## wpiea2019275-print-pdf

## Source details

**Canonical URL:** [wpiea2019275-print-pdf](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019275-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019275-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019275-print-pdf.pdf.json)

---

### I. Motivation
- Maritime trade context:
  - "about 80 percent of global trade by volume (more than 70 percent by value)" was shipped by sea in 2018; "more than 10 billion tons of internationally traded goods were loaded at the world’s ports" (UNCTAD 2018).
- AIS introduction and scope:
  - Automatic Identification System (AIS) introduced by the International Maritime Organization (IMO) in 2004.
  - AIS mandatory for "all major commercial ships (over 300 gross tons)" and broadcasts vessel identification, position, speed, and other safety-related information every few seconds.
- Opportunities from big data:
  - Big data can provide "innovative, real-time, and granular insights" (IMF 2017) and supports early detection of risks, closing data gaps, and improving timeliness of official statistics (IMF 2018).
  - Vessel AIS data allow nowcasting of trade flows in goods (not services), trade volume (not value), gross trade (not re-exports), and trade by broad groups rather than specific goods.
- Limitations and cautions:
  - Vessel data sensitivity: AIS coverage quality (terrestrial or satellite), vessel transponder switching, and mixed cargo origins on container ships.
  - Trade-off: high-frequency big data implies timeliness versus robustness; requires quality control and domain expertise.
- Malta as benchmark:
  - Malta is "a small and open economy, highly dependent on external trade."
  - "About two-thirds of its imports are carried by ship."
  - As an EU member, Malta provides reliable port statistics useful for validation.
- Empirical validation (Malta):
  - "Cargo number" indicator tracks port visits at a higher level.
  - "Cargo load" indicator tracks official trade volume statistics; strong correlation between official and AIS-based estimates.
  - Cargo load by trading partner (using last/next port from AIS) does not match IMF DOTS geographic breakdown.

Policy implications (from Motivation)
- Vessel traffic data can improve timeliness of official trade statistics and help detect emerging risks and turning points—especially for small open economies reliant on seaborne trade.
- Granular ship-by-ship and port-by-port data may reveal emerging patterns in international trade, including those associated with global trade tensions.
- AIS-based data could close data gaps for countries with mostly seaborne trade and weak statistical capacity (e.g., small island states).
- Small states (representing "a quarter of the Fund’s membership") could particularly benefit:
  - Most are island states with seaborne trade predominance.
  - High import shares of consumption make port activity macro-critical for detecting turning points.
  - Limited statistical capacity increases the value of alternative real-time indicators.

### II. AUTOMATIC IDENTIFICATION SYSTEM (AIS): AN INNOVATIVE DATA SOURCE FOR TRACKING VESSEL TRAFFIC IN REAL TIME
- AIS characteristics:
  - AIS is a self-reporting system broadcasting vessel identifying information, characteristics, destination, current location, speed, and heading.
  - Mandatory for international commercial ships "above a certain tonnage, including cargo ships and tankers."
- Commercial providers and coverage:
  - Examples: MarineTraffic, VT explorer, IHS Global, exactEarth, Spire, ORBCOMM, and FleetMon; Bloomberg provides limited historical AIS via IHS Global.
  - "MarineTraffic is currently operating one of the most comprehensive networks; it deploys more than 3,500 terrestrial AIS receiving stations that actively relay AIS information from more than 180 countries (as of end- 2018)."
  - MarineTraffic geospatially defines ports with bounding boxes; "more than 7,300 ports are being monitored."
- From AIS to port call data:
  - Port call registered when vessel records positions inside and outside a port bounding box within a short interval, with vessel speed "below 1 knot" inside the bounding box.
  - Typical port call fields: port name; vessel identifier (IMO, MMSI, or SHIP_ID); gross tonnage; deadweight tonnage; draught; vessel type; arrival/departure timestamps; actual departure from last port; estimated arrival to next port.
- Reasons a port call may not be recorded:
  - AIS coverage low around port area;
  - Vessel switches off AIS transponder;
  - Port is small and not registered in the database.
- Advantages of port call data vs official trade statistics:
  - Timeliness: daily real-time availability vs official trade data often monthly with a "one-to-three month lag" or worse.
  - Granularity: port and vessel-type breakdowns can identify port congestion, infrastructure impacts, or vessel-type shifts (e.g., oil tankers).
  - Potential accuracy improvement where official trade collection is weak (e.g., duty-free imports not recorded at customs).
  - For small island states: trade volume from port call data can proxy real consumption or even real GDP growth when conditions hold: (i) most trade in goods, not services; (ii) most trade seaborne; (iii) high import share of domestic consumption/investment or high export share of domestic production.
- Disadvantages:
  - Commercial providers charge fees for data and historical feeds.
  - Accuracy depends on AIS-receiver coverage near ports.
  - AIS data can be noisy and require specialized filtering techniques and domain experience.

