Which is the safe haven for emerging stock markets, gold or the US dollar?

EMEMAR-00543; No of Pages 22 Emerging Markets Review xxx (2018) xxx–xxx

Contents lists available at ScienceDirect

Emerging Markets Review journal homepage: www.elsevier.com/locate/emr

Which is the safe haven for emerging stock markets, gold or the US dollar? Xiaoqian Wen a,b, Hua Cheng c,⁎ a b c

Institute of Chinese Financial Studies, Southwestern University of Finance and Economics, China Collaborate Innovation Center of Financial Security, Southwestern University of Finance and Economics, China School of Finance, Nankai University, China

a r t i c l e

i n f o

Article history: Received 2 February 2017 Received in revised form 9 July 2017 Accepted 29 December 2017 Available online xxxx JEL classifications: C51 G11 Q43

a b s t r a c t This paper examines whether gold or the US dollar is a safe haven for emerging stocks. By calculating the low-high tail dependence between markets via copulas and the downside risk gains of portfolios, we find that both gold and the US dollar can serve as a safe haven for emerging stocks; that the US dollar is better than gold in most cases, while its superiority in hedging infinitely extreme risks is weakened in the subsample of global financial crisis and the out-of-sample; and that the downside risk gains offered by the US dollar for China and Thailand are very attractive. © 2018 Elsevier B.V. All rights reserved.

Keywords: Emerging stock markets Gold US dollar Safe haven GAS copulas

1. Introduction Emerging economies are commodity-dependent and have become increasingly integrated with the rest of the world in recent decades. With substantial fluctuations of commodity prices, increased financial uncertainty, and an ongoing slowdown in the world economy, emerging stock markets are exposed to multiple global shocks. Managing the risks, especially during extreme market conditions, has become urgent. The gold market is traditionally regarded as a sound place of safety against macroeconomic risk (e.g., Erb and Harvey, 2006; Gorton and Rouwenhorst, 2006; Pukthuanthong and Roll, 2011; Batten et al., 2013; Reboredo, 2013). However, this conclusion is not widely reflected by some recent studies on stock markets, especially on emerging stock markets. Following the rapid financialization of commodities, co-movements between commodity and stock markets are found to be intensified (e.g., Tang and Xiong, 2012; Delatte and Lopez, 2013; Adams and Glück, 2015). As for one particular commodity, gold, Batten et al. (2015) point out that its outstanding over-the-counter (OTC) derivatives account for one fifth of all commodity derivatives in terms of the gross figure in 2013. Also, Bekiros et al. (2017) conclude that the significantly eased investments in gold make gold assets behave more and more like stocks. Regarding emerging stock markets, most studies find that gold cannot serve as a safe haven or is only a weak safe haven (e.g., Baur and McDermott, 2010; Beckmann et al., 2015; Bekiros et al., 2017). These studies highlight the need for alternative safe haven assets for emerging stocks. Given the recent slowdown of emerging economies together with the slump in world commodity prices while an appreciation of the US dollar, we are motivated to expect that the US dollar is a better safe haven asset for emerging stock markets.

⁎ Corresponding author. E-mail addresses: [email protected] (X. Wen), [email protected] (H. Cheng).

https://doi.org/10.1016/j.ememar.2017.12.006 1566-0141/© 2018 Elsevier B.V. All rights reserved.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

0

200

400

600

800

2

Jan 1, 2000

Jan 1, 2005 MSCI EM Gold

Jan 1, 2010 Date

Jan 1, 2015

S&P GSCI Commodity US dollar

Fig. 1. The price trends of MSCI Emerging market index, Standard & Pool GSCI Commodity index, gold and the US dollar. Note. The indices in January 4, 2000 are normalized to 100.

The preceding discussion can be intuitively identified in Fig. 1. In addition to the close co-movements of the emerging stock market index with the commodity index, similar price trends between the emerging stock market index and the world gold spot price are also observed. In contrast to the volatile prices of emerging stocks, commodities and gold, the evolution of the US dollar value is considerably stable—in particular, it slightly increased during the 2001 high-tech bubble collapse and the 2008 global financial crisis. The US dollar value has risen more significantly since 2014, when the emerging stock market index began to slide downward.1 However, these price trends could not fully exclude the possibility of gold as a safe haven, especially during the global financial crisis, given some inconsistencies between the world gold spot price and the emerging stock index price at the end of 2008 and 2011. Thus, this paper rigorously tests the following questions. Is gold a safe haven at least during some periods? If gold is not a safe haven asset, is the US dollar the better safe haven for emerging stock markets? According to the definition used in current studies (e.g., Baur and Lucey, 2010; Baur and McDermott, 2010; Reboredo, 2013; Bekiros et al., 2017), the role of an asset as a safe haven with respect to another asset depends on the link between the two assets in times of extreme market movements. If one asset is unrelated or negatively related to the other asset under extreme market circumstances, then the safe haven property is verified. Here, independence and a negative relationship imply a weak and strong safe haven property, respectively. From this viewpoint, accurately modeling market dependence in extreme situations is crucial. In this respect, copulas are known to be an advantageous tool, and lower tail dependence is a measure usually used to determine the safe haven property. However, such an analytical method is inadequate due to the following limitations. First, as the commonly used copulas only capture extreme price movements in tandem, the lower tail dependence of these copulas can only imply patterns of an extremely small value for one asset together with an extremely small value for another asset. Given that the lower tail dependence is a non-zero probability of observing the extreme price co-movements, it can at most show that, in times of extreme events, the two assets are uncorrelated (i.e. the value of lower tail dependence equals to zero), but cannot show that these two assets are negatively correlated. In this way, patterns of an extremely small value for one asset together with an extremely large value for another asset cannot be detected, and the strong safe haven property fails to be verified. Second, Reboredo and Ugolini (2015) suggest that the tail dependence of copulas offers information about extreme risks but at the limit, given that tail dependence is obtained by making the confidence levels infinitely close to one or zero. From this viewpoint, tail dependence only implies the ability to hedge infinitely extreme risks, but cannot indicate the ability to hedge extreme risks at various confidence levels. To overcome the preceding limitations, our paper contributes to the literature by rotating the commonly used copulas to capture the low (in emerging stock)-high (in gold/the US dollar) tail dependence (such opposite extreme price movements are of interest to investors holding long positions in emerging stocks), and then the strong safe haven property of gold/the US dollar can be detected. Compared with modeling the dependence structure between emerging stock returns and negative gold/US dollar returns, such a method is more flexible in real risk management practice. Simultaneously, based on the dependence structure implied by the optimal copula function, the downside risk gains provided by gold/US dollars for emerging stocks are calculated at various confidence levels, which are regarded as another dimension for detecting the safe haven property. In this way, the hedging ability of gold/the US dollar is more comprehensively examined, and direct implications for risk management are provided. 1 In Fig. 1, the U.S. dollar index is a measure of the value of the US dollar relative to the value of currencies of the majority of the U.S.'s most significant trading partners (including the Euro, Japanese yen, Canadian dollar, British pound, Swedish krona and Swiss franc). This index is similar to other trade-weighted indices, and the Euro holds the most weight, followed by the Japanese yen.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

3

Using the daily data of the MSCI stock index for nine emerging economics (Brazil, Chile, the Czech Republic, Russia, South Africa, China, India, Malaysia, and Thailand), the world gold spot price, and the corresponding exchange rate of the US dollar for each local currency, this paper finds that both gold and the US dollar can serve as a safe haven for emerging stocks. This paper also finds that the US dollar is a superior safe haven asset to gold in most cases, while its superiority in hedging infinitely extreme risks (implied by the low-high tail dependence) is weakened in the subsample of the global financial crisis and in the out-of-sample analysis, and that the US dollar provides attractive downside risk gains for China and Thailand, followed by India and Brazil, while offering relatively less attractive gains for Chile, South Africa, and Russia. Although this paper does not explore the possible reasons for these results in depth, it mentions some reasons that could explain the results. First, based on Baur and McDermott (2016), investors buy gold under stress because of the behavioral bias associated with gold's history as a currency, as a store of value, and as a safe haven asset, even if gold prices are volatile. In our paper, the low (in emerging stock)-high (in gold/US dollars) tail dependence indicates that gold is even superior to the US dollar in slightly more countries during the subsample of the recent global financial crisis. Second, as the US dollar plays a crucial role in the trade and finance of emerging economies, and given the negative relationship of the US dollar value and commodity prices (e.g., Akram, 2009; Nazlioglu and Soytas, 2012; Hammoudeh et al., 2015; Reboredo et al., 2016), the US dollar is generally deemed to be a high-quality currency by emerging countries and serves to hedge against risk. Finally, as noted previously, gold prices are more volatile and closer to stock prices due to commodity financialization, while the value of US dollar is stable, so the US dollar is preferred to gold in most of our analysis. Overall, the results are consistent with the flight-to-quality hypothesis of Caballero and Krishnamurthy (2008), under which investors may rush into buying high-quality assets to alleviate losses when the prices of risky financial assets drop dramatically. The reminder of this study is organized as follows. Section 2 is the literature review. Section 3 describes our dataset. Section 4 introduces the methodology. In Section 5, we present the empirical results and further discuss the subsample and out-of-sample analysis. Section 6 concludes the study.

