Maritime chokepoint selection
In this paper, we define maritime chokepoints as strategic maritime canals, passages, channels or straits within maritime routes where large flows of maritime trade converge. We select 24 chokepoints based on the classification of maritime chokepoints in earlier work20. However, we make some noticeable differences; (i) we do not use the Dardanelles Strait as separate chokepoint given the proximity to the Bosporus Strait, (ii) we do not select the Bering Strait as chokepoint given limited maritime trade flows through this chokepoint, and (iii) we add the Bohai Strait and Ombai Strait given their strategic importance for China and Indonesia, respectively. This leaves us with 24 chokepoints, which are shown in Supplementary Fig. 1.
Maritime trade flows through chokepoints
To derive the dependencies between the maritime chokepoints and the flows of imports and exports of specific countries, we use output of the OxMarTrans global maritime transport model1. The OxMarTrans first utilises a global modal split to estimate the share of maritime trade for every bilateral trade flow in the BACI trade database (on a HS6 level) and disaggregates these maritime trade flows to administrative regions (Administrative Region 1, resulting in around 3500 subnational regions globally). A global maritime transport model, informed by observed vessel movements between 1400 ports globally (to set port-to-port connectivity and capacities) is then used for the flow allocation of each bilateral trade flow, capturing the port of export, possible transhipment ports, and port of import, including the maritime routes taken. We then track, for each bilateral trade flow, whether a maritime chokepoint is passed. This can be multiple chokepoints per trade flow (e.g., trade from China to the Netherlands will pass the Taiwan Strait, Malacca Strait, Bab el-Mandeb, Suez Canal, Gibraltar Strait and Dover Strait). Doing so across all maritime trade flows allows us to capture the quantity and value of maritime trade that flows through the various chokepoints, as well as linking this to the countries these trade flows belong to. The OxMarTrans output in ref. 1 was for the 2015 base year. Here, we update the output to use 2022 as base year by scaling the bilateral trade flows (per sector) from 2015 trade quantities and value to 2022 quantities and values. This is based on data in the BACI trade database21. It should be noted that our global transport model, utilising a least-cost flow allocation algorithm, may not be able to capture all factors that shape routing decisions in complex hinterlands, in particular for landlocked countries with access to multiple port ranges (e.g., landlocked countries in South America, Europe and Central Asia). Hence, for such countries, the chokepoint dependencies modelled should be interpreted with some care, as modelling errors could be present.
To validate the model output for this exercise, we compare our modelled chokepoint throughput estimates for 2022 with the total vessel capacity that flows through the chokepoints based on Automatic Identification System (AIS), extracted from the PortWatch platform (https://portwatch.imf.org/). The correlation coefficient between the modelled and observed data is 0.78, indicating a good alignment and providing confidence in our modelled throughput estimates.
Hazard data
We consider eight hazard types, include five human-induced hazards (piracy, armed conflict, terrorist attacks, blockages, and interstate conflict) and three natural hazards (earthquakes, drought and cyclones). For each hazard, we create a spatial event dataset, which captures the geolocation of the historical events and relevant characteristics of each event (e.g., duration, severity). This allows us, per hazard (h) and chokepoint (c), to derive the following three indicators that we use for the risk quantification: probability (P), severity (S) and duration (D). The hazard probability (P) for each event is defined as the annual exceedance probability:
$${P}_{c,h}=\frac{1}{{T}_{h}}$$
With T being the number of years in the hazard spatial event dataset. The hazard severity is a value between 0 and 1 indicating the severity of the flow duration. A value of 1 indicates that the chokepoint is fully blocked, whereas a value of 0 indicates no flow disruption. The hazard duration is the number of days a chokepoint is in a state of disruption, ranging from 0 to 365 days. Below, we describe per hazard how these three indicators were derived.
We use 46 years of nautical piracy event data (1978–2024) from the Anti-shipping Activity Messages (ASAM) database, compiled by the National Geospatial-Intelligence Agency (NGA)22. This data includes the locations and accounts of hostile acts against ships and mariners. To evaluate the probability, we estimate the annual frequency of events within the chokepoint area, assuming a 50 km buffer around the events, resulting in 820 piracy attacks. The piracy events are shown in green in Supplementary Fig. 9. In line with research evaluating the impacts of piracy events on shipping14, we assume that every piracy event results in 0.5% less trade for a period of 30 days.
Armed conflict events refer to any incident where armed force was used by an organised actor against another organized actor, or against civilians23. Although these conflict events may not happen at sea, conflict events close to the chokepoints can spillover to sea, creating maritime security threats or disrupt port operations in and around the chokepoint. We use the UNCP Georeferenced Event Dataset of conflict events23. We select only specific events of interest; (i) conflict events that are classified as intrastate conflict, and (ii) events that have caused at least 10 fatalities. This to exclude interstate conflict, which we treat separately, and use only major conflict events in the frequency analysis, as they are more likely to affect maritime security. The dataset covers events between 1989 and 2023 (35 years), and we use a 50-kilometre buffer to evaluate probability of conflict events within the chokepoint vicinity, leaving us with 96 armed conflict events (see the red markers in Supplementary Fig. 9). We hereby assume that a major armed conflict event will affect a chokepoint for ten days, disrupting 20% of its trade.
Similar as with the armed conflict events, terrorist attacks in the vicinity of chokepoints indicate the propensity for maritime terrorism risk. It has long been hypothesized that terrorist attacks to maritime transport are a becoming more likely, in particular when targeting the maritime routes important for global energy supply24. To characterise the probability, we extract previous terrorist events from the Global Terrorism Database25, which covers events from 1970 to 2020. We retain only relevant events, including those that have killed >5 people (defined here as major events) and those that covered bombing/explosion, hijacking and facility/infrastructure attacks (see blue markers in Supplementary Fig. 9). Using again a 50-kilometre buffer around the events leaves us with 39 relevant past attacks. We assume that a terrorist attack will affect a chokepoint for ten days, disrupting 20% of trade flowing through it.
The risk of blockage because of maritime incidents is related to the width of the chokepoint. We assume that chokepoints with a width of more than 800 m (approximately twice the length of the largest container vessel) do not face the risk of a disruption, as a maritime accident is not likely to cause a disruption of flows. This therefore only includes the Bosporus Strait, the Suez Canal and the Panama Canal. We evaluate the risks of blockages as the number of major shipping incidents happening in and around these three chokepoints. Major shipping incidents here include those classified as ‘marine incidents’ and ‘very serious marine casualty’ in the Marine Casualties and Incidents database of the Global Integrated Shipping Information System26, an information platform of the International Maritime Organization. The database includes accidents between 2006 and 2024 (see orange markers in Supplementary Fig. 9), of which there are 12 in the three chokepoints considered. We assume that a major incident blocks a chokepoint entirely for three days. This is in line with the disruption of the Suez Canal after the Evergreen ship got stuck in it in 2021.
The interstate conflict indicator captures interstate disputes that result in escalating impacts to the maritime chokepoint. We use data from the Militarized Interstate Dispute (MID) database (v5.0), which has data on MID events between 1816 and 201427. A MID is defined as a sequence of related militarized incidents, each a response to one or multiple previous ones. Disputes are always between two or more interstate system members and are non-routine incidents that are governmentally authorized actions. We only include events happening over the last 50 years in the data records (from 1974 to 2014) and those with hostility level defined as ‘Display of force’, ‘Use of force’ and ‘War’ to exclude events that only included threats to use force. The black markers in Supplementary Fig. 9 show the resulting subset of events. Each event has a representative event duration, which we use as the disruption duration. We assume that a disruption reduces trade flow through the chokepoint by 10% for event classified as ‘Display of force’, 50% for events classified as ‘Use of force’ and 100% for ‘War’ events. We take a 200 km buffer around the geographical location recorded in the database, given the large spatial impact region of interstate conflict events and intersect this with the location of the chokepoints. This results in 114 interstate conflicts found in the vicinity of the chokepoints.
Drought is only a relevant risk for the Panama Canal, which is critically dependent on water availability for the operations of its locks. The Panama Canal has faced multiple droughts in the last decade that impeded operations and transits, including in 2014, 2016 and more recently in 2023–2024. We take the latest drought event as reference event. This drought event affected the Canal for 250 days, reducing trade volume through the canal by a third. Recent analysis has shown that the probability of an event like this occurring is around once in 40 years17, which we take as the representative probability (1/40 per year).
Earthquake risk is also only a relevant risk for the Panama Canal, as it is in the vicinity of two known faults: the Pedro Miguel Fault and Limon Fault28. While an earthquake event has not yet happened, the Canal Authority has designed the dams and abutments for a 2500-year earthquake, with a corresponding peak ground acceleration of 0.97g28. We take this design criteria as our event probability. In line with the risk classification of the Panama Canal Authority29, we assume that a devastating earthquake will affect the Canal for a year, blocking trade flows altogether.
Tropical cyclones can cause extreme weather at sea, including high wind, waves and rain, that disrupt the safe navigation of vessels. Here, we use the synthetic cyclone event set of Bloemendaal et al.30, capturing 10,000 of synthetic cyclone event globally. We track all cyclone events of category-I that were within 250 km of the chokepoint, temporarily disrupting the navigation of vessels through the chokepoint. Per event, we estimate the duration of the disruption based on the number of time periods (per three-hour time slice) that a chokepoint is within the 250 km distance. The median duration is 2.6 days, while the 99th percentile is five days. The event count and annual hit ratio per chokepoint is included in Supplementary Table 5.
Risk quantification
We use the historical hazard event data to quantify the risk of maritime chokepoint disruptions and the economic loss associated with that. We propose two metrics to evaluate the systemic risk, the expected value of trade disrupted (EVTD) and the economic risk (ER) per country (i).
The EVTD (in USD/yr) can be written as follows:
$${{EVTD}}_{i}={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{T}_{i,c}}{365}{\cdot S}_{c,h}\cdot {D}_{c,h}$$
(1)
With P the annual probability of occurrence of the specific hazard (h), D the duration of the disruption (days), S the severity of flow disruption (between 0 and 1), and T the value of maritime trade of country (i) that flows through a specific chokepoint (c). This metric captures the systemic risk exposure, i.e., the amount of trade value that has the potential to be affected each year.
The ER (USD/yr) is a function of the EVTD but captures the various channels of economic losses. We consider five loss channels:
$$E{L}_{i}=E{L}_{d,i}+E{L}_{r,i}+E{L}_{p,i}+E{L}_{t,i}\,+E{L}_{l,i}$$
(2)
With ELd the loss associated with delays, ELr the loss associated with rerouting, ELp the loss associated with increased insurance premiums, ELl the loss associated with trade losses in case delay times exceed inventory levels, and ELt the loss associated with reduced toll revenues for Panama and Egypt because of reduced flows through their respective canals.
Rerouting decision
The occurrence of these different loss channels depends critically on the decision to reroute or not during disruptions of a certain duration, if at all feasible (for the Bohai Strait, Strait of Hormuz, Oresund Strait and Bosporus Strait, no rerouting is possible). The decision to reroute depends on the Dc,h and the rerouting distance \(\Delta {d}_{c}\). Rerouting happens if:
$${D}_{c,h} > \frac{\Delta {d}_{c}}{V}$$
(3)
With V being the average vessel speed, set to 16 knots. Rerouting does not happen when:
$${D}_{c,h}\le \frac{\Delta {d}_{c}}{V}{or\; no\; rerouting\; alternative}$$
(4)
If goods are not rerouted, they are delayed for the duration D.
Economic loss quantification
The economic loss of delays for a country (i) can be written as:
$${{EL}}_{d,i}={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{I}_{i,c}}{365}\cdot {S}_{c,h}\cdot \,{\sum }_{j}^{D}{VOT} \cdot j \quad {if\; no\; rerouting}$$
(5)
$$E{L}_{d,i}={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{I}_{i,c}}{365}\cdot {S}_{c,h}\cdot \,{\sum }_{j}^{\Delta {d}_{c}/V}{VOT}\cdot j \quad {if\; rerouting}$$
(6)
With I the country import value flowing through the chokepoint and VOT the value of time of a day of delay (in % per day). The VOT is assumed to be 1% of trade value per day of delay, in line with ref. 31. The loss associated with delay scales linearly with the duration (in a 7-day delay, 1 day worth of goods is 7 days delayed, 1 day worth of goods is 6 days delayed, etc.).
The economic loss associated with rerouting is only borne in case rerouting is taking place. The costs accrue to the importing country as the additional costs of longer routes (e.g., additional fuel and crew expenditure) is added to the transport costs:
$$E{L}_{r,i}=0 \quad {if\; no\; rerouting}$$
(7)
$$E{L}_{r,i}={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{{IQ}}_{i,c}}{365}\cdot {S}_{c,h}\cdot \Delta {d}_{c}\cdot {Cd}\cdot {D}_{c,h} \quad {if\; rerouting}$$
(8)
With IQ the import quantity (in volume terms) and Cd the maritime costs per t-km shipped, which is set to USD0.003 per t-km1. The import quantity is an output of the OxMarTrans model.
