Vol.:(0123456789)
SN Computer Science (2021) 2:322
https://doi.org/10.1007/s42979-021-00671-z
SN Computer Science
ORIGINAL RESEARCH
A Statistical Test for Detecting Dependency Breakdown in Financial
Markets
Siva Rajesh Kasa
1
· Malay Bhattacharyya
2
Received: 28 December 2020 / Accepted: 28 April 2021 / Published online: 5 June 2021
© The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd 2021
Abstract
Correlations among stock returns during volatile markets difer substantially as compared to those from quieter markets. Dur-
ing times of fnancial crisis, it has been observed that the ‘traditional’ dependencies in global markets break down. However,
such an upheaval in the dependency structure happens over a span of several months, with the breakdown coinciding with a
major bankruptcy or a sovereign default. An important aspect of risk management is to efectively identify the duration of
breakdown and create tailor-made models for these extreme events; nevertheless, there are few statistical methods to identify
such signifcant breakdowns. The purpose of this paper is to propose a simple test to detect such structural changes in global
markets. This test relies on the assumption that asset price follows a Geometric Brownian Motion. We test for a breakdown
in dependence structure using eigenvalue decomposition of the correlation matrix. We derive the asymptotic distribution of
the test statistic under the null hypothesis and apply the test to stock returns. We also compute the power of this proposed test
and compare it with the power of other existing tests. Through extensive experimentation, we show that the proposed test is
able to efectively detect the times of structural breakdown in global stock markets, despite the simplifying assumption of
Geometric Brownian Motion of stock prices.
Keywords Correlation matrix · Fluctuation test · Local power
Introduction
Change point detection is a well-known problem in retro-
spective analysis of time series data. Traditionally, the prob-
lem of change point detection is posed as a hypothesis test-
ing problem with the null-hypothesis indicating structural
stability in the time series and the alternate indicating one
or multiple breakdowns. Change point detection has wide
applicability—from study of genes in cancers [17] to detec-
tion of fnancial fraud [4] to signal processing [11]. Change
point detection is complementary to sequential detection
wherein new data are continually arriving and are adaptively
analyzed.
In this work, we are concerned with the change point
detection specifc to fnancial time series. A signifcant
amount of prior literature has focused on detecting the
changes in the mean [21] and variance of a time series [14].
Another important, often overlooked, problem in statistical
modeling of fnancial time series is to analyze and detect
structural changes in the relationship among stock returns.
The structural dependency among various fnancial time
series is modeled using the Pearson correlation matrix.
Long-term risk-averse investors tend to hold portfolios of
assets whose returns are not positively correlated for diver-
sifcation benefts. However, there is compelling empirical
evidence that the correlation structure among returns of the
assets cannot be assumed to be constant over time, see, e.g.
[9, 15, 24] and [23]. In particular, during periods of fnancial
crisis, correlations among stock returns increase, a phenom-
enon which is sometimes referred to as diversifcation melt-
down. Given that duration of fnancial crisis is spread across
months/years, our work is motivated by the real-life problem
This article is part of the topical collection “Computational
Statistics” guest edited by Anish Gupta, Mike Hinchey, Vincenzo
Puri, Zeev Zalevsky and Wan Abdul Rahim.
* Siva Rajesh Kasa
kasa@u.nus.edu
Malay Bhattacharyya
malay.bhattacharyya@iimkashipur.ac.in
1
School of Computing, National University of Singapore,
Singapore, Singapore
2
Indian Institute of Management Kashipur, Kashipur, India