EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Reinforcement Learning-Based Stock Selection and Trading Using KOSPI 200 Constituents

Sukjun Lee

Applied Sciences · 20.11.2026 · 16 (22) · 11509

DOI: 10.3390/app162211509 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 20.11.2026. Verfasst von Sukjun Lee. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Kwangwoon University, Ministry of Education, National Research Foundation of Korea..

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Equity-market trading systems must account for constraints such as integer-share trading, transaction fees, and taxes. This study proposes a reinforcement learning-based trading system (RLTS) for strategy-based stock selection and subsequent trading simulation in the Korean stock market. Daily price data for stocks included in the Korea Composite Stock Price Index 200 (KOSPI 200) from 2011 to 2021 were used to generate technical-indicator inputs. An asynchronous advantage actor–critic (A3C) algorithm was used to train a shared network across stocks, and stocks with positive mean excess training returns relative to contemporaneous short-term benchmark rates were selected as the selected stock set. Stocks selected using 3-, 6-, 9-, and 12-month training windows were evaluated during subsequent testing periods, and their trading performance was compared with that obtained using the implemented deep Q-network (DQN) and policy-gradient (PG) baselines. The A3C-based configuration achieved positive average trading returns and positive annualized Sharpe ratios based on three-month out-of-sample returns across all training-window widths. The 9-month configuration produced the highest average return (11.43%), whereas the 12-month configuration produced the highest annualized three-month out-of-sample Sharpe ratio (0.57). The A3C-based configuration showed higher observed average trading returns than the implemented DQN and PG baselines across the evaluated training-window settings. Relative to the KOSPI 200 buy-and-hold benchmark, A3C produced higher average returns for the 9- and 12-month configurations but lower average returns for the 3- and 6-month configurations. These findings indicate that the relative performance of the RLTS depends on the evaluated configuration and should not be interpreted as uniform superiority over the market.

Förderung

  • Kwangwoon University
  • Ministry of Education
  • National Research Foundation of Korea.

Zitieren

APA

Lee, S. (2026). Reinforcement Learning-Based Stock Selection and Trading Using KOSPI 200 Constituents. Applied Sciences, 16(22), 11509. https://doi.org/10.3390/app162211509

Vancouver

Lee S. Reinforcement Learning-Based Stock Selection and Trading Using KOSPI 200 Constituents. Applied Sciences. 2026;16(22):11509. doi:10.3390/app162211509

BibTeX
@article{Lee2026,
  author = {Lee, Sukjun},
  title = {Reinforcement Learning-Based Stock Selection and Trading Using KOSPI 200 Constituents},
  journal = {Applied Sciences},
  year = {2026},
  volume = {16},
  number = {22},
  pages = {11509},
  doi = {10.3390/app162211509}
}
Kostenlos starten →