Are Gold and Government Bond Safe‐Haven Assets? An Extremal Quantile Regression Analysis

Date01 June 2020
Published date01 June 2020
AuthorWei‐han Liu
Are Gold and Government Bond
Safe-Haven Assets? An Extremal
Quantile Regression Analysis
School of Business, Southern Universit y of Science and Techn ology, Shenzhen ,
This study reexamines gold and government bonds as potential safe-haven
assets (SHAs) during market turmoil from daily data in 16 international mar-
kets over the past 20 years. We apply the extremal quantile regression model
by Chernozhukov and Chernozhukov and Fernandez-Val for empirical inves-
tigation. The outcomes indicate that a government bond is more likely to be
qualied an active SHA, which can increase in value during market turmoil.
Gold can be generally evaluated as a passive SHA, which is uncorrelated with
market slumps. However, at the extremal 0.001 quantile level, neither asset
can be qualied as a SHA. Since both assets exhibit a similar number of cases
of being qualied as SHAs, we cannot signicantly differentiate the ight-
to-liquidityand ight-to-qualityhypotheses. In terms of market selection,
United States and Singapore are the top two choices while France and Hun-
gary are the least commended markets to invest their local gold market
as SHA.
JEL Codes: C21; G11; G15; G32
Accepted: 8 August 2018
Flights of capital occur frequently during crises (Baur and Lucey 2009). In times
of turbulent markets, investors unload what they perceive to be riskier invest-
ments and purchase safer investments. A safe-haven asset (SHA) is expected to
retain or even increase its value or limit the investors exposure to losses in
times of market turbulence. According to Baur and Lucey (2010), a safe-haven
is dened as an asset that is negatively correlated, or uncorrelated, with another
asset or portfolio in times of market stress. That is, the return series of the SHA
is supposed to move in a reverse or unrelated manner during critical market
moments. The return of the SHA is expected to provide a nonpositive correla-
tion with the aggregate market return at critical market moments. This issue is
crucial especially when most assets slump together during a critical market
moment. The SHA return series is expected to behave in a reverse direction to
© 2018 International Review of Finance Ltd. 2018
International Review of Finance, 20:2, 2020: pp. 451483
DOI: 10.1111/ir.12232
preserve its value. This topic has become increasingly important due to the
recent series of nancial shocks of regional or global scales.
Among assets, gold and government bond are evaluated as the two most
popular candidates of SHA. Among its various functions,
gold has a long his-
tory as an SHA. The market often shows noticeable surges of market demand
for gold during market turmoil, at least over the past two decades. And govern-
ment bonds are generally believed to be default free. However, there have been
questions raised about the SHA property of both candidate assets because of
their noticed change in the market pattern during the recent market turmoil in
some markers, such as the abrupt drop of the price of gold in 2013
and the
outbreak of some European sovereign bond crises since 2009. These scenarios
alert the market participants to reevaluate their respective functions as SHA.
Meanwhile, previous literatures based on the data series prior to those critical
market moments present mixed conclusions about the SHA property of both
assets. Thus, critical market events should be included, and the issue should be
reexamined accordingly.
The qualication of gold and government bonds as SHA(s) has vital policy
implications and nancial applications. In addition, revisiting this issue can
help provide insight to determine which of the two proposed hypotheses best
explains the SHA property: ight-to-quality or ight-to-liquidity. These two
hypotheses help determine which nancial instrument to hold during market
turmoil. Flight-to-quality is the phenomenon in which investors move their
capital away from riskier investment to the safer (the lowest default risk) possi-
ble investment alternatives (Baur and Lucey 2010). Gold is generally regarded
as the most popular choice in this category (World Gold Council 2010). Flight-
to-liquidity occurs when investors sell what they perceive to be less liquid
investment and alternatively buy more liquid investment. Government bonds
are widely considered the top choice for ight-to-liquidity consideration during
market turmoil (Beber et al. 2009; Brunnermeier and Pedersen 2009). That is,
the qualication of SHA property of government bond and gold can help deter-
mine which corresponding hypothesis can hold and decide the investment
asset selection at critical market moments.
