## Tariff Pass-Through and Import Reallocation — Working Paper No. WP/2026/149 (excerpt)

## Source details

**Canonical URL:** [Tariff Pass-Through and Import Reallocation — Working Paper No. WP/2026/149 (excerpt)](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026149-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026149-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026149-source-pdf.pdf.json)

---

### 2.1 Data
- Data sources and price measures:
  - Monthly U.S. import data from the U.S. Census Bureau for 2024-25: import values (in current US$), quantities, and duties by country of origin and HS 10-digit product.
  - Import unit values:
    - FOB prices: values excluding collected duties (ratio of the total customs value of imports for consumption to their physical volume).
    - Tariff-inclusive unit values: values including collected duties (duty-inclusive unit price adds the estimates of calculated duty to the numerator).
  - Note: unit-value approach may mask variation in pricing strategies across firms exporting the same HS 10-digit product from a given country.
- Tariff measures:
  - Effective tariff rates: ratio of calculated duties to the value of imports for consumption; undefined when tariff is ‘prohibitive’ and import values fall to zero.
  - Statutory tariff rates: tariffs implemented by the U.S. administration via official documents.
- Statutory tariff data and adjustments:
  - Source: WTO-IMF Tariff Tracker (WTO-IMF, 2026) — U.S. tariffs at HS 8-digit level with daily dates of implementation.
  - Adjustments:
    - U.S. tariffs on Canada and Mexico scaled down proportionally to observed USMCA preference utilization rate at the product level (updated with a lag of a few months). As of February 2026, USMCA preference utilization rates were calculated using U.S. import data from March to June 2025.
    - Sectoral tariffs on steel and aluminum extended to derivative products using aggregated sectoral input-output coefficients.
  - Daily statutory tariffs aggregated to monthly frequency by taking the modal tariff rate for each month.
  - Baseline statutory tariffs for 2024 assigned to all months before February 2025.
- Preferred measures and instrumentation:
  - Preferred tariff measure: effective tariff rate (captures actual barriers faced).
  - Instrument: statutory tariff rate (τUS,sta) used to instrument effective tariff to address endogeneity.
- Coverage and time span:
  - Analysis focuses on 2024-25 monthly data; regression time unit spans February–December 2025 when U.S. tariffs were modified.

### 2.2 Stylized facts
- Import-weighted average statutory and effective tariff evolution:
  - Solid (time-varying weights) lines lie below dotted (fixed 2024 weights) lines, reflecting reallocation from higher-tariff to lower-tariff goods and countries.
  - Effective (blue) lines lie below statutory (red) lines, reflecting exemptions, implementation complexities, and lags.
- As of December 2025:
  - Average statutory rate using time-varying import weights was around 15.5 percent.
  - This was approximately 6 percentage points higher than the average effective rate.
  - Difference negligible in January 2025 but widened as tariff actions were implemented.
  - Across countries, average effective tariffs below statutory ones; gap particularly large for China (around 10 percentage points).
- Within-product variation:
  - For 2024-25 changes (relative to 2024), median within-product standard deviation of tariff changes around 8 percentage points (57 percent of overall standard deviation).
  - Within-product variation for 2025 accounts for 26 percent of overall tariff variation in the 2025 episode; for 2018-19 within-product share was around 98 percent.
- Aggregate import-price dynamics (BLS):
  - Year-over-year import price inflation rises through mid-2024, then declines thereafter despite new tariffs.
  - By mid-2025, headline import price inflation turns negative.
  - Import prices excluding food and fuels moderate but remain positive.
  - Rising tariffs combined with slowing pre-tariff import price inflation suggests other mechanisms driving decline in aggregate import prices.

### 3 Evidence from the 2025 U.S. tariffs — Empirical approach and identification
- Levels of analysis:
  - Variety level: exporter (i) — HS 10-digit product (g) pair.
  - Product level: how reallocation across varieties shapes tariff pass-through to import prices.
- Core variety-level specification:
  - ∆ ln Y_ig,t = β ∆ ln(1 + τ_US,eff_ig,t) + μ_g,t + η_i + ε_ig,t
  - Y ∈ {import unit values (FOB: P ≡ V/Q), duty-inclusive P* ≡ (V+T)/Q, import values V, quantities Q}; variables measured in year-on-year log differences; month t spans February–December 2025.
  - Fixed effects: HS 10-digit product-time μ_g,t and exporter η_i.
  - Standard errors clustered at HS 8-digit product level.
- Identification:
  - β identified from variation across exporters within an HS 10-digit product.
  - Instrument: statutory tariff changes τ_US,sta_ig,t used in 2SLS for effective tariffs.
  - First-stage shows strong positive correlation between statutory and effective tariffs.

### 3.1 Trade and price effects at the variety level — Baseline results (Table 1)
- Estimation sample details:
  - Year-on-year monthly log changes; sample February–December 2025.
  - Fixed effects: HS 10-digit product-time and country fixed effects in all columns.
  - Observations: 1,262,647 in each reported column.
  - First-stage F-stat: 2195.
- Table 1 key estimates (preserve estimates and standard errors exactly):
  - Column (1) OLS: ∆ ln(1 + τ_US,eff_ig,t) coefficient = -0.478*** (0.059).
  - Column (2) 2SLS (∆ ln V): coefficient = -0.478*** (0.080).
  - Column (3) 2SLS (∆ ln Q): coefficient = -0.0001 (0.054).
  - Column (4) 2SLS (∆ ln P): coefficient = 1.000*** (0.054).
  - Column (1) 2SLS first-stage: ∆ ln(1 + τ_US,sta_ig,t) coefficient = 0.393*** (0.008).
- Main findings and interpretations:
  - Exporters’ FOB unit values show muted response to U.S. tariff changes (no significant reduction in FOB unit values).
  - Duty-inclusive import prices show full pass-through at the variety level (coefficient = 1.000***).
  - Significant negative effects of U.S. tariffs on imports:
    - The 8 percentage point increase in import-weighted average U.S. effective tariff between February and December 2025 would reduce imports by about 3.6 percent on average (trade elasticity in column (3) reported as -0.5 and interpreted as a short-run elasticity).
  - Trade elasticity comparisons:
    - Estimated short-run trade elasticity of -0.5 is close to recent estimates but smaller than estimates from the 2018-19 U.S.-China episode (e.g., about -1.3 in Amiti, Redding and Weinstein (2019)).
  - Robustness:
    - Most restrictive fixed-effects specification raises estimated trade elasticity to -0.85 and slightly reduces tariff pass-through to import prices.
    - In most restrictive specification, exporters’ unit-price effect is negative but statistically insignificant; duty-inclusive prices imply up to 18 percent of tariff increases are absorbed by variety-level exporters’ prices.
  - Pattern interpretation:
    - Substantial trade reallocation across exporters in response to tariffs, with limited downward adjustment of exporters’ FOB prices.
    - Broader scope and greater uncertainty of 2025 tariff episode may explain weaker reallocation relative to 2018-19.

