Time‐Varying Linkage of Possible Safe Haven Assets: A Cross‐Market and Cross‐asset Analysis

• Date: 01 March 2017
• DOI: http://doi.org/10.1111/irfi.12089
• Author: Phong Nguyen,Wei‐han Liu
• Published date: 01 March 2017

Time-Varying Linkage of Possible

Safe Haven Assets: A Cross-Market

and Cross-asset Analysis*

PHONG NGUYEN

†

AND WEI-HAN LIU

‡

†

Faculty of Business, Economics and Law, La Trobe University, Australia and

‡

Department of Quantitative Finance, College of Technology Management, National

Tsing Hua University, Taiwan

ABSTRACT

This paper contributes to applying the time-varying symmetrized Joe–Clayton

copula to study the dynamic linkage among possible safe haven assets (SHAs)

in the major international markets over the past 34years. We re-examine four

major asset types (long-term government bonds, equity indices, oil, and gold)

and test whether they are qualified individually as a safe haven asset against

when paired against each other in a specific market. The empirical analyses in-

dicate that: (1) Government bonds are generally confirmed SHAs. (2) Gold and

oil are overwhelming SHAs against government bond across the markets. (3)

US and East Asian markets (Japan, Australia and New Zealand) have more

SHA options than the other regions against equity index.

JEL Codes: C32; C51; F3; G11

I. INTRODUCTION

Identifying safe haven assets (SHAs) is one of the crucial issues in financial risk

management for the sake of diversification benefits and downside risk reduction.

It is a critical issue especially at times of market turmoil. This issue has important

implications at least in liquidity risk, risk diversification, capital preservation, and

asset allocation. This topic has become increasingly important as the recent

major series of financial crises occur on either a regional or global scale. Their

contagion effects or the co-movement across asset classes during crisis periods

are even more critical (Dornbusch et al. 2000). Although possibly ephemeral,

the linkages among the assets at the critical moments play an essential role for fi-

nancial survivability. The asset selection in the normal market conditions does

not necessarily apply to the critical market conditions. Specifically, SHAs are

expected to provide liquidity, even when the other assets fail, at times of high

market stress (Beber et al. 2009; Brunnermeier and Pedersen 2009). That is, the

* We thank Darren Henry,Head of Department of Economics and Finance, for his helpful discussions

at La Trobe University

[Correction added on 11 November 2016, after first online publication: The affiliation of author Phong

Nguyen has been changed to ‘Faculty of Business, Economicsa ndLaw, La Trobe University, Australia’]

© 2016 International Review of Finance Ltd. 2016

International Review of Finance, 17:1, 2017: pp. 43–76

DOI: 10.1111/irfi.12089

qualification of SHA is contingent on the correlation measure between SHA and

its reference asset during market stress. The return of SHA is expected to provide

non-positive or insignificant correlation with the aggregate market return at crit-

ical market moments (Baur and Lucey 2009). The return of SHA is expected to

move in adverse or unrelated manner during critical market moments. Critical

market moments refer to those particular time points which the overall financial

markets suffer significant downtrend together. Even though the aggregate market

slumps, the return of SHA is expected to be positive or non-negative. Thus, hold-

ing SHA potentially reduces the suffered loss and then enhances the benefit from

portfolio diversification.

Previous studies mostly target gold and government bond as SHAs compared

with equity investments. Gold is one of the top traditional choices at least for

its value preserving and inflation hedging functions. Government bond is

noticed for its secured rate of return and high creditability as being guaranteed

by the central governments. However, these empirical analyses give mixed

outcomes. For example, Baur and Lucey (2010); Baur and McDermott (2010);

Beaudry et al. (2011); Reboredo (2013), and Ciner et al. (2013) give strong support

to gold as an SHA. Hood and Malik (2013) give mild support, but Joy (2011)

rejects this argument. Those articles discuss gold against different asset types in

various markets, for example, oil price, US dollar, and stock index. A similar

situation occurs in discussing the role of government bond, as another compet-

ing candidate as an SHA. Although some articles confirmed government bonds

as an SHA (Longstaff (2004); Connolly et al. (2005), and Baur and McDermott

(2013)), Rankin and Idil (2014) report that government bonds’function shows

significant change since the global financial crisis in 2007. That is, market timing

is also a critical issue in qualifying an SHA. During the recent market turmoil, the

possible SHA candidates reportedly experience significant joint downward price

trend due to the spontaneous asset sale of them (Piplack and Straetmans 2010).

