Publications

Detailed Information

Clustering Approaches for Global Minimum Variance Portfolio : 클러스터링을 이용한 분산 최소화 포트폴리오

DC Field Value Language
dc.contributor.advisor서봉원-
dc.contributor.author박진우-
dc.date.accessioned2020-05-07T05:26:11Z-
dc.date.available2020-05-07T05:26:11Z-
dc.date.issued2020-
dc.identifier.other000000159576-
dc.identifier.urihttp://dcollection.snu.ac.kr/common/orgView/000000159576ko_KR
dc.description학위논문(석사)--서울대학교 대학원 :융합과학기술대학원 디지털정보융합학과,2020. 2. 서봉원.-
dc.description.abstractTo earn higher return, one must bear higher risk, as risk and return are trade-off. However, a portfolio of well diversified assets allows investors to earn the same return at the expense of less amount of risk. It is because price of various assets can move in the opposite way, offsetting each other and thus, resulting in a more stable portfolio with less volatile returns.

Due to recent stock market crashes, portfolio optimization methods which focus on investing safely receive attention from various investors. Global minimum variance portfolio (GMVP) is an investment strategy designed to carry as little variance, which is considered risk in finance, as possible. The only input to attain the portfolio weights of GMVP is the covariance matrix of asset returns. Since the population covariance matrix is not known, investors use historical data to estimate covariance matrix. Even though sample covariance matrix is an unbiased estimator of the population covariance matrix, it includes a great amount of estimation error especially when the number of observed data is not much bigger than number of assets. It is due to the fact that bigger number of variance and covariance parameters need to be estimated and the covariance matrix approaches singularity.

Clustering stocks is proposed to decrease the estimation error contained in the covariance matrix and inverse of it by reducing the number of features. As it is difficult to estimate the covariance matrix with high dimensionality, we can perform portfolio optimization in each cluster first, and then perform portfolio optimization once again between clusters. In this sense, clustering approach to portfolio optimization is divide and conquer.

The motivation of this dissertation is that the estimation error can still remain high even after clustering, if a large amount of stocks is clustered together in a single group. This research, thus, proposes to utilize a bounded clustering method in order to limit the maximum cluster size. The result of one example experiment shows that not only the gap between in-sample volatility and out-of-sample volatility decreases, but also the out-of-sample volatility decreases. Moreover, other risk measures such as downside standard deviation and maximum drawdown tends to diminish. By limiting the maximum cluster size while clustering stocks, investors can better predict the out-of-sample risk based on in-sample counterpart and expect smaller out-of-sample portfolio risks.

There are three academic contributions of this dissertation. Firstly, this research shows that when we utilize clustering approach to portfolio optimization, clustering quality and estimation error are trade-off and maximum clustering size influence both. For example, as maximum clustering size increases, clustering quality improves, whereas the estimation error becomes larger, and vice versa. Secondly, we illustrate that bounded clustering approach is needed to find the best maximum clustering size to find the compromise between the clustering quality and estimation error to achieve the best portfolio performance. Thirdly, portfolio performance improvement from scaling data and applying dimensionality reduction results from taking care of estimation error while clustering stocks.
-
dc.description.abstract최근 주식시장의 붕괴로 인해 안전한 포트폴리오 구성 방법이 각광을 받고 있다. 이 중 하나인 최소 분산 포트폴리오(Global minimum variance portfolio, GMVP)는 위험, 즉 포트폴리오 수익의 변동성을 최소화하는 투자방법이다. 이 포트폴리오를 구성하기 위한 유일한 인풋은 투자 고려 대상인 자산들간의 공분산 행렬이다. 모 공분산의 비편향 추정방법은 샘플 공분산이지만, 투자를 고려하는 주식들(feature)의 수가 가격(observation)과 비슷하거나 더 많다면, 추정 오차가 많이 생기는 문제점이 있다. 클러스터링 방법은 이 추정오차를 줄이기 위한 다양한 방법들 중 하나이다.

이 논문 연구는 클러스터링을 하여도 각 클러스터 내에 모인 주식들의 숫자가 많다면, 여전히 추정오차가 많다는 문제의식에서 시작하였다. 따라서 이 논문은 각 클러스터내에 모일 수 있는 주식의 수를 제한하는 클러스터링 방법, bounded K-means 클러스터링 방법을 이용하여 이 문제를 해결하였다. 실험 결과 다양한 포트폴리오의 성과지수가 개선되는 것이 확인되었다.

이 논문의 연구적 기여는 첫째, 클러스터링 방법을 이용한 포트폴리오 최적화 과정에서 클러스터링의 질과 추정오차간 트레이드 오프가 존재하며 이 둘이 포트폴리오 성과에 영향을 미치기에 둘 다 통제해야 함을 밝혔다는 점이다. 둘째, 이 트레이드 오프 관계는 클러스터에 분류될 수 있는 최대 수(maximum clustering size)와 관계되어 있기에, 이를 조절할 수 있는 알고리즘인 bounded K-means 클러스터링 방법을 통해 포트폴리오 최고 성능을 낼 수 있는 지점을 찾을 수 있다는 점을 확인하였다. 셋째, 데이터 스케일링과 차원축소로 인한 포트폴리오 성능 개선도 추정 오차를 줄이는 관점에서 설명할 수 있다는 점이다.
-
dc.description.tableofcontents1. Introduction 1
1.1 Background and motivation 1
1.2 Research Goal 6
1.3 Experiments 7
1.4 Contributions 7
2. Related Works 9
2.1 Global minimum variance portfolio (GMVP) 9
2.2 Estimation of covariance matrix in portfolio 10
2.3 Clustering 14
3. Methodological Background 16
3.1 Clustering algorithms 16
3.2 Portfolio performance measures 17
4. Experiment methods 20
4.1 Dataset 20
4.2 Experiment design 22
4.3 Experiment procedure 23
5. Results 29
5.1 Model evaluation 29
5.2 Estimation error 32
5.3 Several performance measures 33
6. Discussion 35
6.1 Trade-off between correlation and estimation error 35
6.2 Finding the compromise for the best performance 39
6.3 Effect of feature pre-processing 41
7. Conclusion 47
Works Cited 50
-
dc.language.isoeng-
dc.publisher서울대학교 대학원-
dc.subject.ddc004-
dc.titleClustering Approaches for Global Minimum Variance Portfolio-
dc.title.alternative클러스터링을 이용한 분산 최소화 포트폴리오-
dc.typeThesis-
dc.typeDissertation-
dc.contributor.department융합과학기술대학원 디지털정보융합학과-
dc.description.degreeMaster-
dc.date.awarded2020-02-
dc.identifier.uciI804:11032-000000159576-
dc.identifier.holdings000000000042▲000000000044▲000000159576▲-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share