### Extensive margin specification and results (Table 2)
- Specification:
  - e_ig,t = β ∆ ln(1 + τ_US,sta_ig,t) + μ_g,t + η_i + ε_ig,t
  - e is entry/exit dummy (entry = 1 if imported in t but not t−12; exit = 1 if not imported in t but imported in t−12).
  - Use statutory tariff changes because effective rates undefined for non-imported varieties; regressions estimated via linear probability model (OLS reduced form).
- Reported estimates (preserve coefficients and standard errors exactly):
  - Columns (1)–(3) (samples include varieties imported at t−12) — Exit regressions:
    - Coefficients = 0.107*** (0.006); 0.155*** (0.012); 0.065*** (0.009) across columns.
  - Columns (4)–(6) (samples include varieties not imported at t−12) — Entry regressions:
    - Coefficients = -0.025*** (0.002); -0.036*** (0.003); -0.015*** (0.002) across columns.
  - Observations:
    - Exit columns: 1,723,217; 1,723,183; 1,690,529.
    - Entry columns: 29,255,790; 29,255,790; 29,241,264.
- Key extensive-margin findings:
  - Baseline exit effect (column (1)): the increase in average U.S. statutory tariffs between February and December 2025 (from about 2 to 15.5 percent) raises the probability of exiting the U.S. market by 1.4 percentage points, relative to average exit probability of 26 percent.
  - Baseline entry effect (column (4)): same tariff increase lowers probability of entering the U.S. market by 0.3 percentage points, relative to sample-average entry probability of 1.6 percent.
  - Interpretation: tariff hikes are prohibitive for some varieties, inducing switching to alternative origins and delayed entry.

### 3.2 Price effects at the product level — Methodology and decomposition
- Product-level price and inflation definitions:
  - p_g,t ≡ ∑_i w^V_ig,t p_ig,t with value-share weights w^V_ig,t ≡ V_ig,t / V_g,t and p_ig,t ≡ ln P_ig,t.
  - π_g,t ≡ ∆ p_g,t = ∑_i w^V_ig,t p_ig,t − ∑_i w^V_ig,t−12 p_ig,t−12.
- Log decomposition partitions varieties into continuing (C), entrants (E), and exiters (X) into five components:
  - Within: price changes among incumbents holding import shares fixed.
  - Between: changes in import shares among incumbents evaluated at prices in t−12.
  - Interaction: covariance term capturing simultaneous changes in unit values and import shares among incumbents.
  - Entry: contribution of newly imported varieties relative to product average in t−12.
  - Exit: contribution of exiting varieties relative to product average in t−12.
- Product-level tariff growth measure:
  - ∆ ln(1 + τ^US,k_g,t) ≡ ∑_i w^V_ig,t0 ln(1 + τ^US,k_ig,t) − ∑_i w^V_ig,t0 ln(1 + τ^US,k_ig,t−12), with k∈{eff, sta} and 2024 import value weights w^V_ig,t0 rescaled to sum to one.
- Baseline regression:
  - π_g,t = β ∆ ln(1 + τ^US,eff_g,t) + μ_t + ε_g,t, instrumenting effective tariff by statutory tariff.

### 3.2 Key empirical findings (Table 3 summary)
- Main 2SLS coefficient estimates (standard errors clustered at HS 8-digit):
  - Overall: ∆ ln(1 + τ^US,eff_g,t) → −0.624*** (column (1)).
  - Intensive margin: −0.223*** (column (2)).
  - Extensive margin: −0.401*** (column (3)).
  - Between (intensive decomposition): 0.086*** (column (4)).
  - Interaction (intensive decomposition): −0.127*** (column (5)).
  - Within (intensive decomposition): −0.183*** (column (6)).
  - Entry (extensive decomposition): −0.257*** (column (7)).
  - Exit (extensive decomposition): 0.144*** (column (8)).
- Sample and diagnostics:
  - Obs 171,206
  - F-stat 1st stage71587158715871587158715871587158
- Magnitude interpretation:
  - The 2025 increase in the average ln(1 + τ^US,eff_g,t) (around 8 percentage points) is associated with an approximate 5 percent decline in before-tariff import prices.
- Decomposition and contribution shares:
  - Intensive vs. extensive: around 65 percent of negative product-level price response driven by extensive margin; intensive margin accounts for remaining ~35 percent.
  - Within-term (−0.183***) indicates continuing-variety prices decline in products with larger tariff hikes.
  - Between-term (0.086***) implies incumbents gaining share tended to be relatively higher-priced initially, attenuating within-channel reduction.
  - Interaction (−0.127***) captures joint changes in prices and shares among incumbents.
  - Entry explains about 65 percent of extensive-margin contribution (entry coefficient −0.257***): new varieties tend to be lower-priced.
  - Exit explains about 35 percent of extensive-margin contribution (exit coefficient 0.144***): exiting varieties tend to be relatively higher-priced.
- Additional characterizations:
  - Alternative level decomposition (Appendix Table A.2.1): most negative product-level price effects reflect a ‘mean’ effect—unweighted average prices decreasing with changes in product-level average tariffs—driven by greater entry of lower-priced varieties and exit of higher-priced varieties.
  - Negative covariance effect indicates reallocation away from relatively expensive old varieties toward relatively cheap new varieties.
- Robustness and heterogeneity:
  - Alternative fixed effects produce qualitatively similar or larger coefficients (Table A.1.5).
  - Quarterly aggregation halves magnitudes (Table A.1.6).
  - Excluding AI-related products (AI-related imports grew nearly 73 percent between 2023 and 2025) does not materially change main results (Table A.1.7).
  - HS 8-digit aggregation preserves findings (Table A.1.8).
- Implications:
  - Higher product-level tariff hikes are associated with lower product-level import price inflation driven predominantly by extensive-margin reallocation toward lower-priced varieties.