2. Literature review The safe haven role of gold for stock markets has attracted a great deal of interest from researchers since the recent global financial crisis, and most of the related studies focus on developed countries. For example, Ciner et al. (2013) find that gold acted as a safe haven for US equities around 1990 and the recent global financial crisis; Flavin et al. (2014) find evidence in favor of choosing gold as a safe haven asset for US equity fund managers; and Bredin et al. (2015) show that gold acts as a safe haven for equity investors for long-run horizons of up to one year. However, Baur and Lucey (2010) argue that gold is a safe haven asset only in the very short term, and Lucey and Li (2015) provide evidence that gold is not the strongest or safest haven based on US data. In contrast with these results, studies on hedging extreme price movements of emerging stock markets find that gold cannot serve as a safe haven or is only a weak safe haven. For instance, Baur and McDermott (2010) indicate that gold does not act as a safe haven for BRIC countries; Beckmann et al. (2015) document that gold is not a safe haven for Russia and Indonesia, and only serves as a weak safe haven for emerging economies such as China, Egypt, Korea, South Africa, Turkey, and Thailand; in a more recent study, Bekiros et al. (2017) confirm that gold is not a safe haven asset for BRICS countries (Brazil, Russia, India, China and South Africa). In comparison with the above studies, those addressing the safe haven role of the US dollar on stock markets are relatively limited. Baur and McDermott (2016) show that gold was a particularly strong safe haven in September 2008, and also note the role of the US dollar as a safe haven currency. Furthermore, Liu et al. (2016) provide empirical evidence that both gold and the US dollar can serve as a safe haven for seven developed stock markets in most cases. In fact, although some studies do not focus on the safe haven property of the US dollar, they investigate the dependence structure between equities and the US dollar, which can also provide potential implications for hedging the extreme risks of stocks using the US dollar. For example, with a sample of six industrialized countries, Ning (2010) models the dependence between equity and foreign exchange rate markets and finds evidence of symmetric tail dependence; Wang et al. (2013) conclude that the tail dependence is asymmetric (symmetric) when local currency values against the US dollar and stock returns are negatively (positively) correlated. With a sample of emerging countries, Reboredo et al. (2016) examine the downside and upside risk spillovers between stocks and exchange rate markets, providing evidence of a positive relationship between stocks and currency values in emerging economies with respect to the US dollar, the euro, and bidirectional spillovers of extreme risks. The methods used in most studies are disadvantageous for describing extreme risks. For instance, some studies use a threshold regression model, with the threshold given by a specific quantile of the stock return distributions (see Baur and McDermott, 2010, 2016; Baur and Lucey, 2010; Ciner et al., 2013), and other studies use the multivariate GARCH models (see Ciner et al., 2013; Arouri et al., 2015). Then, market correlations are the main basis used by these studies to examine the safe haven property of assets. However, it is well known that correlation coefficients cannot fully account for the dependence structure between markets, so they fail to sufficiently describe extreme risks. In addition, when setting the specific quantile, there is always discretion, and the multivariate GARCH models often assume a fit to bivariate normality, which is too restrictive in reality and ignores non-linearity in market dependence. To overcome these limitations, copula functions are used in some studies to measure extreme risks (see Ning, 2010; Wang et al., 2013; Mensi et al., 2015; Reboredo et al., 2016; Liu et al., 2016). Copulas have obvious advantages in building multi-distributions of assets returns based on the information of marginal distributions without assuming multivariate normality; they can capture rich patterns of tail dependence without using discretion to define extreme observations, and can Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

4

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

describe both linear and non-linear dependence between markets. Following these latter studies, copulas become our main tool of analysis. The study by Liu et al. (2016) is the study most closely related to ours; it uses a multivariate extended skew-t copula model to investigate the safe haven role of gold and the US dollar for seven developed stock markets. Reboredo (2013) represents an important step in using copulas to detect the safe haven property of assets. However, in addition to the contributions noted in Section 1, our paper is distinguished in the following ways. First, not using the dynamic conditional correlation (DCC) specification of Engle (2002) or the time-varying (TV) specification of Patton (2006), we apply the GAS specification of Creal et al. (2013) in copulas to capture the dynamics of market dependence. Such modeling is still novel and suggested to be more sensitive to the dynamics of market dependence (see Avdulaj and Barunik, 2015; Oh and Patton, 2017). Second, in addition to the full-sample analysis and in-sample estimation, subsample and out-of-sample analyses are conducted to make the study more comprehensive and robust. We specifically examine the performance of gold and the US dollar during the global financial crisis and the recent commodity periods, when the safe haven property of these assets is worthy of more attention. Finally, we focus not on developed countries but rather on emerging countries, which are more sensitive to commodity prices and US dollar value changes (e.g., Reboredo et al., 2016; Bekiros et al., 2017). To our best knowledge, this paper provides the first detailed analysis comparing the safe haven property of gold and the US dollar for emerging stock markets.

3. Data We use the daily prices of the MSCI stock index for each emerging country (Brazil, Chile, the Czech Republic, Russia, South Africa, China, India, Malaysia, and Thailand). These countries are selected based on the trade balance effects of commodity price declines (Commodity Market Review, IMF, 2013). All of the MSCI stock indices are denominated in the local currency. To ensure that gold and the US dollar are denominated in the same currency as the stock indices, we also need the exchange rate of the US dollar for the local currency of each emerging country, measured in terms of units of local currency per US dollar (an exchange rate increase means that the US dollar appreciates). Such specification ensures that our empirical investigation can be applied in real risk management for domestic investors. MSCI stock indices, gold spot prices, and exchange rate data are all taken from the Thomson Reuters DataStream database. Because of the exchange rate reform in China and Malaysia, the sample Table 1 Descriptive statistics. Mean

Std

Skewness

Kurtosis

J-B test

Q(15)

Corr

Panel A: stock index Brazil Chile Czech Russia South Africa China India Malaysia Thailand

0.026 0.021 0.018 0.030 0.039 0.025 0.035 0.019 0.020

1.661 1.003 1.480 2.280 1.261 1.803 1.539 0.794 1.583

−0.093⁎⁎ 0.181⁎⁎⁎ −0.403⁎⁎⁎ −0.479⁎⁎⁎ −0.164⁎⁎⁎ −0.020 −0.268⁎⁎⁎ −0.936⁎⁎⁎ −0.537⁎⁎⁎

5.581⁎⁎⁎ 12.849⁎⁎⁎ 10.556⁎⁎⁎ 13.898⁎⁎⁎ 2.857⁎⁎⁎ 7.077⁎⁎⁎ 7.821⁎⁎⁎ 13.753⁎⁎⁎ 9.081⁎⁎⁎

5595.1⁎⁎⁎ 29,651.6⁎⁎⁎ 20,112.6⁎⁎⁎ 34,829.1⁎⁎⁎ 1484.1⁎⁎⁎ 5968.0⁎⁎⁎ 11,027.9⁎⁎⁎ 22,957.7⁎⁎⁎ 15,006.0⁎⁎⁎

32.5⁎⁎⁎ 133.9⁎⁎⁎ 47.5⁎⁎⁎ 95.5⁎⁎⁎ 45.8⁎⁎⁎ 44.7⁎⁎⁎ 86.4⁎⁎⁎ 48.7⁎⁎⁎ 49.2⁎⁎⁎

– – – – – – – – –

Panel B: gold Brazil Chile Czech Russia South Africa China India Malaysia Thailand

0.050 0.041 0.027 0.056 0.057 0.034 0.046 0.043 0.035

1.401 1.244 1.132 1.266 1.320 1.241 1.132 1.248 1.138

−0.188⁎⁎⁎ −0.085⁎⁎ −0.273⁎⁎⁎ −0.243⁎⁎⁎ 0.084⁎⁎⁎ −0.493⁎⁎⁎ −0.330⁎⁎⁎ −0.387⁎⁎⁎ −0.370⁎⁎⁎

7.325⁎⁎⁎ 4.340⁎⁎⁎ 6.176⁎⁎⁎ 11.568⁎⁎⁎ 4.323⁎⁎⁎ 4.993⁎⁎⁎ 5.052⁎⁎⁎ 4.411⁎⁎⁎ 5.334⁎⁎⁎

9655.4⁎⁎⁎ 3385.9⁎⁎⁎ 6898.7⁎⁎⁎ 24,057.4⁎⁎⁎ 3358.3⁎⁎⁎ 3086.7⁎⁎⁎ 4658.5⁎⁎⁎ 2389.3⁎⁎⁎ 5205.4⁎⁎⁎

43.5⁎⁎⁎ 49.3⁎⁎⁎ 45.9⁎⁎⁎ 40.3⁎⁎⁎ 49.1⁎⁎⁎

−0.173 −0.015 −0.053 −0.058 −0.025 0.072 −0.060 −0.052 −0.034

Panel C: US dollar Brazil Chile Czech Russia South Africa China India Malaysia Thailand

0.014 0.005 −0.009 0.020 0.021 −0.007 0.010 0.002 −0.001

0.959 0.602 0.765 0.774 1.059 0.120 0.388 0.437 0.305

−0.008 0.443⁎⁎⁎

10.598⁎⁎⁎ 5.291⁎⁎⁎ 3.676⁎⁎⁎ 77.536⁎⁎⁎ 5.313⁎⁎⁎ 54.423⁎⁎⁎ 7.530⁎⁎⁎ 4.578⁎⁎⁎ 20.458⁎⁎⁎

20,156.9⁎⁎⁎ 5165.0⁎⁎⁎ 2427.1⁎⁎⁎

−0.054 0.246⁎⁎⁎ 0.268⁎⁎⁎ −0.434⁎⁎⁎ 0.266⁎⁎⁎ −0.351⁎⁎⁎ 0.116⁎⁎⁎

1,078,918.6⁎⁎⁎ 5116.8⁎⁎⁎ 353,046.3⁎⁎⁎ 10,227.3⁎⁎⁎ 2555.8⁎⁎⁎ 75,119.8⁎⁎⁎

21.3 30.4⁎⁎ 27.4⁎⁎ 46.9⁎⁎⁎ 34.3⁎⁎⁎ 89.7⁎⁎⁎ 22.5⁎ 134.8⁎⁎⁎ 30.7⁎⁎⁎ 33.3⁎⁎⁎ 65.6⁎⁎⁎ 17.1 28.6⁎⁎

−0.335 −0.061 −0.126 −0.236 −0.208 −0.100 −0.368 −0.357 −0.209

Note. The total number of daily observations for assets in China and Malaysia is 2860, and that for assets in other countries is 4307. Jarque-Bera (J-B) statistic tests for the null hypothesis of normality in the sample returns distribution. The Q (15) is the Ljung-Box Q test of serial correlation of up to 15 lags in the returns. Corr is the Pearson correlation coefficient of returns of Stock (i) and Gold (i)/the US dollar(i) (i = emerging countries). ⁎⁎⁎, ⁎⁎, ⁎ indicate statistical significance at the 1%, 5% and 10% level, respectively.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

