강화학습을 활용한 주식 포트폴리오 구성 및 트레이딩 시뮬레이션에 관한 연구

Balancing and managing a portfolio is essential for minimizing risks and maximizing returns in investment and asset management, and it is a major concern for financial market research. In recent, with the development of computer hardware technology, studies on applying reinforcement learning algorithms, which have showed excellent performance in various fields, are being actively conducted for the financial field. However, it is difficult to optimize due to the diversity of parameter settings and combinations of entities constituting reinforcement learning, such as environment, policies, and agents, and more diverse studies need to be conducted. This study aim to develop a system for dynamically balancing stock portfolios and trading using reinforcement learning. For this, stock price data and fundamental data were collected from stocks listed on the Korean securities market, and then indexed through a preprocessing, and a model designed using the A3C algorithm was learned. In empirical study, stock portfolios were constructed by selecting stocks with high expected returns, and trading simulations were conducted on the test period to empirically analyze the performance of the model. As a result, the proposed model was 36.64% of the average annual return, which was superior to other comparison algorithms (26.72%), the market average return (32.10%), and KOSPI 200 index fund return (34.08%). In this study, it was found that learning in the entire market could achieve better results than learning in individual stocks, and it was confirmed that the composition of the stock portfolio using reinforcement learning could be effective.