Time series clustering, testing of memory in time series and quantifying dependence in volatility of financial time series using complex network theory
Guide(s)
Bhattacharyya, Malay and Murthy, Rajluxmi V
Department
Decision Sciences
Area
Decision Sciences
University
Indian Institute of Management Bangalore
Place
Bangalore
Publication Date
3-31-2022
Year Awarded
March 2022
Year Completed
March 2022
Year Registered
June 2017
Abstract
The four essays in this dissertation are grounded in the central and interconnected concepts of multifractality and long-range dependence observed in financial time series. We focus our attention particularly to the dependence behaviour of a suitable measure of volatility in financial asset returns. Therefore, the implications of our work are mostly in the area of financial risk management. In the first three essays, we attempt to study the observed dependencies in financial time series from the lens of complex network theory. Specifically, we explore a relatively recent approach called visibility graphs, which involves mapping a time series to a complex network based on a particular geometric criterion. In the first essay, we utilize the concept of visibility graphs, proposed by Lacasa et al. (2008), for developing a method to cluster multiple financial asset return series based on their underlying dependence characteristics. Using simulations, we show that the proposed method exhibits better clustering performance as compared to certain feature-based and model-based time series clustering techniques. In the second essay, we build upon the concept of binary visibility graph, proposed by Ahadpour & Sadra (2012), to develop a reasonably powerful nonparametric test for the presence of memory in time series. The test is predicated on a statistic based on the principle of symbolic entropy. In the third essay, we utilize a scaled version of the symbolic entropy based statistic (developed in the second essay) for estimating the memory in volatility of financial asset returns. We find that the estimator performs well at measuring varying magnitudes of long-range dependence in volatility. In the fourth essay, we augment the Copula-Markov Switching Multifractal (MSM) model (Calvet & Fisher, 2004) by allowing the innovations in the margins to follow leptokurtic and skewed distributions, expecting to obtain better portfolio Value-at-Risk (VaR) forecasts during times of high market volatility. We find that the augmented Copula-MSM models exhibit better forecasting performance as compared to several traditional and popular VaR forecasting methods.
Pagination
107p.
Copyright
Indian Institute of Management Bangalore
Document Type
Dissertation
DAC Chairperson
Bhattacharyya, Malay; Murthy, Rajluxmi V
DAC Members
Anand, Abhinav
Type of Degree
Ph.D.
Recommended Citation
Achari, Giriraj, "Time series clustering, testing of memory in time series and quantifying dependence in volatility of financial time series using complex network theory" (2022). Doctoral Dissertations. 39.
https://research.iimb.ac.in/doc_dissertations/39
Relation
DIS-IIMB-FPM-P22-01
