ADX ESG Intelligence Replication Toolkit: Auditable NLP-Based ESG Disclosure Signals and Investor-Flow Validation for the Abu Dhabi Securities Exchange

Corporate ESG reports are becoming central to investor decision-making, yet they remain difficult to compare because they are lengthy, unstructured, and often mix substantive evidence with promotional sustainability language. This repository provides an open ESG intelligence replication toolkit that converts corporate ESG, sustainability, annual, and integrated reports into auditable disclosure signals for firms listed on the Abu Dhabi Securities Exchange. The toolkit extracts ESG-related text, classifies disclosure content, constructs ESG visibility and credibility measures, links those measures to event-study and regression outputs, and validates the resulting signals using ADX market and investor-type trading data. The framework distinguishes between “green talk,” defined as visible but weakly evidenced ESG communication, and “green proof,” defined as specific, verifiable, evidence-backed ESG disclosure. The package includes Python scripts, configuration files, documentation, notebooks, processed outputs, validation checks, and replication guidance. It is designed to support transparent ESG measurement, reproducible financial-text analysis, and investor-oriented sustainability research in emerging markets. Main Contribution This toolkit contributes to sustainable finance and financial-text analysis by providing an open, auditable, and replicable ESG signal extraction pipeline. Unlike proprietary ESG ratings or closed classification systems, the framework is designed so that ESG disclosure scores can be traced back to the underlying source documents and scoring logic. The toolkit, therefore, supports not only empirical testing but also methodological transparency. The toolkit offers four main contributions: Contribution Area Description Methodological Provides a Python-based workflow for extracting and scoring ESG disclosure signals from corporate reports. Measurement Distinguishes ESG visibility from ESG credibility, allowing researchers to separate promotional ESG communication from evidence-backed disclosure. Empirical Links ESG disclosure signals to ADX event-study outputs, regression results, and investor-type trading summaries. Open Science Provides documentation, code, data dictionaries, output schemas, and validation checks to support replication and extension. Conceptual Framework The toolkit follows the logic below: Corporate ESG Reports ↓PDF Text Extraction ↓ESG Sentence Identification ↓Credibility and Visibility Scoring ↓Firm-Report ESG Signal Construction ↓Event-Study and Regression Merge ↓Investor-Flow Validation ↓Replicable ESG Intelligence Outputs The framework is built around the distinction between the following: ESG_MKT = ESG visibility / communication intensityESG_AI = ESG credibility / evidence-backed disclosure quality This distinction supports a practical investor-facing classification: ESG Visibility ESG Credibility Signal Type Interpretation High High Green proof ESG communication is visible and supported by credible evidence. High Low Green talk ESG communication is prominent but weakly evidenced. Low High Hidden ESG quality Credible ESG evidence exists but is not heavily promoted. Low Low Low ESG information Limited ESG visibility and limited evidence-backed disclosure. Technical Scope The archive includes a structured replication workflow with the following components: config/data/docs/notebooks/output/scripts/README.mdrequirements.txtrun_all.py The code supports: ADX OHLCV data preparationESG report text extractionESG sentence identificationESG credibility and visibility scoringevent-study output constructionregression analysisinvestor-type trading validationquality checks and diagnosticsdocumentation and output manifests The market-data workflow is designed to retrieve daily ADX open, high, low, close, and volume data using the Twelve Data API and an environment variable. The ESG extraction workflow is designed for a Google Colab / Google Drive environment and identifies ESG-related sentences from PDF reports using environmental, social, and governance keyword dictionaries. Main Outputs The toolkit can generate or reproduce the following research outputs: auditable ESG sentence-level signal filesreport-level ESG credibility and visibility scoresESG signal dashboardsevent-study outputs merged with ESG quality measuresregression analysis samplespublication-style regression tableshypothesis summary tablesmarket-wide investor-type flow validation filesPDF extraction diagnosticsdata-quality reportsfile manifests and checksums These outputs are designed to support both empirical research and methodological auditability. Suggested Citation [Veliota, Drakopoulou]. (2026). ADX ESG Intelligence Replication Toolkit [Software and derived data]. Zenodo. https://doi.org/[Zenodo DOI] Data Availability and Use Note This archive contains code, documentation, processed outputs, validation files, and replication materials. Large corporate PDF reports may be stored externally because of file-size limits and third-party redistribution considerations. Users who wish to rebuild the ESG extraction stage should provide a local or Google Drive path to the relevant corporate report folder. Market data and investor-type trading data may be subject to exchange, vendor, or institutional access conditions. Users should verify licensing and redistribution rights before publicly sharing raw market data, raw exchange data, or corporate PDF report files. The investor-type validation component uses market-wide daily ADX investor-type trading summaries. Therefore, investor-flow results should be interpreted as evidence on aggregate individual-investor trading behavior around ESG disclosure dates, not as evidence of firm-level retail investor buying or selling unless security-level investor-flow data are added by the user.