5

period of these two countries starts on July 20, 2005, while for other countries it starts on January 3, 2000, at the beginning of this century. The sample period for all countries ends on July 6, 2016. Table 1 provides the descriptive statistics of the asset returns, with returns calculated as 100 times the difference in the log of prices. In Panel A, stocks in Russia and China are found to be riskier than stocks in the other countries; negative skewness values are observed in all countries except Chile; the high kurtosis statistics indicate fat tails in the return distributions; and the results of the Jarque–Bera (J–B) statistic tests and Ljung–Box Q test suggest that the distributions of the emerging stock returns are nonnormal and serially correlated. Panel B and C report the descriptive statistics for the gold price and the US dollar value in the local currencies, respectively. The gold price denominated in the currencies of Brazil and South Africa has larger fluctuations, while the value of the US dollar to the currencies of South Africa and Brazil is more volatile. Negative skewness values are exhibited in all gold prices, while more cases of positive skewness values are found for the US dollar value. The leptokurtic phenomenon is more evident in the US dollar value than in the gold price. Non-normality and serial correlation are also observed in the returns of both the gold price and US dollar rate. Finally, the values of the Pearson linear correlations between the stock index price and US dollar value are more negative than those between the stock index price and gold price, which preliminarily indicates that the US dollar has a higher potential than gold for risk hedging in emerging stock markets. 4. Methodology Given the advantages of copula functions discussed in Section 2, copula functions are our main tool to model the dependence between returns of emerging stocks and gold/US dollar value. In addition to the commonly used copulas aimed at extreme price movements in tandem, we rotate the copula functions to capture the opposite movements in extreme market environments. Based on estimates of these copula functions, we discuss how to examine the safe haven property of assets. 4.1. Copula specification According to Sklar's theorem, a two-dimensional joint distribution function G with continuous marginal FX and FY has a unique copula representation such that G(x, y) = C(FX(x), FY(y)). A joint distribution function can then be decomposed into marginal distributions, and the dependence structure that can be described by a copula. In this paper, Rstocki,t, Rgoldi,t, and RUS i,t (i = Brazil, Chile, the Czech Republic, Russia, South Africa, China, India, Malaysia, and Thailand) are denoted as the returns of emerging stocks, gold price and the US dollar value at time t, respectively; and Fstocki(Rstocki,t), Fgoldi(Rgoldi,t), and FUSi(RUSi,t) are their respective conditional cumulative distribution functions (CDFs). Then, the joint CDFs of emerging stock market returns and gold price/US dollar value returns can be written as



 G Rstocki;t ; Rgoldi;t ¼ C F stocki Rstocki;t ; F goldi Rgoldi;t ;

ð1Þ





 G Rstocki;t ; RUSi;t ¼ C F stocki Rstocki;t ; F USi RUSi;t :

ð2Þ

Differentiating all of the conditional CDFs, the conditional joint densities are given by


 g Rstocki;t ; Rgoldi;t ¼ cðu; vÞ  f stocki Rstocki;t  f goldi Rgoldi;t ;

ð3Þ




 g Rstocki;t ; RUSi;t ¼ cðu; vÞ  f stocki Rstocki;t  f USi RUSi;t ;

ð4Þ

where c(u,v) = ∂2C(u, v)/∂u∂v is the conditional copula density function of returns of emerging stocks and gold price/US dollar value; u = Fstock i(Rstock i,t) and v = Fgoldi(Rgold i,t) / v = FUS i(RUS i,t) are distributed as continuous uniform variables on (0, 1); and fstock i (Rstocki,t), fgold i (Rgoldi,t) and fUSi (RUSi,t) are the marginal densities of Rstocki,t, Rgoldi,t, and RUSi,t, respectively. Thus, the joint density of emerging stock returns and gold price/US dollar value returns is the product of the copula density and the two marginal densities. Then, the log-likelihood functions are logg ¼ logc þ logf stocki þ logf goldi :

ð5Þ

logg ¼ logc þ logf stocki þ logf USi :

ð6Þ

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

6

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 2 Copula functions. Copula −1

CG(ut,vt; ρ) = Φ(Φ

Gaussian

c

(ut),Φ

−1

Parameter

Dependence structure

(vt))

−1 b ρ b 1

(vt))

−1 b ρ b 1

No tail dependence λU = λL = 0 Symmetric tail dependence pffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi λU ¼ λL ¼ 2t v ð− vc 1−ρ= 1 þ ρÞ N0 Symmetric tail dependence pffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi λU ¼ λL ¼ 2t v ð− vc 1−ρ= 1 þ ρÞ N0 Asymmetric tail dependence λU = 2–21/δ, λL = 0 Asymmetric tail dependence λU = 0, λL = 2–21/δ. Asymmetric tail dependence λU = 2–21/δ, λL = 0

−1

−1

Student-t

CS(ut,vt; ρ, v ) = T(t

270-Degree-rotated Student-t

CRS_270(ut, vt; ρ, vc) = ut − CS(ut,1 − vt; ρ, vc)

−1 b ρ b 1

Gumbel

CG(ut,vt; δ) = exp (−((− log ut)δ + (− log vt)δ)1/δ)

δN1

180-Degree-rotated Gumbel

CRG(ut,vt; δ) = ut + vt − 1 + CG(1 − ut, 1 − vt; δ)

δN1

90-Degree-rotated Gumbel

CRG_90(ut,vt; δ) = vt − CG(1 − ut,vt; δ)

δN1

(ut), t

If we denote the parameters in c, fstock i, and fgold

i/U.S. i

as θc, θstock i and θgold

i/U.S.i.,

respectively, Eqs. (5) and (6) can be written

as

 LðθÞ ¼ Lc ðθc Þ þ Lstocki θstocki þ Lgoldi θgoldi ;

ð7Þ



 LðθÞ ¼ Lc ðθc Þ þ Lstocki θstocki þ LUSi θUSi ;

ð8Þ

where Lk is the log-likelihood function of the copula (k = c), stock (k = stocki) and gold/US dollar (k = goldi / USi) densities. In this paper, we use six kinds of copulas to consider different patterns of tail dependence between markets (see Table 2). The commonly used copulas, including the Gaussian, Student-t, Gumbel, and 180-degree rotated Gumbel copulas, are sensitive to extreme movements in the same direction (i.e., they increase and decrease in tandem); and their lower and upper tail dependence can be written in terms of copulas, respectively, as follows: h i C ðv; vÞ −1 −1 λL ðvÞ ¼ lim P X ≤ F ðvÞjY ≤ F ðvÞ ¼ lim ; v v→0 v→0

ð9Þ

h i 1−2v þ C ðv; vÞ −1 −1 λU ðvÞ ¼ lim P X ≥ F ðvÞjY ≥ F ðvÞ ¼ lim ; 1−v v→1 v→1

ð10Þ

where 0 ≤ λL ≤ 1, 0 ≤ λU ≤ 1. Furthermore, we rotate the Student-t copula by 270 degrees and the Gumbel copula by 90 degrees to capture the extreme movements in an opposite direction. Specifically, we capture the dependence between extreme decreases in emerging stock markets and extreme increases (appreciations) in gold prices (US dollar values). Similar to Eqs. (9) and (10), their lower and upper tail dependence can be written as follows2: h i CR −1 −1 0 λL ðvÞ ¼ lim P X ≥ F ð1−vÞjY ≤ F ðvÞ ¼ lim v→0

v→0

270=90 ðv; vÞ

v

;

h i 1−2v þ C R 270=90 ðv; vÞ −1 −1 0 ; λU ðvÞ ¼ lim P X ≤ F ð1−vÞjY ≥ F ðvÞ ¼ lim v→1 v→1 1−v

ð11Þ

ð12Þ

where 0 ≤ λ'L ≤ 1, 0 ≤ λ'U ≤ 1. Taking into account that dependence between markets can be time-varying, we adopt the GAS model of Creal et al. (2013) to allow the parameters of the copula functions in Table 2 to change over time. Consider a copula with dynamic parameters C(δt(γ)). To keep the parameters within a particular range, the GAS specification applies a strictly increasing transformation (e.g., log, logistic, arc tan) to the copula parameter and models the evolution of the transformed parameter, which is denoted by ft: −1

f t ¼ hðδt Þ⇔δt ¼ h

ð f t Þ;

ð13Þ

2 As the Student-t copula characterizes symmetric tail dependence, so the 90 degree rotation can still capture the movements between extreme decreases in emerging stock markets and extreme increases in gold prices/US dollar rates. However, we do not consider this to avoid repetition.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx −1=2

wheref tþ1 ¼ c þ βGAS f t þ α GAS It

st ¼

st ;

∂ logðut ; vt ; δt Þ; ∂δ

7

ð14Þ

ð15Þ

h 0i It ¼ Et−1 st st ¼ Iðδt Þ:

ð16Þ

Then, the future value of the copula parameter is a function of a constant, the current value, and the score of the copula1/2 likelihood, I− st. As the Gumbel copula and its rotations require the parameter to be greater than one, we follow Patton t (2013) and set the function of δt to be δt = 1 + exp(ft); for the Gaussian and Student-t copulas and their rotation, we set the correlation to be δt = (1 − exp(−ft)) / (1 + exp.(−ft)) to ensure that the correlation lies in the range (−1, 1). 4.2. Marginal distribution models We next discuss the specification of the marginal distribution models. To capture the stylized features of the asset returns, including the fat tails, serial correlation, leverage effects, and conditional heteroscedasticity, we use autoregressive models with generalized autoregressive conditional heteroscedasticity errors for the asset returns. Concretely, the model of the conditional mean for each asset return is given by Rt ¼ μ þ

p X

φ j Rt− j −

j¼1

q X

θ j εt− j þ ε t ;

ð17Þ

j¼1

ε t ¼ σ t zt ;

ð18Þ

zt  skewed−t ðzt jη; ϕÞ:

ð19Þ

where p and q are non-negative integers, εt is the error term, and zt is the standardized residual following the skewed-t distribution of Hansen (1994) (the details of the skewed-t distribution density are provided in Appendix A). In Eq. (18), σt is the square root of the conditional variance for εt. We consider a standard GARCH and a GJR-GARCH for modeling the conditional variance, as they are the most popular GARCH models for capturing the volatilities of asset returns. Based on Brooks, 2008, a GARCH family model with one lag order can sufficiently capture the volatility clustering in asset returns, and few financial studies have considered or used a higher-order model; thus, orders of the GARCH process are all specified as 1 (see, among others, Oh and Patton, 2017; Aloui et al., 2013; Wang et al., 2013; Wen et al., 2012). The GARCH (1, 1) model of conditional variance for each asset return is given by 2

