## Annex I.  Correlation of volume series using the same deflators

## Source details

**Canonical URL:** [Annex I.  Correlation of volume series using the same deflators](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026099-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026099-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026099-source-pdf.pdf.json)

---

### Overview and purpose
- Introduces country-level trade estimates by applying the global nowcasting methodology developed in Arslanalp et al. (2025) and port-level data from IMF PortWatch to nowcast country-level maritime trade values and volumes.
- Produces a monthly series of trade estimates that can be released after 7 working days.
- Validation conducted for pilot countries: Brazil, Jamaica, Japan, Samoa, and the United States.

### Validation framework and main findings
- Four-step validation exercise:
  - Conduct a port review to ensure PortWatch covers all major ports and port boundaries.
  - Validate PortWatch captures port calls at the port-level versus official statistics.
  - Validate PortWatch captures physical volume of incoming and outgoing shipments at port and country level.
  - Assess nowcasting estimates of trade value and statistical trade volume compared to official statistics.
- Performance summary:
  - Brazil, Japan, and the U.S.: nowcasting estimates show a high correlation (correlation coefficient >0.8) and good level fit (normalized RMSE <1).
  - Jamaica and Samoa: lower correlations and higher RMSE; reasons include higher sensitivity to idiosyncratic events and shipments and methodological challenges such as transshipment.

### Key methodological steps (three-step process)
- Overall approach: estimate country trade in value and volume in three steps: from AIS volumes to values to statistical volumes.

- Step 1: Country-level Vessel Shipments in Physical Volume
  - Aggregate metric tons estimated from AIS-derived vessel draught data for IMF PortWatch ports.
  - Vessel groupings: container (containerships, “roll-on, roll-off” (Ro-Ro), and general cargo), dry bulk, liquid bulk.

- Step 2: Country-level Trade Value Index
  - Compute value by multiplying physical volume estimates by unit values per vessel type and country using Equations (1) and (2):
    - V_{c,t}^{i} = sum_{j∈{tn,bc,cs}} U_{c,t}^{i,j} * UV_{c,t}^{i,j},  i∈{x,m}, t∈{t0,t1,..,tT}  (1)
    - UV_{c,t}^{i,j} = UV_{c,t0}^{i,j} * (1 + F_{t}^{i,j}),  i∈{x,m}, j∈{tn,bc,cs}, t∈{t0,t1,..,tT}, F_{t0}^{i,j} = 0  (2)
  - Definitions provided (V, U, UV, F; vessel types tn = tanker, bc = dry bulk carrier, cs = containership).
  - Construction of unit values:
    1) Base year unit values UV_{c,t0}^{i,j} from CEPII BACI mapping; base year used in paper: 2019. Base year updated every 5 years; PortWatch base updated to 2024 in early 2026.
    2) Change in unit values F_{t}^{i,j} from global commodity/manufactured goods price indices:
       - Tankers: percentage change in the fuel price index (excluding coal) from IMF’s Primary Commodity Prices database.
       - Dry bulk carriers: percentage change in the non-fuel commodity price index (including coal but excluding precious metals) from IMF’s Primary Commodity Prices database.
       - Containerships/other cargo: percentage change in the manufactured goods price index compiled by the WTO. For missing months, use bridging data from (i) the US CPI index (excluding food, energy, and services) and (ii) the Cleveland Fed’s US CPI inflation nowcast for the latest month.

- Step 3: Country-level Trade Volume Index
  - Convert value to volume by applying export/import price deflators via Equation (3):
    - Q_{c,t}^{i} = V_{c,t}^{i} / P_{c,t}^{i},  i∈{x,m}, t∈{t0,t1,..,tT}  (3)
  - Two deflator options:
    1) Constructed deflators (Laspeyres-type index, base period t0 = 2019) via Equation (4):
       - P_{c,t}^{i} = (sum_{j∈{tn,bc,cs}} UV_{c,t}^{i,j} * U_{c,t0}^{i,j}) / (sum_{j∈{tn,bc,cs}} UV_{c,t0}^{i,j} * U_{c,t0}^{i,j}),  i∈{x,m}, j∈{tn,bc,cs}  (4)
    2) Official country-level deflators published by national authorities (used when available and appropriate).

### Adjustments implemented for pilot countries
- Implemented where appropriate:
  - Rigorous port review (all pilot countries).
  - Netting adjustment for major transshipment hubs (Port of Kingston).
  - Use of official deflators when constructed deflators do not provide a good fit (Japan).

### Caveats and scope
- Estimates primarily nowcast maritime trade values (based on maritime vessels).
- Interpreting as total trade growth requires maritime share of total trade to be high or broadly unchanged.
- Methodology emphasizes cross-comparability; limits country-specific changes—can diverge from official statistics.
- Nowcasts are complements, not substitutes, to official national estimates.

### Validation sample and coverage
- Sample countries: Brazil, Jamaica, Japan, Samoa, United States.
- Selection criteria: (i) high share of maritime trade in total trade; (ii) availability of port-level data; (iii) not a major international transshipment hub.
- Maritime trade context:
  - Maritime trade accounts for 80 percent of the volume of trade in goods.
  - Sample maritime coverage (volume):
    - Import volume coverage: 87 percent for the U.S. up to over 99 percent for each of Jamaica, Japan, and Samoa.
    - Export volume coverage: 93 percent for Samoa up to over 99 percent for Jamaica.
  - Share of trade value transported by sea is lower for Japan and the U.S. due to higher-share, high unit-value goods transported by alternative modes (e.g., air freight).

### Validation methodology and metrics
- Four validation steps (port review; port-level validation; aggregate checks; country-level assessment).
- Comparison focus: monthly year-over-year changes for January 2020 to December 2024.
- Metrics:
  - Pearson correlation coefficient for trend accuracy.
  - RMSE normalized by standard error of official dataset for level accuracy.
- Benchmarks:
  - Target correlation: above 0.8.
  - RMSE threshold: less than 1.