### III. Proposed approach: filters and indicators
- Two-step methodology:
  1. Develop a filter to identify cargo ships assumed to generate trade activity by unloading and loading goods at port.
  2. Produce two indicators from the filtered ships:
     - "Cargo number" indicator: counts the number of ships visiting ports.
     - "Cargo load" indicator: combines vessel size (deadweight tonnage) and changes in cargo load (draught) from AIS to derive a trade volume index.
- Data and validation:
  - Sample: port call data for Malta from MarineTraffic covering "January 2015 and December" (document cut off).
  - Matlab code used to construct cargo number and cargo load indicators is "available from the authors upon request."
- Empirical observations (Malta):
  - Malta has "only two commercial ports (Valetta and Marsaxlokk)" both covered well by AIS-receiving stations and registered in MarineTraffic.
  - Port call data reduce data volume and complexity relative to tracking full voyages and are more accurate near ports due to terrestrial receiver coverage.

### Malta sample, filtering, and key statistics
- Sample period and ports:
  - Sample period: January 1, 2015, to December 31, 2018.
  - Focused ports: Marsaxlokk and Valletta.
- Aggregate counts:
  - Total AIS-detected port calls in the two ports: 52,863.
  - Two-third of calls (about 35,000) are reported from Marsaxlokk.
- Port characteristics:
  - Marsaxlokk: Freeport container terminal, storage facilities for industrial goods and oil products; offers oil storage, blending, and treatments for third-party countries.
  - Grand Harbor of Valletta: cargo terminals handling ro-ro containers and conventional cargo ships; grain terminals, oil tanks, ship repairs, offshore services.
- Ship-type composition (text): tankers account for more than half of port calls in the two ports; containerships and general cargo ships account for nearly 45 percent of port calls.

Filtering port calls to identify trade-related activity
- Three exclusion rules jointly eliminate 12,050 port calls, which is 22.8 percent of the total sample.
- About half of the ships filtered out are tankers.
- Rationale and calibration:
  - Bunkering and ship-to-ship (STS) transfers in the Marsaxlokk anchorage produce many short-distance movements; these introduce volatility in trade indices.
  - Fuel supplied to foreign vessels should be recorded as exports according to international standards, but data collection is challenging.
  - Stay-duration thresholds (less than 5 hours or more than 60 hours) were calibrated by sensitivity analysis against maritime statistics; future methods may include minimizing RMSE or machine learning (e.g., LSTM).
  - Exceptions: fast ferry services to/from Sicily (perishables) may stay less than five hours; aquaculture exports may require processing longer than 60 hours.

Exclusion rules (summary)
- Rule 1 — Anchorage and bunkering tankers:
  - AIS data show almost 4,000 vessels classified as “bunkering tankers” (about 8 percent of total port calls).
  - These are omitted because inclusion introduces considerable volatility.
- Rule 2 — Ships arrived but not departed:
  - Unmatched arrivals without a corresponding departure are excluded.
  - Unmatched port calls found: 418 (less than 1 percent of the sample).
- Rule 3 — Stay in harbor outside reasonable range for trade activity (threshold: 5 hours–60 hours):
  - Eliminates ships that stayed less than five hours or more than 60 hours in port boundaries.
  - Almost 8,000 ships are eliminated with this rule (about 16 percent of matched port calls), most of which are tankers.
  - Available port statistics indicate almost 80 percent of cargo vessels stay less than one day, and very few exceed three days.

Identification of valid port calls by ship type (Port calls in Marsaxlokk and Valletta. Period: 2015-2018)
- All cargo ships:
  - Total Port Calls (1): 52,863
  - Rule 1. Anchorage and Bunkering Services: 3,861
  - Rule 2. Ships Arrived, but not Departed: 418
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 7,771
  - Invalid Port Calls: 12,050
  - Valid Port Calls (2): 40,813
  - % total (2)/(1): 77.2
- Container ships:
  - Total Port Calls (1): 15,974
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 104
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 509
  - Invalid Port Calls: 613
  - Valid Port Calls (2): 15,361
  - % total (2)/(1): 96.2
- Bulk carriers:
  - Total Port Calls (1): 405
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 23
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 160
  - Invalid Port Calls: 183
  - Valid Port Calls (2): 222
  - % total (2)/(1): 54.8
- Oil tankers:
  - Total Port Calls (1): 18,852
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 97
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 3,001
  - Invalid Port Calls: 3,098
  - Valid Port Calls (2): 15,754
  - % total (2)/(1): 83.6
- Gas tankers:
  - Total Port Calls (1): 177
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 3
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 40
  - Invalid Port Calls: 43
  - Valid Port Calls (2): 134
  - % total (2)/(1): 75.7
- Other tankers:
  - Total Port Calls (1): 9,199
  - Rule 1. Anchorage and Bunkering Services: 3,861
  - Rule 2. Ships Arrived, but not Departed: 66
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 2,034
  - Invalid Port Calls: 5,961
  - Valid Port Calls (2): 3,238
  - % total (2)/(1): 35.2
- General cargo:
  - Total Port Calls (1): 7,238
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 98
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 1,561
  - Invalid Port Calls: 1,659
  - Valid Port Calls (2): 5,579
  - % total (2)/(1): 77.1
- Other types:
  - Total Port Calls (1): 1,018
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 27
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 466
  - Invalid Port Calls: 493
  - Valid Port Calls (2): 525
  - % total (2)/(1): 51.6

Key implications and methodological notes
- Aggregate effect of filters: 12,050 port calls removed; 40,813 port calls retained as valid for trade-activity analysis.
- Bunkering activity between non-resident entities is not part of trade statistics (adjustments for exports and imports of goods are made for BOP and National Accounts purposes).
- Sensitivity and calibration:
  - Thresholds chosen by sensitivity analysis calibrated against maritime statistics used as benchmark.
  - Planned extensions include minimizing Root Mean Squared Forecast Errors (RMSE) and using machine learning techniques, such as Long-Short-Term Memory (LSTM) networks.
- Noted exceptions to stay-duration thresholds: fast ferry services to/from Sicily and aquaculture exports.