During the occurrence of specific hazard that can impact cargo or vessels directly, the insurance premiums that shippers pay on their cargo and vessel value can rapidly spike. For instance, the Russia-Ukraine war spiked insurance premiums for ships crossing the Black Sea to 1–3%32,33, while ships deciding to cross the Bab el-Mandeb Strait during the Houthi rebel attacks faced insurance premiums of 0.6–2%34,35. Historically, vessels crossing regions with high piracy rates faced elevated insurance premiums in the order of 0.1–0.3%36. We characterise the loss due to these insurance rates as:
$$E{L}_{p,i}\,={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{I}_{i,c}}{365}\cdot ({1\,-\,S}_{c,h})\cdot {D}_{c,h}\,\,\cdot {{IP}}_{h}$$
(9)
With IP the increase in insurance premium. We set the IP value to 1% for the interstate conflict and 0.1% for piracy, terrorist and armed conflict. As can be seen, the loss is only attributed to the fraction of trade that still crosses the chokepoint (1-S). We use the value of imports (\({I}_{i,c}\)) of any given country flow through a specific chokepoint to quantify the insurance loss to that country, as the importing economy pays for insurance fees and, hence, faces the loss due to an increase of the insurance rate.
We further add the revenue loss to the Panama and Suez Canal Authority to the estimates:
$$E{L}_{t,i}\,={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{R}_{c,i}}{365}\cdot {S}_{c,h}\cdot {D}_{c,h}$$
(10)
with R the annual toll revenue and i being Egypt for the Bab El-Mandeb and Suez Canal, and Panama for the Panama Canal. The annual revenue is set to USD5 billion per year for the Panama Canal and USD8 billion per year for the Suez Canal.
Both trade delays and rerouting decisions can cause inventories to run out, especially for firms that have adopted just-in-time inventory management practises. Here, we assume that if importing firms run out of inventories, they must stop production and face losses worth the value of imports. This therefore, does not take higher-order supply-chain losses into account. Inventories levels differ considerably per sector, country and even firms within sectors. Here, we split the total trade into three groups (g) with an associated inventory level (Tinv). We do this by approximating the fraction of firms (f) with a given inventory based on the distribution of firm-level inventory data for French firms in Lafrogne-Joussier et al.37 (see Supplementary Table 6). The inventory levels set also align with average sector-specific inventory levels reported for the United Kingdom in Pichler et al.38. We can write the trade losses as:
$$E{L}_{l,i} ={\sum }_{g}{\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{f}_{g}\cdot {I}_{i,c}}{365}\cdot {S}_{c,h}\cdot \left({D}_{c,h}-{T}_{{inv},\,g}\right) \\ {if\; no\; rerouting\; and} \, D > {T}_{{inv}}$$
(11)
$$E{L}_{d,i}=0 \quad \quad {if \; no \; rerouting \; and} \, D < {T}_{{inv}}$$
(12)
$$E{L}_{l,i}= {\sum }_{g}{\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{f}_{g}\cdot {I}_{i,c}}{365}\cdot {S}_{c,h}\cdot \left(\frac{\Delta {d}_{c}}{V}-{T}_{{inv},g}\right) \\ \quad {if\; rerouting\; and}\frac{\Delta {d}_{c}}{V} > {T}_{{inv}}$$
(13)
$$E{L}_{d,i}=0 \, {if\; rerouting\; and}\frac{\Delta {d}_{c}}{V} < {T}_{{inv}}$$
(14)
In other words, losses occur when the delay or rerouting duration exceeds the inventory level, with the economic loss being the value of production that cannot be generated during this period.
Shipping price increase
Apart from the direct economic losses due to maritime chokepoint disruptions, as discussed above, there may also be an indirect economic loss due to rising freight rates (i.e., the price paid to ship goods). This rise in freight rates is attributed to the fact that the rerouting of vessels takes out a certain capacity from the total available fleet capacity, i.e., a supply shock. Freight rates have been shown to spike due to such supply shocks, for instance because of the rerouting of ships due to the Houthi rebel attacks (2023-2025). A global shipping price increase will affect every country reliant on shipping, irrespective if they are directly affected by the chokepoint disruption.