### Baseline performance (January 2020–December 2024)
- Export - Value:
  - CORRELATION: Brazil 0.75, Jamaica 0.36, Japan 0.82, Samoa 0.05, United States 0.90
  - RMSE: Brazil 0.72, Jamaica 1.01, Japan 0.57, Samoa 16.86, United States 0.53
- Import - Value:
  - CORRELATION: Brazil 0.90, Jamaica 0.45, Japan 0.91, Samoa 0.35, United States 0.84
  - RMSE: Brazil 0.47, Jamaica 0.96, Japan 0.45, Samoa 1.54, United States 0.90
- Export - Volume:
  - CORRELATION: Brazil 0.46, Jamaica 0.23, Japan 0.61, Samoa 0.06, United States 0.67
  - RMSE: Brazil 0.96, Jamaica 1.09, Japan 0.87, Samoa 16.87, United States 0.75
- Import - Volume:
  - CORRELATION: Brazil 0.77, Jamaica -0.01, Japan 0.04, Samoa 0.23, United States 0.73
  - RMSE: Brazil 0.64, Jamaica 1.43, Japan 1.37, Samoa 1.79, United States 1.30

### Port review: coverage, boundaries, and offshore platforms
- Port additions (total 64):
  - Brazil 21, Jamaica 2, Japan 37, United States 4.
- Port boundary refinements:
  - Refined boundaries for 18 ports: Brazil - 10; Japan - 5; United States - 2; Jamaica - 1.
  - Example: Port of Santos boundary expanded.
- Offshore platforms:
  - Added 45 offshore platforms for Brazil, 1 for Jamaica, 1 for the U.S.; PortWatch now includes 181 additional offshore platforms across all countries.
  - Rationale: petroleum is Brazil’s second-largest export sector and around 95 percent of production occurs offshore.
  - Identification via MMSI numbers of FPSO, FSU, FSO vessels excluding those within existing port boundaries.

### Port-level and country-level trade adjustments
- Two-way trade and netting adjustment:
  - AIS observes net change in vessel draft; complicates separation of domestic trade and transshipment.
  - Extended netting adjustment to Port of Kingston using official container throughput data to map net imports/exports to gross imports/exports.
  - Arslanalp et al. (2025) incorporated 83 ports into netting adjustment, covering two-thirds of global container trade.
- Specialized export/import ports:
  - For specialized ports, PortWatch sets incoming load factor to zero for export-specialized ports or outgoing load factor to zero for import-specialized ports when AIS-derived load factors misrepresent activity.
  - Examples:
    - Port Rhoades (Jamaica): incoming load factor set to zero for all bulk carriers (bauxite and alumina exports).
    - Samoa (Apia): tanker export estimates set to zero where official statistics indicate no exports of oil or petroleum products.

### Nowcast estimates vs official statistics: deflators and maritime comparison
- Official-deflator hierarchy for PortWatch:
  - i) monthly export/import price deflators where available;
  - ii) quarterly export/import price deflators interpolated to monthly;
  - iii) where no export/import deflators available, use GDP deflators interpolated to monthly.
- Availability in pilot sample:
  - Brazil, Japan, United States: monthly import and export price deflators.
  - Samoa: monthly import deflator not available; interpolate quarterly import deflator to monthly.
  - Jamaica: no import or export price deflators; interpolate annual GDP deflators to monthly.
- Impact and operational considerations:
  - Official deflators materially improve fit for some countries; Japan import volume correlation increases by 0.73 when using official deflators (without other platform changes).
  - Official deflators may be published a few days after PortWatch monthly release; bridge final-month deflator using month-on-month deviations in average period JPY/USD exchange rate as a high-frequency indicator.
  - Criterion to apply official deflators on PortWatch platform:
    - Official deflator must improve correlation of at least one of import or export volume estimates by more than 0.5.
    - Monthly import and export price deflators must be available within a few weeks of month end.
    - Result in pilot sample: apply official deflators only for Japan.
  - Differences explaining Japan case:
    - Official deflator base year 2010 vs constructed 2019; constructed basket ~5000 HS6 products vs official basket of 254 commodities; constructed index in US dollars vs official in yen.
- Maritime comparison group construction:
  - Exclude sectors identified as not transported by vessels and additional non-maritime sectors; apply limited country-specific adjustments.
  - United States validation: constructed maritime value closely tracks U.S. Census maritime trade year-on-year changes — correlation coefficients: export value 0.97; import value 0.98.

### Revised performance (post-adjustments) — Table 3 summary
- Brazil
  - Export - Value: CORRELATION 0.73; RMSE 0.89
  - Import - Value: CORRELATION 0.93; RMSE 0.39
  - Export - Volume: CORRELATION 0.47; RMSE 0.90
  - Import - Volume: CORRELATION 0.84; RMSE 0.59
- Jamaica
  - Export - Value: CORRELATION 0.34; RMSE 1.14
  - Import - Value: CORRELATION 0.74; RMSE 0.74
  - Export - Volume: CORRELATION 0.24; RMSE 1.16
  - Import - Volume: CORRELATION 0.48; RMSE 0.92
- Japan
  - Export - Value: CORRELATION 0.90; RMSE 0.47
  - Import - Value: CORRELATION 0.91; RMSE 0.43
  - Export - Volume: CORRELATION 0.92; RMSE 0.50
  - Import - Volume: CORRELATION 0.73; RMSE 0.82
- Samoa
  - Export - Value: CORRELATION 0.06; RMSE 21.46
  - Import - Value: CORRELATION 0.38; RMSE 1.33
  - Export - Volume: CORRELATION 0.06; RMSE 20.41
  - Import - Volume: CORRELATION 0.39; RMSE 1.41
- United States
  - Export - Value: CORRELATION 0.95; RMSE 0.33
  - Import - Value: CORRELATION 0.86; RMSE 0.62
  - Export - Volume: CORRELATION 0.73; RMSE 0.72
  - Import - Volume: CORRELATION 0.74; RMSE 0.84

