Port Dwell Distributions: Newcastle 2.8×, Houston 13.6×. The Tail Ratio Predicts Demurrage Risk.

Median dwell time tells you what a typical port call looks like. The P90/P50 ratio tells you whether your charterer should buy demurrage insurance.

The Setup

Pull every vessel visit completed in the last 30 days at the world's busy ports, compute the median (P50) and 90th percentile (P90) of total port-stay duration, and divide. The result is a tail-ratio number that ranges from about 2× at single-commodity bulk terminals to 14× at multi-cargo container/general hubs. The ratio is a more useful charterer-risk measure than the median, because the median tells you the planned outcome and the tail tells you what the laytime clock actually has to absorb.

Concrete numbers from the last 30 days:

• Houston Ship Channel — P50 3.2h, P90 43.4h, ratio 13.6×
• Singapore — P50 5.5h, P90 69h, ratio 12.5×
• Norfolk — P50 4.2h, P90 91.6h, ratio 21.8×
• Antwerp — P50 8.8h, P90 146.4h, ratio 16.6×
• Rotterdam — P50 26.7h, P90 194.4h, ratio 7.3×
• Hamburg — P50 32h, P90 166.5h, ratio 5.2×
• English Channel — P50 8.8h, P90 41h, ratio 4.7×
• Newcastle (Australia) — P50 14h, P90 39.1h, ratio 2.8×
• Gladstone — P50 5h, P90 26.8h, ratio 5.4×
• Los Angeles — P50 3.9h, P90 7.9h, ratio 2.0×

LA's 2.0× and Newcastle's 2.8× are the cleanest tail distributions in the dataset. Houston's 13.6× and Norfolk's 21.8× are the messiest.

The Chain

The mechanic is operational diversity. A single-commodity bulk terminal — Newcastle for coal, Gladstone for coal/alumina, LA for containers via PierPass auto-booking — handles a narrow range of vessel sizes, cargo types, and discharge processes. The variance in dwell time is bounded by the physics of cargo handling, which is well-modeled and predictable. A 180,000 DWT Capesize loading coal at Newcastle takes between 12 and 36 hours, almost always. The P90 case is a vessel that arrived with a draft survey complication or had to wait for a tide; it doesn't multiply the median by 10×.

A multi-cargo port like Houston or Antwerp does not have that bound. A small bunker barge call at Houston is a 90-minute affair. A petrochemical tanker waiting for terminal clearance can sit at anchor for 60 hours. A bulk grain vessel waiting for vessel-clearance after loading might sit another 30 hours waiting for pilotage out. The same port produces dwell times that span two orders of magnitude depending on cargo type, terminal, and queue state. The P90 doesn't represent "exceptional cases" — it represents a mode that's structurally always present.

The implication for laytime contracting: a charter party negotiated against a Houston call needs a much wider laytime allowance and demurrage clause than the same vessel routed to Newcastle, even when the median dwell looks similar. The risk is not in the typical case; it's in the right tail. A 13.6× tail ratio means a 1-in-10 call burns 12+ hours more than the median expects, every single voyage.

The Implication

Charterers with multi-port voyages should rank ports not by P50 efficiency but by P50 × P90/P50. That product is the effective cost of laytime risk, and it ranks ports very differently than median dwell does:

• Lowest effective laytime risk: Newcastle (39h × 1.0 = 39 risk-hours), LA (8h), Gladstone (27h)
• Mid-range: English Channel (41h), Houston (43h)
• Highest: Rotterdam (194h), Antwerp (146h), Hamburg (166h), Norfolk (92h)

The Hamburg/Rotterdam/Antwerp tier deserves a note. These ports have long P50 dwell because they are multi-call hubs — vessels typically transit through Antwerp on a Northern European loop with multiple discharge stops. The high P90 is mostly contractual (waiting for next discharge slot), not operational (waiting for cargo handling capacity). For a charterer running a fixed loop, the dwell is built into the schedule and isn't really risk. For a spot-market charterer treating Antwerp as a one-off discharge, that 146h tail is a hard cost.

What to Watch

• Rotterdam dwell-time stability. Rotterdam runs the longest absolute dwell of any port in the dataset; if its P90 starts climbing above 200h, North Sea congestion is recurring and there is downstream pressure on Hamburg and Antwerp slots.
• Houston tail compression. Houston's 13.6× is high but not exceptional for a US Gulf multi-cargo port. If it tightens to 8-10× over the next 30 days, the post-cold-front clearing of the Galveston Bay queue is real; if it widens further, the queue is structural and rates respond.
• Newcastle vs Gladstone divergence. Both are Queensland coal hubs and should track each other. Persistent divergence suggests a terminal-specific issue (industrial action, equipment downtime).
• English Channel P90. The Channel itself isn't a port — it's a transit anchorage where vessels wait for North Sea slots. P90 above 50h indicates downstream congestion is backing into the queue.

Limitations

The dataset is 30 days of vessel visits with both arrived_at and departed_at populated. Visits that haven't closed yet are excluded, which biases toward shorter dwell times — currently in-port vessels are still adding to their stay. The bias is small for ports with median dwell under 24h (most calls have closed) and larger for Rotterdam/Hamburg where the median is already over a day.

Vessel class also matters and is not controlled for here. Ports that handle a higher mix of small vessels — bunker barges, harbor craft — will show shorter P50 by composition, not by efficiency. The Hamburg P50 of 32h is partly a function of the cargo mix being heavily containerized and tanker; a bulk-only Hamburg P50 would be different. A version of this analysis sliced by vessel_class is worth filing as a follow-up.

The "tail ratio" as a charterer signal is a heuristic, not a contractual standard. Lloyd's standard form charter parties speak in days of allowed laytime, not in distribution ratios. The contracts industry will catch up to this kind of analysis when it's clear the ratio predicts demurrage outcomes better than the rules of thumb currently in use.

Data current as of 2026-04-28. Source: Axiom Overwatch vessel_visits table, 30-day completed-visit window, ports with at least 30 completed visits. Underlying AIS port-call detection uses port geofence polygons. axiomoverwatch.io/blog