream tariffs. - Industry-level heterogeneity: - Variation in trade diversion effects is related to: - size of changes of U.S. tariffs on Chinese products, - decrease in U.S. imports from China, - degree of substitutability of Mexico’s products vis-à-vis China. - Some weak evidence that higher global value chain (GVC) integrated industries benefited more. - Industry-level trade diversion effect does not vary according to Mexico’s trade exposure to the U.S. - Event study on 2014 U.S. sanctions on Russia: - Using dynamic regressions and local projection methods with monthly U.S. imports from Mexico at HS 6-digit level. - Finds a positive impact of U.S. sanctions on Mexico’s exports to the U.S., with the size of about 10 percent increase four months after sanctions. ### Data specifics (summary) - INEGI annual input-output table: production structure information of 258 industries at NAICS 4-digit level, over the period 2003–18. - UN Comtrade: granular HS 6-digit annual and/or monthly exports and imports between Mexico and the U.S., and between Mexico and other countries. - WIOD: input-output linkage information from 43 countries, covering 56 industries at ISIC 2-digit level. - Base of constructed database: INEGI dataset (258 industries) for granularity and detailed input-output structure. - Matching procedures: - UN Comtrade (HS 6-digit) matched to NAICS 4-digit industry code following Pierce and Schott (2009). - WIOD (ISIC 2-digit) mapped to NAICS 4-digit using concordance table between 2017 NAICS and ISIC Rev. 4; 215 cases where one NAICS 4-digit code matches multiple ISIC 2-digit codes, summed in those cases. - Additional insights (not used in the core analysis but available from dataset): - granular information on sources of imported inputs for specific industries; - detailed usage of each industry’s product in other industries and countries. - Stylized features highlighted in Appendix II: - Mexican industries rely increasingly on international markets, especially the U.S., for product exporting and input sourcing; - China’s role as a trade partner has been increasing; - Motor vehicle manufacturing is the largest exporting industry in Mexico; - Semiconductors industry is the largest importing industry. ### Paper structure (forward map) - Section 2: dataset. - Section 3: literature review on trade diversion. - Section 4: empirical methodologies. - Section 5: overall trade diversion effects from the 2018 U.S.-China trade tensions. - Section 6: industry-level results from the 2018 U.S.-China trade tensions. - Section 7: comparison of results using nationally sourced data versus cross-country sources. - Section 8: results from the 2014 U.S. sanctions on Russia. - Section 9: conclusion. *IMF Working Papers — "Trade Diversion Effects from Global Tensions—Higher Than We Think", Section 1. Introduction.* ### 3. Literature Review on Trade Diversion and ### 3. Literature Review on Trade Diversion and ### Traditional trade theory and mechanisms - Free trade agreements (FTA) traditionally lead to trade creation through tariff reduction and to trade diversion when member countries replace higher-tariff non-member suppliers with lower-tariff member suppliers. - Trade diversion can also arise from tariff increases between two countries (the opposite of FTA creation). Given the tariff increase on Chinese products during 2018 and early 2019, the U.S. would import less from China and import more from other countries (trade diversion). - Higher trade policy uncertainty can cause trade diversion by reducing trade volumes; the literature documents negative effects of trade policy uncertainty on trade volume (Handley, 2014; Handley and Limao, 2015; Alessandria et al., 2019). - During the 2018 trade tensions, trade policy uncertainty rose to a historically high level (IMF, 2018; Benguria et al., 2022), leading to the U.S. importing less from China and more from a third country, for example Mexico. ### Empirical literature on U.S.-China tariffs and trade diversion - Cigna et al. (2020) - Data: monthly product-level U.S. imports from its top 30 trade partners during 2016/01–2019/05. - Method: difference-in-differences with a time dummy for trade tensions and a product-level treatment dummy for tariff exposure. - Findings: significant decline in U.S. imports from China for tariff-exposed products relative to non-exposed products; no significant short-run trade diversion to third countries (U.S. imports of exposed products from third countries did not increase more than non-exposed products). - Deng (2021) - Data: monthly trade flows during 2014–19 between China and its top 10 trade partners besides the U.S. - Method: country-month level difference-in-differences comparing China–U.S. trade with China–other countries. - Findings: China’s imports were affected more than its exports to the U.S.; magnitude of the decrease in China’s imports from the U.S. is greater than the decrease in China’s exports to the U.S., after controlling for time trend. Trade diversion effects in China’s imports are more pronounced when the third country is a developing country rather than a developed country. - Lovely et al. (2021) - Focus: trade diversion effect of U.S.-China trade tensions on Mexican exports. - Finding: Mexican sales in the U.S. market rose by 3.4 percent on average, with heterogeneity across sectors in Mexico. - Estimation approach: estimate how much trade diversion to an exporting country and an exporting sector depends on the preexisting share of this sector and country in the U.S. market and the size of the tariff change imposed on Chinese products. - Key quantitative result: if the U.S. increases tariffs on one Chinese product by 10 percentage points, a country with a preexisting 10 percent share of the U.S. market for that product would expect a 0.46 percentage point increase in the value of its exports of that product to the U.S. - Approach: panel data of monthly U.S. imports at HS10 product level from the universe of countries; apply estimated coefficients to obtain an in-sample prediction for Mexican exports. - Supporting literature and related findings: - Fajgelbaum et al. (2022) find an increase in Mexico’s exports to the U.S. after the trade tensions, showing a substitution relationship between China’s exports and Mexico’s exports in the U.S. market. - Chiquiar et al. (2007) and Chiquiar and Tobal (2019) find a similar substitution relationship following China’s accession to the WTO. - Conconi et al. (2018) find that rules of origin in NAFTA led to trade diversion of intermediate goods from third countries to NAFTA partners through input-output linkages. ### Contribution of this paper relative to prior work - Methodological difference: this paper uses a cross-industry estimation using only Mexican exports to the U.S. market, exploiting variation in exposure to tariff increases across industries and emphasizing industrial heterogeneity. - Novelty: tariff exposure is measured in forms of both output and input (upstream and downstream) tariffs, accounting for input-output linkages. - Two significant contributions previewed: - (i) Trade diversion effects on Mexico during U.S.