- Overall patterns:
  - Advanced economies and large emerging market (United States, Japan, Brazil) meet benchmarks for several trade flows.
  - Jamaica and Samoa continue to have lower correlations and higher RMSEs though improvements observed after adjustments.
  - Two additional series meet correlation criteria on volumes: Japan export volumes and Brazil import volumes.
  - Jamaica import values, Japan import volumes, and the United States import and export volumes are close to the threshold.
  - RMSE: most value series satisfy criteria; import volumes for Jamaica, Japan and the U.S. moved to below threshold of 1.
  - Samoa: improvements for imports, but exports still low correlation and high RMSE.

### Country-specific key findings
- Brazil
  - Added 21 ports and 45 offshore platforms; offshore platforms account for over 90 percent of Brazil’s oil exports.
  - PortWatch closely tracks official port calls and seasonal dynamics; slightly underestimates levels for imports and exports.
  - Imports: close tracking for values and volumes (low RMSE, high correlation).
  - Exports: reasonable value fit (correlation 0.73; RMSE meets criteria); export volumes weaker.
  - Possible limitation: timing/reporting lags for oil exports in official statistics.

- Jamaica
  - PortWatch port calls average 96 percent of official non-cruise liner port calls (January 2019–December 2024).
  - Netting adjustment for Kingston improved import fits.
  - Imports: good fit for import values (correlation 0.74; normalized RMSE 0.74).
  - Exports: weaker fit (export value correlation 0.34; export volume correlation 0.24; RMSE 1.14 and 1.16).

- Japan
  - Expanded registry to include all 125 ports identified by MLIT (5 strategic international hub ports, 18 international hub ports, 102 major ports).
  - Use of authorities’ import and export price indices significantly improves alignment; official deflators used.
  - Strong fit for trade values and volumes; export and import correlations and RMSEs reflect close mirroring of main trends.

- Samoa
  - PortWatch fits official data loosely; imports fit better than exports.
  - Official import/export values adjusted to remove non-maritime goods; import volumes constructed using published import price index; exports use GDP implicit price deflator as proxy.
  - Exports: nowcast does not accurately capture maritime exports (small and highly volatile); export flows not currently fully incorporated in PortWatch.

- United States
  - PortWatch captures trends across 8 major U.S. ports; overall port call correlation 0.93.
  - Correlation by vessel type: containerships 0.95, tankers 0.75.
  - Physical volumes: good fit, better for exports than imports; overestimation of import volumes driven by container ships where netting adjustment applied.
  - Values: exports correlation 0.95; imports correlation 0.86; low normalized RMSE.
  - Volumes: export 0.73; import 0.74; normalized RMSE 0.55 for exports and 0.89 for imports (meeting level-fit criteria).

### Box 1 — U.S. Imports amid Elevated Trade Tensions (2025 H1)
- Policy context:
  - U.S. effective trade-weighted tariff increased from 2.5 percent in January to a peak of 23 percent in May 2025.
  - Rapid increase in trade policy uncertainty during 2025 H1.
- Official statistics:
  - U.S. Census indicates rapid increase in value of U.S. imports between December 2024 and March 2025—front-loading ahead of tariffs announced between February and April (universal tariff increases announced by the U.S. on April 2nd).
- PortWatch vs official:
  - PortWatch did not capture the front-loading spike in total imports.
  - PortWatch estimates for U.S. imports (January 2024–June 2025) correlation with maritime imports: 0.82.
  - Correlation with total imports: 0.24.
- Interpretation:
  - Front-loading spike driven primarily by air-freight of high-value, small-volume goods (e.g., gold bullion, pharmaceuticals, electronics) excluded from maritime-focused PortWatch.
  - PortWatch is an estimate of maritime trade and provides a less volatile “core” measure useful when transportation-mode composition changes.

### Cross-comparability of PortWatch estimates
- Common methodology and data source applied across countries to provide comparable import/export estimates.
- Divergences arise from national authorities’ differing data and methods; country-specific adjustments could improve fit but would reduce cross-comparability.
- Identified areas of divergence not incorporated in pilot methodology:
  - Price indices:
    - Constructed price indices may not reflect seasonal composition within vessel types or contract-/formula-based pricing.
    - Example: Brazil seasonal composition (soybeans vs iron ore) can bias value estimates when converting from volumes.
    - Contract-based pricing (e.g., Jamaica) can cause divergences; PortWatch does not incorporate contract pricing.
  - Reporting lags:
    - PortWatch records at load/discharge; official statistics may record at customs clearance or other reference periods.
    - Example: potential reporting lag for Brazilian offshore oil exports; legal deadline for petroleum invoices in Brazil: 60 days (may take longer up to 90 days).
    - Reporting lags not incorporated due to lack of public information and cross-country heterogeneity.

- Policy-relevant implications:
  - Global nowcasting approach can be applied at the country level with targeted adjustments (port registry, netting, load-factor adjustments, official deflators).
  - Brazil, Japan and the United States meet performance criteria (correlation >0.8 and normalized RMSE <1) in the majority of statistics across trade flows.
  - Jamaica and Samoa show weaker fits but benefit most from domain-specific adjustments.
  - Timely maritime estimates can provide greater value for policymakers, especially for smaller and developing economies.

### Annex I — formal note (summary)
- Deflating both PortWatch and official trade value series by the same deflator does not guarantee identical correlation properties between value and volume measures.
- Deflator series are time-varying with changing variances and covariances; distribution of volume series can differ from distribution of value series even after quasi-linear log-transformation.
- The paper provides a formal proof of this result.