### 3.3 Product quality and reallocation — Framework and identification
- CES demand framework within HS 10-digit product g over exporter–HS 10-digit varieties:
  - U_g = (∑_i (Θ_ig Q_ig,t)^{(σ_s−1)/σ_s})^{σ_s/(σ_s−1)}, σ_s > 1.
  - Θ_ig is exporter–HS 10-digit product appeal; σ_s is elasticity of substitution associated with HS 6-digit category s.
- Variety share (buyer’s prices, duty-inclusive):
  - w^V_*ig,t = Θ_ig^{σ_s−1} P^*_{ig,t}^{1−σ_s} / Φ^*_{g,t}, Φ^*_{g,t} ≡ ∑_i Θ_ig^{σ_s−1} P^*_{ig,t}^{1−σ_s}.
  - Log-linearized estimating equation:
    - ln w^V_*ig,t + (σ_s−1) ln P^*_{ig,t} = η_ig + μ_{g,t} + ε_ig,t,
    - η_ig identifies ln Θ_ig via ln Θ_ig = η̂_ig / (σ_s−1).
- σ_s obtained from trade elasticities in Fontagné, Guimbard and Orefice (2022); set σ = 1 − ε (their ε is negative elasticity of imports w.r.t. tariffs).
- Appeal estimated assuming time-invariance using 2023–2024 data (pre-2025 tariff hikes).
- CES price indices:
  - Appeal-adjusted: P^{CES}_{g,t} = (∑_i Θ_ig^{σ_s−1} P_{ig,t}^{1−σ_s})^{1/(1−σ_s)}.
  - Non-appeal-adjusted: P^{CES,na}_{g,t} = (∑_i P_{ig,t}^{1−σ_s})^{1/(1−σ_s)}.
- Regress year-on-year log changes in CES indices on log changes in weighted-average tariffs e τ_{g,t}^{US,eff}, instrumenting effective by statutory tariff.

### 3.3 Empirical evidence — 2025 U.S. tariffs (Table 4)
- Key 2SLS estimates on ∆ ln(1 + e τ^{US,eff}_{g,t}) (standard errors in parentheses; F-stat 1st stage reported):
  - Column (1): Dep: ∆ ln P^{CES}_{g,t}, Appeal adjusted: Yes, Duty exclusive: Coefficient = -0.357*** (0.074); Obs: 96,165; F-stat 1st stage: 4428.
  - Column (2): Dep: ∆ ln P^{CES,na}_{g,t}, Appeal adjusted: No, Duty exclusive: Coefficient = -0.642*** (0.064); Obs: 96,165; F-stat 1st stage: 4428.
  - Column (3): Dep: ∆ ln P^{CES}_{g,t}, Appeal adjusted: Yes, Duty inclusive: Coefficient = 0.375*** (0.074); Obs: 96,002; F-stat 1st stage: 4429.
  - Column (4): Dep: ∆ ln P^{CES,na}_{g,t}, Appeal adjusted: No, Duty inclusive: Coefficient = 0.059 (0.065); Obs: 96,002; F-stat 1st stage: 4429.
- Interpretation:
  - Duty-exclusive, appeal-adjusted index (col (1)): tariff increases associated with statistically significant decline in quality-adjusted (appeal-adjusted) before-tariff import prices.
  - Non-appeal-adjusted index (col (2)): more negative coefficient — reallocation toward lower-appeal (lower-quality) varieties is an important channel; appeal-erosion accounts for roughly half of estimated price reduction in 2025.
  - Duty-inclusive indices:
    - Col (4) (no appeal adjustment): positive and statistically insignificant — tariffs had negligible effect on duty-inclusive product-level import prices.
    - Col (3) (appeal adjusted): positive and statistically significant — accounting for appeal differences, tariffs raise duty-inclusive CES import prices, though increases are less than one-for-one with changes in average product-level tariffs.

### Comparison with 2018–19 tariffs (Tables 5 and 6)
- 2018–19 product-level decomposition (Table 5):
  - Sample: May 2018 to December 2019.
  - Overall coefficient: -0.809*** (0.102); Obs: 271,980; F-stat 1st stage: 1330.
  - Intensive margin: 0.076 (0.053).
  - Extensive margin: -0.885*** (0.080).
  - Entry: -0.669*** (0.074); Exit: 0.216*** (0.018).
  - Interpretation: extensive margin predominant; entry among relatively lower-priced varieties and exit among relatively higher-priced varieties.
- 2018–19 quality-adjusted effects (Table 6):
  - Column (1) — ∆ ln P^{CES}_{g,t}, Appeal adjusted, Duty exclusive: -0.421*** (0.121); Obs: 153,697; F-stat 1st stage: 3171.
  - Column (2) — ∆ ln P^{CES,na}_{g,t}, No appeal adjust, Duty exclusive: -1.024*** (0.122); Obs: 153,697; F-stat 1st stage: 3171.
  - Column (3) — ∆ ln P^{CES}_{g,t}, Appeal adjusted, Duty inclusive: 0.274** (0.127); Obs: 153,604; F-stat 1st stage: 3163.
  - Column (4) — ∆ ln P^{CES,na}_{g,t}, No appeal adjust, Duty inclusive: -0.267** (0.125); Obs: 153,604; F-stat 1st stage: 3163.
  - Interpretation: results mirror 2025 patterns — extensive-margin reallocation and appeal-erosion important; after appeal adjustment duty-inclusive CES prices increased with tariffs.

### Substantive findings and implications (summary)
- Variety level:
  - Exporters’ unit values did not respond to 2025 U.S. tariffs; imports decreased significantly.
  - Duty-inclusive variety-level pass-through approximately one-for-one.
- Product level:
  - Duty-exclusive import prices decrease with tariff hikes due to reallocation toward lower-priced varieties, predominantly via extensive margin (entry of lower-priced varieties and exit of higher-priced varieties).
  - A substantial portion of product-level price decline reflects reallocation toward lower-appeal (lower-quality) varieties.
  - Accounting for appeal, U.S. tariff hikes raise duty-inclusive import prices, with pass-through less than one-for-one.
- Welfare and productivity implications:
  - Substitution toward lower-priced varieties cushions nominal price effects but may cause quality downgrading.
  - Appeal-adjusted increases in duty-inclusive import prices imply direct consumer costs and potential indirect productivity costs (firms using lower-quality inputs may face reduced efficiency).
  - Appeal-erosion channel accounted for roughly half of estimated price reduction in 2025, indicating potentially significant welfare and productivity costs.
- Directions for further research:
  - Identify and quantify productivity consequences of importing lower-quality inputs induced by tariffs.
  - Further decompose welfare effects across consumer heterogeneity and firm-level input use.

