Hierarchical Clustering To Improve Portfolio Tail Risk Characteristics
(2021) In Master's Theses in Mathematical Sciences FMSM01 20202Mathematical Statistics
- Abstract
- Many agree that estimating portfolio risks has better estimation possibilities, than estimations on returns. Therefore investors attempts to construct better, more efficient riskmanaged portfolios by diversifying portfolios through factors rather than traditional asset classes. This entails very often in estimations of correlation matrices so complex it cannot be fully analyzed. Hierarchical clustering reduces the complexity, by only focusing on the correlations that matters.
Hierarchical clustering uses graph theory, linked to unsupervised machine learning techniques. Hierarchical clustering is obtained by the suggested data and is a formation of a recursive clustering. Several hierarchical clustering methods are presented and... (More) - Many agree that estimating portfolio risks has better estimation possibilities, than estimations on returns. Therefore investors attempts to construct better, more efficient riskmanaged portfolios by diversifying portfolios through factors rather than traditional asset classes. This entails very often in estimations of correlation matrices so complex it cannot be fully analyzed. Hierarchical clustering reduces the complexity, by only focusing on the correlations that matters.
Hierarchical clustering uses graph theory, linked to unsupervised machine learning techniques. Hierarchical clustering is obtained by the suggested data and is a formation of a recursive clustering. Several hierarchical clustering methods are presented and evaluated against traditional riskbased portfolios with focus on left hand tail risk. A regime shift, based on momentum is applied to minimize drawdowns. The portfolios are tested on simulated data derived from Bootstrapping simulations and on historical data in a Walk forward optimization process.
The results indicate that hierarchical clustering based portfolios are truly diversified and achieve statistically better riskadjusted performances than commonly used portfolio optimization techniques. (Less) - Popular Abstract
- In attempts to construct better and more efficient risk managed portfolios, investors have recently changed their focus towards estimate risk instead of forecasting return. Estimating risk usually involves a correlation matrix, too complex to be fully analyzed. Hierarchical Clustering reduces the complexity and creates diversified portfolios. Big drawdowns could be reduced with a regime shift.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9042910
- author
- Eidenvall, Adam LU
- supervisor
- organization
- alternative title
- Hierarchical Clustering with Regime Shift Reduces big drawdowns
- course
- FMSM01 20202
- year
- 2021
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Hierarchical Clustering, Asset Allocation, Portfolio Construction, Graph Theory, Machine Learning, Risk Parity, Regime Shift, Bootstrapping, Walk Forward
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMS-3409-2021
- ISSN
- 1404-6342
- other publication id
- 2021:E10
- language
- English
- id
- 9042910
- date added to LUP
- 2021-05-06 15:57:11
- date last changed
- 2021-06-03 15:17:10
@misc{9042910, abstract = {{Many agree that estimating portfolio risks has better estimation possibilities, than estimations on returns. Therefore investors attempts to construct better, more efficient riskmanaged portfolios by diversifying portfolios through factors rather than traditional asset classes. This entails very often in estimations of correlation matrices so complex it cannot be fully analyzed. Hierarchical clustering reduces the complexity, by only focusing on the correlations that matters. Hierarchical clustering uses graph theory, linked to unsupervised machine learning techniques. Hierarchical clustering is obtained by the suggested data and is a formation of a recursive clustering. Several hierarchical clustering methods are presented and evaluated against traditional riskbased portfolios with focus on left hand tail risk. A regime shift, based on momentum is applied to minimize drawdowns. The portfolios are tested on simulated data derived from Bootstrapping simulations and on historical data in a Walk forward optimization process. The results indicate that hierarchical clustering based portfolios are truly diversified and achieve statistically better riskadjusted performances than commonly used portfolio optimization techniques.}}, author = {{Eidenvall, Adam}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Hierarchical Clustering To Improve Portfolio Tail Risk Characteristics}}, year = {{2021}}, }