In terms of empirical analysis, previous literatures mostly rely on estimates
of ordinary regression coefcient or correlation structures. For example, Baur
and Lucey (2010) target gold and include the respective return levels of stock
and the possible SHAs at specic lower quantile levels (5%, 2.5%, and 1%) as
additional covariates or dummy variables to explain their relationship with a
dependent gold return series.
They claim that the selected lower quantile levels
are sufcient to represent the impact of market stress and apply the least square
1 Other functions include ination hedge and investment.
2 London Gold Afternoon Gold Fix shows the gold price (USD/Troy ounce) drops from 1963.75
(January 2, 2013) to 1192 (June 28, 2013).
3 The authors assume that the residual terms, after tting the least square regression (LSR), fol-
low an asymmetric GARCH model. The details for model specication or lag number selec-
tion of the asymmetric GARCH model are not given in the paper.
© 2018 International Review of Finance Ltd. 2018452
International Review of Finance
regression (LSR) model for the grouped data series at specic quantile levels.
Ciner et al. (2013) employ a dynamic conditional correlation (DCC) approach
(Engle 2002)
to capture the time-varying correlation structure. Durand
et al. (2010) employ an extreme dependence measure for the signicant non-
linear relationship among extremes.
However, those previous studies provide mixed evidences. For example, in
terms of gold, Baur and Lucey (2010) and Ciner et al. (2013) present supporting
evidence. Hood and Malik (2013) provide conservatively supporting evidence.
Baur and McDermott (2010) highlight the market-dependent picture. Similarly,
a conicting picture appears for government bonds. See Ilmanen (2003), Long-
staff (2004), Connolly et al. (2005), Kim et al. (2006), Durand et al. (2010), and
Baur and McDermott (2013) for details.
Several issues need to be reevaluated before we reach any conclusion. First,
SHA is expected to exhibit signicant SHA property during market turmoil. We
can employ quantile regression (QR) to study the relationship between aggre-
gate market trend and SHA price behavior specically at extremal quantile
levels which represent market distresses. Second, the prespecied lower
quantile levels are not sufciently extremal for this study. For instance, the 1%
quantile level represents 2.5 days in 1 year of 250 trading days. Market turmoil
could de facto occur at a lower probability level. Third, the previously employed
econometric techniques may be inappropriate, as neither LSR nor DCC are
designed for these extremes. The former performs well around the conditional
mean of the data mass. The latter is based on the variancecovariance and
designed not exclusively for the extremes. We need to employ the appropriate
econometric technique designed for extreme quantile levels. Furthermore, the
available multivariate extreme dependence measure does not correctly measure
the dependence in the limiting distribution of maxima and demands an adjust-
ment for this coefcient (Martins and Ferreira 2005). We need to select a more
suitable technique. Third, a sufciently long data series is essential to extract a
sufcient number of extremes and secure the estimation quality of the
extremes. Fourth, SHA property can be market-dependent (Baele et al. 2012;
Santis and Roberto 2012). Both developed and emerging markets should be
included for empirical analysis.
This paper contributes to improve the relevant studies by considering those
four issues. We include a longer data period (20 years of daily data), focus on
the more extreme quantile levels (0.001, 0.01, and 0.05 quantile levels), and
incorporate 16 developed and emerging markets. In terms of econometric tech-
nique, we employ the QR proposed by Koenker and Hallock (2001) and Koen-
ker (2005). This technique generalizes the concept of a univariate quantile to a
conditional quantile given one or more covariates. QR is a more instrumental
alternative in this study than those traditional econometric techniques because
QR can provide the specic pictures at various quantiles. However, the study of
4 DCC is known for its exibility of univariate GARCH models, coupled with parsimonious
parametric models for the correlation structure.
© 2018 International Review of Finance Ltd. 2018 453
Are Gold and Government Bond Safe-Haven Assets?

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT