From news article to expert opinion: teaching AI to decode shipping market dynamics with textual analytics

Human experts achieve robust market forecasts by efficiently distilling actionable insights from complex market dynamics. However, the exponential expansion of market information has made experts’ limited attention even more strained. Although the Artificial Intelligence (AI) technology has shed lights on automated evaluation, a key challenge still remains: how to teach AI to decode shipping market dynamics with textual analytics as human experts do. Thus, our study attempts to pioneer the Large Language Models (LLMs) to emulate human experts in decoding shipping market dynamics from news articles. A corpus of 11,187 TradeWinds articles—including headlines, abstracts, and full texts—was deliberately designed to teach the algorithm through supervise learning. Technically, a Low-Rank Adaptation (LoRA) fine-tuning approach was developed to enhance domain-specific knowledge. Additionally, a Chain-of-Thought (CoT) reasoning mechanism which is rooted in classic shipping economic theory and chartering practice was integrated to guide the LLM in expert-like reasoning processes. Experimental results demonstrate that the LLM-CoT significantly outperforms Natural Language Processing(NLP) and LLM in advancing market news interpretation, capturing market dynamics, and risk monitoring, particularly in context where sentiment cues are ambiguous. Moreover, this study offers policy implications to enable more effective, data-driven decision-making and market risk management.