Here, we derive a simple approach to estimate the shipping price increase as a function of the fleet capacity ‘lost’ due to rerouting. This approach follows as the same rationale as the approach by Arvis et al.39, who relates deviations in the shipping price index to share of the vessel fleet that is affected by a disruption. Given a global fleet, that could (in theory) be reallocated based on changes supply and demand patterns, we can estimate a single coefficient for all chokepoint disruptions, similar as proposed in Arvis et al.39. We calibrate the model on the increase in container freight rates during the period 2023-2024, which saw such elevated prices (e.g., dry bulk and tanker prices spiked less). To do so, we first derive the monthly (m) reduction in daily container fleet capacity (C) going through the Suez Canal and Panama Canal (c) for the period 2023–2024 compared the monthly values in 2022.
$$\varDelta {C}_{m,c}=\,1-\,\frac{{C}_{m,\,c,2023-2024}}{{C}_{m,c,2022}}$$
(15)
We do this based on data provided in the PortWatch platform (https://portwatch.imf.org/). We then construct a metric of the monthly supply shock (S), equal to the reduction in monthly fleet capacity as a fraction of the total fleet capacity and the respective delay (Panama) and rerouting (Suez) days (D):
$${S}_{m}={\sum }_{c}\varDelta {C}_{m,c}\,\cdot \,\frac{{C}_{m,c}\,}{{C}_{m}}\,\cdot \,{D}_{c}$$
(16)
With D set to 15 days for Panama (the average delay observed during the drought restrictions) and 12 days for Suez (the average rerouting duration around the Cape of the Good Hope). We then fit a linear regression between the Sm and the freight cost increase (%FC) using the Drewry World Container Index (R2 = 0.74), as shown in Supplementary Fig. 10. We then apply this approach to the modelled disruptions, replacing D with the rerouting duration, if rerouting is taking place. Using this approach, we predict the %FC per disruption event, after which we can estimate the total economic risk due to freight rate (ELf):
$$E{L}_{f}={\sum }_{h}{\sum }_{c}{P}_{c,h}\cdot \frac{{{IQ}}_{{container}}\cdot d}{365}\cdot {D}_{c,h}\cdot {Cd}\cdot\%F{C}_{c,h}$$
(17)
with IQ the global container volume imported and d the average kilometres shipped per ton of goods. d is set to 9500 km per ton based on the work of Verschuur et al. 1, who estimated that globally 9.4 billion tonnes of maritime trade is moved 90.5 trillion tonnes-km in 2015. We proxy IQ based on the global container fleet capacity (329.5 million deadweight tons) versus the global total maritime fleet capacity (2.35 billion deadweight tons) based on UNCTAD (2024)40. Hence, we set IQ to 14% of the total maritime import volume.
Joint disruption probability
For the above risk calculations, we assume that all hazard-induced chokepoint disruptions are independent from each other. It is possible that hazards have some level of dependence at the chokepoint level (i.e., the occurrence of one hazard can trigger another one) and across chokepoints (i.e., one hazard can affect multiple chokepoints simultaneously). We therefore evaluate whether this assumption, which is necessary to make the risk calculation tractable, is indeed justified. Earthquake and drought hazards are not considered here, given that they only affect one chokepoint (Panama Canal). Similarly, blockage events are also excluded for the analysis, given that we consider them random in nature.
To test the dependency between hazard-induced disruptions, we perform two sets of further analysis. First, at each chokepoint, we test whether there is a statistically significant relationship between two hazards occurring within the same year. We only do this for the four hazards where we expect such as potential relationship: armed conflict, piracy, terrorist attacks and interstate conflict. For all overlapping years between each combination of hazard, we classify whether a given year faced a hazard or not (binary 1 or 0). We then run Fisher’s exact test41 to test whether binary time series are considered independent or not by calculating the contingency table between hazard incidences. In total, this results in 6 combinations ((4 × 3)/2) per chokepoint, so 144 possible combinations in total. Given the sparsity of events, we deem a joint-occurrence of events to be statically significant if (i) we find a p-value smaller than 0.05, and (ii) joint-occurrence is observed at least twice. In Supplementary Table 2, we report all the statistically significant combinations of hazards affecting a specific maritime chokepoint.
Second, we test whether there is a statistical dependence between hazards across chokepoints. This could be because an event can affect two chokepoints at the same time, such as a cyclone event, or because of some other spatial interdependency (e.g., an armed conflict spilling over to another region). For cyclone events, we evaluate the number of times an event affects multiple chokepoints, within the synthetic event dataset we use (see ‘Tropical Cyclones’ section). The results are reported in Supplementary Table 4 (see below). For armed conflict, piracy, terrorist attacks and interstate conflict events, we again analyse the joint occurrence based on the binary time series, but now across chokepoints and across hazards (including similar hazards). Per hazard combination considered, this yields 276 chokepoint combinations (24 × 23/2), totally 4416 ((4 × 4) x 276) combinations. Here, we define a joint-occurrence of events to be statistically significant if (i) we find a p-value smaller than 0.05, and (ii) the joint incidence occurs at least five times. In Supplementary Table 3, we report all the statistically significant combinations of hazards across different chokepoints.