2

2

σ t ¼ ω þ αεt−1 þ βσ t−1 ;

ð20Þ

where α represents the ARCH term that measures the effect of past innovations on current variance and β represents the GARCH term that measures the effect of past variance on current variance. The degree of persistence of the variance shock is measured by the sum of the ARCH and GARCH parameters (α + β). Considering the asymmetry in the conditional variance process, the GJR-GARCH (1, 1) model is given by 2

2

2

2

σ t ¼ ω þ αεt−1 þ βσ t−1 þ γεt−1 It−1 ;

ð21Þ

where, in comparison with Eq. (20), the additional variable It-1 in Eq. (21) is a dummy variable that measures the asymmetric response of the conditional variance to shocks. The dummy variable takes a value of 1 in response to negative shocks and 0 in response to non-negative shocks. If the asymmetric coefficient of the conditional variance γ is significantly positive, a negative shock leads to a greater rise in future conditional variance than a positive shock of the same magnitude. The order of the AR and MA terms in the conditional mean model and the type of GARCH formulation are selected by the Schwarz information criterion (SIC), which is known to lead to a parsimonious specification. Overall, the parameters of marginal distribution models and copula are estimated by a two-stage estimation procedure called as the inference for the margins (IFM) proposed by Joe and Xu (1996). Concretely, at the first step, we estimate the parameters in the marginal models by maximizing the log-likelihood; at the second step, given the estimated marginal parameters, the marginal Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

8

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 3 Estimates of marginal distributions.

Panel A: stock index Conditional mean μ

Brazil

Chile

Czech

Russia

South Africa

China

India

Malaysia

Thailand

0.017 (0.016)

0.015 (0.012) 0.184⁎⁎⁎

0.030 (0.021)

0.047⁎⁎ (0.022) 0.040⁎⁎

0.028⁎ (0.016)

0.056 (0.035)

0.068⁎⁎⁎ (0.018) 0.065⁎⁎⁎

0.015 (0.021) 0.090⁎⁎⁎

0.061⁎⁎⁎ (0.019) 0.049⁎⁎⁎

(0.016)

(0.019)

(0.016) 0.008 (0.019) −0.019 (0.023)

φ1

(0.016)

(0.018)

φ2 φ3 Conditional variance ω

η

0.049⁎⁎⁎ (0.015) 0.016 (0.011) 0.921⁎⁎⁎ (0.015) 0.086⁎⁎⁎ (0.018) 9.270⁎⁎⁎

ϕ

(1.287) −0.049⁎⁎⁎

α β γ

Goodness-of-fit tests K-S A-D Panel B: gold Conditional mean μ

0.030⁎⁎⁎ (0.006) 0.038⁎⁎⁎

0.069⁎⁎⁎ (0.016) 0.053⁎⁎⁎

0.058⁎⁎⁎ (0.015) 0.072⁎⁎⁎

0.027⁎⁎⁎ (0.007) 0.019⁎⁎

0.027⁎⁎⁎ (0.010) 0.073⁎⁎⁎

0.035⁎⁎⁎ (0.008) 0.102⁎⁎⁎

0.008⁎⁎⁎ (0.003) 0.066⁎⁎⁎

0.047⁎⁎⁎ (0.012) 0.101⁎⁎⁎

(0.011) 0.868⁎⁎⁎ (0.016) 0.120⁎⁎⁎ (0.019) 9.499⁎⁎⁎

(0.010) 0.871⁎⁎⁎ (0.019) 0.079⁎⁎⁎ (0.022) 6.960⁎⁎⁎

(0.011) 0.885⁎⁎⁎ (0.013) 0.068⁎⁎⁎ (0.019) 5.859⁎⁎⁎

(0.009) 0.908⁎⁎⁎ (0.014) 0.111⁎⁎⁎ (0.016) 11.567⁎⁎⁎

(0.012) 0.920⁎⁎⁎ (0.013)

(0.012) 0.883⁎⁎⁎ (0.013)

(0.012) 0.885⁎⁎⁎ (0.014)

(0.688) −0.051⁎

(0.535) −0.051⁎⁎⁎

(1.885) −0.096⁎⁎⁎

6.996⁎⁎⁎ (0.765) −0.071⁎⁎⁎

(0.018)

(1.344) −0.025 (0.022)

(0.030)

(0.021)

(0.021)

6.442⁎⁎⁎ (0.816) −0.036 (0.029)

(0.015) 0.891⁎⁎⁎ (0.022) 0.066⁎⁎⁎ (0.023) 5.104⁎⁎⁎

(0.018)

(0.521) −0.060 (0.040)

0.193 0.119

0.545 0.293

0.373 0.278

0.133 0.338

0.532 0.454

0.003 0.066

0.059 0.075

0.153 0.091

0.013 0.076

0.054⁎⁎⁎ (0.017)

0.049⁎⁎⁎ (0.016)

0.019⁎⁎ (0.009) −0.039⁎⁎⁎

0.056 (0.600)

0.060⁎⁎⁎ (0.017)

0.040⁎⁎⁎ (0.017)

0.044⁎⁎⁎ (0.014) −0.068⁎⁎⁎

0.041⁎⁎ (0.020)

0.042 (0.051) −0.092⁎⁎⁎ (0.015)

φ1

(0.015) Conditional variance ω

(0.014)

5.560⁎⁎⁎ (0.566) 0.033⁎ (0.019)

0.023⁎⁎⁎ (0.007) 0.088⁎⁎⁎ (0.014) 0.933⁎⁎⁎

0.014⁎⁎⁎ (0.005) 0.067⁎⁎⁎ (0.010) 0.948⁎⁎⁎

0.017⁎⁎⁎ (0.005) 0.055⁎⁎⁎ (0.009) 0.933⁎⁎⁎

0.013⁎ (0.164) 0.107 (0.246) 0.927⁎⁎⁎

0.021⁎⁎⁎ (0.007) 0.091⁎⁎⁎ (0.015) 0.930⁎⁎⁎

0.012⁎⁎⁎ (0.004) 0.039⁎⁎⁎ (0.006) 0.956⁎⁎⁎

0.010⁎⁎⁎ (0.003) 0.050⁎⁎⁎ (0.008) 0.945⁎⁎⁎

0.016⁎⁎⁎ (0.005) 0.048⁎⁎⁎ (0.008) 0.944⁎⁎⁎

0.013⁎ (0.008) 0.041⁎⁎⁎ (0.009) 0.950⁎⁎⁎

(0.012) −0.065⁎⁎⁎ (0.012) 6.451⁎⁎⁎ (0.648) 0.021 (0.020)

(0.008) −0.044⁎⁎⁎

(0.011)

(0.013) −0.065⁎⁎⁎

(0.006)

(0.009)

(0.009)

(0.014)

(0.012) 5.941⁎⁎⁎ (0.561) −0.009 (0.017)

(0.012) 7.072⁎⁎⁎ (0.794) 0.057⁎⁎⁎

4.318⁎⁎⁎ (0.376) −0.037⁎

−0.021 (0.103) 4.568⁎⁎⁎

(0.022)

(0.019)

4.600⁎⁎⁎ (0.386) −0.005 (0.018)

−0.046⁎⁎ (0.022) 5.454⁎⁎⁎

(0.010)

(0.211) −0.074 (0.219) 4.620⁎ (2.814) 0.005 (1.728)

(0.577)

(0.358)

0.764 0.473

0.844 0.445

0.902 0.443

0.288 0.265

0.663 0.535

0.666 0.153

0.164 0.154

0.869 0.204

0.279 0.227

−0.001 (0.009) 0.102⁎⁎⁎ (0.016)

−0.000 (0.007) 0.129⁎⁎⁎ (0.015)

−0.016 (0.010)

0.011⁎⁎⁎ (0.003) −0.046⁎ (0.024)

0.029⁎⁎⁎ (0.012)

−0.005 (0.003) −0.012 (0.024) −0.042⁎⁎ (0.019)

−0.001 (0.002)

−0.002 (0.004)

−0.008⁎⁎⁎ (0.003) 0.089⁎⁎⁎ (0.016) 0.016 (0.015)

β

0.009⁎⁎⁎ (0.003) 0.176⁎⁎⁎ (0.024) 0.870⁎⁎⁎

0.004⁎⁎⁎ (0.002) 0.100⁎⁎⁎ (0.018) 0.909⁎⁎⁎

0.002⁎⁎ (0.001) 0.040⁎⁎⁎ (0.007) 0.958⁎⁎⁎

0.000 (0.000) 0.242 (0.216) 0.902⁎⁎⁎

0.009⁎⁎⁎ (0.003) 0.080⁎⁎⁎ (0.014) 0.932⁎⁎⁎

0.000 (0.000) 0.125⁎⁎⁎ (0.000) 0.854⁎⁎⁎

0.000⁎⁎⁎ (0.000) 0.130⁎⁎⁎ (0.013) 0.870⁎⁎⁎

0.000 (0.000) 0.112⁎⁎⁎ (0.010) 0.888⁎⁎⁎

0.002⁎⁎⁎ (0.001) 0.278⁎⁎⁎ (0.034) 0.777⁎⁎⁎

(0.016) −0.101⁎⁎⁎

(0.015) −0.032⁎⁎ (0.016) 6.534⁎⁎⁎ (0.722)

(0.005) −0.001⁎⁎⁎ (0.008) 8.014⁎⁎⁎ (1.334)

(0.015) 0.401 (0.358) 2.161⁎⁎⁎ (0.152)

(0.012) −0.036⁎⁎⁎ (0.012) 8.348⁎⁎⁎ (0.930)

(0.075) 0.168⁎⁎⁎ (0.038) 2.803⁎⁎⁎ (0.378)

(0.015)

(0.011)

γ

4.319⁎⁎⁎ (0.212)

5.354⁎⁎⁎ (0.357)

(0.021) −0.035 (0.032) 3.863⁎⁎⁎ (0.275)

α β γ η ϕ Goodness-of-fit tests K-S A-D Panel C: US dollar Conditional mean μ φ1

4.946⁎⁎⁎ (0.384) −0.018⁎

φ2 Conditional variance ω α

η

(0.021) 7.268⁎⁎⁎ (0.742)

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

9

Table 3 (continued)

ϕ Goodness-of-fit tests K-S A-D

Brazil

Chile

Czech

Russia

South Africa

China

India

Malaysia

Thailand

0.031 (0.021)

0.003 (0.022)

−0.014 (0.019)

0.107⁎⁎⁎ (0.039)

0.079⁎⁎⁎ (0.019)

−0.128⁎⁎ (0.065)

0.046⁎⁎⁎ (0.015)

−0.030 (0.020)

−0.022 (0.016)

0.319 0.471

0.225 0.447

0.481 0.658

0.076 0.068

0.260 0.515

0.037 0.057

0.139 0.047

0.004 0.072

0.185 0.058

Note. This table provides parameter estimates of marginal distribution models with standard errors in parentheses. Parameters of marginal distribution models are defined in Eqs. (17)–(21). For K-S test and A-D test, any p-value less than 0.05 indicate a rejection of the null hypothesis that the particular model is well specified at the 5% level. ⁎⁎⁎, ⁎⁎, ⁎ indicate statistical significance at the 1%, 5% and 10% level respectively.

^ t and ^ CDFs are applied to the standardized residuals, and u vt for t = 1 to T are computed. We estimate the copula parameters by maximizing the log-likelihood: ^θ ¼ arg max c θc

T X

^ t ; ^vt ; θc Þ: lncðu

ð22Þ

t¼1

4.3. Safe haven property 4.3.1. Low-high tail dependence As the Gaussian, Student-t, Gumbel, and 180-degree rotated Gumbel copulas capture extreme movements in tandem, and their lower tail dependence is a non-zero probability, we formulate the hypothesis λL = 0 (see Eq. (9)) to determine whether gold/the US dollar can serve as a weak safe haven asset for emerging stocks markets. Also, as the 270-degree rotated Student-t copula and 90-degree rotated Gumbel copula can capture extreme movements in an opposite direction, we formulate the hypothesis λ′U N 0 (see Eq. (12)) to determine whether gold/the US dollar can serve as a strong safe haven asset for emerging stock markets; the λ′U is also noted as the low(in emerging stocks)-high(in gold/US dollars) tail dependence in this paper. If the commonly used copulas (including the Gaussian, Student-t, Gumbel, and 180-degree rotated Gumbel copula) are obviously superior to the 270-degree rotated Student-t copula and 90-degree rotated Gumbel copula, and the hypothesis λL = 0 cannot be rejected, then gold/the US dollar can serve as a weak safe haven asset for emerging stocks. If the 270-degree rotated Student-t copula and 90-degree rotated Gumbel copula are obviously superior to the commonly used copulas and the hypothesis λ′U N 0 cannot be rejected, then gold/the US dollar can serve as a strong safe haven asset for emerging stocks. If the optimality of the commonly used copulas and that of the 270-degree rotated Student-t copula and 90-degree rotated Gumbel copula are quite similar or equal to each other, we first need to test the hypothesis λ′U N 0. If such a hypothesis cannot be rejected, then the strong safe haven property is verified (and then the weak safe haven property does not need to be tested); if it is rejected, the hypothesis λL = 0 for the weak safe haven property needs to be further tested. 4.3.2. Downside risk gains As noted in Section 1, tail dependence implies only the ability to hedge infinitely extreme risks; to examine the safe haven property comprehensively, extreme risk hedging at various confidence levels should also be investigated. In this subsection, we specifically calculate the extent to which the extreme risks of emerging stocks can be reduced by gold/the US dollar, which is noted as the downside risk gains provided by gold/the US dollar for emerging stocks. To this end, portfolios consisting of emerging stocks and gold/the US dollar need to be established first. Given their popularity among private investors, fund and index providers, pension funds, endowments, and other long-term investors, the naïve diversification rule and risk parity approach are used (e.g., Huberman and Jiang, 2006; Anderson et al., 2012; Bessler and Wolff, 2015). These strategies are simple to implement and eliminate the costs induced by adjusting the dynamic weights of assets over time. 1=σ 2 The weight of each asset under the naïve diversification rule is 1/2, and is wi ¼ 2 i 2 , with σ2i being the variance of returns of ∑i¼1 ð1=σ Þ asset i (for each emerging country, i is stock and gold/stock and the US dollar) under ithe risk parity approach. As for characterizing extreme risks, the popular measures of VaR (Value-at-Risk) and ES (Expected Shortfall) are used. Here, VaR for asset returns is defined as follows at the confidence level of (1 − p) at time t: PrðRt bVaRt jψt−1 Þ ¼ p;

ð23Þ

where ψt − 1 is the information set at time t. According to Eq. (23), given a time horizon, we have (1 − p) confidence that the loss is no larger than VaR. However, as this measure gives only a probability of extreme loss and not the magnitude of these losses, ES is also measured and given by: ES ¼ E½Rt jRt bVaRt ðpÞ;

ð24Þ

We use a Monte Carlo simulation to calculate the VaR and ES of portfolio returns. We simulate 5000 values of the uniform variates of the best-fitted copula model at each time (t = 1 to T), and invert the marginal cumulated distribution function to obtain standardized residuals for each asset. These simulated standardized residuals are then used to compute portfolio returns, with the Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

10

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 4 Estimates of the optimal static and GAS dynamic copulas. Panel A: stock index - gold price

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Static Student-t Copula ρ v −0.129⁎⁎⁎ 9.861⁎⁎⁎ (0.015) (1.697) −0.002 15.949⁎⁎⁎ (0.016) −0.009 (0.016) −0.066⁎⁎⁎ (0.015) 0.010 (0.017) 0.092⁎⁎⁎ (0.020) −0.078⁎⁎⁎ (0.015) −0.078⁎⁎⁎ (0.020) −0.023 (0.016)

(4.216) 11.685⁎⁎⁎

AIC −106.363

8.349

−12.695

13.573

(2.392) 9.711⁎⁎⁎ (1.674) 7.143⁎⁎⁎ (0.918) 11.170⁎⁎⁎ (2.650) 11.043⁎⁎⁎ (2.130) 9.084⁎⁎⁎ (1.977) 15.726⁎⁎⁎

−23.143

28.221

−52.439

39.133

−74.262

22.479

−40.954

27.617

−51.231

20.680

−37.355

8.900

−13.796

(4.149)

GAS dynamic Student-t Copula βGAS c αGAS −0.003 0.030 0.987⁎⁎⁎ (0.003) (0.021) (0.000) 0.000 0.033⁎⁎⁎ 0.974⁎⁎⁎ (0.001) 0.000 (0.000) −0.001 (0.000) 0.000 (0.001) 0.000 (0.000) −0.001⁎⁎ (0.000) 0.000 (0.000) −0.001 (0.001)

LL 55.183

v 10.493⁎⁎⁎ (0.009) 16.313⁎⁎⁎

(0.002) 0.015⁎⁎⁎ (0.001) 0.026⁎⁎⁎ (0.001) 0.045⁎⁎⁎

(0.000) 0.997⁎⁎⁎ (0.000) 0.998⁎⁎⁎ (0.000) 0.993⁎⁎⁎

(0.002) 0.016⁎⁎⁎

(0.000) 0.999⁎⁎⁎

(0.035) 13.210⁎⁎⁎ (0.001) 11.848⁎⁎⁎ (0.000) 10.101 (0.002) 15.823⁎⁎⁎

(0.001) 0.018⁎⁎⁎ (0.005) 0.010⁎⁎⁎

(0.000) 0.997⁎⁎⁎ (0.000) 0.999⁎⁎⁎

(0.013) 13.298⁎⁎⁎ (0.019) 9.872⁎⁎⁎

(0.003) 0.021⁎⁎⁎

(0.000) 0.990⁎⁎⁎

(0.000) 16.155⁎⁎⁎

(0.008)

(0.000)

(0.059)

LL 72.020

AIC −136.031

17.343

−26.676

28.717

−49.426

96.498

−184.986

115.121

−222.233

38.116

−68.217

49.320

−90.632

36.848

−65.682

19.244

−30.479

LL 260.690

AIC −517.377

22.468

−40.933

19.334

Static 270-degree-rotated Student-t Copula ρ v 0.129⁎⁎⁎ 9.862⁎⁎⁎ (0.016) (1.756) 0.002 15.968⁎⁎⁎ (0.016) 0.009 (0.016) 0.066⁎⁎⁎ (0.013) −0.010 (0.015) −0.092⁎⁎⁎ (0.020) 0.078⁎⁎⁎ (0.016) 0.078⁎⁎⁎ (0.020) 0.023 (0.016)

(5.131) 11.694⁎⁎⁎ (2.156) 9.717⁎⁎⁎ (1.367) 7.143⁎⁎⁎ (1.134) 11.172⁎⁎⁎ (2.605) 11.045⁎⁎⁎ (1.514) 9.084⁎⁎⁎ (0.338) 15.748⁎⁎⁎

LL 55.183

AIC −106.363

8.349

−12.695

13.573

−23.143

28.221

−52.439

39.133

−74.262

22.479

−40.954

27.617

−51.231

20.680

−37.355

8.900

−13.796

(3.411)

GAS dynamic 270-degree-rotated Student-t Copula c αGAS βGAS v LL 0.003 0.031 0.987⁎⁎⁎ 10.482⁎⁎⁎ 71.998 (0.001) (0.004) (0.002) (0.017) 0.000 0.034⁎⁎⁎ 0.974⁎⁎⁎ 16.129⁎⁎⁎ 17.276 (0.000) (0.003) (0.017) (0.036) 0.000 0.016⁎⁎⁎ 0.997⁎⁎⁎ 13.158⁎⁎⁎ 28.796 (0.000) (0.001) (0.000) (0.026) 0.999⁎⁎⁎ 12.048⁎⁎⁎ 96.488 0.000 0.025⁎⁎⁎ (0.004) (0.000) (0.008) (0.000) 0.000 0.045⁎⁎⁎ 0.994⁎⁎⁎ 10.309 114.977 (0.000) (0.015) (0.003) (0.020) 0.000 0.015⁎⁎⁎ 0.999⁎⁎⁎ 15.873⁎⁎⁎ 38.098 (0.000) (0.016) (0.000) (0.031) 0.001⁎⁎ 0.017⁎⁎⁎ 0.997⁎⁎⁎ 13.333⁎⁎⁎ 49.223 (0.000) (0.005) (0.002) (0.013) 0.001 0.009 0.999⁎⁎⁎ 9.709⁎⁎⁎ 36.684 (0.001) (0.009) (0.000) (0.076) 0.001 0.022⁎⁎⁎ 0.990⁎⁎⁎ 15.873⁎⁎⁎ 19.383 (0.000) (0.004) (0.000) (0.010)

