## wpiea2026013-source-pdf

## Source details

**Canonical URL:** [wpiea2026013-source-pdf](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026013-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026013-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026013-source-pdf.pdf.json)

---

### Objective and scope
- Objective: understand how frontloading was achieved by identifying the channels that facilitated this adjustment over multiple months, and to contrast these dynamics with those observed during the 2018 US-China trade tensions.
- Scope exclusion: disentangle underlying mechanisms of export frontloading; not to quantify macroeconomic impacts, which lie beyond the scope of this analysis.

### Methodological approach: factor models and latent export baskets
- Core innovation: novel application of factor models to identify latent export baskets at the product level to capture dynamics of China’s export flows over time and across destinations.
- Conceptual distinction: endogenously identifies latent groupings of products that co-move across destinations, differing from pre-defined mappings of product categories to end uses.
- Model representation:
  - Exports at time t represented as X_t = Λ_t F_t + ε_t, where:
    - X_t is an H×N matrix of exports across HS2 categories and destinations at time t.
    - Λ_t is an H×P matrix of basket loadings mapping HS2 categories into P latent export baskets.
    - F_t is a P×N matrix of basket-level export quantities across destinations.
    - ε_t is an H×N matrix of idiosyncratic residuals.
  - Interpretation:
    - Loadings describe basket composition (weights across product categories).
    - Factor realizations capture scaling (total value of each basket exported to each destination).
  - Decomposition at any point in time:
    - Export value from a basket to a destination = (i) total value of that basket exported to all destinations (extensive margin) × (ii) allocation share of that basket assigned to that destination (intensive margin), with allocation shares summing to one.
- Empirical implementation:
  - Monthly Chinese customs data at the Harmonized System 2-digit (HS2) classification level.
  - Panel dimension: H×N×T with H = 92 HS2 product categories and N = 42 export destinations.
  - The monthly HS2 dataset is released about 15 days after the reference month.
  - A three-factor specification explains the bulk of cross-sectional variation in exports across HS2 categories.

### Key empirical findings: latent baskets and their roles
- Three latent baskets recovered; first two account for majority of China’s total export value and dynamics, third basket has relatively low and stable total value and allocation shares.
- Characterization of main baskets:
  - Basket 1: China’s general export basket — highest concentration of consumer and final goods; shipped to major advanced economies such as the US, EU, and Japan.
  - Basket 2: China’s secondary export basket — highest concentration of industrial supplies and intermediate goods used in manufacturing; primarily shipped to Vietnam, Korea, Japan, and other ASEAN economies.
  - Basket 3: relatively low and stable total value; concentrated shipments to Hong Kong SAR (and Singapore to a lesser extent).

### Timeline and dynamics of the 2024–25 frontloading episode
- Second half of 2024:
  - Shipments of basket 2 declined relative to the first half of the year.
  - Shipments of basket 1 accelerated.
  - Shift coincided with ramp-up in domestic production and possible diversion from third destinations that share demand for intermediate inputs.
  - Vietnam saw an increase in allocations of basket 2, largely offsetting the overall decline in intermediate exports along the extensive margin.
  - Exporters began prioritizing shipments of basket 1 to the US (intensive margin), and this US prioritization persisted through March 2025 despite initial phases of US-China tariff rate increases.
- January 2025:
  - Total shipment value of basket 1 declined sharply; timing consistent with typical duration of sea-freight shipments (exports departing in January would no longer reach the US before the start of the new administration).
  - Shipments of basket 2 accelerated to Vietnam and, to a lesser extent, other ASEAN economies — interpreted as relocation of production activity from China to Vietnam and more broadly across ASEAN.
- After sharp escalation of reciprocal tariffs in April:
  - Allocations of basket 1 to the US dropped steeply; allocations to other destinations rose and offset much of the shortfall.
  - Non-US destinations’ ability to absorb decreased US allocations related to deficits accumulated during the frontloading phase; largest increases in allocations occurred in destinations whose import profiles from China closely resembled that of the US and that had experienced reduced allocations during frontloading.
  - Evidence of backloading to third destinations complemented initial frontloading to the US and later facilitated absorption during tariff implementation — described as cross-destination intertemporal reallocation in export shipments.

### Comparison with the 2018 episode
- 2018 episode did not exhibit similar adjustments:
  - No ramp-up of domestic production observed in 2018.
  - No subsequent redirection of intermediate inputs to ASEAN economies like in 2024–25.
- Interpretation: producers and exporters may have learned from 2018 and invested in strategies (logistical preparation and production relocation capability) to respond to a renewed escalation in trade tensions.

### Evidence and interpretation around production relocation vs transshipment
- Production relocation hypothesis is economically distinct from simple transshipment or re-exporting.
- While aggregate trade data can make them appear similar, product-level export flows can disentangle these hypotheses.
- Additional evidence presented supports the production relocation hypothesis and suggests the recent increase in exports to Asia is likely to be sustained over the medium term.

