Estimating the competence of combined method of principal component analysis and K-means clustering: Evidence from hedge funds

No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
School of Business | Master's thesis
Degree programme
Objective of the study The purpose of this thesis is to study combine method of principal component analysis and K-means clustering in hedge fund data and analysis if the method is suitable tool to find drivers behind hedge fund returns. This study provides risk factors presented in previous literature to describe clusters that the method forms. Focus of this thesis is to test if the clusters formed by the method are meaningful and insightful when comparing hedge fund strategies and predictable based on findings in previous literature. In addition, I test if the method can be used to construct reliable portfolios for investor. Data and methodology I use data from Lipper Tass hedge fund database and hedge fund risk factors created by Fung William and Hsieh David. Data is from 2007-2018 and criteria is used to make the data suitable for study, I construct two datasets, one for each test. In general I ably principal component analysis to the data and retain predetermine number of components and funds loadings to these components. After that I perform K-means cluster analysis based on the loadings of the funds. Number of clusters is chosen with criteria. Firstly, I test if the clusters membership is dependable on hedge fund strategy. Secondly, I form investment portfolios with the method and compare predictability to benchmark portfolios. Results This study uses both statistical analysis and practical investing case to determine if the used method is competent tool with hedge fund return data. This thesis main contribution is the finding that the cluster formed with method are describable with hedge fund risk factors and membership of the cluster is dependable on hedge fund strategy. The implementation for investors and researcher is practical and the model can help pinpoint hedge funds that differs from its main strategy or help to find new risk factors and niche hedge fund strategies.
Thesis advisor
Suominen, Matti
principal component analysis, K-means, cluster analysis, clustering analysis, hedge funds, risk factors, portfolio constructing
Other note