### Appendix A — Selected robustness and decomposition results (highlights)
- Baseline level-decomposition 2SLS (Feb–Dec 2025):
  - Table A.2.1 Overall: ∆ ln(1 + eτ^US,eff_{g,t}) = -0.705*** (0.050).
  - Decomposition: Mean = -0.544*** (0.057); Cov = -0.162*** (0.052).
  - Textual share: Mean accounts for about 77 percent and Cov about 23 percent of overall tariff effect on import prices.
- OLS and reduced-form level decomposition (Table A.2.2):
  - OLS overall: -0.453*** (0.037); Mean = -0.302*** (0.041); Cov = -0.152*** (0.038).
  - OLS with statutory rates: -0.513*** (0.036) overall; alternate column -0.734*** (0.044) overall with Cov = 0.221*** (0.036).
- BEA End-Use heterogeneity (Table A.2.3) — selected 2SLS estimates (standard errors in parentheses):
  - BEA 0 Foods/feeds/beverages: Overall = -0.448*** (0.118); Mean = -0.294** (0.142); Cov = -0.154 (0.129).
  - BEA 1 Industrial supplies/materials: Overall = -0.580*** (0.060); Mean = -0.381*** (0.076); Cov = -0.200*** (0.065).
  - BEA 2 Capital goods excl. auto: Overall = -1.059*** (0.199); Mean = -1.236*** (0.205); Cov = 0.177 (0.178).
  - BEA 3 Auto vehicles/parts/engines: Overall = -1.130*** (0.447); Mean = -0.266 (0.452); Cov = -0.864** (0.424).
  - BEA 4 Consumer goods excl. food/auto: Overall = -1.063*** (0.121); Mean = -0.725*** (0.126); Cov = -0.338*** (0.127).
- Alternative fixed effects (Table A.2.4):
  - HS 8-digit product-time FE: Overall = -1.338*** (0.130); Mean = -1.016*** (0.123); Cov = -0.322*** (0.115).
  - HS 8-digit product FE & Time FE: Overall = -1.304*** (0.080); Mean = -0.918*** (0.091); Cov = -0.386*** (0.081).
- Quarterly specification (Table A.2.5; Q1–Q4 2025):
  - 2SLS quarterly: Overall = -0.470*** (0.062); Mean = -0.245*** (0.079); Cov = -0.225*** (0.074).
- Excluding AI-related products (Table A.2.6):
  - 2SLS: Overall = -0.711*** (0.051); Mean = -0.542*** (0.057); Cov = -0.169*** (0.052).
- HS 8-digit aggregation (Table A.2.7):
  - 2SLS (HS 8-digit): Overall = -0.657*** (0.062); Mean = -0.423*** (0.074); Cov = -0.235*** (0.069).
- Historical comparison (2018–19; Table A.2.8):
  - 2SLS: Overall = -1.123*** (0.105); Mean = -0.502*** (0.105); Cov = -0.621*** (0.095) — covariance term relatively more important than in 2025 episode.

*Italic: Source: wpiea2026149-source-pdf*

### 2.1    Data

### 2.1    Data

### Data sources and price measures
- Monthly U.S. import data: U.S. Census Bureau, reporting import values (in current US$), quantities, and duties by country of origin and HS 10-digit product for 2024-25.
- Import unit values:
  - FOB prices: values excluding collected duties (calculated as the ratio of the total customs value of imports for consumption to their physical volume).
  - Tariff-inclusive unit values: values including collected duties (duty-inclusive unit price adds the estimates of calculated duty to the numerator).
- Note: The unit-value approach is standard in the empirical literature but can mask variation in pricing strategies across firms exporting the same HS 10-digit product from a given country.

### Tariff measures used
- Effective tariff rates:
  - Computed as the ratio of calculated duties to the value of imports for consumption.
  - Reflect tariffs actually faced by importers at the border.
  - Can deviate from statutory rates due to implementation practices, exemptions not fully captured in statutory schedules, shipment lags, and behavioral responses (e.g., changes in use of preferential rates).
  - Undefined when the tariff is ‘prohibitive’ and import values fall to zero.
- Statutory tariff rates:
  - Defined as tariffs implemented by the U.S. administration through official documents (e.g., Federal Register).
  - Measure the trade policy policymakers intend to implement.

### Statutory tariff data and adjustments
- Statutory tariffs and time variation obtained from the WTO-IMF Tariff Tracker (WTO-IMF, 2026), which provides U.S. tariffs at the HS 8-digit level with daily dates of implementation.
- Two adjustments applied to better reflect prescribed implementation:
  - U.S. tariffs on Canada and Mexico scaled down proportionally to the observed USMCA preference utilization rate at the product level—updated with a lag of a few months—to proxy for USMCA eligibility and exemption from U.S. import tariffs. (As of February 2026, the USMCA preference utilization rates were calculated using U.S. import data from March to June 2025.)
  - Sectoral tariffs on steel and aluminum extended to derivative products for their steel or aluminum content using aggregated sectoral input-output coefficients to approximate steel and aluminum contents of downstream products.
- Daily statutory tariffs aggregated to monthly frequency by taking the modal tariff rate for each month.
- Baseline statutory tariffs for 2024 assigned to all months before February 2025.

### Preferred measures and instrumentation
- Effective tariff rate is the preferred tariff measure in regressions because it captures actual tariff barriers faced by importers and exporters.
- To address potential endogeneity of effective tariffs, the effective tariff is instrumented with the statutory tariff rate (τUS,sta), leveraging the statutory rate’s exogeneity with respect to trade responses.

### Data coverage and time span
- Analysis focuses on 2024-25 monthly data; the regression time unit spans February–December 2025 when U.S. tariffs were modified.

---

### 2.2    Stylized facts

- Import-weighted average U.S. statutory and effective tariff rates evolution:
  - Figure 1 (described): solid lines use time-varying weights; dotted lines use fixed 2024 weights.
  - For both statutory and effective rates, solid lines lie below dotted lines, reflecting reallocation of imports from higher-tariff to lower-tariff goods and countries.
  - Across weighting schemes, effective (blue) lines lie below statutory (red) lines, reflecting exemptions, implementation complexities, and lags.
- As of December 2025:
  - Average statutory rate using time-varying import weights was around 15.5 percent.
  - This was approximately 6 percentage points higher than the average effective rate.
  - The difference was negligible in January 2025 but widened as tariff actions were implemented.
  - Across countries, average effective tariffs are below statutory ones; the gap is particularly large for China (around 10 percentage points).
- Within-product (HS 8-digit) variation in statutory tariff changes:
  - For 2024-25 changes (relative to 2024), the median within-product standard deviation of tariff changes is around 8 percentage points (57 percent of the overall standard deviation).
  - The within-product variation for 2025 accounts for 26 percent of overall tariff variation in the 2025 episode; for 2018-19 the within-product share was around 98 percent.
  - The 2018-19 episode shows higher normalized within-product standard deviation than 2024-25 (tariffs in 2018-19 targeted mainly China).
- Aggregate U.S. import price index (Bureau of Labor Statistics) dynamics:
  - Year-over-year import price inflation for all imports rises through mid-2024, then declines thereafter despite new tariffs.
  - By mid-2025, headline import price inflation turns negative.
  - Import prices excluding food and fuels moderate but remain positive.
  - The combination of rising tariffs and slowing pre-tariff import price inflation suggests other mechanisms may be driving the decline in aggregate import prices.

---

### 3    Evidence from the 2025 U.S. tariffs

### Empirical approach and identification
- Levels of analysis:
  - Variety level: exporter (i) — HS 10-digit product (g) pair for price responses and reallocation across exporters.
  - Product level: examine how reallocation across varieties shapes tariff pass-through to import prices.
- Core empirical specification (variety level):
  - ∆ ln Y_ig,t = β ∆ ln(1 + τ_US,eff_ig,t) + μ_g,t + η_i + ε_ig,t
  - Y is import unit values (FOB: P ≡ V/Q; duty-inclusive: P* ≡ (V+T)/Q), import values (V), or quantities (Q).
  - τ_US,eff_ig,t is the U.S. effective tariff rate; variables measured in year-on-year log differences.
  - Time unit: month t spanning February–December 2025.
  - Fixed effects: HS 10-digit product-time μ_g,t and exporter η_i.
  - Standard errors clustered at HS 8-digit product level.
- Identification:
  - β identified from variation in import outcomes and tariff changes across exporters within an HS 10-digit product.
  - Instrumental variables: statutory tariff changes τ_US,sta_ig,t used as instruments for effective tariffs in 2SLS to address potential endogeneity.
  - Exclusion restriction: statutory tariffs affect trade outcomes only through effective tariffs.
  - First-stage shows a strong positive correlation between statutory and effective tariffs.

### Interpretation of coefficients
- When dependent variable is import unit values (FOB), β captures exporters’ price responses.
- In duty-inclusive price regressions, β = 1 implies full pass-through to import prices; β < 1 implies incomplete pass-through.
- Quantity and value regressions allow decomposition of effects into intensive-margin adjustments and price responses.

---

### 3.1    Trade and price effects at the variety level — Baseline results (Table 1)
- Estimation sample and details:
  - Year-on-year monthly log changes.
  - Sample includes changes from February to December 2025.
  - All columns include HS 10-digit product-time and country fixed effects.
  - Observations: 1,262,647 in each reported column.
  - First-stage F-stat: 2195.
- Table 1 key estimates (preserve estimates and standard errors as reported):
  - Column (1) OLS: ∆ ln(1 + τ_US,eff_ig,t) coefficient = -0.478*** (0.059).
  - Column (2) 2SLS (∆ ln V): coefficient = -0.478*** (0.080).
  - Column (3) 2SLS (∆ ln Q): coefficient = -0.0001 (0.054).
  - Column (4) 2SLS (∆ ln P): coefficient = 1.000*** (0.054).
  - Column (1) 2SLS first-stage: ∆ ln(1 + τ_US,sta_ig,t) coefficient = 0.393*** (0.008).
- Main findings:
  - U.S. tariff changes have a muted effect on exporters’ unit values (column (4) indicates no significant reduction in FOB unit values).
  - Duty-inclusive import prices show full pass-through of tariffs at the variety level (column (5): coefficient = 1.000***).
  - Significant negative effects of U.S. tariffs on imports:
    - The 8 percentage point increase in import-weighted average U.S. effective tariff between February and December 2025 would reduce imports by about 3.6 percent on average (trade elasticity in column (3) reported as -0.5 and interpreted as a short-run elasticity).
  - Trade elasticity comparisons:
    - The estimated short-run trade elasticity of -0.5 is close to recent estimates but smaller than estimates from the 2018-19 U.S.-China episode (e.g., about -1.3 in Amiti, Redding and Weinstein (2019)).
  - Robustness:
    - Alternative fixed-effects specifications (most restrictive adds country-time and country–HS 6-digit sector fixed effects) raise the estimated trade elasticity to -0.85 and slightly reduce tariff pass-through to import prices.
    - In the most restrictive specification, exporters’ unit-price effect is negative but statistically insignificant; duty-inclusive prices imply up to 18 percent of tariff increases are absorbed by variety-level exporters’ prices.
  - Interpretation of pattern:
    - Results indicate substantial trade reallocation across exporters in response to tariffs, with limited downward adjustment of exporters’ FOB prices.
    - The broader scope and greater uncertainty of the 2025 tariff episode may explain weaker reallocation relative to 2018-19.

### Extensive margin specification
- To capture entry and exit responses, estimate:
  - e_ig,t = β ∆ ln(1 + τ_US,sta_ig,t) + μ_g,t + η_i + ε_ig,t
  - e is an entry/exit dummy: entry = 1 if imported in t but not t−12; exit = 1 if not imported in t but imported in t−12.
  - Because effective rates are undefined for non-imported varieties, statutory tariff changes are used; regressions estimated via linear probability model (OLS reduced form).

---

### Extensive margin results (Table 2)
- Sample and estimation notes:
  - Entry and exit regressions use year-on-year monthly log changes in statutory tariffs.
  - Sample includes changes from February to December 2025.
  - Standard errors clustered at HS 8-digit product level.
- Reported estimates (columns indicate alternative fixed-effect sets; preserve reported coefficients and standard errors):
  - Columns (1)–(3) (samples include varieties imported at t−12):
    - Exit regressions: ∆ ln(1 + τ_US,sta_ig,t) coefficient = 0.107*** (0.006); 0.155*** (0.012); 0.065*** (0.009) across columns.
  - Columns (4)–(6) (samples include varieties not imported at t−12):
    - Entry regressions: ∆ ln(1 + τ_US,sta_ig,t) coefficient = -0.025*** (0.002); -0.036*** (0.003); -0.015*** (0.002) across columns.
  - Observations:
    - Exit columns: 1,723,217; 1,723,183; 1,690,529.
    - Entry columns: 29,255,790; 29,255,790; 29,241,264.
- Key extensive-margin findings:
  - Baseline exit effect (column (1)): The increase in average U.S. statutory tariffs between February and December 2025 (from about 2 to 15.5 percent) raises the probability of exiting the U.S. market by 1.4 percentage points, relative to an average exit probability of 26 percent.
  - Baseline entry effect (column (4)): The same tariff increase lowers the probability of entering the U.S. market by 0.3 percentage points, relative to a sample-average entry probability of 1.6 percent.
  - Interpretation: Tariff hikes are prohibitive for some varieties, inducing switching to alternative origins and delayed entry.

---

*Italic: Source: wpiea2026149-source-pdf - 2.1    Data*

### 3.2    Price effects at the product level

### 3.2    Price effects at the product level

### Methodology and decomposition approach
- Product-level log aggregate price approximation:
  - p_g,t ≡ ∑_i w^V_ig,t p_ig,t with value-share weights w^V_ig,t ≡ V_ig,t / V_g,t and p_ig,t ≡ ln P_ig,t.
- Year-on-year product-level import price inflation:
  - π_g,t ≡ ∆ p_g,t = ∑_i w^V_ig,t p_ig,t − ∑_i w^V_ig,t−12 p_ig,t−12.
- Log decomposition partitions varieties into continuing (C), entrants (E), and exiters (X), yielding five components:
  - Within: price changes among incumbent varieties, holding import shares fixed.
  - Between: changes in import shares among incumbents, evaluated at prices in t−12.
  - Interaction: covariance term capturing simultaneous changes in unit values and import shares among incumbents.
  - Entry: contribution of newly imported varieties relative to the product average in t−12.
  - Exit: contribution of exiting varieties relative to the product average in t−12.
- Product-level tariff growth measure:
  - ∆ ln(1 + τ^US,k_g,t) ≡ ∑_i w^V_ig,t0 ln(1 + τ^US,k_ig,t) − ∑_i w^V_ig,t0 ln(1 + τ^US,k_ig,t−12), with k∈{eff, sta} and 2024 import value weights w^V_ig,t0 rescaled to sum to one across observed varieties.
- Baseline regression (product level):
  - π_g,t = β ∆ ln(1 + τ^US,eff_g,t) + μ_t + ε_g,t, with the effective tariff instrumented by the statutory tariff.

