Conference Agenda

Containerized maritime transport is a core pillar of global trade, yet persistent imbalances between import and export flows continuously generate spatial mismatches in container availability. To manage these imbalances, liner shipping companies rely on empty container repositioning (ECR), a complex and cost-intensive process that ensures equipment availability across global networks. In recent years, this task has become increasingly challenging due to recurring global disruptions, including pandemics, geopolitical conflicts, and supply chain volatility.

ECR decisions depend on accurate forecasts to estimate future container demand and support operational planning. Increased uncertainty during trade crises gives leverage to human planners during the forecasting process, as they review and adjust system-generated predictions using contextual and operational knowledge. Understanding how users interact with these forecasts, particularly under varying levels of uncertainty and across different forecast horizons, is therefore crucial for improving decision quality in operational logistics environments.

This study examines user adjustment behavior and its effectiveness in an operational empty container repositioning forecasting environment at a major liner shipping company through a large-scale field study. The analysis is based on a comprehensive dataset covering 2020–2024 with more than 1.3 million adjusted forecasts across multiple container types. It includes system-generated forecasts, predominantly produced by AI-based and statistical models, realized demand, and user-adjusted forecasts submitted by practitioners for empty container releases at depots and terminals serving customer export shipments.

The empirical setting distinguishes between a crisis period (2020–2022), marked by severe disruptions such as the COVID-19 pandemic and a more stable period (2023–2024). Methodologically, the study combines established Forecast Value Added metrics (e.g., RelAME, MAPE, bias) with regression-based models that examine the determinants of forecast adjustment behavior and its impact on forecast accuracy across forecast horizons and different levels of environmental uncertainty. The paper aims to answer the following question: (1) Do human planners improve forecast accuracy greater during times of crises (and if so, in which means) (2) does the forecast horizon influence adjustment behavior and performance, and (3) does the influence of the forecast horizon change between stable and unstable times?

This study contributes to research in several ways. First, it extends existing research by providing a large-scale empirical analysis of human adjustment behavior in maritime demand forecasting. Second, it builds on prior work by examining human responses to AI-generated forecasts in an ECR setting across multiple forecast horizons. Finally, the findings offer insights into the role of human judgment in expert-supported forecasting processes under conditions of operational disruption and uncertainty.

Results show a pronounced shift in user behavior during crisis conditions, particularly in the magnitude of forecast adjustments. While forecast value added is higher at shorter forecast horizons, overall human interventions tend to reduce forecast accuracy on average. During crisis periods, however, planners contribute value primarily through downward adjustments, indicating that human judgment can partially offset systematic over forecasting by system forecasts under uncertain conditions.