AIC −135.987 −26.543 −49.583 −184.967 −221.945 −68.182 −90.436 −65.354 −30.757

Panel B: stock index – US dollar

Brazil

Static Student-t Copula ρ v −0.331⁎⁎⁎ 12.772⁎⁎⁎

Czech

(0.014) −0.038 (1.000) −0.051⁎⁎⁎

(2.785) 10.000⁎⁎⁎ (1.000) 11.761⁎⁎⁎

Russia

(0.016) −0.303⁎⁎⁎

(2.403) 11.652⁎⁎⁎

India

(0.013) −0.191⁎⁎⁎ (0.016) −0.139⁎⁎⁎ (0.018) −0.328⁎⁎⁎

(2.482) 5.901⁎⁎⁎ (0.623) 39.001⁎⁎⁎ (23.742) 8.462⁎⁎⁎

Malaysia

(0.013) −0.349⁎⁎⁎

(1.185) 9.975⁎⁎⁎

(0.015) −0.194⁎⁎⁎ (0.015)

(1.396) 8.392⁎⁎⁎ (1.332)

Chile

South Africa China

Thailand

Brazil

GAS dynamic Student-t Copula βGAS c αGAS −0.002⁎⁎ 0.016⁎⁎⁎ 0.996⁎⁎⁎ (0.001) (0.001) (0.000)

209.888

LL 260.690

AIC −517.377

22.519

−41.035

19.334

−34.664

209.888

−415.773

131.320

−258.637

−34.664

(0.014) 0.038⁎⁎ (0.016) 0.051⁎⁎⁎

(2.822) 9.449⁎⁎⁎ (1.535) 11.771⁎⁎⁎

−415.773

(0.016) 0.303⁎⁎⁎

(2.866) 11.662⁎⁎⁎ (0.651) 5.900⁎⁎⁎ (1.113) 38.920 (28.276) 8.454⁎⁎⁎

30.396

−56.787

302.107

−600.212

217.226

−430.448

131.320

−258.637

30.396

−56.787

302.107

−600.212

(0.042) 0.191⁎⁎⁎ (0.016) 0.139⁎⁎⁎ (0.019) 0.328⁎⁎⁎

−430.448

(0.014) 0.349⁎⁎⁎

(1.051) 9.950⁎⁎⁎ (1.994) 8.395⁎⁎⁎ (1.324)

217.226

v 13.947⁎⁎⁎ (0.000)

Static 270-degree-rotated Student-t Copula ρ v 0.331⁎⁎⁎ 12.776⁎⁎⁎

115.023

−226.043

(0.016) 0.194⁎⁎⁎ (0.016)

115.023

−226.043

LL 278.639

AIC −549.269

GAS dynamic 270-degree-rotated Student-t Copula c αGAS βGAS v LL 0.680⁎⁎⁎ −0.034 0.007 12.821 261.350 (0.046) (0.065) (0.015) (0.056)

AIC −514.690

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

11

Table 4 (continued) Panel A: stock index - gold price Chile Czech Russia South Africa China India Malaysia Thailand

−0.078 (0.114) 0.000 (0.000) −0.002⁎⁎⁎ (0.000) −0.001⁎⁎⁎

0.027 (0.033) 0.023⁎⁎⁎ (0.002) 0.034⁎⁎⁎ (0.003) 0.051⁎⁎⁎

0 (0.823) 0.997⁎⁎⁎ (0.000) 0.997⁎⁎⁎ (0.000) 0.996⁎⁎⁎

9.515⁎⁎⁎ (0.019) 15.456⁎⁎⁎ (0.001) 16.393⁎⁎⁎ (0.001) 13.812⁎⁎⁎

(0.000) −0.063⁎⁎⁎ (0.004) −0.001 (0.001) −0.034⁎⁎⁎

(0.002) 0.090⁎⁎⁎ (0.006) 0.020⁎⁎⁎ (0.002) 0.035⁎⁎⁎

(0.000) 0.789⁎⁎⁎ (0.024) 0.998⁎⁎⁎ (0.000) 0.952⁎⁎⁎

(0.001) 39.683⁎⁎⁎ (0.008) 10.560⁎⁎⁎ (0.000) 10.010⁎⁎⁎

(0.001) −0.002 (0.007)

(0.001) 0.016 (0.129)

(0.000) 0.996⁎⁎⁎

(0.010) 8.842⁎⁎⁎

(0.000)

(0.008)

22.847

−37.685

69.248

−130.488

288.223

−568.437

316.737

−625.464

0.071⁎⁎⁎ (0.000) 0.000 (0.000) 0.003⁎⁎⁎ (0.001) 0.001⁎⁎⁎

0.027⁎⁎⁎ (0.000) 0.024⁎⁎⁎ (0.004) 0.035⁎⁎⁎ (0.003) 0.051⁎⁎⁎

0.087⁎⁎⁎ (0.013) 0.997⁎⁎⁎ (0.000) 0.996⁎⁎⁎ (0.000) 0.996⁎⁎⁎

9.515⁎⁎⁎ (0.018) 15.385⁎⁎⁎ (0.010) 16.129⁎⁎⁎ (0.016) 13.514⁎⁎⁎

(0.000) 0.794⁎⁎⁎ (0.039) 0.998⁎⁎⁎ (0.000) 0.953⁎⁎⁎ (0.000) 0.996⁎⁎⁎

(0.004) 40.000⁎⁎⁎ (0.007) 10.417⁎⁎⁎ (0.017) 10.000 (0.103) 8.772⁎⁎⁎

(0.000)

(0.025)

39.940

−71.866

360.140

−712.270

221.456

−434.898

(0.000) 0.061⁎⁎⁎ (0.020) 0.002⁎⁎⁎ (0.000) 0.034⁎⁎⁎

−253.239

(0.014) 0.002⁎⁎⁎

(0.006) 0.094⁎⁎⁎ (0.009) 0.023⁎ (0.012) 0.035 (0.476) 0.016⁎⁎⁎

(0.000)

(0.003)

130.624

22.846

−37.683

69.417

−130.824

287.720

−567.431

317.013

−626.017

39.937

−71.859

360.013

−712.017

221.476

−434.938

130.652

−253.294

Note. This table provides parameter estimates of the optimal static and GAS dynamic copulas with standard errors in parentheses. Log-likelihood (LL) and Akaike information criterion (AIC) values adjusted for small-sample bias are provided for the copula models. ⁎⁎⁎, ⁎⁎, ⁎ indicate statistical significance at the 1%, 5% and 10% level respectively.

estimates of marginal distribution models, and the portfolio weights under the naïve diversification rule and risk parity approach. Finally, we measure VaR as the pth quantile of the empirical distribution of these simulated portfolio returns, and ES as the expected loss exceeding VaR. The downside risk gains offered by gold/the US dollar are concretely calculated as the average of VaR and ES decreases given by Eqs. (25) and (26), respectively.3 Positive values of Eq. (25) and (26) indicate that gold/the US dollar decreases the VaR and ES of emerging stocks, and thus the safe haven property of gold/the US dollar for the emerging stocks is implied4: VaRDec:portfolio ¼

ESDec:portfolio ¼

1 T VaRstock;t −VaRportfolio;t ∑t¼1 ; T VaRstock;t

1 T ESstock;t −ESportfolio;t ∑t¼1 : T ESstock;t

ð25Þ

ð26Þ

5. Empirical results 5.1. Estimates of marginal distributions Table 3 presents the estimates of the marginal distribution models for emerging stock returns, gold price returns, and US dollar value changes. For most of the asset returns, the optimal marginal distribution model is the AR (p)-GJR-GARCH (1, 1) model. The highly significant coefficients β and α in almost all cases indicate that the volatility depends not only on past volatility but also on contemporaneous shocks to returns. The leverage effect exists in more than half of all asset returns. Similarly, the asymmetry parameter ϕ is also significant in more than half of all cases, while the kurtosis parameter η is significant in almost all cases, thus confirming the suitability of the skewed-t distribution in modeling the leptokurtic, fat-tailed, and skewed behavior of the asset returns. Based on the results of goodness-of-fit checks to test whether the standardized residuals are i.i.d. uniform (0,1), we reject the null hypothesis of correct model specification for only 4 out of the 27 cases in the K-S (Kolmogorov-Smirnov) test and for only one out of the 27 cases in the A-D (Anderson-Darling) test, and none of the assets reject the null hypothesis in either the K-S or A-D test. Thus, overall, the marginal distribution models adequately describe the distributions of individual asset returns, and the copula functions can be used to capture the dependence structure between emerging stocks and gold/US dollar prices. 5.2. Estimates of copulas Following prior studies (see, among others, Wen et al., 2012; Reboredo, 2013; Wang et al., 2013; Reboredo et al., 2016), we use the values of the log-likelihood ratios and AICs to select the optimal copula function for paired assets of emerging stocks and gold/the US dollar. 3 The VaR of individual emerging stock returns is calculated as: VaRt = μt + F−1 skew − t(η,ϕ)(p)σt, where μt and σt is the conditional mean and volatility of emerging stock returns, respectively; F−1 skew − t(η,ϕ)(p) is the pth quantile of the skewed-t distribution with the kurtosis parameter η and the asymmetry parameter ϕ. Given the VaR, the ES of the emerging stock is the expected loss that exceeds its VaR. 4 The strong/weak safe haven property is not distinguished from the viewpoint of downside risk gains.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

12

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 5 Descriptive statistics of the low-high tail dependence. Panel A: stock index - gold price

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

0.0173 0.0012 0.0037 0.0116 0.0127 0.0009 0.0062 0.0187 0.0015

147.1 101.2 122.7 54.8 67.2 44.4 103.1 99.9 112.6

0.0077 0.0008 0.0020 0.0139 0.0124 0.0010 0.0040 0.0100 0.0009

0.0038 0.0001 0.0004 0.0002 0.0001 0.0001 0.0004 0.0060 0.0002

0.0476 0.0054 0.0111 0.0895 0.0768 0.0064 0.0247 0.0558 0.0056

Panel B: stock index - US dollar

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

t (US dollar N gold)

0.0245 0.0153 0.0036 0.0169 0.0205 0.0000 0.0506 0.0528 0.0400

718.4 / 58.4 63.9 61.7 32.0 112.7 251.4 192.5

0.0022 0 0.0040 0.0174 0.0218 0.0000 0.0295 0.0112 0.0136

0.0185 0.0153 0.0000 0.0003 0.0000 0.0000 0.0050 0.0256 0.0121

0.0392 0.0153 0.0210 0.0877 0.1275 0.0002 0.1559 0.0865 0.0834

56.6 / −1.7 21.4 26.5 −44.1 103.3 121.1 186.0

Note. This table provides summary statistics for λU′ from the 270-degree rotated Student-t copula (i.e., the low (in emerging stocks)-high (in gold/the US dollar) tail dependence), including the mean, standard deviation, minimum, maximum of λU′ and t statistic for testing λU′ N 0; t (US dollar N gold) is the t statistic for the hypothesis that the λU′ of emerging stock–the US dollar pair is larger than that of emerging stock–gold pair.

Table 4 reports the estimates of the optimal static and GAS dynamic copulas.5 Among the static copulas, the static Student-t copula and the static 270-degree rotated Student-t copula are both the best-fit copulas for paired assets of emerging stocks and gold, as they have the same log-likelihood ratios and AIC; such a conclusion also holds for paired assets of emerging stocks and the US dollar, with the only exception being Chile, where the latter copula is slightly better than the former one. As the static Student-t copula captures price movements in tandem while the static 270-degree rotated Student-t copula captures price movements in an opposite direction, the sign of their parameter ρ is found to be opposite. We find a more significant and more negative relationship between the returns of emerging stocks and the US dollar values, which is consistent with the results shown in Table 1. As for the GAS dynamic copulas, given the similar log-likelihood ratios and AICs, both the GAS dynamic Student-t copula and GAS dynamic 270-degree rotated Student-t are found to be both optimal for all paired assets. Compared with the corresponding static versions of these two copulas, the GAS dynamic versions are much preferred for all paired assets, excepting the stock-US dollar pair of Chile. The significance of the parameters in the GAS dynamic copulas in almost all cases also implies that the dynamic copulas are more suitable than the constant ones. Given the similar performance of the GAS dynamic Student-t copula and the GAS dynamic 270-degree rotated Student-t copula, and based on the discussion in Subsection 4.3.1, we firstly use the λ′U of the GAS dynamic 270-degree rotated Student-t copula to examine the strong safe haven property of gold/the US dollar for the emerging stock markets in the following analysis. Simultaneously, according to the dependence structure depicted by the GAS dynamic Student-t copula, we calculate the downside risk gains provided by gold/the US dollar, given that this copula function is still slightly better than the GAS dynamic 270-degree rotated Student-t copula in most cases. Regarding the exception of Chile, we use the static Student-t copula to characterize the dependence structure of the returns of stock prices and US dollar values for calculating the downside risk gains, and use the static 270-degree rotated Student-t copula to examine their opposite extreme movements. 5.3. Safe haven property 5.3.1. Low-high tail dependence Table 5 reports the descriptive statistics and hypothesis results of the λ′U of the 270-degree- rotated Student-t copula. The t statistics for testing the hypothesis of λU′ N 0 show that this hypothesis cannot be rejected because the t statistics values are very large and positive for all paired assets. Therefore, both gold and the US dollar can serve as strong safe haven assets for emerging stock markets.6 5

To save space, we only present the estimates of the optimal copulas. The estimates for the other copulas in Table 2 are available from the authors upon request. Because the hypothesis of λU′ N 0 cannot be rejected, that of λL = 0 does not need to be tested. However, for robustness, we still tested and find that of λL = 0 indeed cannot be rejected either. To save space, we do not present the results, but it is available from the authors upon request. 6

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 Brazil

.02 .01

Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 Chile

stock-dollar

stock-gold

.006 stock-dollar

.08 stock-dollar

stock-gold

stock-dollar

.04 0

.04 0 stock-gold

Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 China

.08

stock-gold

.1 .05 0 Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 India

stock-dollar

0

.05 0 Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 South Africa

stock-dollar

.15

stock-gold

stock-gold

.003

.1

.08 .04 0 Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 Russia

Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 Czech

stock-dollar

.15

stock-gold

13

0

0

.02

.04

0 .005 .01 .015

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 Malaysia stock-gold

stock-dollar

Jan 1, 2000 Jan 1, 2005 Jan 1, 2010 Jan 1, 2015 Thailand stock-gold

stock-dollar

Fig. 2. The development of the low(in emerging stocks)–high(in gold/the US dollar) tail dependence.

Fig. 2 shows the development of λU′ between returns of emerging stocks and gold price/US dollar values. In all emerging countries except China, the λU′ for emerging stock prices and US dollar rates is higher than that for emerging stock and gold prices. However, Table 5 shows that, in addition to China, the λU′ of the stock-US dollar pair in the Czech Republic is slightly lower than that of the stock-gold pair because the statistics of t (US dollar N gold) is −1.7. In contrast, in the other countries the former λU′ is significantly larger than the latter λU′ because the statistics of t (US dollar N gold) are large and positive. Thus, the US dollar is a better safe haven for emerging stock markets than gold in most cases.

5.3.2. Downside risk gains Table 6 shows the downside risk gains provided by gold/the US dollar for emerging stock markets at the 99%, 99.5%, and 99.9% confidence levels. For convenience of comparison, the portfolios under the naïve diversification rule and the risk parity approach are noted as EW and ERC, respectively.7 From this table, the ERC strategy is superior to the EW strategy, as the former strategy brings larger downside risk gains in most cases. Under the ERC strategy, all of the VaR(p) Dec and ES(p) Dec (p = 1%, 0.5%, and 0.1%) are positive, indicating that both gold and the US dollar decrease the VaR and ES of stocks in all emerging countries; thus, they can serve as safe haven assets for emerging stock markets. In addition, the US dollar provides larger downside risk gains than gold in almost all cases, thus again confirming that the US dollar is a better safe haven than gold.8 With the ERC strategy, the US dollar provides the highest downside risk gains for Thailand at all confidence levels, and the lowest for South Africa and Russia at the 99.5%/99% and 99.9% confidence levels, respectively. As for emerging countries with a shorter sample period, including China and Malaysia, the downside risk gains provided by the US dollar for Chinese stocks are attractive, while those for Malaysian stocks are much lower. Accordingly, the VaR/ES reductions offered by the US dollar for stocks of Thailand and China can be as high as 0.8 and 0.92, respectively, while those for Russia can be as low as 0.069, and reductions for the other emerging countries range between 0.39 and 0.75. 7 The naïve diversification rule is traditionally regarded as the equally weighted portfolio strategy, so it is noted as EW; the risk-parity approach is also called the equally weighted risk contribution in Maillard et al., 2010, so it is noted as ERC. 8 We tested the accuracy of the VaR for each portfolio using the likelihood ratio test for “correct conditional coverage” proposed by Christoffersen, 1998. The results show that the model can accurately calculate the VaR of the portfolios in almost all cases. To save space, these figures are not provided. More details are available from the authors upon request.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

14

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 6 Downside risk gains. VaR (1%) Dec

VaR (0.5%) Dec

VaR (0.1%) Dec

ES (1%) Dec

ES (0.5%) Dec

ES (0.1%) Dec

Panel A: stock index - gold price Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.383 0.393 0.123 0.173 0.344 0.353 0.409 0.446 0.288 0.288 0.323 0.322 0.363 0.358 0.003 0.157 0.346 0.345

0.393 0.390 0.173 0.172 0.353 0.353 0.446 0.444 0.288 0.288 0.336 0.320 0.359 0.366 0.020 0.172 0.345 0.345

0.364 0.369 0.059 0.128 0.346 0.358 0.389 0.415 0.274 0.277 0.356 0.288 0.371 0.367 0.042 0.186 0.318 0.297

0.372 0.376 0.082 0.148 0.344 0.351 0.403 0.432 0.278 0.280 0.340 0.898 0.367 0.360 0.018 0.486 0.306 0.316

0.364 0.364 0.052 0.129 0.346 0.350 0.396 0.422 0.270 0.273 0.349 0.294 0.368 0.360 0.024 0.189 0.275 0.295

0.323 0.302 −0.108 0.034 0.351 0.330 0.340 0.389 0.222 0.230 0.351 0.278 0.364 0.322 −0.005 0.203 0.073 0.172

Panel B: Stock Index - US dollar Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.504 0.601 0.406 0.449 0.435 0.523 0.468 0.628 0.420 0.431 0.496 0.917 0.509 0.747 0.475 0.483 0.496 0.774

0.508 0.603 0.409 0.454 0.434 0.531 0.481 0.613 0.426 0.437 0.504 0.911 0.505 0.745 0.485 0.469 0.491 0.781

0.515 0.553 0.409 0.400 0.442 0.511 0.427 0.069 0.441 0.387 0.574 0.871 0.520 0.626 0.506 0.497 0.483 0.754

0.509 0.604 0.408 0.457 0.440 0.542 0.441 0.544 0.430 0.441 0.532 0.898 0.511 0.729 0.493 0.486 0.447 0.782

0.511 0.606 0.408 0.462 0.444 0.553 0.417 0.488 0.436 0.447 0.557 0.886 0.513 0.717 0.503 0.494 0.414 0.785

0.511 0.611 0.404 0.475 0.472 0.586 0.255 0.223 0.447 0.459 0.630 0.843 0.531 0.683 0.522 0.538 0.199 0.796

Note. This table shows the evaluations of downside risk gains provided by gold/the US dollar for emerging stock markets. VaR (p) Dec. is the averaged VaR decrease between the VaR of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate VaR decrease); ES (p) Dec. is the averaged ES decrease between the ES of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate ES decrease). The numbers written in bold are the highest and lowest reductions in VaR/ES with the US dollar as the safe haven asset and under the ERC strategy.

