EUVIMEDEUVIMED entdecken
Crossref

Design of a Text-Mining Model Based on ParsBERT and a Financial Sentiment Lexicon for Extracting Stock-Selection Signals from Board of Directors’ Activity Reports: Evidence from the Tehran Stock Exchange

Hossein Nabieiboroujeni, Hamidreza Jafari Dehkordi

Dynamic Management in Business Analysis · 2026 · 1-22

DOI: 10.61838/dmbaj.500 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Dynamic Management in Business Analysis, 2026. Verfasst von Hossein Nabieiboroujeni und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Objective

This study aimed to design and validate a multilayer text-mining model for converting board of directors’ activity reports of Tehran Stock Exchange-listed firms into quantitative stock-selection signals and to test the economic value of combining these signals with accounting information. Methodology: This applied, ex post facto study examined 330 firms listed on the Tehran Stock Exchange, from which 161 firms were selected through systematic elimination for the 2009–2023 period. The four-layer model comprised Persian-text preprocessing using Hazm, development of a domain-specific financial sentiment lexicon incorporating negation and intensification controls, polarity classification using ParsBERT, and conversion of textual indicators into calibrated probability scores using a support vector machine with Platt calibration. Three portfolios were constructed using textual information, accounting information, and their combination. Portfolio efficiency was measured using the slacks-based measure of data envelopment analysis. The study hypothesis was tested using Welch’s t-test, a 95% confidence interval, and Cohen’s d effect size.

Findings

Welch’s t-test demonstrated that the hybrid portfolio achieved significantly greater efficiency than the text-based portfolio, with a mean efficiency difference of 0.354 (t=2.48, df=19.8, p=0.022). The 95% confidence interval for the mean difference ranged from 0.056 to 0.652 and excluded zero. Cohen’s d was 1.06, indicating a large effect and supporting the research hypothesis. Sensitivity analyses involving alternative risk-free rates, alternative risk definitions, and exclusion of 2020 and 2021 did not alter the direction of the findings.

Conclusion

Textual signals extracted from board activity reports provide greater economic value when integrated with accounting information rather than used independently; accordingly, combining ParsBERT-based natural language processing with fundamental accounting indicators can provide a more effective framework for stock-selection decisions in the Iranian capital market.

Zitieren

APA

Nabieiboroujeni, H., & Dehkordi, H. J. (2026). Design of a Text-Mining Model Based on ParsBERT and a Financial Sentiment Lexicon for Extracting Stock-Selection Signals from Board of Directors’ Activity Reports: Evidence from the Tehran Stock Exchange. Dynamic Management in Business Analysis, 0(0), 1-22. https://doi.org/10.61838/dmbaj.500

Vancouver

Nabieiboroujeni H, Dehkordi HJ. Design of a Text-Mining Model Based on ParsBERT and a Financial Sentiment Lexicon for Extracting Stock-Selection Signals from Board of Directors’ Activity Reports: Evidence from the Tehran Stock Exchange. Dynamic Management in Business Analysis. 2026;0(0):1-22. doi:10.61838/dmbaj.500

BibTeX
@article{Nabieiboroujeni2026,
  author = {Nabieiboroujeni, Hossein and Dehkordi, Hamidreza Jafari},
  title = {Design of a Text-Mining Model Based on ParsBERT and a Financial Sentiment Lexicon for Extracting Stock-Selection Signals from Board of Directors’ Activity Reports: Evidence from the Tehran Stock Exchange},
  journal = {Dynamic Management in Business Analysis},
  year = {2026},
  volume = {0},
  number = {0},
  pages = {1--22},
  doi = {10.61838/dmbaj.500}
}
Kostenlos starten →