Financial analysts rely on identifying key factors and opinions relevant to their targeted assets to make timely and informed investment decisions. However, the overwhelming volume of market analysis reports and the dynamic nature of financial markets make it challenging to manually track and summarize these critical insights efficiently. To address this real-world need, we introduce the task of asset-centric Factor-Opinion Pair Extraction (FOPE) from market analysis reports. This task is designed to unsupervisedly identify the most relevant factors and associated opinions for given assets, enabling analysts to stay informed and make data-driven decisions. We propose a novel four-stage framework that integrates advanced large language models with domain-specific expertise to achieve this goal. Experiments on a dataset of 11,560 articles spanning eight macro asset classes demonstrate the effectiveness of our method in terms of coverage, fluency, coherence, relevance, and faithfulness. This work provides a practical solution to a pressing real-world problem, enabling efficient extraction of critical factor-opinion pairs for informed financial decision-making.