### Box 1. Tracking the Arrival of a Liquefied Natural Gas Tanker in Malta
- Vessel: Armada Mediterrana, a liquefied natural gas (LNG) tanker built in 1985 and transformed into a floating storage unit (FSU).
- Purpose: Providing storage and supplying gas to an on-shore electricity power station as part of Malta’s energy policy.
- Key AIS-detected events (local times reported as GMT in source):
  - 10/09/2016 7:51 AM — Arrival (A) at Malta anchorage (from the Suez Canal).
  - 10/10/2016 5:19 AM — Departure (D) for Marsaxlokk port.
  - 10/10/2016 5:35 AM — Arrival (A) at Marsaxlokk port.
  - 10/10/2016 8:58 AM — Departure (D).
  - 10/10/2016 9:17 AM — Arrival (A).
  - 10/12/2016 2:22 PM — Departure (D).
  - 10/12/2016 2:37 PM — Arrival (A).
  - 10/12/2016 5:50 PM — Departure (D).
  - 10/12/2016 6:03 PM — Arrival (A).
  - 10/17/2016 6:02 AM — Departure (D).
  - 10/17/2016 6:41 AM — Arrival (A).
  - 10/17/2016 8:13 PM — Departure (D).
  - 10/17/2016 8:26 PM — Arrival (A).
  - 10/18/2016 4:05 PM — Departure (D).
  - 10/18/2016 4:25 PM — Arrival (A).
  - January 2017 — Tanker/storage unit permanently moored at a terminal jetty in Marsaxlokk.
- Observations:
  - Back-and-forth movements between Marsaxlokk and anchorage attributed to drifting events and recalibration of anchorage boundaries.
- Filtering relevance:
  - Offshore storage vessels empty on first arrival should be excluded from valid port call samples.
  - Back-and-forth movements that do not generate trade activity should be excluded.
- Measurement caveats:
  - AIS draught is manually entered and subject to inaccuracy and delay.
  - Less than 5 percent of filtered ships had no change in draught in Malta’s sample.

Deriving weekly indicators of vessel traffic (Malta)
- Weekly aggregation: first week of 2015 to last week of 2018 (211 weeks).
- Cargo number indicator:
  - Counts incoming ships.
  - Moves around an average level of about 100 ships a week.
  - Seasonal trough at beginning of year; peaks around weeks 24-25.
  - Weekly series contains substantial noise; a five-term (centered) moving average used to smooth noise.
- Cargo load indicator:
  - Combines deadweight tonnage (DWT) and reported draught (arrival draught Aitd and departure draught Ditd) to approximate cargo load.
  - Formula (as presented):
    - CWIDWT_{t} = Σ_i DWT_{it} * max(|A_{it}d − D_{it}d| / max(d_i),  )
    - (where max(d_i) is the maximum draught reported by the ship in the sample data; the resulting ratio is ≤ 1 and interpretable as capacity utilization)
  - Shortcomings:
    - max(d_i) is a local maximum, not the ship’s designed maximum draught; may underestimate cargo load.
    - Assumes no trade when there is no change in draught (counterexample: equal weights unloaded and loaded).
    - Draught measurement errors mitigated by large samples.
  - Empirical movement:
    - Peak between last week of May and second week of August 2016.
    - Steep fall and return to 2015 level at start of 2017; moderate upward trend through end of 2018.
- Ancillary indicator:
  - Gross tonnage–based indicator developed and showed similar movements to cargo number for Malta (not shown in paper).

### Quality assurance: comparison with official statistics
- Maritime statistics comparison:
  - Official sources: National Statistics Office (NSO) of Malta; Eurostat Transport database.
  - AIS-based port calls consistently exceed official numbers by about 20-30 percent.
  - Correlation between AIS cargo number indicator and official port statistics: 0.75 for the whole sample.
  - Timeliness: Port call data available almost in real time and can be used to forecast vessel traffic for up to two-three quarters.
- Trade statistics comparison:
  - Cargo load indicator compared with official trade in volume (NSO of Malta).
  - Cargo load indicator closely follows Malta’s trade volumes (sum of imports and exports).
  - Correlation between cargo load indicator and official trade index: 0.65.
  - Official trade index shows higher short-term volatility; differences due to coverage (e.g., air trade) and methodology.
- Trade-by-partner geographic distribution differences:
  - DOTS top five exporters to Malta (2015-18): Italy, Germany, United Kingdom, France, and Spain (about 55-60 percent of total nominal trade in goods of Malta).
  - AIS-based "country of last port" distribution: Italy and France remain in top five; AIS top partners include Spain, Turkey, and Egypt; Germany and the United Kingdom not recorded as direct trading partners in AIS port-call data.
  - Reasons: AIS records last port before arriving in Malta, container ships carry goods from landlocked exporters, multi-stop schedules and geographic choke points, oil/tanker sales and storage/resale onshore.