*IMF Working Paper — Nowcasting Country-level Trade Estimates using IMF PortWatch (Working Paper No. WP/2026/099; sections: Annex I; Chapters 3–5; Section 4).*

### Annex I.  Correlation of volume series using the same deflators ........................................................

### Annex I.  Correlation of volume series using the same deflators

### Overview and purpose
- The paper introduces country-level trade estimates by applying the global nowcasting methodology developed in Arslanalp et al. (2025) and port-level data from IMF PortWatch to nowcast country-level maritime trade values and volumes.
- The methodology produces a monthly series of trade estimates that can be released after 7 working days, providing timely estimates ahead of many official statistics.
- Validation is conducted for pilot countries: Brazil, Jamaica, Japan, Samoa, and the United States.

### Validation framework and main findings
- Validation exercise four steps:
  - Conduct a port review to ensure PortWatch covers all major ports and port boundaries.
  - Validate PortWatch captures port calls at the port-level versus official statistics.
  - Validate PortWatch captures physical volume of incoming and outgoing shipments at port and country level.
  - Assess nowcasting estimates of trade value and statistical trade volume compared to official statistics.
- Performance summary:
  - For Brazil, Japan, and the U.S., nowcasting estimates show a high correlation (correlation coefficient >0.8) and good level fit (normalized RMSE <1).
  - For Jamaica and Samoa (smaller emerging markets and low-income countries, especially small island developing states), nowcast estimates show lower correlations and higher RMSE.
  - Reasons for weaker fit in smaller countries include higher sensitivity to idiosyncratic events and shipments and methodological challenges such as transshipment.

### Key methodological steps (three-step process)
- The approach estimates country trade in both value and volume terms in three steps: from AIS volumes to values to statistical volumes.

- Step 1: Country-level Vessel Shipments in Physical Volume
  - Aggregate metric tons of goods shipped by vessel types estimated from AIS-derived vessel draught data for all ports monitored by IMF PortWatch.
  - Three main groupings: container (containerships, “roll-on, roll-off” (Ro-Ro), and general cargo), dry bulk, liquid bulk.

- Step 2: Country-level Trade Value Index
  - Calculate country-level trade value estimates by multiplying physical trade volume estimates by relevant unit values for each vessel type and country as described in Equations 1 and 2:
    - V_{c,t}^{i} = sum_{j∈{tn,bc,cs}} U_{c,t}^{i,j} * UV_{c,t}^{i,j},  i∈{x,m}, t∈{t0,t1,..,tT}  (1)
    - UV_{c,t}^{i,j} = UV_{c,t0}^{i,j} * (1 + F_{t}^{i,j}),  i∈{x,m}, j∈{tn,bc,cs}, t∈{t0,t1,..,tT}, F_{t0}^{i,j} = 0  (2)
  - Definitions (as given):
    - V_{c,t}^{x} and V_{c,t}^{m} are exports and imports in value terms at period t for country c.
    - U_{c,t}^{x,j} and U_{c,t}^{m,j} are exports and imports in physical volume by vessel type j at period t for country c.
    - UV_{c,t}^{x,j} and UV_{c,t}^{m,j} are average unit values (US$/metric ton) of goods exported and imported by vessel type j at period t for country c.
    - F_{t}^{x,j} and F_{t}^{m,j} are percentage changes in the average unit value of goods exported and imported by vessel type j between the base period (t0) and period t.
    - Vessel type notation: tn = tanker, bc = dry bulk carrier, cs = containership.

  - Construction of unit values:
    1) Base year unit values (UV_{c,t0}^{i,j}): use CEPII BACI database and mapping between vessel types and HS codes to calculate average unit value by vessel type for base period.
       - The base year for estimates presented in this paper was 2019. The base year will be updated every 5 years to create a chained measure of prices. The base year for estimates on the PortWatch platform was updated to 2024 in early 2026.
    2) Change in unit values (F_{t}^{i,j}): update base-year unit values using changes in commodity prices and manufactured goods prices (global, not country specific):
       - Tankers: percentage change in the fuel price index (excluding coal) from the IMF’s Primary Commodity Prices database.
       - Dry bulk carriers: percentage change in the non-fuel commodity price index (including coal but excluding precious metals) from the IMF’s Primary Commodity Prices database.
       - Containerships and other cargo vessels: percentage change in the manufactured goods price index compiled by the WTO. Data typically available with a lag of 3 months; for missing months use bridging data from (i) the US CPI index (excluding food, energy, and services) and (ii) the Cleveland Fed’s US CPI inflation nowcast for the latest month.

- Step 3: Country-level Trade Volume Index
  - Convert the estimated country-level trade value into a country-level trade volume index by applying export/import price deflators:
    - Q_{c,t}^{i} = V_{c,t}^{i} / P_{c,t}^{i},  i∈{x,m}, t∈{t0,t1,..,tT}  (3)
    - Q_{c,t}^{x} and Q_{c,t}^{m} are exports and imports in volume terms at period t for country c.
    - P_{c,t}^{x} and P_{c,t}^{m} are export and import deflators at period t for country c.

  - Two types of deflators considered:
    1) Constructed deflators: Laspeyres-type index using base-period quantities, following Equation 4:
       - P_{c,t}^{i} = (sum_{j∈{tn,bc,cs}} UV_{c,t}^{i,j} * U_{c,t0}^{i,j}) / (sum_{j∈{tn,bc,cs}} UV_{c,t0}^{i,j} * U_{c,t0}^{i,j}),  i∈{x,m}, j∈{tn,bc,cs}  (4)
       - t0 stands for the base period (2019).
    2) Official country-level deflators: price deflators published by national authorities; may better reflect country-specific pricing dynamics and are sometimes available, especially for advanced economies.

