EUVIMEDEUVIMED entdecken
Crossref

Design and Evaluation of a Hybrid Natural Language Processing and Machine Learning Model for Economic News Sentiment Analysis to Predict Price Fluctuations in the Iran Mercantile Exchange

Amin Jamali, Hamid Reza Jamali

Accounting, Finance and Computational Intelligence · 2027 · 5 (5) · 1-25

DOI: 10.61838/jafci.485 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Accounting, Finance and Computational Intelligence, 2027. Verfasst von Amin Jamali und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study aimed to design and evaluate a hybrid Persian natural language processing and machine learning model for extracting economic news sentiment and predicting steel price fluctuations in the Iran Mercantile Exchange. This applied-developmental study employed a descriptive-analytical design based on secondary textual and time-series data. The dataset included 45,836 economic, industrial, and financial news articles published between January 2019 and January 2026 and 1,486 synchronized daily observations of steel billet and sheet transactions. A fine-tuned ParsBERT model classified news as positive, negative, or neutral, and a daily sentiment index was calculated using source credibility, publication timing, and news intensity. Numerical variables included closing prices, trading volume, supply, demand, the informal exchange rate, the global iron ore price, and technical indicators. Random Forest, LSTM, XGBoost, ParsBERT-LSTM, and ParsBERT-XGBoost were compared using walk-forward validation. ParsBERT achieved a sentiment-classification accuracy of 89.4% and an F1 score of 0.887. Granger causality analysis showed that the two-business-day lag of news sentiment significantly predicted price returns, with F = 8.75 and p = 0.003. ParsBERT-XGBoost produced the best performance, with an RMSE of 710 and an MAE of 520 Iranian rials per kilogram. Compared with numerical XGBoost, the hybrid model reduced RMSE by 27.6%. It also achieved a directional accuracy of 72.6%, a win rate of 67.4%, a Sharpe ratio of 1.76, and a maximum drawdown of 9.2%. The informal exchange rate, news sentiment index, and global iron ore price were the most influential predictors. Integrating Persian news embeddings with fundamental, transactional, and technical variables through the ParsBERT-XGBoost architecture improved forecasting accuracy and risk-adjusted performance and may provide an effective decision-support and early-warning tool for the Iran Mercantile Exchange.

Zitieren

APA

Jamali, A., & Jamali, H. R. (2027). Design and Evaluation of a Hybrid Natural Language Processing and Machine Learning Model for Economic News Sentiment Analysis to Predict Price Fluctuations in the Iran Mercantile Exchange. Accounting, Finance and Computational Intelligence, 5(5), 1-25. https://doi.org/10.61838/jafci.485

Vancouver

Jamali A, Jamali HR. Design and Evaluation of a Hybrid Natural Language Processing and Machine Learning Model for Economic News Sentiment Analysis to Predict Price Fluctuations in the Iran Mercantile Exchange. Accounting, Finance and Computational Intelligence. 2027;5(5):1-25. doi:10.61838/jafci.485

BibTeX
@article{Jamali2027,
  author = {Jamali, Amin and Jamali, Hamid Reza},
  title = {Design and Evaluation of a Hybrid Natural Language Processing and Machine Learning Model for Economic News Sentiment Analysis to Predict Price Fluctuations in the Iran Mercantile Exchange},
  journal = {Accounting, Finance and Computational Intelligence},
  year = {2027},
  volume = {5},
  number = {5},
  pages = {1--25},
  doi = {10.61838/jafci.485}
}
Kostenlos starten →