Splash247: Shipping’s AI forecasters run into the limits of history

Published by Splash247

Machine learning is improving freight-rate forecasting, but shipping’s volatility remains beyond the reach of any model. A review of 28 studies published between 2012 and 2024 found growing use of vessel supply, commodity demand, bunker prices and economic indicators to predict markets. Yet it also concluded that geopolitical shocks, regulation and sudden supply disruptions cannot be captured reliably from historical data.

Industry figures agree that AI is best treated as decision support. Arrow’s Burak Cetinok told sister title SplashTech that modern tools process datasets faster, but weak data quality and limited coverage still constrain accuracy. Panos Patsadas of Trans Global Projects argued shipping remains insufficiently digitised and warned that past cycles are poor guides to future behaviour.

Sea’s Christoffer Svärd said output depends heavily on configuration, prompting, context and human interpretation. 

Roar Adland of SSY was more sceptical, arguing that large language models add little to the machine-learning systems already used for more than a decade and remain unsuitable for autonomous commercial decisions.

Experts identified three drivers of freight rates: predictable fundamentals, unpredictable shocks and short-term herd behaviour. AI may improve analysis of the first, but does little for the second and could amplify the third. If participants use similar models and data, they may crowd into the same positions, accelerating market moves and undermining forecasts.

The consensus is that AI can sharpen analysis, reveal patterns and improve timing, but it will not remove shipping cycles or replace judgement.