-China trade tensions are stronger than those found in Lovely et al. (2021) when input-output linkages are accounted for. - (ii) The effect is stronger when country-sourced richer input-output tables are used, as opposed to a common cross-country source database with limited information than country-sourced data. ### Empirical strategy (overview and identification) - Identification: difference-in-differences methods exploit variation in tariff exposure across industries. If the U.S. imposes a higher import tariff on Chinese products in one industry, Mexico would export more products of this industry to the U.S. relative to another Mexican industry where U.S. tariffs on Chinese products were lesser. - Tariff exposure captures both direct (output) and indirect (upstream, downstream) channels. - Estimation uses an industry-month panel data set to identify average trade diversion effects. - Alternative exercises: document differentiated trade diversion effects across industries and explore correlation between trade diversion magnitude and characteristics such as tariff exposure, trade structure, product substitutability, and GVC integration. - Parallel trends assumption: required for difference-in-differences. The paper tests this by checking if estimated monthly trade diversion effects are close to zero before the announcement of the tariff increase (section 5.2). - Caveats and robustness: Chaisemartin et al. (2022) discuss potential biases and alternative estimators (e.g., Gardner (2022) method). Gardner’s method is not applicable here because most industries were treated after three months, limiting the time period to estimate treatment effects with that method. The paper conducts numerous robustness checks and exercises to support results. ### Construction of the tariff exposure measure - Tariff exposure in each industry j is measured by a continuous variable ∆τ_j which is the size of the tariff increase in industry j in 2018. - Calculation inputs: - Tariffs on products at HS 6-digit level. - Input-output linkage from INEGI to capture output, upstream, and downstream tariffs. - Definitions: - Upstream tariff: incorporates the indirect effect of tariff increases on a product through tariffs imposed on inputs of this product via a limited/enlarged input availability channel. - Downstream tariff: considers the indirect effect on a product due to its increased demand as inputs in other products. - Formula (as presented): - ∆τ_j = ∆τ_j^output + ∆τ_j^upstream + ∆τ_j^downstream - ∆τ_j = (τ_{j,post}^{output} − τ_{j,pre}^{output}) + (τ_{j,post}^{upstream} − τ_{j,pre}^{upstream}) + (τ_{j,post}^{downstream} − τ_{j,pre}^{downstream}) - Output tariff τ_{j,post}^{output} is the direct U.S. import tariff imposed on Chinese products in industry j after the tariff change during the trade tensions. - τ_{j,pre}^{output} is the tariff in 2017, chosen as the MFN tariff rates on January 1, 2017. *IMF WORKING PAPERS — Trade Diversion Effects from Global Tensions—Higher Than We Think* ### 2018. In other words, output or direct tariffs are essentially what is reported as the tariff of a particular ### wpiea2023234-print-pdf - 2018. In other words, output or direct tariffs are essentially what is reported as the tariff of a particular ### Tariff measures: output, upstream, downstream, total, net - Output (direct) tariff: the reported tariff of a particular HS 6-digit item/product aggregated to industry j by a weighted average, with weight being the import value share of the product among all products in industry j. - Upstream tariff (τ_{j,up}): captures the indirect impact of U.S.-China tariffs on Mexican products via the input availability channel. It aggregates tariffs on all upstream industries k relative to industry j as a weighted average where the weight contains: - the input-output linkage d~_{k→j} (total input value share of domestic products in industry k used in production of industry j, obtained after applying the Leontief-type inverse matrix on the input-output table from INEGI), and - U.S. demand s_{k,US→MX} (the value of U.S. imports from Mexico in industry k in 2017 divided by the total imports of the U.S. from Mexico in 2017). - Downstream tariff (τ_{j,down}): captures the indirect impact via indirect input demand (demand for industry j’s output as input into downstream industries). It is computed as the weighted average of tariffs on all downstream industries k relative to industry j, with weight parts: - input-output linkage d~_{j→k} (total input value share of domestic products in industry j used in production of industry k, element in the Leontief-type inverse matrix), and - U.S. demand s_{k,US→MX} (same demand concept as for upstream). - Total tariff: the sum of output, upstream, and downstream tariffs. - Net tariff: output plus downstream minus upstream tariffs. ### Key numeric and data facts preserved exactly - U.S. import value of product p and q from China in 2017 are used to construct product shares for weighting. - The U.S. demand share s_{k,US→MX} is computed as s_{k,US→MX} = 100 * (value of U.S. imports from Mexico in industry k in 2017) / (total U.S. imports from Mexico in 2017). - Input-output table year used: 2015 (to alleviate potential endogeneity in input-output structure). - Example illustrative numbers for motor vehicle parts: - Before tariff increases: U.S. imports 100 motor vehicle parts from China and 50 from Mexico. - After tariff increases: U.S. imports 50 from China and 100 from Mexico. - Trade data aggregation period used in regressions: 2016/01–2019/05. ### Matrix and Leontief-type method for indirect tariffs - The upstream and downstream tariff calculations use a Leontief-type inverse matrix to account for infinite iterations of input–output linkages (how many units of product k are required to produce one unit of product j). - Notation and matrix facts preserved: - B̃ is the N×N matrix with element (j,k) being d_{k→j} (or d_{j→k} depending on upstream/downstream context). - L̃ = (I − B̃)^{-1} is the Leontief-type inverse (or L = (I − B)^{-1} in the downstream formulation). - d̃_{k→j} (or d̃_{j→k}) used as weights are elements of the Leontief-type inverse matrix. - The matrix method solves for τ_{·,up} and τ_{·,down} by multiplying the Leontief-type inverse by the vector of (industry output tariffs × U.S. demand shares). ### Conceptual channels and countervailing effects (preserved language) - Upstream channel has two possible consequences when U.S. demand for Mexican product k increases: - Negative channel: limited short-run domestic supply of product k reduces availability of input k for industry j, lessening production in j. - Positive channel: increased production of product k to meet U.S. demand may raise availability of inputs for industry j if production surpasses U.S. demand. - Aggregate impact depends on whether domestic production of input k can grow or not. "The former negative channel dominates if production reacts. Otherwise, the latter positive channel dominates." - Downstream channel: higher U.S. demand for downstream industry k (which uses industry j’s output as input) increases indirect demand for inputs from industry j, boosting j’s production and exports. This effect is stronger when U.S. demand for product k is higher. ### Construction of treatment variables (dummy and continuous) - Dummy variables (treatment indicators): - Output tariff dummy T^{out}_j = 1(Δτ^{out}_j ≥ 0) denotes output tariff increases. - Upstream tariff dummy T^{up}_j = 1(Δτ^{up}_j ≥ 0). - Downstream tariff dummy T^{down}_j = 1(Δτ^{down}_j ≥ 0). - Total tariff dummy T_j = 1(Δτ^{total}_j ≥ 0) where Δτ^{total}_j is the sum of output, upstream, and downstream tariff changes. - Net tariff dummy T^{net}_j = 1(Δτ^{net}_j ≥ 0) where τ^{net}_j = τ^{out}_j + τ^{down}_j − τ^{up}_j. - Continuous variables (sizes of tariff changes): - Output tariff change: Δτ^{out}_j = τ^{out}_{j,after} − τ^{out}_{j,before}. - Upstream tariff change: Δτ^{up}_j = τ^{up}_{j,after} − τ^{up}_{j,before}. - Downstream tariff change: Δτ^{down}_j = τ^{down}_{j,after} − τ^{down}_{j,before}. - Total tariff change: Δτ^{total}_j = Δτ^{out}_j + Δτ^{up}_j + Δτ^{down}_j. - Net tariff change: Δτ^{net}_j = Δτ^{out}_j − Δτ^{up}_j + Δτ^{down}_j. - Additional treatment definition: T_j = 1 if Δτ_j ≥ median to separate higher-exposed industries from lower-exposed ones. ### Empirical strategy: difference-in-differences (DiD) and specifications - Identification: variation in tariff exposure across industries, combined with before/after timing of tariff increases, used in a difference-in-differences framework (industry × time). - Baseline DiD regression (level specification with dynamics): - Dependent variable Y_{j,t}: U.S. imports from Mexico (in logarithms) in sector j in month t, covering 2016/01–2019/05. - Baseline regression form: Y_{j,t} = α + β T_j · Post_{j,t} + ρ Y_{j,t−1} + η_j + ξ_t + ε_{j,t}. - Post_{j,t} is a time dummy equal to 1 after tariff increases for industry j; industries are treated in different months (three rounds of tariffs), and if affected multiple times the first month affected is picked as treatment. - T_j is the industry-level treatment dummy (defined above). - Continuous-size specification: - Y_{j,t} = α + β Δτ_j · Post_{j,t} + ρ Y_{j,t−1} + η_j + ξ_t + ε_{j,t}. - Three-tariff decomposition specification: - Y_{j,t} = α + β_1 Δτ^{out}_j · Post_{j,t} + β_2 Δτ^{down}_j · Post_{j,t} + β_3 Δτ^{up}_j · Post_{j,t} + ρ Y_{j,t−1} + η_j + ξ_t + ε_{j,t}. - Dynamic panel estimation and robustness: - Lagged dependent variable included to consider persistence; Arellano-Bond estimator (generalized method of moments) used to address endogeneity from lagged dependent variables. - Static panel models with lagged dependent variable also used as robustness and report qualitatively similar results. - Alternative dependent variable for robustness: U.S. import growth rate from Mexico in sector j in month t. - Time-varying treatment (event-study style) specification: - Y_{j,t} = α + Σ_{τ=−T}^{T} β_τ T_j · 1_{t=τ} + η_j + ξ_t + ε_{j,t}, estimating β_τ for each month to verify pre-trends and dynamic treatment effects. Expectation: β_τ = 0 before trade tensions and positive after 2018. ### Illustrative example (motor vehicle parts) - Direct (output) diversion example: - Before tariffs: U.S. imports 100 from China and 50 from Mexico. - After tariffs: U.S. imports 50 from China and 100 from Mexico. - This arbitrary-number example illustrates the direct trade diversion captured by the output tariff. - Indirect (upstream) example (rubber input): - Imposing a higher tariff on Chinese rubber (τ̄) increases Mexican rubber exports to the U.S.; this can either: - Reduce availability of domestically produced rubbers for motor vehicle parts (negative effect on motor vehicle parts production), or - Increase domestic rubber production to meet U.S. demand and thereby expand availability of inputs for motor vehicle parts (positive effect). - Upstream tariff aggregates impacts across all inputs (rubber, glass, steel, etc.) using input shares d_{k→j} and U.S. demand shares s_{k,US→MX}. - Indirect (downstream) example (motor vehicle manufacturing downstream): - Higher tariff on Chinese motor vehicle manufacturing increases U.S. demand for Mexican motor vehicles, raising demand for motor vehicle parts as inputs and boosting production and exports of motor vehicle parts. Downstream tariff aggregates across downstream industries with weights d_{j→k} and s_{k,US→MX}. *IMF Working Paper excerpt: Trade Diversion Effects from Global Tensions—Higher Than We Think (excerpt covering tariff measure definitions, Leontief-type aggregation, illustrative examples, treatment construction, and DiD estimation strategy).* ### 4.2 An Alternative Approach Exploiting Industry-Level Trade Diversion Effects ### 4.2 An Alternative Approach Exploiting Industry-Level Trade Diversion Effects ### Methodology: industry-level monthly regressions (2016/01–2019/05) - For each industry j, the paper estimates the monthly time-series regression: Y_{j,t} = α_j + β_j Post_t + ρ_j Y_{j,t−1} + γ_j X_{j,t} + ε_{j,t} - Y_{j,t} is U.S. imports from Mexico (in logarithms) in sector j in month t. - Post_t is a time dummy equal to 1 after tariffs increased. - Control variables X_{j,t} include GDP growth rate of the U.S. and Mexico, CPI growth rate (inflation rate) of the two countries, and exchange rate of peso against dollars. Lagged imports are included. - The equation is applied to both industries affected and unaffected by tariffs; observed increases in unaffected industries could indicate indirect effects via input-output linkages (upstream and downstream). - Given only one industry had no total tariff change, the authors use tariff change size (∆τ_j) instead of a binary treatment dummy in equation (2) and regard that result as the main finding. - After estimating β̂_j for each industry, the authors regress β̂_j on industry tariff exposure measures: - β̂_j = α + β T_j + ε_j (using tariff increase dummy T_j) - β̂_j = α + β ∆τ_j + ε_j (using tariff increase size ∆τ_j) - They consider output, upstream, downstream, total, and net tariffs as explanatory variables separately. ### Tariff exposure: summary statistics (Table 1) - Sample: Obs = 258 industries. - Output tariff 2017 τ_{j,2017}: Mean = 1.17; S.D. = 3.23; Min = 0; Max = 34.05 - Output tariff 2018 τ_{j,2018}: Mean = 4.61; S.D. = 7.16; Min = 0; Max = 38.93 - Output tariff difference ∆τ_j: Mean = 3.43; S.D. = 5.56; Min = 0; Max = 25.00 - Upstream tariff 2017 τ_{j,2017}^{up}: Mean = 0.46; S.D. = 1.56; Min = 0; Max = 15.77 - Upstream tariff 2018 τ_{j,2018}^{up}: Mean = 1.34; S.D. = 3.59; Min = 0; Max = 37.91 - Upstream tariff difference ∆τ_j^{up}: Mean = 0.88; S.D. = 2.24; Min = 0; Max = 176.35 - Downstream tariff 2017 τ_{j,2017}^{down}: Mean = 0.33; S.D. = 1.34; Min = 0; Max = 16.24 - Downstream tariff 2018 τ_{j,2018}^{down}: Mean = 1.30; S.D. = 2.82; Min = 0; Max = 40.36 - Downstream tariff difference ∆τ_j^{down}: Mean = 0.97; S.D. = 1.73; Min = 0; Max = 24.12 - Total tariff 2017 τ_{j,2017}^{total}: Mean = 1.97; S.D. = 5.36; Min = 0; Max = 62.08 - Total tariff 2018 τ_{j,2018}^{total}: Mean = 7.25; S.D. = 11.14; Min = 0; Max = 117.20 - Total tariff difference ∆τ_j^{total}: Mean = 5.28; S.D. = 7.74; Min = 0; Max = 70.44 - Net tariff 2017 τ_{j,2017}^{net}: Mean = 1.31; S.D. = 3.32; Min = -1.96; Max = 33.98 - Net tariff 2018 τ_{j,2018}^{net}: Mean = 4.65; S.D. = 7.53; Min = -2.78; Max = 37.84 - Net tariff difference ∆τ_j^{net}: Mean = 3.35; S.D. = 5.80; Min = -1.95; Max = 25.77 - Dummy measures (∆τ_j > 0): - Dummy output tariff increase 1(∆τ_j^{output}>0): Mean = 0.38; S.D. = 0.49 - Dummy upstream tariff increase 1(∆τ_j^{up}>0): Mean = 0.97; S.D. = 0.17 - Dummy downstream tariff increase 1(∆τ_j^{down}>0): Mean = 1.00; S.D. = 0.06 - Dummy total tariff increase 1(∆τ_j^{total}>0): Mean = 1.00; S.D. = 0.06 - Dummy net tariff increase 1(∆τ_j^{net}>0): Mean = 0.46; S.D. = 0.50 - Correlation structure (selected reported correlations): - Correlation coefficients among changes in output, upstream, and downstream tariffs range from 0.05 to 0.77. - Correlation between ∆τ_j^{output} and ∆τ_j^{up} = 0.38. - Correlation between ∆τ_j^{output} and ∆τ_j^{down} = 0.32. - Correlation between ∆τ_j^{up} and ∆τ_j^{down} = 0.77. - Additional discrete-indicator correlations reported: 0.05, 0.14, 0.35 for particular binary comparisons. ### Aggregate trade diversion effect: difference-in-differences results (Table 2 summary) - Sample: monthly data from 258 INEGI industries during 2016/01–2019/05. - Estimation approach: regression (1) with standard difference-in-differences, varying treatment time by industry; dependent variable in logarithm so (100 β̂) gives percent change. Arellano-Bond dynamic panel estimator used when lagged log imports included. - Main findings (reported β̂ coefficients interpreted as percent differences in U.S. imports after trade tensions): - Output tariff exposure (dummy): industries affected by output tariff increase experienced on average 16.1 percent larger increase in U.S. imports after the trade tensions (Column 1). - Upstream and downstream tariffs separately (Columns 2 and 3): both suggest increased exports, but upstream tariff effect is not statistically significant while downstream shows positive impact. - Joint specification of output, upstream, downstream (Column 4): positive effects for output and downstream tariffs; mildly negative effect for upstream tariffs; only output tariffs statistically significant individually. The Wald test for joint significance of the three variables rejects the null with significance 1 percent. - Total tariff (dummy, Column 5): industries exposed to total tariff change had a larger increase in exports to the U.S. by 8.5 percent compared to industries with no increase in all three tariffs. - Net tariff (dummy, Column 6): industries exposed to a positive net tariff change increased their exports to the U.S. by 16.3 percent higher compared to industries exposed to no net tariff increase; the net tariff effect is statistically significant while total tariff effect is not (individually). - Above-median exposure (Columns 7–12): - Because almost all industries experienced some upstream/downstream/total tariff increase (only one or three industries with zero change in some measures), the authors also compare industries with tariff increase above median versus below median. - Results similar to dummy specifications: output tariffs and downstream tariffs are statistically significant individually; when all three are included, only output tariffs remain significant individually, but Wald tests support joint significance. - Example: Column (11) (treatment = total tariff increase above median) — industries with larger total tariff increase experienced 18.0 percent larger increase in exports to the U.S., which is significant. - Column (12) (net tariff above median) — compared to industries with lower net tariff change, those with net tariff changes above the median increased their exports to the U.S. by (text truncated in source). ### Interpretation and implications from these results - Positive trade diversion effects are observed, mainly driven by output tariffs and with additional evidence for downstream tariffs. - Upstream tariff effects are ambiguous: individual coefficient often not statistically significant and can be mildly negative in joint specifications, though joint tests indicate upstream/downstream/output variables are jointly significant. - Considering upstream and downstream tariffs (through input-output linkages) is important; omitting these could bias estimated trade diversion effects. - Using tariff change size (∆τ_j) becomes necessary when a binary control group is insufficient (only one industry with no total tariff change). *Source: IMF Working Paper chapter "4.2 An Alternative Approach Exploiting Industry-Level Trade Diversion Effects" from the provided PDF content.* ### 17.4 percent more. We also used different cutoffs in tariff changes to test the robustness of this set of results, ### wpiea2023234-print-pdf - 17.4 percent more. We also used different cutoffs in tariff changes to test the robustness of this set of results, ### Continuous treatment results (Table 3 and related text) - Regression results using continuous tariff changes indicate: - Output tariff: coefficient 0.013*** (0.005) in column (1) and 0.015** (0.007) in column (2) (Table 3). - Upstream tariff: coefficient 0.023 (0.015) in column (2) and -0.034 (0.038) when placed with other tariff variables (Table 3). - Downstream tariff: coefficient 0.021 (0.014) in column (3) and 0.009 (0.038) when placed with other tariff variables (Table 3). - Total tariff: coefficient 0.008*** (0.002) in column (5) (Table 3). - Net tariff: coefficient 0.011*** (0.002) in column (6) (Table 3). - Lagged ln(usimports): coefficients 0.158** (0.074) to 0.160** (0.074) across columns (Table 3). - Magnitudes (from main text based on one standard deviation changes): - If one industry’s increase in output tariff of U.S. on Chinese product is one standard deviation higher (5.6 percentage points), then this industry’s exports to the U.S. is estimated to go up by 7.2 percentage points after the start of the trade tensions (column 1). - Upstream: one standard deviation higher (1.73) → larger but insignificant increase in exports to the U.S. by 4.0 percent (column 2). - Downstream: one standard deviation higher (2.24) → exports to the U.S. estimated to increase 4.7 percent (insignificantly). - Total tariff: one standard deviation higher (7.74) → exports to the U.S. will increase 6.2 percent more after the trade tensions (column 5). - Net tariff: one standard deviation (5.80) → associated with 6.4 percent increase in exports (column 6). - Interpretation: - Output tariff increases drive the strongest and most consistent positive trade diversion effects. - Upstream tariffs show negative/mixed effects; downstream tariffs show positive but often insignificant effects. - Wald tests support joint significance of the tariff variables even where individual upstream/downstream coefficients are insignificant. ### Dummy treatment results (Table 2 and related text) - Dummy-based regressions indicate: - Output dummy: coefficient 0.161*** (0.045) in column (1) and 0.158*** (0.049) in column (2) (Table 2). - Downstream dummy: 0.699*** (0.269) in its specification; other downstream entries include 0.075 (0.137) and 0.185*** (0.062) depending on column (Table 2). - Upstream dummy: reported estimates include 0.851 (0.852), -0.000 (0.090), -0.015 (0.126), -0.049 (0.120) across specifications (Table 2). - Net and total dummies: significant coefficients reported in specific columns (Table 2). - Lagged ln(usimports): 0.156** (0.073) to 0.160** (0.074) across specifications (Table 2). - Specification details: - Dynamic panel model with lagged export value term, industry and month fixed effects, and Arellano-Bond estimator. - Panel: 258 industries × 41 months; observations 10,062. ### Time-varying treatment effect (Section 5.2 and Figure 3) - Findings: - Positive trade diversion effect began to appear after the announcement of the trade tensions. - Using a flexible monthly specification for 2016/01–2019/05, the treatment dummy (industry net tariff change above the median) yields coefficients: - βt close to zero and insignificant before 2018. - βt becomes positive upon announcement of trade tensions and significantly positive during the trade tension period. - Evidence of an anticipation effect: statistically significant trade diversion effect appears earlier than formal policy implementation dates. ### Robustness checks (Section 5.3) - Static panel (excluding lagged term) with industry-clustered standard errors: - If one industry’s total tariffs imposed by the U.S. on Chinese products increases by one standard deviation, Mexican exports to the U.S. of this industry go up by 4.6 percent more after the trade tensions (significant). - Estimated effect slightly smaller when lagged terms are controlled. - Alternative dependent variable: monthly growth rate of Mexican exports to the U.S. (including lagged export to control convergence): - One standard deviation increase in total tariffs → growth rate of Mexican exports to the U.S. goes up by 0.04 percentage points more after the trade tensions (significant). - Other exposure measures imply positive (negative) trade diversion effect of output and downstream (upstream) tariff changes. - Robustness results reported in Appendix III (tables A1–A4 referenced in source). ### Industry-level variation in trade diversion effects (Section 6, Table 4, Table 5, Figure 4) - Overall industry outcomes: - Most industries’ exports to the U.S. in Mexico grew after the trade tensions, with an average growth rate of 6.25 percent. - Table 4 summary statistics of estimated industry-specific trade diversion effects: - U.S. import level (258 obs.): Mean 0.0625, S.D. 0.4327, Min -1.5549, Max 5.6514. - Non-zeros (97 obs.): Mean 0.1663, S.D. 0.6955, Min -1.5549, Max 5.6514. - Distributional note: blue bar in Figure 4 includes industries that never exported during the sample period (mostly services) with estimated coefficients equal to zeros; red bar excludes these zeros. - More than 75 percent of industries’ exports to the U.S. expanded during the trade tension period after controlling for macro variables. - Correlates of industry-specific trade diversion (Table 5: correlation coefficient and p-value): - Change in U.S. imports from China: correlation coefficient -0.7181, p-value 0.0000. - Net tariff change: correlation coefficient 0.2500, p-value 0.0196. - Output tariff change: correlation coefficient 0.2742, p-value 0.0102. - Export share to the U.S. in 2017: correlation coefficient 0.0149, p-value 0.8716. - Imported input value share in production in 2016: correlation coefficient 0.0655, p-value 0.2956. - Export share in sales in 2016: correlation coefficient 0.0624, p-value 0.3182. - Product substitutability (σ): correlation coefficient 0.2320, p-value 0.0002. - Interpretation: - Positive trade diversion effect across industries is more strongly associated with decreases in U.S. imports from China, net and output tariff changes, and product substitutability with Chinese products. - Little or no correlation with industry-level trade exposure to the U.S., imported input share, or export share in sales. - Positive, though insignificant, correlation with GVC integration measures. ### Key specifications and data notes - Panel: 258 industries over 41 months (2016/01–2019/05), observations 10,062. - Estimation approaches: dynamic panel Arellano-Bond estimator (primary), static panel with clustered SEs (robustness), difference-in-differences with flexible monthly specification for time-varying effects. - Dependent variable: U.S. import value from Mexico (log) and alternative specifications using monthly growth rate. - Controls: GDP growth rates (U.S. and Mexico), CPI growth rates, exchange rate (peso/dollar), lagged imports, industry and month fixed effects. *Source: wpiea2023234-print-pdf — https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023234-print-pdf.pdf* ### 6.1 Decrease in Imports from China ### 6.1 Decrease in Imports from China ### Trade diversion from China to Mexico - When U.S. imports from China fell in an industry, U.S. imports from Mexico in that industry rose; this is estimated using industry-level regression (equation 5). - The estimated industry-level correlation between the change in U.S. imports from China and the increase in U.S. imports from Mexico is -0.718 (negative and significant). - Figure 5 (as described) ranks industries by the change in U.S. imports from China (black line, right y-axis) and plots for each industry the estimated trade diversion effect—the increase in exports to the U.S. after trade tensions, 훽̂_{j}^{M} —with 90 percent confidence intervals (blue line and area, left y-axis). - Several important Mexican industries are highlighted, including: - motor vehicle manufacturing, - motor vehicle parts manufacturing, - motor vehicle body and trailers manufacturing, - semiconductor and other electronic component manufacturing. ### Key statistic - Correlation coefficient (industry-level export growth vs. change in U.S. import from China): -0.718. --- ### Tariff changes and trade diversion - The industry-level trade diversion effect is positively correlated with tariff exposures; output tariff has a larger impact than upstream, downstream, total, or net tariff. - Table 6 reports regressions of the estimated trade diversion effect (dependent variable = estimated 훽̂_{j}^{M} from regression (5)) on different tariff treatment measures. Each entry is from a separate regression. - Column (1): Tariff increase dummy - Output tariff: 0.166** (0.071) - Upstream tariff: 0.063** (0.027) - Downstream tariff: 0.065** (0.028) - Total tariff: 0.063** (0.027) - Net tariff: 0.128** (0.058) - Column (2): Tariff increase dummy of median - Output tariff: 0.166** (0.071) - Upstream tariff: 0.033 (0.054) - Downstream tariff: 0.114** (0.054) - Total tariff: 0.113** (0.054) - Net tariff: 0.125** (0.053) - Column (3): Tariff increase levels - Output tariff: 0.012** (0.005) - Upstream tariff: 0.007** (0.004) - Downstream tariff: 0.007 (0.004) - Total tariff: 0.007** (0.003) - Net tariff: 0.011** (0.005) - Additional notes: - Positive coefficients imply industries more exposed to U.S.-China trade tensions experienced larger increases in U.S. imports from Mexico. - When regressing output, upstream, and downstream tariff changes simultaneously, output tariff changes retain a positive correlation; upstream is negative but insignificant; downstream is positive but insignificant. - The positive correlation between industry-level trade diversion and net tariff change is illustrated in Figure 6; the correlation coefficient is 0.25 (positive and significant). --- ### Mexico’s industry-level exposure to the U.S. market - The positive trade diversion effect is not explained by pre-existing industry exposure of Mexico to the U.S. market. - Figure 7 orders industries by their export share to the U.S. in 2017 (black line) and plots estimated 훽̂_{j}^{M} from regression (5) with 90 percent confidence intervals (blue). - Correlation coefficient between the size of the increase in export value to the U.S. and the 2017 export share: 0.01 (close to zero, not significant). --- ### GVC integration in Mexico - Industry-level global value chain (GVC) integration measures: - (i) share of imported input in total production cost; - (ii) share of exports in total production. - Measures constructed from INEGI input-output table for 2016. - Correlation between GVC integration and estimated increase in exports from Mexico to the U.S. after U.S. tariffs on Chinese products: 0.06 for both measures (positive, though insignificant). --- ### Product substitutability - Industries with higher product substitutability benefited more from trade diversion. - Substitutability measure: elasticity of substitution between varieties from different countries (second layer of nested CES demand), estimated at HS 3-digit level (Broda and Weinstein (2006)) and aggregated to industry level by weighted average using 2016 import shares. - As industry substitutability increases (left to right in Figure 10), the increase in exports from Mexico to the U.S. after trade tensions becomes larger. - Correlation coefficient (trade diversion vs. product substitutability): 0.23 (positive and significant). --- ### 7. Using WIOD and UNComtrade Data Instead ### Comparison of INEGI and WIOD results - The matched INEGI-UN Comtrade dataset is more granular (258 industries) than the WIOD-UN Comtrade matched dataset (56 industries); tariff change variation across industries is larger in INEGI. - Qualitative findings are the same across datasets; quantitatively, estimated trade diversion effects are larger using INEGI than using WIOD. - Key quantitative comparison: - If an industry experienced a one standard deviation increase in net tariff change, the exports from this industry to the U.S. increase by: - 6.4 percent using INEGI, - 1.4 percent using WIOD. - Table 7 (summary): reported percentage changes in exports to the U.S. for a one S.D. increase in tariff exposure show larger magnitudes in INEGI across output, upstream, downstream, total, and net tariff measures (examples: Output: 7.2*** INEGI vs. 0.7 WIOD; Total: 6.2*** INEGI vs. 1.6 WIOD). - Dynamic pattern: - Estimated monthly trade diversion effects (Figure 11) using both INEGI (blue) and WIOD (orange) show the increase in exports from Mexico to the U.S. beginning in early 2018 (upon announcement of tariff increases) and remaining positive afterwards. - Magnitudes over time are comparable between the two datasets. *Source: IMF Working Paper — chapter section "6.1 Decrease in Imports from China" from the provided PDF content.* ### 7.2 Industry-Level Trade Diversion Effect ### 7.2 Industry-Level Trade Diversion Effect ### Comparison of INEGI and WIOD industry-level estimates - Using WIOD data, out of 56 industries, 25 industries experienced changes in exports to the U.S. after the trade tensions and 23 industries had exports to the U.S. increased. - Mean increase in exports using WIOD: - Industry-level trade diversion effect (Obs. 56): Mean (percent) 4.24 - Non-zeros (Obs. 25): Mean (percent) 9.49 - Mean increase in exports using INEGI: - Industry-level trade diversion effect (Obs. 258): Mean (percent) 6.25 - Non-zeros (Obs. 97): Mean (percent) 16.63 - Note: Table 8 lists the mean of the estimated industry-specific trade diversion effects from (5) with lagged terms for U.S. import value, using INEGI and WIOD data. ### Correlation between industry-level trade diversion effect and tariff change exposures - The estimated industry-level trade diversion effect is positively correlated with the tariff change exposure at the industry level. - Correlation results (Table 9): each entry is the coefficient from regressions of the estimated 훽̂ from regression (5) on various tariff-exposure measures; results shown for INEGI and WIOD. - Tariff increase dummy (column (1)): - Output: INEGI 0.166**, WIOD 0.095*** - Upstream: INEGI 0.063**, WIOD 0.045*** - Downstream: INEGI 0.065**, WIOD 0.047*** - Total: INEGI 0.063**, WIOD 0.045*** - Net: INEGI 0.128**, WIOD 0.074*** - Tariff increase dummy of median (column (2)): - Output: INEGI 0.166**, WIOD 0.095*** - Upstream: INEGI 0.033, WIOD 0.029 - Downstream: INEGI 0.114**, WIOD 0.055*** - Total: INEGI 0.113**, WIOD 0.085*** - Net: INEGI 0.125**, WIOD 0.085*** - Tariff increase levels (column (3)): - Output: INEGI 0.012**, WIOD 0.010*** - Upstream: INEGI 0.007**, WIOD 0.081*** - Downstream: INEGI 0.007, WIOD 0.022 - Total: INEGI 0.007**, WIOD 0.009*** - Net: INEGI 0.011**, WIOD 0.009*** - Interpretation: The positive correlation between industrial trade diversion effect and tariff exposure measure is more pronounced using INEGI, though both INEGI and WIOD correlations are significant in many cases. ### Event study: Effect of U.S. sanctions on Russia (2014) on U.S. imports from Mexico - Context and approach: - Aim: estimate whether U.S. sanctions on Russia in 2014 led to diverted U.S. imports toward Mexico. - Data: product-level U.S. import flows at HS 6-digit, monthly 2012/01–2016/12; U.S. tariff level at product level controlled. - Regression (7) specification: logs of nominal U.S. import value of product i from Mexico in month t, with product fixed effects and product-level U.S. tariffs; parameter 훽τ measures average U.S. imports from Mexico over time relative to January 2012. - Standard errors clustered at the product level. - Stylized aggregate facts (2012–16): - U.S. imports from Russia declined, especially after 2014; U.S. imports from Mexico increased in aggregate following sanctions. - Table 10 — average annual nominal U.S. imports (2012–14 vs 2015–16) and changes: - Mexico: - Pre-sanction U.S. imports (billion $, percent in 2012 GDP): 23.89, 1.99% - Post-sanction U.S. imports (billion $, percent in 2012 GDP): 24.74, 2.06% - Change in U.S. imports (billion $, percent in 2012 GDP): 0.85, 0.07% - Russian Federation: - Pre-sanction U.S. imports (billion $, percent in 2012 GDP): 2.29, 0.10% - Post-sanction U.S. imports (billion $, percent in 2012 GDP): 1.36, 0.06% - Change in U.S. imports (billion $, percent in 2012 GDP): -0.93, -0.04% - Notes: imports measured in billion dollars nominal; second row for each country shows imports normalized by 2012 GDP. - Dynamic regression results (equation (7)): - Estimated time-dummy coefficients (훽τ) increase in the months following March 2014 and December 2014, indicating positive trade diversion effects toward Mexico. - 훽τ measures the average level (in percentage change) of U.S. import value from Mexico in period τ relative to January 2012. - Local projection results (equation (8)): - Specification estimates accumulated changes in log U.S. imports from Mexico over horizons k = 0,...,6 months after sanction shocks, controlling for lagged import changes, Mexico’s GDP changes, and U.S. tariffs on Mexico at product level. - Impulse response findings: - Nominal exports from Mexico increase by about 10 percent four months after the sanction (coefficient and confidence intervals plotted; horizon k = 4 shows ~10 percent increase). ### Broader conclusions and implications (from sections 8–9) - Main findings across episodes (U.S.-China trade tensions 2018 and U.S. sanctions on Russia 2014): - Positive trade diversion effects for Mexico’s exports to the U.S., emanating mainly from output tariffs and also from downstream tariffs. - Effects are stronger when nationally sourced input-output data are used (INEGI) compared to cross-country sources (WIOD). - Determinants of industry-level trade diversion magnitude: - Depends on U.S. tariff changes on Chinese goods, decrease in U.S. imports from China, product substitutability with Chinese products, and (weakly) on Mexico’s GVC integration. - Easily substitutable products may gain U.S. market share during global trade tensions even if initial exposure to U.S. market was not substantial. - Policy-relevant messages: - Differential short-term impacts across countries and industries can arise from policy-driven geoeconomic fragmentation (GEF); some countries/industries may benefit in the short term but higher tariffs and geopolitical tensions reduce global welfare. - Proper accounting of supply linkages matters: improving industry-level coverage and comparing country-specific versus cross-country input-output sources is important for future work. - Relation to other studies: - Findings are in line with Freund et al. (2023), Alfaro et al. (2023), and Fajgelbaum et al. (2023) that identify Mexico as among winners during global trade tensions and find evidence of nearshoring. *IMF Working Paper — Trade Diversion Effects from Global Tensions—Higher Than We Think (sections 7.2, 8, and 9 summarized from the provided content)* ### References ### References and Appendices (wpiea2023234-print-pdf) ### References - Comprehensive bibliography listing empirical and theoretical works on trade policy, trade wars, tariffs, global value chains, sanctions, and input-output methods. Notable entries include: - Abadie, A., A. Diamond, and J. Hainmueller, 2010. “Synthetic control methods for comparative case studies: Estimating the effect of California’s Tobacco control program.” Journal of the American Statistical Association 105 (490): 493–505. - Ahn, Daniel P., and Rodney D. Ludema, 2017. Measuring Smartness: Understanding the Economic Impact of Targeted Sanctions. United States Department of State, Office of the Chief Economist. - Aiyar, Shekhar, et al., 2023. "Geo-Economic Fragmentation and the Future of Multilateralism." IMF Staff Discussion Notes, No. 001. - Amiti, Mary; Redding, Stephen J.; Weinstein, David E., 2019. "The Impact of the 2018 Tariffs on Prices and Welfare." Journal of Economic Perspectives 33(4): 187-210. - Fajgelbaum, Pablo D., and Amit K. Khandelwal, 2022. "The Economic Impacts of the U.S.