Analysis of the Applicability of Dimensionality Reduction Techniques in Stock Portfolio Construction

This study aims to overcome the limitations of previous research that utilized only a small number of financial indicators and proposes a method to comprehensively utilize all available financial data at the time of investment. To implement this effectively, we propose a Sliced Inverse Regression (SIR) model, a form of sufficient dimension reduction. In this study, we compare the returns of portfolios constructed using dimensionality reduction with those of market indices (KOSPI, KOSDAQ) and portfolios based on previous research (small-cap stocks, low PBR). The results indicate that the portfolio constructed using SIR achieved higher returns (46.1% cumulative over three years) compared to the benchmark portfolios and market indices, demonstrating strong potential for practical investment application. This suggests that the principal components extracted via SIR contain more critical information that can influence future returns than other comparative indicators. However, limitations of this study include the difficulty in providing a financial interpretation of the principal components extracted through dimensionality reduction.