Work detail

Portfolio Construction Using Hierarchical Clustering

Author: Mgr. Vojtěch Fučík
Year: 2017 - summer
Leaders: doc. PhDr. Ladislav Krištoufek Ph.D.
Consultants:
Work type: Economic Theory
Masters
Language: English
Pages: 80
Awards and prizes:
Link:
Abstract: The main objective of this thesis is to summarize and mainly interconnect the existing
methodology on correlation matrix filtering, graph algorithms utilized in the minimum
spanning trees, hierarchical clustering and principal components analysis in order to create
quantitative investment strategies. Instead of traditional usage of stocks returns series, factor
models residuals are utilized. Residuals are then an ultimate input for all the algorithms to
arrive at probability of centrality (PoC) -- an impure probability where values near 1 signalize
high probability of a stock being central in the network. Several investment strategies are
created based on PoC and tested on data from major US stock market indices. It cannot be
imperatively argued that peripheral-based strategies are always better than central-based
strategies. Both central and peripheral-based strategies share high upside profit potential at
the cost of high volatility whereas traditional Markowitz's optimization process yields stable
profits with moderate upside potential.

Partners

ČSOB
Deloitte
McKinsey & Company

Sponsors

CRIF