Bunker Insights: External Methodology

1. Introduction

Bunker fuel is the single largest voyage cost for most commercial vessels, typically 50-60% of total voyage expenses. Yet for most operators, buying bunker fuel is still an opaque process: relationships with brokers and suppliers, lagged price assessments, and educated guesswork. The market lacks an independent, rigorous benchmark providing a fair market price at a specific port on a specific day.

Bunker Insights delivers daily bunker pricing estimates across 1,300+ ports. It's built on proprietary data science modelling that brings together crude settlement data with hundreds of thousands of aggregated and anonymized bunker transactions. The result is an independent price benchmark that operators can use to validate quotes, inform negotiations, plan voyages, and manage fuel cost exposure with confidence.

The core advantage behind Bunker Insights is that price estimates are generated by a model trained on real, aggregated and anonymized bunker transaction data. This is what makes it fundamentally different from broker assessments, observational or survey-based alternatives.

2. Data Sources

Bunker Insights is built on transaction data, not market opinion. Every price estimate the system generates is trained on bunker transactions executed by real operators, not broker assessments, indicative supplier quotes, or price-reporting-agency surveys, all of which introduce some degree of approximation, selection bias, or commercial interest. Our training data reflects what vessels actually paid, where, for what fuel, and in what quantity.

Our model uses the following as input data:

• Bunker transactions: hundreds of thousands of actual executed bunker transactions, recorded from 2012 onwards, including the final agreed price paid. This dataset grows continuously, so the model's accuracy improves over time as more transaction history accumulates. Data is used to power Bunker Insights in aggregated and anonymized form; outputs reflect broad market patterns and are not traceable to any individual transaction or location.

• Crude settlement prices: daily closing spot settlement prices sourced from CME, used as the fundamental market driver of bunker fuel costs.

• Port and routing data: navigable maritime distances between ports, used to model how prices propagate across the global bunkering network.

3. How the Data is Compiled

Bunker Insights uses a proprietary deep-learning model that combines transaction history, crude oil market data, and a model of how prices propagate across the global port network to generate daily probabilistic price estimates for bunkering ports worldwide.

Fuel type normalization. Fuel type labels in real procurement data are not always consistent: the same physical fuel can appear under several different names depending on the supplier or how the record was entered. Bunker Insights maps this variation into three primary types:

Before any transaction enters the model, it passes through a cleaning process: only contracts with a confirmed procurement status are retained, quantity outliers are excluded, and automated controls guard against any single source disproportionately influencing the output.

Port hub network. A small number of major ports (Singapore, Rotterdam, Fujairah, Houston, and a handful of others) dominate the global bunkering market and effectively set prices for the broader network. Smaller ports typically price at a premium above their nearest hub, reflecting additional logistics costs. Hub rankings are derived each year from actual transaction volumes rather than a fixed list, and every other port is assigned to its reference hub based on navigable maritime distance. That distance feeds into the price estimate as a proxy for the logistics premium more remote ports carry. Because hub rankings are recalculated annually, historical predictions always reflect the hub structure that was accurate at the time, not today's structure applied retrospectively.

Market sentiment. An additional input captures whether prices are currently running above or below their historical average, derived from recent transaction data.

Refresh cadence. The model is retrained daily to incorporate the latest transaction and to learn new data patterns.

Scenario and forward-looking views. Once calibrated, the model can represent the relationship between crude price and bunker price under different conditions. This supports two related capabilities:

• Estimates consistent with crude futures: The model can generate a series of forward bunker prices consistent with crude futures prices. This supports multi-month voyage planning and financial risk management decisions.

• Scenario analysis: any hypothetical crude price can be used as a model input, allowing users to model bunker cost implications across different market environments. Available for ~9,500 ports at any quantity volume. This is useful for stress-testing voyage economics, annual budget planning, or assessing exposure to crude price movements.

4. Data Quality & Validation

Bunker Insights maintains ~95% accuracy across all three fuel types (HFO, LFO, MDO). How that figure is measured matters as much as the figure itself.

Backtesting. Accuracy is measured the same way the product is used in practice: predicting prices that aren't yet known using only information available prior to the model training date. This way, every component of the system uses only information that would genuinely have been available on the date a prediction was made: hub rankings use only prior-year data, and crude oil prices use only the prior day's close. No future data is used to inform a past prediction, using a strict point-in-time architecture. In practice, to assess historical performance, the system loads the model as it stood the day before each test date and compares its predictions against transaction prices realized afterwards. The resulting accuracy figures reflect the errors the system would genuinely have made in production, not a retrospective fit to data the model already saw.

Confidence intervals. Every estimate is delivered with a confidence interval rather than a single price point, reflecting two sources of uncertainty: how consistent the model is for a given port and market condition, and how sensitive the estimate is to recent crude price and market sentiment volatility. Users can select a narrower interval (focused on the most likely outcomes, useful for quote validation and negotiation) or a wider one (covering more extreme cases, useful for budget and risk planning). The central estimate stays the same; only the width of the range changes.

5. Coverage & Scope

• Port coverage: daily price estimates for 1,300+ ports worldwide.

• Historical time series: available for approximately 20 major bunkering hubs, from January 2018 onwards. This does not cover the full port list available for current-day estimates.

• Scenario analysis: ~9,500 ports, any crude quantity, any fuel quantity.

• Fuel types covered: HFO, LFO, and MDO (see Section 3 for market-standard equivalents). Alternative fuels will be added as market adoption and available transaction history grow.

• Quantity tiers: Predictions are generated across a range of standard procurement quantities for each fuel type, producing a volume-price curve that reflects realistic volume-discount effects at each port. This means buyers can assess price sensitivity to quantity, not just a single point estimate for a fixed stem size.

Disclaimer

CME Group market data is used under license as a source of information for certain Veson Nautical LLC (Veson) products. CME Group has no other connection to Veson products and services and does not sponsor, endorse, recommend or promote any Veson products or services. CME Group has no obligation or liability in connection with the Veson products and services. CME Group does not guarantee the accuracy and/or the completeness of any market data licensed to Veson and shall not have any liability for any errors, omissions, or interruptions therein. There are no third-party beneficiaries of any agreements or arrangements between CME Group and Veson.