–China Trade War." Annual Review of Economics 14: 205-228. - Multiple IMF World Economic Outlook citations: 2018; 2019; 2022; 2023 (Geoeconomic Fragmentation and the Future of Multilateralism). - The references support empirical methods (e.g., synthetic control, two-way fixed effects, differences-in-differences with heterogeneous treatment effects), IO/Leontief analysis, and data concordances (NAICS/ISIC/HS mappings). ### Appendix I. Dataset Matching Methods - Purpose: Describe matching procedure between WIOD and INEGI datasets and assumptions for constructing industry input and trade structures; present methods for matching UN Comtrade and WIOD to INEGI at NAICS 4-digit level; present Leontief inverse formalism. - A1.1 Matching WIOD and INEGI for Stylized Facts on Industry Input and Trade Structure - Data coding: - INEGI uses NAICS 4-digit 2017 version. - WIOD adapts ISIC Rev 4 at 2-digit level. - Matching uses concordance table between ISIC Rev. 4 to 2017 NAICS from U.S. Census website and averages when encountering aggregation issues. - Steps to construct input and import structure for an industry (motor vehicle example): - Step 1: Use INEGI IO tables (total values). Keep column for motor vehicle industry; rows 1–258 represent input industries; remaining rows aggregate domestic, imports, and total production. Calculate total production cost by summing inputs from 258 industries; compute value share of each input and rank to find top 10 inputs. - Step 2: Use INEGI IO tables with imported values and total values. Combine tables to calculate imported value share per input as ratio of imported value to total value. - Step 3: Use WIOD IO table to find top sourcing countries for each input by matching NAICS 4-digit inputs in INEGI to sourcing countries in WIOD. - Step 3.0: - Match NAICS and ISIC Rev 2 to merge databases; prepare WIOD by linking WIOD codes to ISIC Rev. 4 using UN Statistics Division documents. - Four details: - Match at IndustryCode and Country level using WIOD lists to ensure unique WIOD rows and full coverage of Countries and ISIC industries. - After matching, panel at ISIC and Country level may map several ISIC codes to one WIOD IndustryCode; evenly distribute input values to each ISIC code in such cases. - Reshape WIOD matrix to long panel with output, input, and country as three dimensions. - After reshaping, link output code with ISIC Rev.4 2-digit codes without averaging or aggregation as in distribution step. - Step 3.1: - For production industry (e.g., NAICS 3361 motor vehicles), find corresponding WIOD ISIC output industries. When multiple ISIC industries match, take average of input values among these industries by industry and country. - Step 3.2: - For each INEGI input industry, find corresponding WIOD industries and match import structure. - Steps: match INEGI to NAICS 4-digit; match NAICS 4-digit to ISIC 2-digit; match ISIC to WIOD IO tables from Step 3.0. - When one NAICS code matches multiple ISIC codes, sum input values and imported input values of all matched ISIC industries for each country, then calculate imported share as an average. - Concordance references: - U.S. Census concordances: https://www.census.gov/eos/www/naics/concordances/concordances.html (used for 2017 NAICS to ISIC Rev. 4). - UN Statistics Division ISIC Rev.4 documentation: https://unstats.un.org/unsd/publication/seriesm/seriesm_4rev4e.pdf. - A1.2 Imports from the U.S. in Each Industry - Method: Calculate exports to U.S. in each industry at NAICS 4-digit level by matching imports from the U.S. from UN Comtrade trade flow data at HS 6-digit level. - Matching steps with UN Comtrade: - Step 1: For each industry j from INEGI and each year, calculate imports from the U.S. from UN Comtrade at HS 6-digit level. Requires: - Step 1.1: Match NAICS 4-digit code with INEGI industry codes. - Step 1.2: Match NAICS 4-digit with HS 6-digit as used in UN Comtrade (Pierce and Schott (2009) mapping). - A1.3 Exports to the U.S. in Each Industry - Method: Calculate exports to U.S. in each industry at NAICS 4-digit by matching exports to the U.S. from WIOD and UN Comtrade. - Matching with UN Comtrade: - Step 1: For each INEGI industry j and year, calculate exports to the U.S. from UN Comtrade at HS 6-digit level: - Step 1.1: Match NAICS 4-digit with INEGI industry codes. - Step 1.2: Match NAICS 4-digit with HS 6-digit (Pierce and Schott (2009)). - Matching with WIOD (distinguishing intermediate input and final goods): - Step 1: For each INEGI industry j and year, calculate exports to the U.S. from WIOD IO table via: - Step 1.1: Match ISIC 2-digit code with WIOD industry codes; for each ISIC 2-digit industry, calculate exports to U.S. by summing exports across industries in the U.S. (WIOD distinguishes exports of intermediate inputs and final goods). - Step 1.2: Match NAICS 4-digit code with INEGI industry codes for both input and output industries. - Step 1.3: Match NAICS 4-digit industry j from INEGI with ISIC 2-digit industry j from WIOD; when multiple ISIC industries map to one NAICS industry, average across WIOD industries. - A1.4 Leontief Inverse of Input-Output Table - Accounting identity and fixed technical coefficients: - X_i = sum_j Z_ij + Y_i = sum_j m_ij X_j + Y_i, m = 1, ... , m, - where X_i is total amount produced, Z_ij is interindustry flow from sector m to sector j, and Y_i is final demand for product m. - Fixed technical coefficients: m_ij = Z_ij / X_j (Miller and Blair, 2009). - Matrix form: - X = A X + Y - (I − A) X = Y - X = (I − A)^{-1} Y = L Y - Assume (I − A) is invertible and Y ≥ 0. L = (I − A)^{-1} is the Leontief inverse. - Interpretation: Leontief inverse answers amounts of each product needed to meet final demand given input-output linkages. ### Appendix II. Aggregate Trade Structure in Mexico (2003–18) - Overview: Examines evolution of Mexico’s production and trade structure across 2003–18, including external market dependence, linkage with U.S. market, top trading partners and top trading sectors. Key finding: Mexico is increasingly dependent on external markets for both exports and input sourcing, especially the U.S.; China’s importance has grown over time. - A2.1 Gross Exports and Imports, as Share of Production - Empirical findings: - Export share increased from 14 percent in 2003 to 22 percent in 2018. - Import share in production increased from 26 percent in 2003 to 42 percent in 2018. - Interpretation: Rising export and import shares imply greater integration of Mexico into global markets and international trade. - Data source: INEGI input-output tables. - A2.2 Share of the U.S. in Exports and Imports - Empirical finding (partial text provided): - The U.S. is the largest destination and supplier for Mexico, accounting for 79 percent of exports and 44 percent of imports in *Source: IMF Working Paper — Trade Diversion Effects from Global Tensions—Higher Than We Think (References and Appendices as provided in wpiea2023234-print-pdf).