### Key empirical findings (Table 3 summary)
- Main coefficient estimates (2SLS; standard errors clustered at HS 8-digit):
  - Overall: ∆ ln(1 + τ^US,eff_g,t) → −0.624*** (column (1))
  - Intensive margin: −0.223*** (column (2))
  - Extensive margin: −0.401*** (column (3))
  - Between (intensive decomposition): 0.086*** (column (4))
  - Interaction (intensive decomposition): −0.127*** (column (5))
  - Within (intensive decomposition): −0.183*** (column (6))
  - Entry (extensive decomposition): −0.257*** (column (7))
  - Exit (extensive decomposition): 0.144*** (column (8))
- Sample and diagnostics as reported:
  - Obs 171,206
  - F-stat 1st stage71587158715871587158715871587158
- Magnitude interpretation reported in the text:
  - The 2025 increase in the average ln(1 + τ^US,eff_g,t) (around 8 percentage points) is associated with an approximate 5 percent decline in before-tariff import prices.

### Decomposition and contribution shares
- Intensive vs. extensive margins:
  - The majority of the estimated negative product-level price response (around 65 percent) is driven by the extensive margin; the intensive margin accounts for the remaining ~35 percent.
- Within intensive decomposition:
  - Within-term coefficient (−0.183***) indicates continuing-variety prices decline in products with larger tariff hikes.
  - Between-term coefficient (0.086***) implies incumbent varieties gaining market share tended to be relatively higher-priced initially in products with larger tariff hikes, attenuating the within-channel price reduction.
  - Interaction term (−0.127***) captures joint changes in prices and shares among incumbents.
- Extensive decomposition:
  - Entry channel explains about 65 percent of the extensive-margin contribution (entry coefficient −0.257***): new varieties tend to be lower-priced.
  - Exit channel explains about 35 percent of the extensive-margin contribution (exit coefficient 0.144***): exiting varieties tend to be relatively higher-priced.
- Additional characterization:
  - An alternative level decomposition (Appendix Table A.2.1) shows most negative product-level price effects reflect a ‘mean’ effect—unweighted average prices decreasing with changes in product-level average tariffs—driven by greater entry of lower-priced varieties and exit of higher-priced varieties.
  - A negative effect on the covariance term indicates reallocation away from old relatively expensive varieties toward new relatively cheap varieties in products facing higher tariff hikes.
  - Reallocation among existing varieties has a smaller and, if anything, price-increasing effect (consistent with positive between coefficient).

### Robustness and heterogeneity checks
- Alternative fixed effects:
  - Replacing baseline time fixed effects with HS 8-digit and time fixed effects or HS 8-digit-time fixed effects yields qualitatively similar coefficients, if anything larger in magnitude (Table A.1.5).
- Quarterly data:
  - Aggregation to quarterly frequency produces broadly consistent results with coefficients about halved in magnitude (Table A.1.6), suggesting some monthly effects reflect high-frequency variation.
- Excluding AI-related products:
  - AI-related imports grew nearly 73 percent between 2023 and 2025 and faced relatively low effective tariffs; excluding these products does not materially change main results (Table A.1.7).
- Aggregation at HS 8-digit:
  - Using HS 8-digit product classification (level of statutory tariffs in the WTO-IMF Tariff Tracker) preserves the findings (Table A.1.8), indicating results are not driven by within-HS 8-digit compositional effects.

### Implications and interpretation
- Product-level import price inflation decreases with U.S. tariff changes: higher product-level tariff hikes are associated with lower product-level import price inflation.
- Mechanisms:
  - Extensive-margin adjustments (entry of lower-priced varieties and exit of higher-priced varieties) are the predominant channel for the negative pass-through from tariffs to product-level import prices.
  - Intensive-margin dynamics also matter: continuing-variety price declines occur, but reallocation toward incumbents that initially were higher-priced can offset some within-channel price reduction.
- Sectoral heterogeneity (summary from Appendix Table A.1.3):
  - Capital and consumer goods, including autos, exhibit larger product-level price declines in response to tariffs.
  - Foods and industrial goods show smaller negative responses.
  - The extensive margin is particularly important for consumer goods and autos.

_Italic: Source — IMF working paper section “3.2    Price effects at the product level” (wpiea2026149-source-pdf)._

### 3.3    Product quality and reallocation

### 3.3    Product quality and reallocation

### Framework and identification
- Assumes CES preferences within an aggregated product category g (HS 10-digit) over exporter–HS 10-digit varieties:
  - U_g = (∑_i (Θ_ig Q_ig,t)^{(σ_s−1)/σ_s})^{σ_s/(σ_s−1)}, σ_s > 1.
  - Θ_ig is exporter–HS 10-digit product appeal; Q_ig,t is total import quantity.
  - σ_s is elasticity of substitution associated with HS 6-digit category s containing HS 10-digit product g.
- Variety share (buyer’s prices, duty-inclusive notation with “∗”):
  - w^V_*ig,t = Θ_ig^{σ_s−1} P^*_{ig,t}^{1−σ_s} / Φ^*_{g,t}, with Φ^*_{g,t} ≡ ∑_i Θ_ig^{σ_s−1} P^*_{ig,t}^{1−σ_s}.
  - Log-linearized: ln w^V_*ig,t = (1−σ_s) ln P^*_{ig,t} + (σ_s−1) ln Θ_ig − ln Φ^*_{g,t}.
- Empirical estimable equation:
  - ln w^V_*ig,t + (σ_s−1) ln P^*_{ig,t} = η_ig + μ_{g,t} + ε_ig,t,
    where η_ig (exporter–HS 10-digit FE) identifies ln Θ_ig via ln Θ_ig = η̂_ig / (σ_s−1).
- σ_s obtained from trade elasticities in Fontagné, Guimbard and Orefice (2022); set σ = 1 − ε (their ε is negative elasticity of imports w.r.t. tariffs).
- Appeal is estimated assuming time-invariance over estimation window; estimated using 2023–2024 data (pre-2025 tariff hikes).