### Conclusions and policy implications
- Main empirical findings:
  - Cargo number indicator records a larger number of ships relative to official port statistics, suggesting exhaustive AIS coverage.
  - Cargo load indicator is a good proxy for Malta’s trade volumes and detects major activity peaks (e.g., mid-2016).
- Policy implications and potential uses:
  - Real-time vessel data can be used to nowcast trade flows, offering prompt insights into the health of economies—especially small open states dependent on maritime trade.
  - Methodology can be extended to other countries with significant maritime trade and aggregated to generate a real-time worldwide indicator of trade in goods (not services).
  - AIS-derived indicators can complement official statistics by improving timeliness, granularity, and validation of trade volume indicators and national accounts components.
  - Potential use for detecting macroeconomic risks and assessing effects of trade protectionism on global trade.
- Limitations emphasized:
  - AIS-based indicators do not replace customs-based official trade statistics.
  - AIS indicators are restricted chiefly to goods transported by ship and provide limited detail on types of traded goods (except broad categories like oil and gas).

### Proposed extensions and future research
- Test robustness with longer time spans and additional countries.
- Improve cargo load indicator by exploiting statistical relationships between actual cargo size and AIS-reported draught (Jia and others 2015).
- Use dynamic AIS data (speed and course over ground) to isolate ships moving around port boundaries versus actually entering ports.
- Explore aggregating port-call data across countries to create a comparable, high-frequency leading indicator of trade for regions (e.g., Europe’s largest maritime hubs).

### Appendix 1. Glossary of maritime terms (selected)
- Gross tonnage (GT): a measure of a ship’s volume.
- Deadweight tonnage (DWT): a measure of how much weight (in tons) a ship can carry.
- Draught (or draft): the depth to which a vessel is immersed in water; vertical distance (in meters) between the waterline and the bottom of the vessel’s hull.
- Maximum draught (or draft): special marking (Plimsoll line) indicating legal loading limit.
- IMO number: unique seven-digit reference number assigned when constructed; permanent.
- MMSI number: nine-digit Maritime Mobile Service Identity; may change over time; first three digits are MID (Maritime Identification Digits).
- Ship_Id: proprietary vessel identifier introduced by MarineTraffic in 2014.
- Port call: discrete event representing a vessel arriving at or departing from a port; triggered when two positions inside and outside port boundaries are recorded within a short interval.

*Source: Excerpt from wpiea2019275-print-pdf - References ________________________________________________________________33*

### References ________________________________________________________________33

### wpiea2019275-print-pdf - References ________________________________________________________________33

### I. MOTIVATION
- Maritime trade: "about 80 percent of global trade by volume (more than 70 percent by value)" was shipped by sea in 2018; "more than 10 billion tons of internationally traded goods were loaded at the world’s ports" (UNCTAD 2018).
- AIS introduction and scope:
  - Automatic Identification System (AIS) introduced by the International Maritime Organization (IMO) in 2004.
  - AIS mandatory for "all major commercial ships (over 300 gross tons)" and broadcasts vessel identification, position, speed, and other safety-related information every few seconds.
- Opportunities from big data:
  - Big data can provide "innovative, real-time, and granular insights" (IMF 2017) and supports early detection of risks, closing data gaps, and improving timeliness of official statistics (IMF 2018).
  - Vessel AIS data allow nowcasting of trade flows in goods (not services), trade volume (not value), gross trade (not re-exports), and trade by broad groups rather than specific goods.
- Limitations and cautions:
  - Vessel data are sensitive to AIS coverage quality (terrestrial or satellite), vessel transponder switching, and shipping patterns (e.g., mixed cargo origins on container ships).
  - High-frequency big data imply a trade-off between timeliness (efficiency) and robustness (reliability); statistics derived from real-time data require quality control and domain expertise.
- Malta as benchmark:
  - Malta is "a small and open economy, highly dependent on external trade."
  - "About two-thirds of its imports are carried by ship."
  - As an EU member, Malta provides reliable port statistics useful for validation.
- Empirical validation (Malta):
  - "Cargo number" indicator tracks well the number of port visits in Malta’s maritime statistics, albeit at a higher level.
  - "Cargo load" indicator tracks well Malta’s official trade volume statistics; strong correlation between official and AIS-based estimates.
  - Cargo load by trading partner (using last/next port from AIS) does not match IMF DOTS geographic breakdown, as country of origin of goods does not necessarily coincide with country of last port.