### Adjustments implemented for pilot countries
- Important adjustments to improve country-level trade estimates implemented where appropriate:
  - Rigorous port review to ensure PortWatch captures all major ports, port boundaries, and offshore platforms (implemented for all pilot countries).
  - Netting adjustment for ports that are major transshipment hubs (implemented for the Port of Kingston).
  - Application of official deflators when constructed deflators do not provide a good fit (implemented for Japan).

### Caveats and scope
- Estimates are primarily a nowcast of maritime trade values as they are based on data for maritime vessels.
  - These estimates could only be interpreted as the growth in total trade if the share of maritime trade in total trade was high or broadly unchanged over time.
- Methodology prioritizes cross-comparability across countries and limits country-specific changes to the methodology; this can lead to divergences from official statistics where country-specific pricing dynamics or reporting lags exist.
- Nowcast estimates are intended as high-frequency complements and not substitutes to official estimates produced by national authorities.

### Related methodological context
- The paper builds directly on Arslanalp et al. (2025) and the broader literature that uses vessel-level big data to develop trade estimates.
- Nowcasting methods referenced include dynamic factor models, machine learning techniques, and regression models used in prior global trade and GDP nowcasting work.

*IMF Working Paper — Nowcasting Country-level Trade Estimates using IMF PortWatch*

### 3.  Development of country-level trade

### 3.  Development of country-level trade estimates

### Validation sample and coverage
- Sample countries: Brazil, Jamaica, Japan, Samoa, and the United States.
- Selection criteria:
  - (i) a high share of maritime trade in total trade;
  - (ii) the availability of port-level data;
  - (iii) not a major international transshipment hub.
- Global context:
  - Maritime trade accounts for 80 percent of the volume of trade in goods.
  - In the sample countries, maritime coverage is higher:
    - Import volume coverage: 87 percent for the U.S. up to over 99 percent for each of Jamaica, Japan, and Samoa.
    - Export volume coverage: 93 percent for Samoa up to over 99 percent for Jamaica.
  - Share of trade value transported by sea is lower for Japan and the U.S. due to higher-share, high unit-value goods transported by alternative modes (e.g., air freight).

### Validation methodology
- Four validation steps:
  1. Port review to ensure coverage of major ports and refine port boundaries.
  2. Port-level validation by matching port calls and physical shipment volumes.
  3. Check that aggregate port calls and physical shipments match country-level official estimates.
  4. Country-level assessment comparing PortWatch nowcasts with official statistics.
- Performance assessment details:
  - Comparison focus: monthly year-over-year changes for January 2020 to December 2024.
  - Metrics used:
    - Pearson correlation coefficient to capture trend accuracy.
    - Root-mean-squared error (RMSE) normalized by the standard error of the official dataset to capture level accuracy.
  - Benchmarks adopted from global nowcasting study:
    - Target correlation: above 0.8.
    - RMSE threshold: less than 1 (one standard deviation).

### Baseline performance (replication of global nowcasting at country level)
- Key observations:
  - Advanced economies and a large emerging market (United States, Japan, Brazil) meet benchmarks for several trade flows.
  - Smaller emerging markets (Samoa, Jamaica) generally fall short.
  - Japan’s trade volume estimates show very low correlation with official statistics for some flows, potentially due to constructed deflator composition or seasonal variation differences.
  - Larger economies tend to have higher correlations, possibly due to smoother aggregated data.
- Baseline assessment table (selected correlations and RMSEs for January 2020–December 2024):
  - Export - Value
    - CORRELATION: Brazil 0.75, Jamaica 0.36, Japan 0.82, Samoa 0.05, United States 0.90
    - RMSE: Brazil 0.72, Jamaica 1.01, Japan 0.57, Samoa 16.86, United States 0.53
  - Import - Value
    - CORRELATION: Brazil 0.90, Jamaica 0.45, Japan 0.91, Samoa 0.35, United States 0.84
    - RMSE: Brazil 0.47, Jamaica 0.96, Japan 0.45, Samoa 1.54, United States 0.90
  - Export - Volume
    - CORRELATION: Brazil 0.46, Jamaica 0.23, Japan 0.61, Samoa 0.06, United States 0.67
    - RMSE: Brazil 0.96, Jamaica 1.09, Japan 0.87, Samoa 16.87, United States 0.75
  - Import - Volume
    - CORRELATION: Brazil 0.77, Jamaica -0.01, Japan 0.04, Samoa 0.23, United States 0.73
    - RMSE: Brazil 0.64, Jamaica 1.43, Japan 1.37, Samoa 1.79, United States 1.30

### Port review: coverage, boundaries, and offshore platforms
- Port coverage adjustments:
  - Added 64 ports in total:
    - 21 for Brazil (separate from offshore platforms),
    - 2 for Jamaica,
    - 37 for Japan (covering all officially defined “major ports” in Japan),
    - 4 for the United States.
- Port boundary refinements:
  - Refined boundaries for 18 ports across the sample.
  - Country-specific counts of adapted port boundaries: Brazil - 10; Japan - 5; United States - 2; Jamaica - 1.
  - Example: Port of Santos boundary expanded to include additional berths previously outside the polygon.
- Offshore platforms:
  - Identified offshore oil and gas platforms as additional port polygons where relevant (e.g., Brazil).
  - Rationale: petroleum is Brazil’s second-largest export sector and around 95 percent of production occurs offshore.
  - Pilot additions: 45 offshore platforms added for Brazil, 1 for Jamaica, and 1 for the U.S.
  - General extension: PortWatch now includes 181 additional offshore platforms across all countries.
  - Offshore platform identification method: MMSI numbers of vessels classified as FPSO, FSU, FSO, excluding those within existing PortWatch port boundaries.

