2024 年 2024 巻 FIN-033 号 p. 163-168
Domain-adaptation causal knowledge plays an important role in an ever-evolving frontier of intelligent inference that references expert cognition in addressing complex decision-making problems. With the recent increase in uncertainty, cognitive biases may induce managers to overestimate their decision-making abilities and the probability of success, potentially leading to irrational decisions. It is both academically and practically important to develop methods from a semantic perspective to directly measure managerial cognitive patterns, apply standardized processes to larger samples, and provide empirical evidence of their impact on corporate decision-making. In this study, we provide a financial expert-annotated dataset to extract CC-BizEnv (Causal Cognition of Business and Environment) of managers from Japanese annual reports. This dataset is based on more granular "cause-effect" annotations and clue expressions for fine-tuning of Pre-trained Natural Language Model (NLM). As a result, according to pre-set ``clues'', CC-BizEnv allows the extraction of sentences containing causal rationale and classification of ``internal cause'', ``external cause'', ``positive effect'' and ``negative effect'' from narratives with 87.4% F1 score. We apply this approach to measure managerial cognitive patterns and identify cognitive biases, such as self-serving attributional bias.