### Policy implications (from Motivation)
- The use of vessel traffic data can improve the timeliness of official trade statistics, helping detect emerging risks in trade flows and identify turning points in the business cycle—especially for small open economies reliant on seaborne trade.
- Granular ship-by-ship and port-by-port data may reveal emerging patterns in international trade, including those associated with global trade tensions.
- AIS-based data could close data gaps for countries with mostly seaborne trade and weak statistical capacity (e.g., small island states).
- Small states (representing "a quarter of the Fund’s membership") could particularly benefit:
  - Most are island states with seaborne trade predominance.
  - High import shares of consumption make port activity macro-critical for detecting turning points.
  - Limited statistical capacity increases the value of alternative real-time indicators.

### II. AUTOMATIC IDENTIFICATION SYSTEM (AIS): AN INNOVATIVE DATA SOURCE FOR TRACKING VESSEL TRAFFIC IN REAL TIME
- AIS characteristics:
  - AIS is a self-reporting system broadcasting vessel identifying information, characteristics, destination, current location, speed, and heading.
  - Mandatory for international commercial ships "above a certain tonnage, including cargo ships and tankers."
- Commercial data providers aggregate AIS into datasets; examples cited include MarineTraffic, VT explorer, IHS Global, exactEarth, Spire, ORBCOMM, and FleetMon; Bloomberg provides limited historical AIS via IHS Global.
- MarineTraffic specifics:
  - "MarineTraffic is currently operating one of the most comprehensive networks; it deploys more than 3,500 terrestrial AIS receiving stations that actively relay AIS information from more than 180 countries (as of end- 2018)."
  - MarineTraffic geospatially defines ports with bounding boxes; "more than 7,300 ports are being monitored."
- From AIS to port call data:
  - Port call registered when vessel records positions inside and outside a port bounding box within a short interval, with vessel speed "below 1 knot" inside the bounding box.
  - Typical port call entry fields: port name; vessel identifier (IMO, MMSI, or SHIP_ID); gross tonnage; deadweight tonnage; draught; vessel type; arrival/departure timestamps; actual departure from last port; estimated arrival to next port.
- Reasons a port call may not be recorded:
  - AIS coverage low around port area;
  - Vessel switches off AIS transponder;
  - Port is small and not registered in the database.
- Advantages of port call data versus official trade statistics:
  - Timeliness and periodicity: port call data available on a daily basis in real time versus official trade data often monthly with a "one-to-three month lag" or worse for countries with weak capacity.
  - Granularity: port and vessel-type breakdowns can identify port congestion, infrastructure impacts, or vessel-type shifts (e.g., oil tankers).
  - Potential accuracy improvement where official trade collection is weak (e.g., duty-free imports not recorded at customs).
  - For small island states with high import shares (sometimes "70-80 percent" of consumption), trade volume from port call data can proxy real consumption or even real GDP growth when conditions hold: (i) most trade in goods, not services; (ii) most trade seaborne; (iii) high import share of domestic consumption/investment or high export share of domestic production.
- Disadvantages:
  - Commercial providers charge fees for data and historical feeds.
  - Accuracy depends on AIS-receiver coverage near ports.
  - AIS data can be noisy and require specialized filtering techniques and domain experience.

### III. OUR PROPOSED APPROACH
- Two-step methodology to derive high-frequency indicators from AIS port call data:
  1. Develop a filter to identify cargo ships assumed to generate trade activity by unloading and loading goods at port.
  2. Produce two indicators from the filtered ships:
     - "Cargo number" indicator: counts the number of ships visiting ports.
     - "Cargo load" indicator: combines vessel size (deadweight tonnage) and changes in cargo load (draught) from AIS to derive a trade volume index.
- Data and validation:
  - Sample: port call data for Malta from MarineTraffic covering "January 2015 and December" (document cut off at this point).
  - Appendix 1 provides a glossary of maritime terms; Matlab code used to construct cargo number and cargo load indicators is "available from the authors upon request."
- Empirical observations (Malta):
  - Port call data coverage is adequate: Malta has "only two commercial ports (Valetta and Marsaxlokk)" both covered well by AIS-receiving stations and registered in MarineTraffic.
  - Port call data reduce data volume and complexity relative to tracking full voyages and tend to be more accurate near ports because terrestrial receivers have better coverage than satellites in open waters.

*Source: Excerpt from wpiea2019275-print-pdf - References ________________________________________________________________33*

### 2018. The sample starts in 2015 because data availability and reliability improved

### wpiea2019275-print-pdf - 2018. The sample starts in 2015 because data availability and reliability improved

### Background: ports and dataset
- Sample period: January 1, 2015, to December 31, 2018.
- Focused ports: Marsaxlokk and Valletta.
- Total AIS-detected port calls in the two ports: 52,863.
- Two-third of calls (about 35,000) are reported from Marsaxlokk.
- Port characteristics:
  - Port of Marsaxlokk: Freeport container terminal, storage facilities for industrial goods and oil products; offers oil storage, blending, and treatments for third-party countries.
  - Grand Harbor of Valletta: cargo terminals handling ro-ro containers and conventional cargo ships; grain terminals, oil tanks, ship repairs, offshore services.
- Ship-type composition (text): tankers account for more than half of port calls in the two ports; containerships and general cargo ships account for nearly 45 percent of port calls.

### Filtering port calls to identify trade-related activity
- Three exclusion rules jointly eliminate 12,050 port calls, which is 22.8 percent of the total sample.
- About half of the ships filtered out are tankers.
- Rationale and notes:
  - Bunkering and ship-to-ship (STS) transfers in the Marsaxlokk anchorage produce many short-distance movements; these introduce volatility in trade indices.
  - Fuel supplied to foreign vessels should be recorded as exports according to international standards, but data collection is challenging.
  - The chosen stay-duration thresholds (less than 5 hours or more than 60 hours) were calibrated by sensitivity analysis against maritime statistics; future methods may include minimizing RMSE or machine learning (e.g., LSTM).
  - Exceptions noted: fast ferry services to/from Sicily (perishables) may stay less than five hours; aquaculture exports may require processing longer than 60 hours.

### Exclusion rules (detailed)
- Rule 1 — Anchorage and bunkering tankers:
  - AIS data show almost 4,000 vessels classified as “bunkering tankers” (about 8 percent of total port calls).
  - Bunkering tankers are omitted from valid port calls because inclusion introduces considerable volatility to indices.
- Rule 2 — Ships arrived but not departed:
  - The AIS variable flags movement as arrival ("0") or departure ("1"); unmatched arrivals without a corresponding departure are excluded.
  - Unmatched port calls found: 418 (less than 1 percent of the sample).
- Rule 3 — Stay in harbor outside reasonable range for trade activity (threshold: 5 hours–60 hours):
  - Eliminates ships that stayed less than five hours or more than 60 hours in port boundaries.
  - Almost 8,000 ships are eliminated with this rule (about 16 percent of matched port calls), most of which are tankers.
  - Available port statistics indicate almost 80 percent of cargo vessels stay less than one day, and very few exceed three days.

### Table of identification of valid port calls by type of cargo ships (Port calls in Marsaxlokk and Valletta. Period: 2015-2018)
- All cargo ships
  - Total Port Calls (1): 52,863
  - Rule 1. Anchorage and Bunkering Services: 3,861
  - Rule 2. Ships Arrived, but not Departed: 418
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 7,771
  - Invalid Port Calls: 12,050
  - Valid Port Calls (2): 40,813
  - % total (2)/(1): 77.2
- Container ships
  - Total Port Calls (1): 15,974
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 104
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 509
  - Invalid Port Calls: 613
  - Valid Port Calls (2): 15,361
  - % total (2)/(1): 96.2
- Bulk carriers
  - Total Port Calls (1): 405
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 23
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 160
  - Invalid Port Calls: 183
  - Valid Port Calls (2): 222
  - % total (2)/(1): 54.8
- Oil tankers
  - Total Port Calls (1): 18,852
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 97
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 3,001
  - Invalid Port Calls: 3,098
  - Valid Port Calls (2): 15,754
  - % total (2)/(1): 83.6
- Gas tankers
  - Total Port Calls (1): 177
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 3
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 40
  - Invalid Port Calls: 43
  - Valid Port Calls (2): 134
  - % total (2)/(1): 75.7
- Other tankers
  - Total Port Calls (1): 9,199
  - Rule 1. Anchorage and Bunkering Services: 3,861
  - Rule 2. Ships Arrived, but not Departed: 66
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 2,034
  - Invalid Port Calls: 5,961
  - Valid Port Calls (2): 3,238
  - % total (2)/(1): 35.2
- General cargo
  - Total Port Calls (1): 7,238
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 98
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 1,561
  - Invalid Port Calls: 1,659
  - Valid Port Calls (2): 5,579
  - % total (2)/(1): 77.1
- Other types
  - Total Port Calls (1): 1,018
  - Rule 1. Anchorage and Bunkering Services: 0
  - Rule 2. Ships Arrived, but not Departed: 27
  - Rule 3. Stay in Harbor Outside Range (5h-60h): 466
  - Invalid Port Calls: 493
  - Valid Port Calls (2): 525
  - % total (2)/(1): 51.6

### Key implications and methodological notes
- Aggregate effect of filters: 12,050 port calls removed; 40,813 port calls retained as valid for trade-activity analysis.
- Bunkering activity between non-resident entities is not part of trade statistics (adjustments for exports and imports of goods are made for BOP and National Accounts purposes).
- Sensitivity and calibration:
  - Thresholds for stay duration were chosen by sensitivity analysis calibrated against maritime statistics used as benchmark for the indicators.
  - Planned methodological extensions include minimizing Root Mean Squared Forecast Errors (RMSE) ala Stock and Watson (2003) and using machine learning techniques, such as Long-Short-Term Memory (LSTM) networks.
- Noted exceptions to stay-duration thresholds: fast ferry services to/from Sicily and aquaculture exports.