### Quality-adjusted price indices
- Appeal-adjusted CES price index (within HS 10-digit product g):
  - P^{CES}_{g,t} = (∑_i Θ_ig^{σ_s−1} P_{ig,t}^{1−σ_s})^{1/(1−σ_s)}.
- Non-appeal-adjusted CES price index:
  - P^{CES,na}_{g,t} = (∑_i P_{ig,t}^{1−σ_s})^{1/(1−σ_s)}.
- Analysis conducted with both duty-exclusive and duty-inclusive prices (P or P^*).
- Regressions: year-on-year monthly log changes in CES price index on log changes in (1 + weighted-average tariffs) e τ_{g,t}^{US,eff} ≡ ∑_i w^V_{ig,t}^0 τ^{US, k}_{ig,t}, controlling for time fixed effects.
  - CES index changes incorporate extensive-margin (entry/exit) and intensive-margin channels.

### Empirical evidence — 2025 U.S. tariffs (Table 4)
- Regression setup: Dependent variables are ∆ ln P^{CES}_{g,t} and ∆ ln P^{CES,na}_{g,t}; estimator 2SLS; sample includes changes from February to December 2025; standard errors clustered at HS 8-digit product level.
- Key coefficient estimates on ∆ ln(1 + e τ^{US,eff}_{g,t}):
  - Column (1) — Dep: ∆ ln P^{CES}_{g,t}, Appeal adjusted: Yes, Duty inclusive: No:
    - Coefficient: -0.357*** (0.074)
    - Obs: 96,165
    - F-stat 1st stage: 4428
  - Column (2) — Dep: ∆ ln P^{CES,na}_{g,t}, Appeal adjusted: No, Duty inclusive: No:
    - Coefficient: -0.642*** (0.064)
    - Obs: 96,165
    - F-stat 1st stage: 4428
  - Column (3) — Dep: ∆ ln P^{CES}_{g,t}, Appeal adjusted: Yes, Duty inclusive: Yes:
    - Coefficient: 0.375*** (0.074)
    - Obs: 96,002
    - F-stat 1st stage: 4429
  - Column (4) — Dep: ∆ ln P^{CES,na}_{g,t}, Appeal adjusted: No, Duty inclusive: Yes:
    - Coefficient: 0.059 (0.065)
    - Obs: 96,002
    - F-stat 1st stage: 4429
- Interpretation:
  - Duty-exclusive, appeal-adjusted index (col (1)): tariff increases associated with a statistically significant decline in quality-adjusted prices — i.e., before-tariff cost of imported goods fell even after accounting for appeal differences.
  - Non-appeal-adjusted index (col (2)): more negative coefficient, indicating that reallocation toward lower-appeal (lower-quality) varieties is an important channel; appeal-erosion accounts for roughly half of the estimated price reduction in 2025.
  - Duty-inclusive indices (cols (3) and (4)):
    - Without appeal adjustment (col (4)): positive and statistically insignificant coefficient — tariffs had a negligible effect on duty-inclusive product-level import prices.
    - With appeal adjustment (col (3)): positive and statistically significant coefficient — once accounting for appeal differences, tariffs raise duty-inclusive CES import prices, but increases are less than one-for-one with changes in average product-level tariffs.

### Comparison with 2018–19 U.S. tariffs (Tables 5 and 6)
- Table 5 — Product-level pass-through decomposition (2018–19):
  - Sample: May 2018 (relative to 2017) to December 2019; year-on-year monthly log changes.
  - Overall coefficient on ∆ ln(1 + τ^{US,eff}_{g,t}): -0.809*** (0.102), Obs: 271,980, F-stat 1st stage: 1330.
  - Intensive margin coefficient: 0.076 (0.053).
  - Extensive margin coefficient: -0.885*** (0.080).
  - Between, Interaction, Within, Entry, Exit columns show additional decomposition (selected example coefficients):
    - Entry: -0.669*** (0.074).
    - Exit: 0.216*** (0.018).
  - Interpretation: predominant role for the extensive margin in driving negative product-level price responses; entry concentrated among relatively lower-priced varieties and exit among relatively higher-priced varieties.
- Table 6 — Quality-adjusted price effects (2018–19):
  - Regression setup like Table 4; sample May 2018 to December 2019.
  - Key coefficient estimates on ∆ ln(1 + e τ^{US,eff}_{g,t}):
    - Column (1) — ∆ ln P^{CES}_{g,t}, Appeal adjusted: Yes, Duty inclusive: No:
      - Coefficient: -0.421*** (0.121)
      - Obs: 153,697
      - F-stat 1st stage: 3171
    - Column (2) — ∆ ln P^{CES,na}_{g,t}, Appeal adjusted: No, Duty inclusive: No:
      - Coefficient: -1.024*** (0.122)
      - Obs: 153,697
      - F-stat 1st stage: 3171
    - Column (3) — ∆ ln P^{CES}_{g,t}, Appeal adjusted: Yes, Duty inclusive: Yes:
      - Coefficient: 0.274** (0.127)
      - Obs: 153,604
      - F-stat 1st stage: 3163
    - Column (4) — ∆ ln P^{CES,na}_{g,t}, Appeal adjusted: No, Duty inclusive: Yes:
      - Coefficient: -0.267** (0.125)
      - Obs: 153,604
      - F-stat 1st stage: 3163
  - Interpretation:
    - Duty-exclusive, appeal-adjusted index (col (1)): higher tariffs associated with a genuine decline in quality-adjusted import prices at the product level.
    - Non-appeal-adjusted index (col (2)): much stronger negative coefficient — more than half of the tariff-induced decline driven by reallocation toward lower-appeal exporters.
    - Duty-inclusive appeal-adjusted index (col (3)): positive and statistically significant — appeal-adjusted, duty-inclusive CES import prices increased with tariff changes in 2018-19.
  - Conclusion: accounting for differences in quality and appeal is crucial for welfare/productivity assessment; results for 2018–19 mirror the role of reallocation and quality downgrading observed for 2025.

### Substantive findings and implications
- At the exporter–product (variety) level:
  - Exporters’ unit values did not respond to U.S. tariffs imposed during 2025 (consistent with literature).
  - Imports decrease significantly with U.S. tariff hikes.
- At the product (HS 10-digit) level:
  - Duty-exclusive import prices decrease significantly with tariff hikes due to reallocation toward lower-priced varieties.
  - This reallocation operates strongly along the extensive margin: higher tariffs → more entry of lower-priced varieties and greater exit of higher-priced varieties.
  - A substantial portion of the measured product-level price decline reflects reallocation toward lower-appeal (lower-quality) varieties rather than pure reallocation among varieties of the same appeal.
  - Once differences in appeal are accounted for, U.S. tariff hikes raise duty-inclusive import prices, though pass-through is less than one-for-one.
- Welfare and productivity implications:
  - Substitution toward lower-priced varieties cushions nominal price effects of tariffs but may entail quality downgrading.
  - Appeal-adjusted increases in duty-inclusive import prices imply direct consumer costs and potential indirect productivity costs (firms using lower-quality inputs may face reduced efficiency).
  - The appeal-erosion channel accounted for roughly half of the estimated price reduction in the 2025 episode, indicating potentially significant welfare and productivity costs.