### Data choice and trade-offs
- Data source: China’s General Administration of Customs (GACC) monthly export and import values disaggregated by HS2 product categories and destination economies.
- Trade-off motivating HS2 choice:
  - Higher-granularity flows are available but released with substantial delays or at lower frequency.
  - Monthly HS2 dataset offers a middle ground: timeliness (about 15 days after reference month) and meaningful product-destination detail, though cannot fully resolve within-category distinctions.

### Estimation procedure (3.3 Estimation Procedure)
- Estimation of basket compositions:
  - Input data: 12-month rolling average of the export-flow matrix at time t, defined as X̄_t = 1/12 ∑_{i=0}^{11} X_{t−i}.
  - Estimator: Non-negative Matrix Factorization (NMF): ˆΛ_t = NMF_P(X̄_t) with X̄_t ≈ ˆΛ_t  F̄′_t.
  - Implementation uses the NMFroutine from the scikit-learn package.
- Estimation of basket intensities:
  - Second-step estimator: Non-Negative Regression (NNR): ˆF_t = NNR(X_t | ˆΛ_t) with X_t ≈ ˆΛ_t ˆF′_t + e_t.
  - Implementation uses the nnlssolver from the scipy.optimize module.
- Decomposition of basket intensities:
  - ˆF_t = diag(ˆq_t) ˆΘ_t where ˆq_t = ˆF′_t 1 and ˆΘ_t = diag(ˆq_t)^{−1} ˆF_t.
  - ˆq_t captures the extensive margin; ˆΘ_t captures the intensive margin.
- Robustness note:
  - PCA recovers a similar structure and spans the same subspace but allows negative weights which complicate interpretation.

### Baseline estimation (June 2024) and basket compositions
- Baseline period: June 2024 (12-month average ending in June 2024 used as baseline).
- Model fit:
  - Overall R^2 of the factor model, averaged across HS2 categories and destinations, is 0.79 (leaving 21 percent of the variation unexplained).
- Three distinct export baskets identified; normalization sets estimated weights to range from zero to one.
- Baseline allocation highlights:
  - Basket 1: largest recipient US, followed by Germany, the United Kingdom, Japan, and the Netherlands.
  - Basket 2: largest recipient Vietnam, with other ASEAN economies significant.
  - Basket 3: shipped overwhelmingly to Hong Kong SAR, Singapore a distant second.
- BEC composition of baskets (exact values preserved)
  - Food & beverages — Basket 1: 0.110 — Basket 2: 0.160 — Basket 3: 0.490
  - Industrial supplies — Basket 1: 0.420 — Basket 2: 0.710 — Basket 3: 0.250
  - Fuels & lubricants — Basket 1: 0.004 — Basket 2: 0.005 — Basket 3: 0.019
  - Capital goods — Basket 1: 0.060 — Basket 2: 0.030 — Basket 3: 0.060
  - Transport equipment — Basket 1: 0.040 — Basket 2: 0.010 — Basket 3: 0.090
  - Consumer goods — Basket 1: 0.370 — Basket 2: 0.080 — Basket 3: 0.090
  - Goods n.e.s. — Basket 1: 0.004 — Basket 2: 0.001 — Basket 3: 0.012
- Model fit by HS2:
  - Explanatory power high across most HS2 categories; exceptions concentrated in low-value export segments (example: HS2 category 43 represents only 0.03% of China’s total export value and shows relatively lower model fit).

### Full-sample estimation (2016 — May 2025) — allocation and extensive-margin dynamics
- Sample period: 2016 through May 2025.
- Allocation-share dynamics:
  - Basket 1:
    - US remains dominant destination.
    - US share fell with onset of first US–China trade tensions in 2018, rebounded in 2021–2022, trended downward since 2023.
    - EU’s combined allocation share comparable in scale to US.
  - Basket 2:
    - Allocation shares more evenly distributed across destinations.
    - Sustained increase in Vietnam’s share since 2018.
  - Basket 3:
    - Remains heavily concentrated in Hong Kong SAR, declined steadily since 2020; Singapore increased share modestly.
- Extensive-margin dynamics:
  - Basket 1:
    - Dominant component of China’s export flows and principal driver of China’s export growth in 2024.
    - Positive comovement with domestic production indicators.
    - Correlation statistics:
      - Basket 1 vs 2 total value growth 01/2023 - 05/2025: Corr = -0.346, p = 0.066
      - Basket 1 total value vs IP growth 01/2023 - 05/2025: Corr = 0.348, p = 0.064
  - Basket 2:
    - Increasing share of export growth since 2018, accelerated during COVID-19 when domestic production contracted.
    - Some negative monthly comovement observed between shipments of basket 1 and basket 2.
  - Basket 3:
    - Total nominal value relatively stable over time.

### The anatomy of frontloading — margin decomposition and quantitative contributions
- Deviation metric for frontloading analysis:
  - ∆F_t = F_t − F̄, where F̄ is the pre-frontloading baseline given by the 12-month average ending in June 2024.
- Margin decomposition definitions:
  - ExtMargin_t = diag(q_t − q̄) θ̄
  - IntMargin_t = diag(q̄) (θ_t − θ̄)
  - Interaction_t = diag(q_t − q̄) (θ_t − θ̄)
  - Approximation used (interaction term omitted): ∆F_t ≈ ExtMargin_t + IntMargin_t
- Average absolute contributions (application-specific summary):
  - Intensive margin: roughly 70 percent of the total change.
  - Extensive margin: roughly 27 percent of the total change.
  - Interaction term: roughly 2 percent of the total change.

### Adjustment along the extensive margin (frontloading episode)
- Time patterns (Q3 2024 — early 2025):
  - Basket 1:
    - Sustained acceleration from Q3 2024, peaking in December 2024, then reversing in early 2025.
    - Extensive-margin frontloading phase ends in January 2025.
    - Sharp decline in February 2025 coinciding with start of new US administration and Lunar New Year.
    - Gradual recovery through May 2025 despite escalation of reciprocal tariffs.
  - Basket 2:
    - Sharp contraction in September 2024, recovery in late 2024, surge in early 2025.
    - Surge timing coincides with decline in basket 1 shipments, consistent with input allocation hypothesis.
- Interpretation: dynamics consistent with deceleration in Chinese domestic production and reallocation toward Vietnam and other ASEAN economies; rapidly adapting supply chains during China’s export frontloading.

### Adjustment along the intensive margin (frontloading episode)
- Allocation-share driven adjustments:
  - Basket 1:
    - Prioritization of US-bound shipments begins in July 2024 and persists through late 2024.
    - Reallocation along intensive margin largely offset declines on the extensive margin during early tariff escalation.
    - By April 2025 allocation shares for US-bound shipments in basket 1 drop sharply.
  - Basket 2:
    - Reallocation toward Vietnam evident in September 2024, continues through November–December 2024.
    - Pace slows in January–February 2025, accelerates again from March 2025 onward.
- Interpretation: intensive margin adjustments underscore exporters’ and producers’ flexibility to dynamically reallocate shipments in response to shifting trade policy expectations.

### Cross-destination intertemporal reallocation of shipments (4.3)
- Deviations in allocation shares across frontloading phases:
  - Analysis compares pre-tariff frontloading phase (10/2024–12/2024) with post-tariff phase (01/2025–05/2025) relative to a linear trend estimated from 2023 onward.
  - Basket 1:
    - Strong negative relationship between deviations in allocation shares in the pre-tariff phase and the post-tariff period.
    - Spearman rank corr = -0.82.
    - The US received larger allocation shares in the pre-tariff phase and correspondingly lower shares in the post-tariff period, especially in April and May.
  - Basket 2:
    - Relationship weaker: Spearman rank corr = -0.64.
- Mechanism:
  - Economies facilitated US frontloading by receiving lower allocation shares during pre-tariff phase and then higher allocation shares after tariff implementation to replenish inventories and absorb excess capacity.
  - Germany, the United Kingdom, Japan, and the Netherlands exhibited adjustments opposite to the United States.
- Basket-specific constraints:
  - Buffer mechanism primarily relevant for basket 1 (final goods).
  - For basket 2 (intermediate goods), less evidence of cross-destination intertemporal reallocation—possible reasons include immediate processing needs and stronger competition from alternative suppliers.
- Capacity and inventory considerations:
  - Indirect evidence suggests buffers (commercial inventories, household-level inventories, durable vs non-durable consumption distinctions) enabled flexibility to support frontloading; direct investigation of cross-destination inventory dynamics is left for future research.
- Manufacturing capacity utilization (reported):
  - 74.5 percent in 24H1
  - 75.8 percent in 24H2
  - 74.2 percent in 25H1
  - Peak of 78.2 percent in 21H1
- Interpretation: manufacturing slack decreased during the period most relevant for export frontloading, consistent with intertemporal reallocation hypothesis, although overall utilization remained below its 21H1 peak.

### Value added, tariff mitigation incentives, and policy implications (5.3)
- Aggregate indicators and comovement:
  - Figure 15 indices normalized to 100 in 2018 show:
    - Marked acceleration in Vietnam’s manufacturing value added in the first half of 2025, with growth reaching 20% on a quarter-on-quarter annualized basis.
    - Robust positive comovement between exports and manufacturing value added, inconsistent with a pure transshipment narrative.
  - Comparative dynamics since 2018:
    - Exports and imports increased proportionally since 2018.
    - Manufacturing value added has grown by a smaller margin, implying a decline in the marginal domestic value added per unit of exports.
    - Decline in marginal domestic value added per unit of exports appears to have accelerated in the first half of 2025.
- Value added continuum and relocation incentives:
  - Value added ranges between end-stage manufacturing (typically around one-fifth of final value) and zero value-added transshipment.
  - When share of value added is too low, cost savings unlikely to justify investment in infrastructure required to relocate production to Vietnam.
  - Observed positive comovement and increased manufacturing VA are suggestive of meaningful product transformation rather than pure transshipment.
- Tariff wedge effects:
  - Opening of a tariff rate wedge lowers profitability threshold for relocation, making previously too-low-margin activities viable to relocate to Vietnam.
  - Expectation of expansion of Vietnam’s end-stage manufacturing, especially at the margin of lower value-added activities.
- Scenario on tariffs targeting transshipment:
  - If recent export strength to Vietnam were driven primarily by transshipment, a US tariff specifically targeting such activity could trigger a sharp reversal.
  - Authors view such a sharp reversal as unlikely given evidence pointing to meaningful product transformation in Vietnam.
- Policy implications and suggested research directions:
  - Evidence tentative that increased shipments from China to Vietnam in 2025 reflect acceleration of production relocation rather than pure transshipment.
  - A widening tariff differential is likely to accelerate production relocation.
  - Suggested research: firm-level analysis of FDI and input–output linkages; role of vertical integration; cross-destination inventory adjustments; dynamic modeling of supply chain responses; joint modeling of export and import flows across economies.