*Source: Authors’ estimates based on MarineTraffic data.*

### Box 1. Tracking the Arrival of a Liquefied Natural Gas Tanker in Malta

### Box 1. Tracking the Arrival of a Liquefied Natural Gas Tanker in Malta

### Arrival and AIS tracking of Armada Mediterrana
- Vessel: Armada Mediterrana, a liquefied natural gas (LNG) tanker built in 1985 and transformed into a floating storage unit (FSU) through a 17-month conversion process in Singapore.
- Purpose in Malta: Providing storage facility and supplying gas to an on-shore electricity power station on a continuous basis as part of the Malta government’s energy policy to reduce reliance on oil products for electricity production.
- Key AIS-detected events (local times reported as GMT in source):
  - 10/09/2016 7:51 AM — Arrival (A) at Malta anchorage (from the Suez Canal).
  - 10/10/2016 5:19 AM — Departure (D) for Marsaxlokk port.
  - 10/10/2016 5:35 AM — Arrival (A) at Marsaxlokk port.
  - 10/10/2016 8:58 AM — Departure (D).
  - 10/10/2016 9:17 AM — Arrival (A).
  - 10/12/2016 2:22 PM — Departure (D).
  - 10/12/2016 2:37 PM — Arrival (A).
  - 10/12/2016 5:50 PM — Departure (D).
  - 10/12/2016 6:03 PM — Arrival (A).
  - 10/17/2016 6:02 AM — Departure (D).
  - 10/17/2016 6:41 AM — Arrival (A).
  - 10/17/2016 8:13 PM — Departure (D).
  - 10/17/2016 8:26 PM — Arrival (A).
  - 10/18/2016 4:05 PM — Departure (D).
  - 10/18/2016 4:25 PM — Arrival (A).
  - January 2017 — Tanker/storage unit permanently moored at a terminal jetty in Marsaxlokk.
- Observations:
  - Local media coverage and government press conference followed the arrival.
  - Several back-and-forth movements between Marsaxlokk and anchorage were observed; experts at MarineTraffic attribute these to drifting events and recalibration of anchorage boundaries.

### Filtering and relevance for trade measurement
- Two filtering necessities highlighted:
  - Offshore storage vessels such as Armada Mediterrana were empty on first arrival and should be excluded from the valid sample of port call data when generating genuine trade activity.
  - Back-and-forth movements between anchorage and port that do not generate trade activity should be excluded.
- Measurement caveats:
  - AIS draught is manually entered and subject to inaccuracy and delay.
  - Instances where draught is updated before approaching next destination (not at departure) can affect measurement; in Malta these instances were not significant.
  - Less than 5 percent of filtered ships had no change in draught in Malta’s sample.

### Deriving weekly indicators of vessel traffic (Malta)
- Weekly aggregation used to exploit date and time of port events; advantage over monthly/quarterly official customs-based statistics.
- Two indicators calculated from filtered container and cargo ships (sample period and specifics):
  - Cargo number indicator:
    - Counts the number of incoming ships.
    - Chart covers the first week of 2015 to the last week of 2018 (211 weeks).
    - Indicator moves around an average level of about 100 ships a week.
    - Seasonal trough at the beginning of the year; seasonal peaks concentrated around weeks 24-25 (the second half of June).
    - Weekly series contains substantial noise; a five-term (centered) moving average is used to smooth noise.
    - Does not carry information about ship size or cargo load (all ships are equal in this indicator).
  - Cargo load indicator:
    - Combines deadweight tonnage (DWT) and reported draught (arrival draught ,Aitd, and departure draught ,Ditd) to approximate cargo load.
    - Formula (as presented):
      - CWIDWT_{t} = Σ_i DWT_{it} * max(|A_{it}d − D_{it}d| / max(d_i),  )
      - (where max(d_i) is the maximum draught reported by the ship in the sample data; the resulting ratio is ≤ 1 and interpretable as capacity utilization)
    - Shortcomings noted:
      - max(d_i) is a local maximum from port calls, not the ship’s designed maximum draught; may underestimate cargo load when local maximum is far from design maximum.
      - Assumes no trade when there is no change in draught (counterexample: equal weights unloaded and loaded), though such cases were limited in Malta.
      - Draught measurement errors can be mitigated by working with large samples.
    - Empirical movement:
      - Shows an evident peak of activity between the last week of May and the second week of August 2016.
      - Followed by a steep fall and return to 2015 level at the start of 2017, then a moderate upward trend through end of 2018.
- Ancillary indicator:
  - An indicator based on gross tonnage was developed (Appendix 1) and showed similar movements to the cargo number indicator for Malta; not shown in the paper.