### Port-level and country-level trade adjustments
- Two-way trade and netting adjustment:
  - AIS-based estimates observe only net change in vessel draft, complicating separation of domestic trade and transshipment, especially for containerized vessels in transshipment hubs (example: Port of Kingston in Jamaica).
  - Extended the netting adjustment from Arslanalp et al. (2025) to Port of Kingston using official container throughput data to estimate netting effect and create a scaling factor mapping net imports/exports into gross imports/exports.
  - Effect: scales PortWatch physical volume estimates of container ships; improves nowcasting fits where transshipment cargo trends differ from domestic cargo.
  - Arslanalp et al. (2025) incorporated 83 ports into the netting adjustment, covering two-thirds of global container trade.
- Specialized export and import ports:
  - Some ports specialize in exports or imports for specific commodities (e.g., bulk carriers for mining, tankers for energy).
  - Where AIS-derived load factors misrepresent specialized ports, PortWatch sets incoming load factor to zero for specialized export ports or outgoing load factor to zero for specialized import ports.
  - Examples:
    - Port Rhoades (Jamaica): incoming load factor set to zero for all bulk carriers to reflect specialized export of bauxite and alumina; primarily improves level fit.
    - Samoa: official statistics indicate no exports of oil or petroleum products; PortWatch sets a handful of tanker port calls with non-zero export estimates to zero to align with official classifications.

### Nowcast estimates vs official statistics: deflators and maritime comparison
- Deflators:
  - Option to use official country deflators (monthly, interpolated quarterly, or interpolated annual GDP deflators) instead of constructed global-price-based deflators.
  - Hierarchy for official-deflator use:
    - i) monthly export/import price deflators where available;
    - ii) quarterly export/import price deflators interpolated to monthly;
    - iii) where no export/import price deflators available, use GDP deflators interpolated to monthly.
  - Availability in pilot sample:
    - Brazil, Japan, and the U.S.: monthly import and export price deflators.
    - Samoa: monthly import deflator not available, interpolate quarterly import deflator to monthly.
    - Jamaica: no import or export price deflators available, interpolate annual GDP deflators to monthly.
  - Impact:
    - Application of official deflators materially improves fit for some countries.
    - Japan: correlation for import volume increases by 0.73 when using official deflators (without other platform changes).
  - Implementation operational consideration:
    - Official deflators may be published a few days after PortWatch monthly release window.
    - To bridge timing gap, proxy final-month price deflator using month-on-month deviations in average period JPY/USD exchange rate as a high-frequency indicator.
  - Criterion to use official deflators on PortWatch platform:
    - Official deflator must improve correlation of at least one of import or export volume estimates for a country by more than 0.5.
    - Monthly import and export price deflators must be available within a few weeks of the end of a month.
    - Result in pilot sample: apply official deflators only for Japan.
  - Notes on constructed vs official deflator differences for Japan:
    - Possible factors: base year differences (official deflator base year 2010 vs constructed 2019), number of commodities in basket (constructed ~5000 HS6 products vs official basket of 254 commodities), currency of index (constructed in US dollars vs official in yen).
- Maritime comparison group (benchmarking maritime-only trade):
  - PortWatch estimates maritime trade; headline official statistics include all transport modes.
  - Constructed maritime comparison benchmark by:
    - (i) excluding sectors identified by Arslanalp et al. (2025) as not transported by vessels (precious metals and stones; aircraft and spacecraft; ships, boats, and floating structures; electrical energy);
    - (ii) excluding additional non-maritime sectors (pharmaceuticals products; electronics; gold bullion);
    - (iii) applying limited country-specific adjustments based on domain knowledge (e.g., Samoa: remove re-exports that include sale of oil to ships and aircraft).
  - Validation for the United States:
    - Constructed maritime value closely tracks official U.S. Census maritime trade year-on-year changes.
    - Correlation coefficients: export value 0.97; import value 0.98.

*IMF Working Paper — Nowcasting Country-level Trade Estimates using IMF PortWatch (chapter: 3. Development of country-level trade estimates)*

### 4. Performance of nowcast

### 4. Performance of nowcast

### Assessment of revised results
- Adjustments incorporated:
  - Implementing changes to the PortWatch platform and estimates from the port review, netting adjustment and adjustments for specialized ports.
  - Comparing the PortWatch estimates to maritime data taken from official sources (where available) or removing non-maritime sectors from official trade data.
  - Comparing volumes estimates based on constructed deflators, other than for Japan where official deflators are used.
- Table 2 summary of main adjustments by country (as reported):
  - Port coverage – additional ports: Brazil 21, Jamaica 2, Japan 37, Samoa 0, United States 4.
  - Port boundaries refined: Brazil 10, Jamaica 1, Japan 5, Samoa 0, United States 2.
  - Offshore platforms: Brazil 45, Jamaica 1, Japan 0, Samoa 0, United States 1.
  - Additional netting adjustment: Jamaica 1 addition – Kingston; others 0.
  - Specialized import/export port adjustment: Jamaica 1 – Port Rhoades; Samoa 1 - Apia; others 0.
  - Deflators: Constructed for Brazil, Jamaica, Samoa, United States; Official for Japan.
  - Maritime data source: Brazil Official – COMEX STAT; Jamaica N/A; Japan Constructed; Samoa Constructed; United States Official - U.S. Census.
- Overall performance patterns:
  - Performance remains better in advanced economies and large emerging markets with high correlations and low RMSEs.
  - Jamaica and Samoa continue to have lower correlations and higher RMSEs; estimates more volatile for smaller countries.
  - Comparing baseline (Table 1) and revised (Table 3) results, fit improved for most country-flow groups.
  - On volumes, two additional series now meet correlation criteria: Japan export volumes and Brazil import volumes.
  - Jamaica import values, Japan import volumes, and the United States import and export volumes are close to the threshold.
  - RMSE results: most value series satisfy the criteria; import volumes for Jamaica, Japan and the U.S. moved to below the threshold of 1.
  - Samoa shows improvement for the fit for imports, though still low correlation and high RMSE for exports.