*Source: wpiea2026013-source-pdf*

### 6.6 billion USD higher than a year earlier, an increase of 15.6 percent, and a notable acceleration

### 6.6 billion USD higher than a year earlier, an increase of 15.6 percent, and a notable acceleration

### Objective and scope
- Objective: understand how frontloading was achieved by identifying the channels that facilitated this adjustment over multiple months, and to contrast these dynamics with those observed during the 2018 US-China trade tensions.
- Scope exclusion: disentangle underlying mechanisms of export frontloading; not to quantify macroeconomic impacts, which lie beyond the scope of this analysis.

### Methodological approach: factor models and latent export baskets
- Core innovation: novel application of factor models to identify latent export baskets at the product level to capture dynamics of China’s export flows over time and across destinations.
- Conceptual distinction: endogenously identifies latent groupings of products that co-move across destinations, differing from pre-defined mappings of product categories to end uses.
- Model representation:
  - Exports at time t represented as X_t = Λ_t F_t + ε_t, where:
    - X_t is an H×N matrix of exports across HS2 categories and destinations at time t.
    - Λ_t is an H×P matrix of basket loadings mapping HS2 categories into P latent export baskets.
    - F_t is a P×N matrix of basket-level export quantities across destinations.
    - ε_t is an H×N matrix of idiosyncratic residuals.
  - Interpretation:
    - Loadings describe basket composition (weights across product categories).
    - Factor realizations capture scaling (total value of each basket exported to each destination).
  - Decomposition at any point in time:
    - Export value from a basket to a destination = (i) total value of that basket exported to all destinations (extensive margin) × (ii) allocation share of that basket assigned to that destination (intensive margin), with allocation shares summing to one.
- Empirical implementation:
  - Monthly Chinese customs data at the Harmonized System 2-digit (HS2) classification level.
  - Panel dimension: H×N×T with H = 92 HS2 product categories and N = 42 export destinations.
  - The monthly HS2 dataset is released about 15 days after the reference month.
  - A three-factor specification explains the bulk of cross-sectional variation in exports across HS2 categories.

### Key empirical findings: latent baskets and their roles
- Three latent baskets recovered; first two account for majority of China’s total export value and dynamics, third basket has relatively low and stable total value and allocation shares.
- Characterization of main baskets:
  - Basket 1: China’s general export basket — highest concentration of consumer and final goods; shipped to major advanced economies such as the US, EU, and Japan.
  - Basket 2: China’s secondary export basket — highest concentration of industrial supplies and intermediate goods used in manufacturing; primarily shipped to Vietnam, Korea, Japan, and other ASEAN economies.

### Timeline and dynamics of the 2024–25 frontloading episode
- Second half of 2024:
  - Shipments of basket 2 declined relative to the first half of the year.
  - Shipments of basket 1 accelerated.
  - Shift coincided with ramp-up in domestic production and possible diversion from third destinations that share demand for intermediate inputs.
  - Vietnam saw an increase in allocations of basket 2, largely offsetting the overall decline in intermediate exports along the extensive margin.
  - Exporters began prioritizing shipments of basket 1 to the US (intensive margin), and this US prioritization persisted through March 2025 despite initial phases of US-China tariff rate increases.
- January 2025:
  - Total shipment value of basket 1 declined sharply; timing consistent with typical duration of sea-freight shipments (exports departing in January would no longer reach the US before the start of the new administration).
  - Shipments of basket 2 accelerated to Vietnam and, to a lesser extent, other ASEAN economies — interpreted as relocation of production activity from China to Vietnam and more broadly across ASEAN.
- After sharp escalation of reciprocal tariffs in April:
  - Allocations of basket 1 to the US dropped steeply; allocations to other destinations rose and offset much of the shortfall.
  - Non-US destinations’ ability to absorb decreased US allocations related to deficits accumulated during the frontloading phase; largest increases in allocations occurred in destinations whose import profiles from China closely resembled that of the US and that had experienced reduced allocations during frontloading.
  - Evidence of backloading to third destinations complemented initial frontloading to the US and later facilitated absorption during tariff implementation — described as cross-destination intertemporal reallocation in export shipments.