### Directions for further research
- Identify and quantify the productivity consequences of importing lower-quality inputs induced by tariffs.
- Further decompose welfare effects across consumer heterogeneity and firm-level input use.

*Source: wpiea2026149-source-pdf - 3.3    Product quality and reallocation.*

### References

### wpiea2026149-source-pdf - References

### Appendix A: Empirical results and decomposition approach (major themes)
- Product-level import price defined as P_{g,t} = ∑_i V_{ig,t} / ∑_i Q_{ig,t} = ∑_i w^Q_{ig,t} P_{ig,t} with quantity-share weights w^Q_{ig,t} ≡ Q_{ig,t}/Q_{g,t} and Q_{g,t} = ∑_i Q_{ig,t}.
- Product-level tariff defined as eτ^US,k_{g,t} ≡ ∑_i w^V_{ig,t,0} τ^US,k_{ig,t}, with base year t_0 = 2024 and k ∈ {eff, sta}.
- Log decomposition: ∆ ln P_{g,t} = ∆ ln P̄_{g,t} + ∆ ln Z_{g,t}, where P̄_{g,t} is the simple mean across varieties and Z_{g,t} captures covariance between prices and quantity shares.

- Baseline level-decomposition 2SLS estimates (sample: changes from February to December 2025; time fixed effects; standard errors clustered at HS 8-digit product level):
  - Table A.2.1 (Dep. Variable: Overall): coefficient on ∆ ln(1 + eτ^US,eff_{g,t}) = -0.705*** (0.050).
  - Table A.2.1 decomposition: Mean term coefficient = -0.544*** (0.057); Cov term coefficient = -0.162*** (0.052).
  - The simple average (Mean) accounts for about 77 percent and the covariance (Cov) about 23 percent of the overall tariff effect on import prices (textual statement in Appendix A.2).

- Robustness and alternative estimators/fixed effects (preserve reported estimates exactly):
  - OLS and reduced-form level decomposition (Table A.2.2):
    - OLS: ∆ ln(1 + eτ^US,eff_{g,t}) = -0.453*** (0.037) overall; Mean = -0.302*** (0.041); Cov = -0.152*** (0.038).
    - OLS with statutory rates: ∆ ln(1 + eτ^US,sta_{g,t}) = -0.513*** (0.036) overall; ∆ ln(1 + eτ^US,sta_{g,t}) = -0.734*** (0.044) overall in alternate columns; corresponding Cov = 0.221*** (0.036) in that column.
  - BEA End-Use category heterogeneity (Table A.2.3) — representative 2SLS estimates (standard errors in parentheses):
    - BEA 0: Foods/feeds/beverages: Overall = -0.448*** (0.118); Mean = -0.294** (0.142); Cov = -0.154 (0.129).
    - BEA 1: Industrial supplies/materials: Overall = -0.580*** (0.060); Mean = -0.381*** (0.076); Cov = -0.200*** (0.065).
    - BEA 2: Capital goods excl. auto: Overall = -1.059*** (0.199); Mean = -1.236*** (0.205); Cov = 0.177 (0.178).
    - BEA 3: Auto vehicles/parts/engines: Overall = -1.130*** (0.447); Mean = -0.266 (0.452); Cov = -0.864** (0.424).
    - BEA 4: Consumer goods excl. food/auto: Overall = -1.063*** (0.121); Mean = -0.725*** (0.126); Cov = -0.338*** (0.127).
    - BEA 5: Other goods: Overall = -0.722 (1.537); Mean = -2.371* (1.291); Cov = 1.649 (1.120).
  - Alternative fixed effects (Table A.2.4):
    - With HS 8-digit product-time FE (columns 1-3): Overall = -1.338*** (0.130); Mean = -1.016*** (0.123); Cov = -0.322*** (0.115).
    - With HS 8-digit product FE & Time FE (columns 4-6): Overall = -1.304*** (0.080); Mean = -0.918*** (0.091); Cov = -0.386*** (0.081).
  - Quarterly specification (Table A.2.5; sample: Q1 to Q4 2025):
    - 2SLS quarterly: Overall = -0.470*** (0.062); Mean = -0.245*** (0.079); Cov = -0.225*** (0.074).
  - Excluding AI-related products (Table A.2.6; classification à la Waugh (2026)):
    - 2SLS: Overall = -0.711*** (0.051); Mean = -0.542*** (0.057); Cov = -0.169*** (0.052).
  - HS 8-digit aggregation (Table A.2.7):
    - 2SLS (HS 8-digit): Overall = -0.657*** (0.062); Mean = -0.423*** (0.074); Cov = -0.235*** (0.069).

- Historical comparison: 2018-19 tariffs episode (Table A.2.8; sample: May 2018 to December 2019):
  - 2SLS: Overall = -1.123*** (0.105); Mean = -0.502*** (0.105); Cov = -0.621*** (0.095).
  - Note: For 2018-19 the covariance term is relatively more important than in the 2025 episode.

- Econometric and sample notes (preserved exactly as stated):
  - Estimations use year-on-year monthly log changes in import and tariff variables (sample includes changes from February to December 2025, unless otherwise specified).
  - Time fixed effects included in baseline regressions.
  - Standard errors clustered at the HS 8-digit product level in parentheses (some tables note clustering at HS 6-digit level where specified).
  - Significance levels: *10 %, **5%, ***1%.
  - First-stage F-statistics reported in tables (exact values preserved in source tables).

### References (selected citations as listed in the source)
- Ahn, JaeBin, and Brandon Tan. 2025. “Supply chain diversification and resilience.” International Monetary Fund Washington, DC, USA.
- Alfaro, Laura, Mariya Brussevich, Camelia Minoiu, and Andrea F Presbitero. 2025. “Bank financing of global supply chains.” National Bureau of Economic Research.
- Amiti, Mary, Chris Flanagan, Sebastian Heise, and David E. Weinstein. 2026. “Who Is Paying for the 2025 U.S. Tariffs?” Liberty Street Economics. Liberty Street Economics Blog, February 12.
- Amiti, Mary, Stephen J Redding, and David E Weinstein. 2019. “The impact of the 2018 tariffs on prices and welfare.” Journal of Economic Perspectives, 33(4): 187–210.
- Fajgelbaum, Pablo D, and Amit Khandelwal. 2026. “Tariffs in 2025: Short-Run Impacts on the US Economy.” National Bureau of Economic Research.
- Gopinath, Gita, and Brent Neiman. 2026. “The Incidence of Tariffs: Rates and Reality.” National Bureau of Economic Research.
- Waugh, Michael E. 2026. “Trade in AI-Related Products.” National Bureau of Economic Research.
- WTO-IMF. 2026. “Methodology Note for the WTO-IMF Tariff Tracker.” World Trade Organization and International Monetary Fund.

*Tariff Pass-Through and Import Reallocation — Working Paper No. WP/2026/149*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026149-source-pdf.pdf_