MD&A-ESP: Dataset and Code for Attributing Predictive Content Across Sentiment, Structure, and Fundamentals
Published: 4 June 2026| Version 1 | DOI: 10.17632/gg6t9xgwhb.1
Contributors:
Mengchao Sun, Description
This study builds a modular decision support system (MD&A-ESP) that decomposes the predictive content of Management Discussion and Analysis (MD&A) text across three information layers—sentiment tone, text structure, and financial fundamentals. We conduct a full factorial evaluation of 144 configurations (4 sentiment extractors × 3 feature configurations × 6 prediction models × 2 targets) on 43,232 MD&A documents from 4,745 Chinese A-share firms (2016–2024).
Files
Institutions
- China University of Geosciences (Beijing)Beijing, Beijing
Categories
Computer Science, Business, Management, Economics