### Comparison with the 2018 episode
- 2018 episode did not exhibit similar adjustments:
  - No ramp-up of domestic production observed in 2018.
  - No subsequent redirection of intermediate inputs to ASEAN economies like in 2024–25.
- Interpretation: producers and exporters may have learned from 2018 and invested in strategies (logistical preparation and production relocation capability) to respond to a renewed escalation in trade tensions.

### Evidence and interpretation around production relocation vs transshipment
- Production relocation hypothesis is economically distinct from simple transshipment or re-exporting.
- While aggregate trade data can make them appear similar, product-level export flows can disentangle these hypotheses.
- Additional evidence presented in the paper supports the production relocation hypothesis and suggests the recent increase in exports to Asia is likely to be sustained over the medium term.

### Broader implications and mechanisms
- Supply-chain adaptability:
  - China’s vertical integration along the value-added chain may have facilitated reallocation of intermediate goods to support domestic production in late 2024.
  - Horizontal integration across economies may mitigate trade tariffs.
- Incentives for relocation:
  - Relocation concentrated in sectors benefiting from lower labor costs.
  - Emergence of a tariff-rate wedge between China and ASEAN economies provided cost-saving incentives and may explain accelerated relocation.
  - Firms face incentives to invest preemptively in external production capacity as a hedge against future trade policy risk because such investments typically require multiple years to materialize.
- Policy-relevant question: sustainability of China’s exports to Asia if the US imposes a tariff specifically targeting transshipments.

### Data choice and trade-offs
- Data source: China’s General Administration of Customs (GACC) monthly export and import values disaggregated by HS2 product categories and destination economies.
- Trade-off motivating HS2 choice:
  - Higher-granularity flows are available but released with substantial delays or at lower frequency.
  - Monthly HS2 dataset offers a middle ground: timeliness (about 15 days after reference month) and meaningful product-destination detail, though cannot fully resolve within-category distinctions.

*Source: wpiea2026013-source-pdf*

### 3.3    Estimation Procedure

### 3.3    Estimation Procedure

### Estimation of Basket Compositions
- Input data: 12-month rolling average of the export-flow matrix at time t, defined as
  - X̄_t = 1/12 ∑_{i=0}^{11} X_{t−i}
  - Purpose: smooth monthly volatility and residual seasonality to capture stable patterns in export composition.
- Estimator: Non-negative Matrix Factorization (NMF)
  - ˆΛ_t = NMF_P(X̄_t)
  - Decomposition relation: X̄_t ≈ ˆΛ_t  F̄′_t
  - Interpretation:
    - ˆΛ_t: estimated basket loadings (contribution of each HS category to each basket).
    - F̄_t: corresponding basket intensities associated with the 12-month rolling average at time t.
- Implementation note:
  - NMF imposes non-negativity on factors and loadings (see Lee & Seung (1999)); implementation uses the NMFroutine from the scikit-learn package (as described in source).

### Estimation of Basket Intensities
- Second-step estimator: Non-Negative Regression (NNR)
  - ˆF_t = NNR(X_t | ˆΛ_t)
  - Reconstruction relation: X_t ≈ ˆΛ_t ˆF′_t + e_t
  - ˆF_t: estimated monthly export intensities for each basket across destinations at time t.
  - e_t: matrix of model errors.
- Interpretation and robustness:
  - Non-negativity in both ˆΛ_t and ˆF_t aids economic interpretability (each basket is a physically plausible grouping of HS2 flows).
  - Results are robust to choice of factor model: PCA recovers a similar structure and spans the same subspace of variation in the HS2-destination panel, but PCA allows negative weights which complicate interpretation.
- Implementation note:
  - NNR solves the OLS least-squares problem with non-negativity constraints; implemented using the nnlssolver from the scipy.optimize module (per source).

### Decomposition of Basket Intensities
- Decomposition: ˆF_t = diag(ˆq_t) ˆΘ_t
  - ˆq_t: vector of total basket shipments (scale of each basket’s total export intensity).
  - ˆΘ_t: matrix of normalized destination profiles (allocation shares across destinations).
- Definitions:
  - ˆq_t = ˆF′_t 1, where 1 is a column vector of ones (column-wise sum of ˆF_t).
  - ˆΘ_t = diag(ˆq_t)^{−1} ˆF_t so each column of ˆΘ_t sums to one.
- Economic interpretation:
  - ˆq_t captures the extensive margin (overall magnitude of each basket’s exports).
  - ˆΘ_t captures the intensive margin (distribution of those exports across destinations).

### 3.4    Baseline Estimation (June 2024)
- Baseline period: June 2024 (12-month average ending in June 2024 used as baseline).
- Model fit:
  - Overall R^2 of the factor model, averaged across HS2 categories and destinations, is 0.79 (leaving 21 percent of the variation unexplained).
- Factor solution:
  - Three distinct export baskets identified; three baskets sufficient to explain the majority of variation in the sample.
  - Normalization: cross-destination variation in export flows within each HS2 category standardized so estimated weights range from zero to one.
- Baseline allocation highlights:
  - Basket 1: largest recipient US, followed by Germany, the United Kingdom, Japan, and the Netherlands.
  - Basket 2: largest recipient Vietnam, with other ASEAN economies significant.
  - Basket 3: shipped overwhelmingly to Hong Kong SAR, Singapore a distant second.
- BEC composition of baskets (Table 1 exact values preserved)
  - BEC Category — Basket 1 — Basket 2 — Basket 3
    - Food & beverages — 0.110 — 0.160 — 0.490
    - Industrial supplies — 0.420 — 0.710 — 0.250
    - Fuels & lubricants — 0.004 — 0.005 — 0.019
    - Capital goods — 0.060 — 0.030 — 0.060
    - Transport equipment — 0.040 — 0.010 — 0.090
    - Consumer goods — 0.370 — 0.080 — 0.090
    - Goods n.e.s. — 0.004 — 0.001 — 0.012
- Descriptions:
  - Basket 1: China’s general export basket, highest relative share of consumer goods, shipped primarily to advanced economies.
  - Basket 2: secondary export basket, greater weight on materials and components used in manufacturing, shipped primarily to ASEAN economies.
  - Basket 3: highest relative weight in food and beverages, fuels and lubricants, and transport equipment; reflects goods typically shipped to Hong Kong SAR.
- Model fit by HS2:
  - Explanatory power high across most HS2 categories; exceptions concentrated in low-value export segments.
  - Example: HS2 category 43 represents only 0.03% of China’s total export value and shows relatively lower model fit.

### 3.5    Full-Sample Estimation (2016 — May 2025)
- Sample period: 2016 through May 2025.
- Allocation-share dynamics (Figure 5 observations):
  - Basket 1:
    - US remains dominant destination.
    - US share fell with onset of first US–China trade tensions in 2018, rebounded in 2021–2022, trended downward since 2023.
    - EU’s combined allocation share comparable in scale to US (dataset disaggregates EU into member economies).
  - Basket 2:
    - Allocation shares more evenly distributed across destinations.
    - Sustained increase in Vietnam’s share since 2018 (consistent with relocation of Chinese manufacturing to Vietnam).
  - Basket 3:
    - Remains heavily concentrated in Hong Kong SAR, declined steadily since 2020 after US reclassification of Hong Kong SAR.
    - Singapore increased share as a re-routing export hub; magnitude of shifts modest relative to baskets 1 and 2.
- Extensive-margin dynamics (Figure 6 and Figure 7 observations):
  - Basket 1:
    - Remains the dominant component of China’s export flows and principal driver of China’s export growth in 2024.
    - High-frequency dynamics exhibit positive comovement with domestic production indicators (example: industrial production growth; correlation reported in Figure 7).
    - Correlation statistics in figure captions (preserved):
      - Basket 1 vs 2 total value growth 01/2023 - 05/2025: Corr = -0.346, p = 0.066
      - Basket 1 total value vs IP growth 01/2023 - 05/2025: Corr = 0.348, p = 0.064
  - Basket 2:
    - Increasing share of export growth since 2018, accelerated during COVID-19 when domestic production contracted.
    - Some negative monthly comovement observed between shipments of basket 1 and basket 2 (consistent with input allocation hypothesis).
  - Basket 3:
    - Total nominal value relatively stable over time; not a major driver of recent export growth.
- Full-sample figures:
  - Figure 6: Total Export Value by Basket from 01-2016 to 05-2025 (shows monthly export values in US$ million by basket).
  - Figure 7: Growth rates and correlations between baskets and industrial production (mom % series and correlations as above).

### 4    The Anatomy of Frontloading — Methodological Setup
- Deviation metric for frontloading analysis:
  - ∆F_t = F_t − F̄, where F̄ is the pre-frontloading baseline given by the 12-month average ending in June 2024.
- Margin decomposition:
  - Decompose change in export intensities ˆF_t into:
    - Extensive margin: changes in total value exported for each basket (holding allocation shares constant).
    - Intensive margin: changes from reallocation of trade across destinations (conditional on fixed basket export values).
    - Interaction term: second-order effects from simultaneous changes in basket sizes and allocation shares.
  - Formal definitions:
    - ExtMargin_t = diag(q_t − q̄) θ̄
    - IntMargin_t = diag(q̄) (θ_t − θ̄)
    - Interaction_t = diag(q_t − q̄) (θ_t − θ̄)
    - q_t: vector of total export values by basket at time t.
    - θ_t: matrix of allocation shares across destinations.
    - q̄ and θ̄: baseline values.
  - Approximation used (interaction term omitted as second-order):
    - ∆F_t ≈ ExtMargin_t + IntMargin_t
  - Average absolute contributions (application-specific summary):
    - Intensive margin: roughly 70 percent of the total change.
    - Extensive margin: roughly 27 percent of the total change.
    - Interaction term: roughly 2 percent of the total change.