- Table 3 — Revised assessment table (correlation and RMSE by country and flow)
  - Brazil
    - Export - Value: CORRELATION 0.73; RMSE 0.89
    - Import - Value: CORRELATION 0.93; RMSE 0.39
    - Export - Volume: CORRELATION 0.47; RMSE 0.90
    - Import - Volume: CORRELATION 0.84; RMSE 0.59
  - Jamaica
    - Export - Value: CORRELATION 0.34; RMSE 1.14
    - Import - Value: CORRELATION 0.74; RMSE 0.74
    - Export - Volume: CORRELATION 0.24; RMSE 1.16
    - Import - Volume: CORRELATION 0.48; RMSE 0.92
  - Japan
    - Export - Value: CORRELATION 0.90; RMSE 0.47
    - Import - Value: CORRELATION 0.91; RMSE 0.43
    - Export - Volume: CORRELATION 0.92; RMSE 0.50
    - Import - Volume: CORRELATION 0.73; RMSE 0.82
  - Samoa
    - Export - Value: CORRELATION 0.06; RMSE 21.46
    - Import - Value: CORRELATION 0.38; RMSE 1.33
    - Export - Volume: CORRELATION 0.06; RMSE 20.41
    - Import - Volume: CORRELATION 0.39; RMSE 1.41
  - United States
    - Export - Value: CORRELATION 0.95; RMSE 0.33
    - Import - Value: CORRELATION 0.86; RMSE 0.62
    - Export - Volume: CORRELATION 0.73; RMSE 0.72
    - Import - Volume: CORRELATION 0.74; RMSE 0.84

### Brazil — key findings
- Port registry and offshore platform additions:
  - Added 21 ports and 45 offshore platforms; offshore platforms account for over 90 percent of Brazil’s oil exports.
- PortWatch performance:
  - Port calls: PortWatch closely tracks official estimates in levels and trends.
  - Physical trade volumes: Captures seasonal dynamics but slightly underestimates levels for imports and exports.
  - Imports: Nowcasting estimates closely track official statistics for import values and volumes; reflected in low RMSE and high correlation, with slightly better performance for import values.
  - Exports: Export values show reasonable fit (correlation 0.73, RMSE meets criteria); export volumes weaker—good fit on average rate of growth but lower correlation on monthly dynamics.
- Possible limitation:
  - Deterioration from baseline for exports may reflect timing of reporting of oil exports in official statistics.

### Jamaica — key findings
- Port calls and physical volumes:
  - PortWatch captures aggregate number of port calls; average monthly PortWatch port calls average 96 percent of official non-cruise liner port calls (January 2019–December 2024).
  - Good fit to level of official physical import and export volumes after methodological changes, notably a netting adjustment for Kingston.
- Nowcast performance:
  - Imports: Good fit for import values (correlation 0.74; normalized RMSE 0.74); improvement from baseline due to netting adjustment for container imports.
  - Exports: Weaker fit — export value correlation 0.34 and export volume correlation 0.24; normalized RMSE 1.14 and 1.16 respectively.
- Interpretation:
  - Weaker export nowcast fit may result from commodity pricing dynamics not captured by PortWatch despite good fit to physical export volumes; challenges matching tanker exports from processing at Kingston and dry bulk exports from Ocho Rios.

### Japan — key findings
- Port registry expansion:
  - PortWatch expanded registry to include all 125 ports identified by MLIT: 5 strategic international hub ports, 18 international hub ports, and 102 major ports.
- Deflator usage:
  - Applying authorities’ import and export price indices significantly improves alignment; for Japan official deflators used rather than constructed ones.
- Nowcast performance:
  - Port calls: Aggregate international port calls follow trends in authorities’ statistics though levels differ due to inclusion of passenger vessels in official data.
  - Physical trade volumes: PortWatch underestimates levels due to differences in definition of international trade flows and coverage of cargo vessel types, but tracks cyclical movements well.
  - Trade values and volumes: Strong fit with official data for both imports and exports; PortWatch closely mirrors main trends.

### Samoa — key findings
- General performance:
  - PortWatch estimates fit official data only loosely; further AIS-based methodological refinements may be needed for small island economies.
- Comparability adjustments:
  - Official import and export values adjusted to remove goods not typically transported by sea.
  - Official import volumes constructed using published import price index.
  - For exports, no official export price index: GDP implicit price deflator used as proxy to derive export volumes.
- Nowcast performance:
  - Imports: Fit noticeably better than exports; import series more stable and larger in absolute terms.
  - Exports: Nowcast export series does not accurately capture maritime exports (small and highly volatile); such export flows not currently incorporated in the PortWatch platform.
- Data issues:
  - Official and PortWatch physical trade volumes show large discrepancies; PortWatch export volumes are more volatile than official exports.

### United States — key findings
- Port calls and physical volumes:
  - PortWatch captures trends in official port call records across 8 major U.S. ports; overall correlation coefficient 0.93.
  - Correlation by vessel type: containerships 0.95, tankers 0.75.
- Physical volumes:
  - Good fit with official data; better for exports than imports.
  - Overestimation of import volumes driven by container ships in ports where netting adjustment for transshipment is used; may reflect goods arriving without clearing customs due to onward non-maritime transport.
- Nowcast performance (values and volumes):
  - Values: Very strong fit — exports correlation 0.95, imports correlation 0.86; low normalized RMSE.
  - Volumes: Correlations just below threshold — export volumes 0.73, import volumes 0.74; normalized RMSE 0.55 for exports and 0.89 for imports (meeting level-fit criteria).
- Application during 2025 H1 trade tensions:
  - PortWatch nowcast captures maritime trade dynamics and excludes volatile non-maritime trade movements.

