상세 보기
기대수익률 추정 방법에 따른 평균-분산 포트폴리오의 성과 비교
초록
Mean-variance optimization is sensitive to estimation error in expected returns, which results in poor out-of-sample performance. State-space approaches that treat expected returns as unobservable state variables have been proposed as alternatives to the sample mean. This study estimates latent expected returns for individual assets with a Kalman filter based on a local level model and incorporates the estimates into mean-variance optimization. For a fixed universe of 29 constituent stocks of the Dow Jones Industrial Average, the Kalman filter strategy is compared with a rolling sample-mean strategy, the global minimum variance (GMV) portfolio and the equal-weight (1/N) portfolio under long-only, 10% weight-cap, and monthly rebalancing conditions. The results show that the Kalman filter strategy achieves a Sharpe ratio of 0.23, exceeding the sample-mean strategy (-0.15). Both estimation-based strategies, however, are outperformed by GMV (0.51) and the equal-weight benchmark (0.46). An ablation analysis identifies the weight cap as the dominant design component, and a shrinkage analysis shows that removing the cross-sectional information in expected return estimates (δ=0) performs best; these conclusions are insensitive to the choice of the state noise variance and the risk aversion coefficient. These findings suggest that Kalman filtering is an alternative to the sample mean when expected return estimates are used, and that excluding expected returns in favor of covariance-based risk minimization remains an effective strategy.
키워드
- 제목
- 기대수익률 추정 방법에 따른 평균-분산 포트폴리오의 성과 비교
- 제목 (타언어)
- Performance Comparison of Mean-Variance Portfolios by Expected Return Estimation Methods
- 저자
- 이우식
- 발행일
- 2026-10
- 유형
- Y
- 저널명
- 한국산업융합학회논문집
- 권
- 29
- 호
- 5
- 페이지
- 1261 ~ 1271
- 언어
- KOR
- 출판사
- 한국산업융합학회
- 발행국가
- 대한민국
- 분량
- 11 페이지
- ISSN
- E 2765-5415
P 1226-833x