### 4.1    Adjustment Along the Extensive Margin (Frontloading episode)
- Observed time patterns (third quarter 2024 — early 2025):
  - Basket 1:
    - Sustained acceleration from Q3 2024, peaking in December 2024, then reversing in early 2025.
    - Extensive-margin frontloading phase ends in January 2025.
    - Sharp decline in February 2025 coinciding with start of new US administration and Lunar New Year.
    - Gradual recovery through May 2025 despite escalation of reciprocal tariffs.
    - Note: third quarter 2024 pattern may reflect precautionary frontloading ahead of potential tariff escalation.
  - Basket 2:
    - Sharp contraction in September 2024, recovery in late 2024, surge in early 2025.
    - Surge timing coincides with decline in basket 1 shipments, consistent with input allocation hypothesis (reallocation toward external manufacturing hubs).
- Interpretation:
  - Dynamics consistent with deceleration in Chinese domestic production and reallocation toward Vietnam and other ASEAN economies.
  - Aggregate narrative: rapidly adapting supply chains during China’s export frontloading.
- Figure 8: Frontloading Along the Extensive Margin (changes in US$ mn by destination and rest of world).

### 4.2    Adjustment Along the Intensive Margin (Frontloading episode)
- Observed time patterns (allocation-share driven adjustments):
  - Basket 1:
    - Prioritization of US-bound shipments throughout second half of 2024.
    - Allocation shares begin shifting decisively toward the United States in July 2024.
    - Reallocation along intensive margin largely offset declines on the extensive margin during early tariff escalation.
    - By April 2025 allocation shares for US-bound shipments in basket 1 drop sharply (effects of tariff escalation fully evident).
  - Basket 2:
    - Reallocation toward Vietnam becomes evident in September 2024 and continues through November–December 2024.
    - Pace of reallocation slows in January–February 2025 (initial implementation of tariff increases), accelerates again from March 2025 onward amid rising trade tensions.
- Interpretation:
  - Intensive margin adjustments underscore exporters’ and producers’ flexibility to dynamically reallocate shipments in response to shifting trade policy expectations.
  - Prioritization of US-bound exports helped offset value declines along the extensive margin in the first three months of 2025.
- Figure 9: Frontloading Along the Intensive Margin (changes in US$ mn by destination and rest of world).

*Source: wpiea2026013-source-pdf - 3.3    Estimation Procedure*

### 4.3    Cross-Destination Intertemporal Reallocation of Shipments

### 4.3    Cross-Destination Intertemporal Reallocation of Shipments

### Deviations in allocation shares across frontloading phases
- Analysis focuses on deviations in allocation shares relative to a linear trend estimated from the sample starting in 2023, comparing the pre-tariff frontloading phase (10/2024–12/2024) with the post-tariff phase (01/2025–05/2025).
- For basket 1 (general goods):
  - Strong negative relationship between deviations in allocation shares in the pre-tariff frontloading phase and the post-tariff period (01/2025–05/2025).
  - Spearman rank corr = -0.82.
  - The US received larger allocation shares in the pre-tariff phase and correspondingly lower shares in the post-tariff period, especially in April and May.
- For basket 2 (intermediate goods):
  - Relationship is weaker and less robust.
  - Spearman rank corr = -0.64.

- Spearman rank correlation coefficient definition provided: "The Spearman rank correlation coefficient is defined as the Pearson correlation between the ranked values of two variables. It measures the strength and direction of a monotonic relationship and is more robust to outliers and non-normality than the Pearson correlation."

### Cross-destination intertemporal adjustment mechanism
- Evidence indicates a pervasive negative relationship across destinations consistent with cross-destination intertemporal reallocation:
  - Economies facilitated US frontloading by receiving a lower allocation share during the pre-tariff phase and then received a higher allocation share after tariff implementation to replenish depleted inventories and absorb excess capacity from reduced US demand.
- The next four economies with the largest allocation shares of the general export basket—Germany, the United Kingdom, Japan, and the Netherlands—each exhibited the opposite adjustment to that of the United States.
- Together, this pattern is consistent with:
  - Frontloading to the US (intensive shipments advanced).
  - Backloading from third destinations (post-tariff increases in their allocation shares) helping absorb the payback and reducing the magnitude of subsequent declines in exports to the US.

### Basket-specific interpretation and constraints
- The buffer mechanism appears primarily relevant for the basket of final goods (basket 1).
- For intermediate goods (basket 2):
  - Less evidence of cross-destination intertemporal reallocation.
  - Possible reasons: intermediate goods are more likely to be processed immediately (less scope for intertemporal reallocation) or face stronger international competition from alternative suppliers.

### Capacity and inventory/ buffering considerations
- For frontloading to mitigate tariffs effectively, capacity must exist to retain earlier shipments for later use—a buffer that smooths inflows and outflows.
- Prior literature cited (Alessandria et al. (2010), Alessandria et al. (2021), Alessandria et al. (2024), among others) documents how commercial inventory stocks can facilitate this buffering mechanism; similar effects may arise from household-level inventory accumulation or models distinguishing durable and non-durable consumption.
- Indirect evidence is provided that such buffers enabled additional flexibility to support frontloading of exports to the US; however, direct investigation of cross-destination inventory dynamics is beyond the scope of this analysis and left for future research.

