Dry-Bulk Shipping Dynamics: A Cohort-Based Model of Fleet Adjustment and Freight Rates with History-Fitted and Language-Model Shipowners
Year: 2026
Type: preprint
Abstract: Shipping investment is a canonical case of decisions under delayed capacity adjustment, yet shipowners’ decision rules are hard to identify from a few aggregate cycles. We ask whether decision rules elicited from large language model (LLM) agents can be estimated and evaluated inside a dynamic market model. A cohortbased model of the world dry-bulk market links age-structured fleets, distributed delivery lags, scrapping and convex freight clearing. Each of two owner types is represented by an equation calibrated to history and by an equation estimated from LLM decisions on predetermined states. Conditional on observed utilisation, the freight block fits earnings closely; this is a conditional fit, not a forecast. The short, conflicting public contracting series gives limited support for a positive profitability response within the maintained specification. Across 1,860 elicited decisions, the estimated rules reproduce the agents on frozen test states and are stable across repeated runs. A factorial brief attributes most of the owner-type difference to the stated time horizon; removing the delivery lag halves the orderbook response; incorrect rules of thumb leave the estimated rules unchanged; a disguised, non-shipping version yields a similar pattern. Elicited scrapping tracks observed demolition, but elicited ordering does not track observed contracting. In simulations, extrapolative expectations amplify earnings fluctuations and strengthen a long-period component, while individual peaks occur more often. Internal fidelity and stability therefore do not establish external validity. All data are public; the model, data and experiment record are released with an interface for licensed industry data.
Author: Sarastov, Zlatin
Institution: University of National and World Economy
Language: English
Topic: Maritime Ports and Logistics
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225326691