The possible co-crash of assets introduces new challenges to the qualification of

those traditional SHA candidates. In fact, there are some other major candidates

of SHA, for example, oil and equity index. There are possibilities that those

candidate assets can be SHA to each other. Yet there are not much literatures

investigating those possibilities. Although vital, an agreement on the list of the

SHAs still has not been reached.

The major reasons for the mixed conclusion can be discussed in the aspects of

data period, market, asset type, and employed econometric technique. We

attempt to make a contribution via these attributes to attain a more subjective

assessment. To investigate the SHA property the data period needs to be lengthy.

Accordingly, it should contain a sufficient number of extreme observations to

attain an accurate estimate. We need to use a data series as long as possible. More

importantly, the dependence between financial assets is not fixed but varying

over time (Manner and Reznikova 2012). If this is true, then the SHA property will

change over time because this property is contingent on linkage. Each market has

its own properties and the investors within that market exhibit their specific pref-

erence pattern toward risk and asset types. SHA property may vary across markets.

International Review of Finance

© 2016 International Review of Finance Ltd. 201644

While we investigate the candidate assets across the major markets, this study fo-

cuses on those SHA candidate assets, not the markets.

1

It is advisable to consider

both the major developed and developing markets across regions to get a more ap-

propriate assessment. In addition to gold and government bonds, we include

crude oil as another candidate SHA. The crude oil market is well known for its vol-

atility level and volatility spillover.Crude oil has interesting properties in terms of

leverage. Crude oil has a low long-term correlation with traditional asset classes

(Chevallier and Ielpo 2013). These favorable properties could qualify crude oil

as a possible candidate for an SHA. An equity index usually serves as the aggregate

indicator of economic condition in the relevant literatures. Equity indices are thus

included in this study.

Further, existing literatures are mainly based on the following major economet-

ric methods: the generalized autoregressive conditional heteroskedasticity models

(GARCH) (Engle 1982), dynamic conditional correlation (DCC) model (Engle

2002), quantile regression (QR) (Koenker and Hallock 2001; Koenker 2005), and

vector autoregressive model (VAR) (Johansen 1991). GARCH models and DCC are

popularmodels for modeling time-varyingcorrelation structures(Tsay 2012). These

time series analysis models provide conventional linear-based correlations, but

their applications are restrictive. This is due to the certain parametric assumption

on the residuals anda certain requirement of linear relationships among variables.

In spite of their dynamicproperty,these two models are not essentiallydesigned for

capturing the extremes but specifically the general trendof the mean or the volatil-

ity level. QR generalizes the concept of a univariate quantile to a conditional

quantile given one or more covariates. QR is more instrumental than DCC when

lower quantilesare of interest because QR can give more specificpicturesatvarious

quantiles. However, Koenker and Hallock (2001) and Koenker (2005) notethat tra-

ditional QR performs well enough for the quantileshigher than 15% or lower than

85%. That is, QR is appropriate for moderate quantile levels butnot suitable for this

study. SHA property investigation targets on to the moments of major financial cri-

ses, for example, extremal quantiles like 0.001, 0.01, and 0.05 levels. Vector

autoregressivegives a concise method to summary the relationship among the var-

iables but its performance deteriorates when there is significant correlation among

the time series and its lagged terms (Lütkepohl 2005). It is expected to employ a

more appropriateeconometric technique to study thetime-varying linkage among

the SHA candidates at critical market moments.

The dependence among the extremes are highly nonlinear, asymmetric, and

time-varying (Cherubini et al. 2011; Joe 2014). Copula function is widely

accepted as one of the most appropriate tools for this purpose. To assess the

changing nonlinear dependence structures of extremes over time, we contribute

to employ econometric techniques that can capture the asymmetric dependence

structure among extremes under a time-varying framework. Manner and

Reznikova (2012) recommend the copulas that feature tail dependence and

1 The market characteristics are not considered in this study whose focus lies on SHA

qualification.

Time-varying Linkage of Possible Safe Haven Assets

© 2016 International Review of Finance Ltd. 2016 45