### Quality assurance: comparison with official statistics
- Maritime statistics comparison:
  - Official data sources: National Statistics Office (NSO) of Malta; port-level data via Eurostat Transport database.
  - Focus ports: Marsaxlokk and Valletta.
  - AIS-based port calls consistently exceed official numbers by about 20-30 percent.
  - Interpretation: AIS port calls provide exhaustive coverage; possible multiple visits or transiting ships captured by AIS; port statistics may be incomplete.
  - Correlation between AIS cargo number indicator and official port statistics: 0.75 for the whole sample.
  - Timeliness advantage: Port call data from AIS are available almost in real time and can be used to forecast vessel traffic for up to two-three quarters.
- Trade statistics comparison:
  - Cargo load indicator compared with official trade in volume (NSO of Malta).
  - Cargo load indicator closely follows Malta’s trade volume (sum of imports and exports).
  - The June 2016 peak and subsequent deceleration are detected by the cargo load indicator.
  - Correlation between cargo load indicator and official trade index: 0.65.
  - Official trade index shows higher short-term volatility; differences attributable to coverage (e.g., air trade included in official index) and methodology.
- Trade-by-partner geographic distribution:
  - DOTS top five exporters to Malta (2015-18): Italy, Germany, United Kingdom, France, and Spain (about 55-60 percent of total nominal trade in goods of Malta).
  - AIS-based “country of last port” distribution differs: Italy and France remain in top five, but AIS top partners include Spain, Turkey, and Egypt; Germany and the United Kingdom not recorded as direct trading partners in AIS port-call data.
  - Reasons for differences:
    - AIS records last known port before arriving in Malta, not original loading port.
    - Container ships carry goods from landlocked exporters which are not recorded in port-call data.
    - Multi-stop schedules and geographic choke points (Gibraltar/Algeciras/Tangiers; Suez/Port-Said; Istanbul/Dardanelles) distort geographic distributions.
    - Oil and tanker trades may involve sales and ownership transfers while at sea and storage/resale onshore, complicating country-pair monitoring.

### Conclusions and policy implications
- Main empirical findings:
  - The cargo number indicator records a larger number of ships relative to official port statistics, suggesting exhaustive AIS coverage.
  - The cargo load indicator is a good proxy for Malta’s trade volumes and effectively detects major trade activity peaks (e.g., mid-2016).
- Policy implications and potential uses:
  - Real-time vessel data can be used to nowcast trade flows, offering prompt insights into the health of economies—especially small open states dependent on maritime trade.
  - Methodology can be extended to other countries with significant maritime trade and aggregated to generate a real-time worldwide indicator of trade in goods (not services).
  - AIS-derived indicators can complement official statistics by improving timeliness, granularity, and validation of volume trade indicators and national accounts components.
  - Potential use for detecting macroeconomic risks and assessing effects of trade protectionism on global trade.
- Limitations emphasized:
  - AIS-based indicators do not replace customs-based official trade statistics, which remain the most accurate and comprehensive measurement of trade flows.
  - AIS indicators are restricted chiefly to goods transported by ship and provide limited detail on types of traded goods (except broad categories like oil and gas).

### Proposed extensions and future research
- Test robustness with longer time spans and additional countries.
- Improve cargo load indicator by exploiting statistical relationships between actual cargo size and AIS-reported draught (Jia and others 2015).
- Use dynamic AIS data (speed and course over ground) to isolate ships moving around port boundaries versus actually entering ports.
- Explore aggregating port-call data across countries to create a comparable, high-frequency leading indicator of trade for regions (e.g., Europe’s largest maritime hubs).

*Prepared by Jung Eun Yoon (Box 1) — excerpted from the source PDF content.*

### Appendix 1. Glossary of Maritime Terms

### Appendix 1. Glossary of Maritime Terms

### Definitions of vessel measurements
- Gross tonnage (GT) is a measure of a ship’s volume.
- Deadweight tonnage (DWT) is a measure of how much weight (in tons) a ship can carry.
- Draught (or draft) is the depth to which a vessel is immersed in water. It is the vertical distance (in meters) between the waterline and the bottom of the vessel’s hull.
- Maximum draught (or draft) is a special marking on the hull of a ship, also known as the international load line or Plimsoll line, that indicates the legal limit to which a ship may be loaded to safely maintain buoyancy.

### Vessel identifiers
- IMO number is unique seven-digit reference number assigned to each ship when constructed. It follows the ship identification system introduced by the International Maritime Organization (IMO) in 1987 to improve maritime safety and to prevent maritime fraud. The IMO number is permanently associated with the hull of a ship and would not change regardless of a change in the ships name, owner, or flag. The IMO number is part of the static AIS information provided by the vessels crew and can be transmitted additionally, but not all vessels have an IMO number.
- Maritime Mobile Service Identity (MMSI) number is used as an alternate to the IMO number. It is a nine-digit number programmed into AIS systems on board of the vessel. While every vessel has an MMSI, it may change over time as it is not permanent like the IMO number. In particular, a ship could have a different MMSI number when its owner or flag changes.
  - The first three digits of the MMSI are called MID (Maritime Identification Digits). These digits are assigned by the International Telecommunications Union (ITU) to denote the country of registration of a vessel. Hence, the MMSI number of a vessel could change when it is sold or reflagged.
- Ship_Id is a proprietary vessel identifier introduced by MarineTraffic in 2014 to provide a more reliable and unique identifier for each vessel, in addition to IMO or MMSI numbers. Using the Ship_Id, MarineTraffic further ensures that AIS information is attributed to the correct vessel regardless of whether the vessel has an IMO number or if the MMSI number changes with a sale and purchase.

### Port event
- Port call is a discrete event representing a vessel arriving at or departing from a port. It is triggered when two positions of a vessel inside and outside of the port's defined boundaries are recorded within a short interval.

*Source: Appendix 1. Glossary of Maritime Terms (from wpiea2019275-print-pdf).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019275-print-pdf.pdf_