Indonesian Corporate Board Facial Features Dataset (2021–2023): Firm-Year Aggregated Facial Action Units and Perceived Traits from Annual Report Photographs

This dataset contains firm-year aggregated facial features extracted from the photographs of directors, commissioners, and other officers published in the annual reports of Indonesian publicly listed companies for fiscal years 2021–2023. It accompanies the paper "Predicting Relative Stock Performance from Automated Facial Expression Analysis of Indonesian Annual Report Photographs." The underlying source material comprises 30,553 facial photographs extracted from 2,140 annual reports obtained from the Indonesia Stock Exchange (IDX) and company websites. Facial Action Units (AUs) were extracted using the open-source library Py-Feat, perceived personality traits (competence, dominance, likeability, facial maturity, trustworthiness, and attractiveness, rated on a 1–7 scale) were scored using a large language model's vision capabilities following Rule and Ambady (2008), and composite smile-authenticity metrics (Duchenne smile indicators, fake-smile flags, smile tightness ratios, and related measures) were engineered as described in the paper. Each row represents one company-year, identified by the IDX ticker (kode_emiten) and fiscal year (tahun). For each of three board roles (Commissioner, Director, Other), every underlying facial feature is aggregated using four statistics: mean, minimum, maximum, and standard deviation. Feature names follow the pattern Role_feature_statistic (e.g., Director_AU12_mean). Features suffixed with _year_z are z-score normalized cross-sectionally within each fiscal year to remove cohort effects. Additional columns include estimated age statistics, discrete emotion probabilities (anger, disgust, fear, happiness, sadness, surprise, neutral), the proportion of female members per role (Role_female_ratio), and indicator columns for role presence (has_Commissioner, has_Director, has_Other). To protect individual privacy, only firm-level aggregates are released; no individual-level records, names, or images are included. All source photographs originate from public regulatory disclosures in which the individuals appear in their official corporate capacity. If you use this dataset, please cite the accompanying paper.