* ### 2020. Over time, the U.S. share in Mexico’s exports has declined from 88 percent in 2003 to 79 percent in ### wpiea2023234-print-pdf - 2020. Over time, the U.S. share in Mexico’s exports has declined from 88 percent in 2003 to 79 percent in 2020, while the U.S. share in Mexico’s imports has decreased from 62 percent in 2003 to 44 percent in 2020. ### Trade shares and trends - U.S. share in Mexico’s exports: - 88 percent in 2003 - 79 percent in 2020 - U.S. share in Mexico’s imports: - 62 percent in 2003 - 44 percent in 2020 - Figures plotting gross exports and imports (values in millions of pesos, current price) are shown for 2003–2020 with USA share (%) axes ranging (exports top figure: USA Share (%) axis includes 78, 80, 82, 84, 86, 88; imports bottom figure: USA Share (%) axis includes 45, 50, 55, 60, 65). ### Top five trade partners (as in 2020) - Top five export destinations in 2020: U.S., Canada, China, Germany, Japan. - Top five import sourcing countries in 2020: U.S., China, South Korea, Japan, Germany. - China trade share changes (2003 → 2020): - Export share to China: 0.6 percent in 2003 to 1.9 percent in 2020 (tripled). - Import share from China: 5.5 percent in 2003 to 19.2 percent in 2020 (nearly fourfold). - Note: Figures fix top five destinations/sources as in 2020 while plotting 2003–20; main message: declining U.S. share and growing shares for Asian economies, especially China. China became one of Mexico’s top 5 destinations since 2008. ### Top five trading sectors (industry-level 2003–18, INEGI matched with UN Comtrade) - Method: 258 industries from INEGI matched to UN Comtrade via Pierce and Schott (2009) concordance; matched series cover 2003–18. - Exports (top five exporting industries fixed as in 2018; shares plotted 2003–18): - Motor vehicle related industries (motor vehicle manufacturing, motor vehicle parts manufacturing, motor vehicle body and trailers manufacturing) together account for 32 percent of exports in 2018 (increasing trend since 2003). - Semiconductor and other electronic component manufacturing, and computer and peripheral equipment manufacturing together comprise 11 percent of exports. - Semiconductor sector joined top five exporters since 2012 and has been growing, replacing the share of oil and gas extraction. - Imports (top five importing industries fixed as in 2018; shares plotted 2003–18): - Semiconductor industry is the largest importing industry. - Mexico imports semiconductor and other electrical component manufacturing the most, accounting for about 9 percent of total imports in 2018. - Motor vehicle parts contribute 8 percent of total imports in 2018. - Rising trend in import share for petroleum and coal industry; other electrical equipment and engine, turbine, and power transmission equipment manufacturing are relatively stable. - Petroleum and Coal sector joined top 5 since 2007; petroleum and coal imports from the U.S. rose from less than 1% in 2003 to 11% and became the top importing industry in 2018 (noted for U.S. market). - Interpretation: - Main inputs for motor vehicle production: motor vehicle parts; engine, turbine, and power transmission; semiconductor and other electrical component manufacturing. - Imports are mainly for production rather than consumption; high export and import shares in motor vehicles and electrical equipment imply Mexico’s positioning near the bottom of the global value chain in these sectors, focusing on auto parts assembly. - Transportation equipment (auto parts and assembly) represents 13 percent of Mexico’s manufacturing value added (Bolio et al. (2014) cited in text). - Mexico’s auto assembly plants are considered world-class with faster expansion; Mexico occupies the most labor-intensive and least value-added tasks in the auto production chain (Crossa and Ebner (2020) cited in text). ### Trade diversion effects — regression summaries (Appendix III) - Panel setup: - Industry-month panel with 258 industries and 41 months (observations reported as 10,578 for level regressions; 10,320 for growth-rate regressions depending on specification). - Models adopt static fixed effect panel with industry and month fixed effects and clustered standard errors at industry level. - Table A1 (Dummy treatment variable, dependent variable: exports values to the U.S., 258 industries × 41 months, Obs. 10,578, Ind. = 258): - Output: coefficient 0.171*** (0.052) - Upstream (Up): coefficient 0.061*** (0.015) in one specification; other Up coefficients reported including 0.002 (0.030) and -0.002 (0.026) across columns. - Downstream (Down): coefficient 0.063*** (0.016) in one specification; other Down coefficients include 0.008 (0.010), 0.059** (0.030), 0.002 (0.030). - Total and Net reported coefficients in some columns: Total 0.061*** (0.015); Net 0.109*** (0.031) and 0.119*** (0.029) in other columns. - Constant: 6.584*** (0.024) across columns. - R-squared values reported range around 0.011–0.017 depending on column. - Table A2 (Continuous treatment variable, dependent variable: exports values to the U.S., Obs. 10,578): - Output Δ휏: 0.012*** (0.004) in column (1); 0.013*** (0.005) in column (2). - Upstream: 0.006 (0.004) and -0.014 (0.010) across columns. - Downstream: 0.006 (0.004) in one column; omitted or not significant in others. - Total: 0.006*** (0.002) in one column. - Net: 0.010*** (0.002) in column (6). - Constant: 6.584*** (0.024). - R-squared values reported 0.011–0.015. - Table A3 (Dummy treatment, dependent variable: exports growth rates to the U.S., 258 industries × 41 months, Obs. 10,320): - Output: 0.144*** (0.041) in several columns; variants 0.145*** (0.045). - Upstream: 0.051*** (0.012) in one specification; other Up coefficients include 0.000 (0.000), 0.001 (0.026), -0.003 (0.023). - Downstream: 0.053*** (0.012) in one specification; other Down coefficients include 0.006 (0.009), 0.048* (0.025), 0.000 (0.026). - Total: 0.051*** (0.012) in one column; 0.076*** (0.024) in another. - Net: 0.091*** (0.025) and 0.100*** (0.023) in specified columns. - L.lnimp (lagged ln imports) coefficients: -0.850*** (0.066) to -0.846*** (0.067) across columns. - Constant values around 5.706*** to 5.731*** with standard errors 0.473–0.480. - R-squared around 0.423–0.425. - Table A4 (Continuous treatment, dependent variable: exports growth rates, Obs. 10,320): - Output Δ휏: 0.010*** (0.003) and 0.011*** (0.004) in columns (1) and (2). - Upstream: 0.004 (0.004) and -0.012 (0.008) in columns (2) and (3). - Downstream: 0.004 (0.003) and 0.003 (0.006) reported. - Total: 0.005*** (0.001) in column (5). - Net: 0.008*** (0.002) in column (6). - L.lnusimports (lagged ln US imports) coefficients: -0.848*** (0.067) to -0.846*** (0.067). - Constant values 5.722*** (0.478) to 5.718*** (0.479). - R-squared values around 0.423–0.424. ### U.S. tariff changes during 2012–16 (Appendix IV) - Figure A5: U.S. monthly aggregate import tariffs on Mexico (solid) and Russia (dashed), normalized to be one in January 2012, plotted for 2012–2017 with y-axis labeled "Aggregate import tariffs, relative to 2012/01 level" and x-axis months 2012–2017. - Source for tariffs: USITC. *International Monetary Fund. Trade Diversion Effects from Global Tensions—Higher Than We Think. Working Paper No. WP/2023/234* --- _Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023234-print-pdf.pdf_