Use this URL to cite or link to this record in EThOS:
Title: Dynamical filtered graphs in finance
Author: Musmeci, Nicolo
ISNI:       0000 0004 5988 9043
Awarding Body: King's College London
Current Institution: King's College London (University of London)
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Financial markets are complex systems characterised by the interaction of several heterogeneous agents. The associated dependence structure is non-trivial and exhibits high levels of non-stationarity and non-linearity. These features make the understanding and forecasting of financial risk very challenging, since regularities observed from historical data do not necessarily mirror future behaviours. The main aim of this thesis is to investigate the complexity of the dependence structure through network filtering and clustering techniques. We have relied on these tools because they are data driven, model-independent and lend themselves to dynamical analyses. In particular, we have proposed a novel volatility forecasting tool based on network filtering. Furthermore, we have applied the Directed Bubble Hierarchical Tree (DBHT) clustering method for the first time to financial data, highlighting its advantages over other clustering techniques. We have performed statistical hypothesis tests on the dynamical DBHT clustering, in order to track the evolution of each cluster and how their industry-related information is affected by the market regime. We have studied the evolution of correlation-based filtered networks topology by means of data mining and time series techniques, investigating long-term memory properties and their relation with market risk. We have investigated how different measures of dependence perform and compare in terms of network topology, by combining multiplex tools and network filtering for the first time. We have found that the 2007 financial crisis marks a phase transition between two different regimes of dependence, which display deep dissimilarities in terms of industrial information and remain well distinct for years after the crisis. We have found that different clustering methods display different sensitivity to these structural changes. Moreover we have shown that correlation-based filtered networks display peculiar patterns in their evolution, notably long-term memory and possibly early-warning signals. After having found that a significant interplay exists between dependence structure variations and volatility, we have introduced a novel volatility forecasting tool which relies on this empirical feature. This new tool overcomes the curse of dimensionality, which limits traditional econometric models to porfolios of few assets. The multiplex analysis has revealed that it is crucial to monitor financial dependence with more than one measure at a time, as linear measures turn out to provide an incomplete picture of the dependence structure, especially during financial crises.
Supervisor: Di Matteo, Tiziana Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available