### Manufacturing capacity utilization evidence relevant to reallocation necessity
- For reallocation to be necessary, limited slack in China’s manufacturing capacity is required.
- China’s manufacturing capacity utilization data reported:
  - 74.5 percent in 24H1
  - 75.8 percent in 24H2
  - 74.2 percent in 25H1
  - Peak of 78.2 percent in 21H1
- Interpretation: manufacturing slack decreased during the period most relevant for export frontloading, consistent with the intertemporal reallocation hypothesis, although overall utilization remained below its 21H1 peak.

*Italic source: IMF Working Paper content unit — 4.3 Cross-Destination Intertemporal Reallocation of Shipments*

### 5.3    Value Added, Tariff Mitigation Incentives, and Policy Implications

### 5.3    Value Added, Tariff Mitigation Incentives, and Policy Implications

### Aggregate indicators and comovement
- Figure 15 presents seasonally adjusted indices of Vietnam’s manufacturing value added, exports, and imports, each normalized to 100 in 2018.
- Key observations:
  - A marked acceleration in Vietnam’s manufacturing value added in the first half of 2025, with growth reaching 20% on a quarter-on-quarter annualized basis.
  - Robust positive comovement between exports and manufacturing value added, both historically and in recent quarters.
  - This comovement is described as inconsistent with the transshipment narrative because transshipment would imply no impact on domestic value added.
- Comparative dynamics since 2018:
  - Exports and imports have increased proportionally since 2018.
  - Manufacturing value added has grown by a smaller margin, suggesting a decline in the marginal domestic value added per unit of exports.
  - The decline in marginal domestic value added per unit of exports appears to have accelerated in the first half of 2025, as export and import growth outpaced the strong increase in manufacturing value added.

### Value added continuum and relocation incentives
- Conceptual framing:
  - Value added lies along a continuum between conventional end-stage manufacturing (typically around one-fifth of the final value) and zero value-added transshipment.
  - Outsourced manufacturing can fall anywhere along this spectrum.
  - When the share of value added is too low, cost savings are unlikely to justify investment in infrastructure required to relocate production to Vietnam.
- Implication:
  - Observed patterns (positive comovement and increased manufacturing VA) are suggestive of meaningful product transformation rather than pure transshipment.

### Tariff wedge effects and export implications
- Tariff rate wedge between Vietnam and China:
  - Opening of a tariff rate wedge lowers the profitability threshold for relocation.
  - Manufacturing activities that were previously too low-margin to justify outsourcing may become profitable to relocate to Vietnam.
  - Expectation of an expansion of Vietnam’s end-stage manufacturing, especially at the margin of lower value-added activities.
- Overlapping predictions:
  - Both production relocation and transshipment respond to tariff rate differentials, producing overlapping empirical predictions.
  - The authors’ key takeaway: the underlying trend of production relocation appears well established and ongoing, and a widening tariff differential is likely to accelerate this process.
- Scenario regarding tariffs targeting transshipment:
  - If recent export strength to Vietnam were driven primarily by transshipment, a US tariff specifically targeting such activity could trigger a sharp reversal, with exports from China to Vietnam falling abruptly as the tariff wedge is eliminated or narrowed.
  - The authors view this sharp reversal as unlikely given evidence pointing to meaningful product transformation in Vietnam.

### Policy implications and research directions
- Immediate policy-relevant conclusions:
  - Evidence is tentative that increased shipments from China to Vietnam in 2025 reflect an acceleration of production relocation rather than a surge in transshipment.
  - A widening tariff differential is likely to accelerate production relocation.
- Suggested avenues for future research to inform policy analysis:
  - Horizontal supply chain integration:
    - To what extent were investments in production capacity in Vietnam driven by the incentive to hedge against future tariff risks, as opposed to fundamentals and comparative advantage?
    - Were FDI inflows disproportionately concentrated in industries most exposed to the 2018 tariffs?
    - Further analysis of firm-level data and changes in input–output linkages could provide evidence on cross-border reorganization of production.
  - Role of vertical integration:
    - How did vertical integration enable firms to retain intermediates for domestic use, and how did this interact with incentives for horizontal integration through international supply chains to mitigate tariff risks?
    - Did this integration disadvantage producers dependent on China for intermediate inputs, potentially incentivizing supplier diversification as a hedge against future risks?
  - Cross-destination inventory adjustments:
    - Investigate whether there is evidence of cross-destination inventory adjustments consistent with intertemporal reallocation across destinations.
  - Dynamic modeling of supply chain responses:
    - Develop models that capture dynamic supply chain and exporter responses, allowing frontloading behavior to emerge endogenously and providing a framework for policy analysis.
  - Joint modeling of export and import flows:
    - A promising extension is to jointly model export and import flows across multiple economies to identify common factors driving global trade dynamics (for example, combining China’s export data with US import data to trace indirect export flows from China through ASEAN to the US).
    - Such a model could distinguish global, regional, and destination-specific components of trade variation, offering a more comprehensive understanding of how trade shocks propagate across economies.

*Source: 5.3    Value Added, Tariff Mitigation Incentives, and Policy Implications (from the supplied content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026013-source-pdf.pdf_