### Box 1 — U.S. Imports amid Elevated Trade Tensions (2025 H1)
- Policy context and measured changes:
  - U.S. effective trade-weighted tariff increased from 2.5 percent in January to a peak of 23 percent in May 2025.
  - Rapid increase in trade policy uncertainty during 2025 H1.
- Observed behavior in official statistics:
  - U.S. Census indicates rapid increase in value of U.S. imports between December 2024 and March 2025—documented as front-loading ahead of tariffs announced between February and April (universal tariff increases announced by the U.S. on April 2nd).
- PortWatch vs official series:
  - PortWatch nowcast did not capture the front-loading spike seen in total imports.
  - PortWatch estimates for U.S. imports (January 2024–June 2025) correlation with maritime imports: 0.82.
  - Correlation with total imports: 0.24.
- Interpretation and policy implication:
  - Front-loading spike driven primarily by air-freight of high-value, small-volume goods (e.g., gold bullion, pharmaceuticals, electronics) which maritime-focused PortWatch excludes.
  - PortWatch should be considered an estimate of maritime trade rather than total goods trade, especially when transportation-mode composition changes.
  - PortWatch provides information on less volatile components of international trade (longer shipping times), analogous to a “core” measure: it strips out more volatile transport-mode-driven components and can help policymakers understand underlying trade dynamics.

*Source: IMF Working Paper — Nowcasting Country-level Trade Estimates using IMF PortWatch (Section 4: Performance of nowcast).*

### 5. Cross-comparability of PortWatch estimates

### 5. Cross-comparability of PortWatch estimates

### Methodology and comparability
- PortWatch nowcasts provide comparable estimates for imports and exports across countries by applying the same methodology and same underlying source of data for each country.
- Official import and export statistics are produced by each national authority and may differ in data and methodologies (for example reporting timeframes and the definitions).
- The common methodology can lead to divergences with official data. Additional country-specific adjustments could improve fit with official statistics for particular countries, but idiosyncratic adjustments would reduce cross-comparability.
- In the pilot phase, the following areas of divergence are identified and are not incorporated into the nowcasting methodology.

### Price indices
- PortWatch converts physical volumes into trade values using price indices; concerns arise whether the constructed price indices reflect actual country pricing behavior.
- Two identified reasons for deviation:
  - Seasonal composition of trade within a vessel type:
    - The PortWatch price index does not account for seasonal composition of trade within a vessel type.
    - For countries with highly seasonal composition and different average unit values across seasons, this can cause under/over-estimation of trade values when converting from physical volumes to trade values.
    - Example: Brazil — main bulk commodities are soybeans and iron ore exported at different seasons, leading to under-estimation of trade value during the higher unit value soybean exporting season and over-estimation during periods when iron ore dominates exports.
    - PortWatch does not create seasonal price indices for individual countries because a cross-comparable approach for all countries is not currently available.
    - Instead, results are presented based on year-on-year changes, which remove much of the seasonality and are comparable to approaches taken with other economic indicators, such as CPI estimates.
  - Contract- or formula-based pricing:
    - The PortWatch price index is based on contemporaneous prices for commodities.
    - Many commodity prices are set using long-term bilateral contracts, formulas linked to reference prices, negotiated benchmarks and transfer-pricing arrangements or futures prices.
    - Contract-based pricing arrangements may cause divergences relative to contemporaneous-price-based estimates for countries with high shares of commodity exports (example: Jamaica).
    - As pricing dynamics differ across commodities and countries, PortWatch does not incorporate contract pricing into the price index.

### Reporting lags
- PortWatch records trade at the time a vessel is loaded or discharged.
- Official statistics may record transactions at different times, such as when a good clears customs or another reference period, creating reporting lags.
- Reporting lags can be significant when customs processes take extended periods (for instance goods moved into bonded warehousing).
- Pilot identified a potential reporting lag for Brazilian oil exports from offshore platforms due to a window from vessel departure to export being recorded; incorporating a lag for Brazil’s oil exports would improve the correlation coefficient of Brazil’s export value with official statistics.
- Information on reporting lags is not always publicly available and may differ across countries; therefore reporting lags are not incorporated into PortWatch estimates.
- Legal deadline example for petroleum invoices in Brazil: 60 days, although in practice this may take longer up to 90 days.

### Performance across countries and policy-relevant implications
- The global nowcasting approach from Arslanalp et al. (2025) can be applied at the country level with country-specific adjustments:
  - Domain-specific knowledge can improve underlying data (ensuring major ports and offshore platforms are captured; appropriate port boundaries).
  - Methodological improvements for specific ports (e.g., netting adjustment where there is significant two-way trade; adjusting load factors for specialized import or export ports).
  - Using official country-specific price deflators to create statistical volume estimates when this significantly improves the fit.
- Pilot sample and performance:
  - Sample of 5 countries across a range of income groups.
  - Brazil, Japan and the United States meet performance criteria of a correlation coefficient of greater than 0.8 and normalized RMSE of less than 1 in the majority of the statistics across trade flows.
  - Jamaica and Samoa (smaller and developing economies) had a weaker fit between nowcast estimates and official statistics, but showed the greatest improvement from incorporating domain-specific knowledge and adjustments.
- Policy implication:
  - Even where fit is weaker, the availability of timely estimates of trade for smaller and developing economies could provide greater value for policymakers in understanding economic developments.

### Annex I — Correlation of volume series using the same deflators (summary)
- Deflating both PortWatch and official trade value series by the same deflator does not guarantee identical correlation properties between value and volume measures.
- Deflator series are time-varying with changing variances and covariances, meaning the distribution of volume series can differ from the distribution of value series, even after a quasi-linear log-transformation.
- The paper provides a formal proof of this result.

*Source: Nowcasting Country-level Trade Estimates using IMF PortWatch, Working Paper No. WP/2026/099*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026099-source-pdf.pdf_