본문 바로가기
  • Home

Prediction of Stock Returns from News Article‘s Recommended Stocks Using XGBoost and LightGBM Models

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2024, 29(2), pp.51-59
  • DOI : 10.9708/jksci.2024.29.02.051
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : January 15, 2024
  • Accepted : February 13, 2024
  • Published : February 29, 2024

Yoo-jin Hwang 1 Seung-yeon Son 1 Zoon-Ky Lee 2

1연세대학교 정보대학원
2연세대학교

Accredited

ABSTRACT

This study examines the relationship between the release of the news and the individual stock returns. Investors utilize a variety of information sources to maximize stock returns when establishing investment strategies. News companies publish their articles based on stock recommendation reports of analysts, enhancing the reliability of the information. Defining release of a stock-recommendation news article as an event, we examine its economic impacts and propose a binary classification model that predicts the stock return 10 days after the event. XGBoost and LightGBM models are applied for the study with accuracy of 75%, 71% respectively. In addition, after categorizing the recommended stocks based on the listed market(KOSPI/KOSDAQ) and market capitalization(Big/Small), this study verifies difference in the accuracy of models across four sub-datasets. Finally, by conducting SHAP(Shapley Additive exPlanations) analysis, we identify the key variables in each model, reinforcing the interpretability of models.

Citation status

* References for papers published after 2023 are currently being built.

This paper was written with support from the National Research Foundation of Korea.