The prominent superiority of the US dollar in hedging the downside risks of stocks in Thailand may be attributed to the 1997 Asian Financial Crisis and Thailand's subsequent inclination to choose this powerful currency rather than gold to hedge against extreme risks. As for China, the depreciation pressure faced by the Chinese yuan may be the reason for the high downside risk gains of the US dollar for Chinese stocks. Also, Russian-U.S. relations have worsened in recent years, which might limit the potential of the US dollar as a safe haven for Russia. In addition, according to Bekiros et al. (2017), the increased interest from emerging countries in gold as a diversifying asset during the past decade has strengthened the co-movement between gold and stock markets, which should be the reason that the US dollar is superior to gold in almost all cases. 5.4. Further discussions As is shown in Fig. 1, the emerging stock index closely co-moves with the world commodity price, with both experiencing sharp falls during the 2008 global financial crisis and significant slumps since the beginning of 2014. As the examination of safe haven assets must focus on times of crisis or turmoil (Baur and McDermott, 2016), we focus on these two subsamples in this section.9 In addition, the in-sample estimation implicitly assumes a perfect forecast of expected asset returns, volatilities, and asset correlations, while the out-of-sample analysis is based only on the information available at present. Thus, the portfolio benefits of in-sample analysis may be unobservable in an out-of-sample analysis (e.g., Daskalaki and Skiadopoulos, 2011; Bessler and Wolff, 2015). For this reason, an out-of-sample estimation is also conducted. 9 According to Liu et al. (2016), it is also necessary to test whether gold (the US dollar) can serve a safe haven asset for stock market crash when the US dollar (gold) performs extremely poorly. For this reason, we pay a special attention to the period of the recent global financial crisis when various assets prices tend to simultaneously slump and be more volatile due to the contagion effect.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

15

5.4.1. Subsample analysis For all of the emerging countries in our sample, the first subsample, the Global Financial Crisis, starts on July 1, 2008 and ends on July 1, 2011, and the second subsample, the Global Commodity Price Slump, starts on January 1, 2014 and runs to the end of the full sample on July 6, 2016. In the subsample analysis, to keep the model consistent and for convenient comparison with the full sample analysis, we use the same marginal distribution models and copulas as for the full sample. Table 7 reports the descriptive statistics and hypothesis results of the λ’U, the low (in emerging stocks)-high (in gold/US dollars) tail dependence, from the 270-degree rotated Student-t copula during these two subsamples. The t statistics for testing the hypothesis of λU′ N 0 show that this hypothesis cannot be rejected because the t statistics values are large and positive for all paired assets during these two subsamples. Therefore, both gold and the US dollar can serve as strong safe haven assets for emerging stocks in these subsamples. However, unlike the results in the full sample, during the global financial crisis, more

Table 7 Descriptive statistics of the low-high tail dependence in subsamples. Panel A: stock index - gold price during global financial crisis

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

0.0192 0.0000 0.0043 0.0000 0.0113 0.0000 0.0045 0.0017 0.0003

128.6 59.4 168.8 31.5 268.1 43.9 101.1 332.5 56.5

0.0042 0.0000 0.0007 0.0000 0.0012 0.0000 0.0013 0.0001 0.0001

0.0101 0.0000 0.0017 0.0000 0.0082 0.0000 0.0018 0.0012 0.0001

0.0322 0.0000 0.0072 0.0002 0.0178 0.0000 0.0081 0.0022 0.0008

Panel B: stock index - US dollar during global financial crisis

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

t (US dollar N gold)

0.0312 0.0447 0.0008 0.0011 0.0000 0.0024 0.0006 0.0003 0.0002

175.7 / 113.4 76.8 15.7 66.3 27.5 97.5 79.9

0.0050 0 0.0002 0.0004 0.0000 0.0010 0.0007 0.0001 0.0001

0.0170 0.0447 0.0002 0.0001 0.0000 0.0002 0.0000 0.0001 0.0000

0.0404 0.0447 0.0015 0.0021 0.0000 0.0079 0.0043 0.0005 0.0005

64.5 / −150.0 74.6 −270.0 66.2 −82.5 −270.0 −11.7

Panel C: Stock Index - Gold Price during Global Commodity Price Slump

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

0.0006 0.0002 0.0086 0.0382 0.0009 0.0006 0.0044 0.0046 0.0125

89.0 55.8 125.1 128.0 39.2 68.3 659.3 81.7 73.2

0.0002 0.0001 0.0018 0.0077 0.0006 0.0002 0.0002 0.0014 0.0044

0.0003 0.0001 0.0037 0.0212 0.0002 0.0001 0.0039 0.0020 0.0032

0.0013 0.0011 0.0138 0.0639 0.0030 0.0012 0.0053 0.0146 0.0237

Panel D: Stock Index - US dollar during Global Commodity Price Slump

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

t (US dollar N gold)

0.0211 0.0027 0.0000 0.0386 0.0180 0.0088 0.0421 0.0051 0.0255

67.4 / 11.2 72.0 124.8 489.2 827.4 109.4 133.2

0.0080 0 0.0000 0.0137 0.0037 0.0005 0.0013 0.0012 0.0049

0.0099 0.0027 0.0000 0.0075 0.0060 0.0081 0.0379 0.0014 0.0090

0.0726 0.0027 0.0000 0.0813 0.0303 0.0110 0.0459 0.0092 0.0386

66.3 / −130.0 0.8 119.6 360.1 719.7 6.2 74.9

Note. This table provides summary statistics for λU′ from the 270-degree rotated Student-t copula (i.e., the low (in emerging stocks)-high (in gold/the US dollar) tail dependence) in the two sub-samples, with the upper 2 panels being the subsample for the financial crisis period and the lower 2 panels being the subsample for the commodity slump period, including the mean, standard deviation, minimum, maximum of λU′ and t statistic for testing λU′ N 0; t (US dollar N gold) is the t statistic for the hypothesis that the λU′ of emerging stock-the US dollar pair is larger than that of emerging stock–gold pair.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Jul 1, 2008

Jul 1, 2009 Jul 1, 2010 Brazil stock-dollar

.003 .006 Jul 1, 2009 Jul 1, 2010 Chile stock-gold

stock-dollar

Jul 1, 2009 Jul 1, 2010 South Africa stock-gold

Jul 1, 2009 Jul 1, 2010 India stock-gold

stock-dollar

Jul 1, 2008

stock/dollar

Jul 1, 2009 Jul 1, 2010 China stock-gold

stock-dollar

0 .0003.0006

0

.001

.004 0 Jul 1, 2008

Jul 1, 2008

stock-dollar

.002

.008

stock-gold

Jul 1, 2008

stock/gold

0

.01 0 Jul 1, 2009 Jul 1, 2010 Russia

01jul2009 01jul2010 Czech

.003 .006

.002 .001 0 Jul 1, 2008

01jul2008

stock-dollar

.02

stock-gold

Jul 1, 2008

0

0

0

.02

.02

.04

.04

16

Jul 1, 2009 Jul 1, 2010 Malaysia stock-gold

stock-dollar

Jul 1, 2008

Jul 1, 2009 Jul 1, 2010 Thailand stock-gold

stock-dollar

Fig. 3. The development of the low(in emerging stocks)–high(in gold/the US dollar) tail dependence during Global Financial Crisis.

countries, including the Czech Republic, South Africa, India, Malaysia, and Thailand, exhibit negative values of t (US dollar N gold) statistics. This means that, in these countries, the low-high tail dependences of paired assets of emerging stocks and the US dollar are lower than those of paired assets of emerging stocks and gold in times of financial crisis; then, gold is superior to the US dollar in hedging the infinitely extreme risks of these emerging stock markets. According to Baur and McDermott (2016), the phenomenon of investors under stress buying gold is rooted in the behavioral bias associated with gold's history as a currency, as a store of value, and as a safe haven asset. However, during the recent world commodity slump, the US dollar was still a better safe haven asset for all emerging stock markets, excepting the Czech Republic. This can be explained by the observed market data in Fig. 1, in which the value of the US dollar is found to have risen significantly since 2014. Figs. 3 and 4 show the λU′ for the paired assets during the Global Financial Crisis and the Global Commodity Price Slump, respectively. These two figures confirm the results shown in Table 7. Correspondingly, Tables 8 and 9 show the downside risk gains offered by gold/the US dollar for emerging stocks during the Global Financial Crisis and Global Commodity Price Slump, respectively. These two tables confirm that both gold and the US dollar can serve as safe haven assets for emerging stocks, and that the US dollar is a superior safe haven to gold, as it provides larger downside risk gains for emerging stock markets in almost all cases. Under the generally preferred portfolio strategy, the ERC strategy, China benefits the most from using the US dollar as a safe haven asset, followed by Thailand, while Chile gains the least. Accordingly, the downside risk gains offered by the US dollar for China, Thailand, and Chile are around 0.95, 0.87, and 0.49 at the most, respectively.

5.4.2. Out-of-sample analysis In the out-of-sample analysis, we use the same marginal distribution models and copulas used for the full sample to keep the model consistent, and for convenience of comparison with the full sample analysis. For all of the emerging countries except China and Malaysia, we estimate the marginal distribution models and copulas using an in-sample period (from January 4, 2000 to December 31, 2010, including 2869 observations) and evaluate it on the remaining 1438 observations (the out-of-sample period including 4307–2869 = 1438 observations). For China and Malaysia, we estimate the marginal distribution models and copulas using an in-sample period (from July 21, 2005 to December 31, 2012, including 1943 observations) and evaluate it on the remaining observations (the out-of-sample period including 2860–1943 = 917 observations). We use the “fixed window” estimation, whereby the models (including the marginal distribution models and optimal copula) are estimated once, with the data from [1, 1943] for China and Malaysia and [1, 2869] for the other sampled emerging countries; Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

Jan 1, 2014

Jan 1, 2015 Brazil

Jan 1, 2014

stock-dollar

Jan 1, 2015 Chile stock-gold

Jan 1, 2016

Jan 1, 2014

stock-dollar

stock-gold

Jan 1, 2015 India stock-gold

Jan 1, 2016 stock-dollar

Jan 1, 2014

Jan 1, 2014

stock-dollar

Jan 1, 2015 China

stock-dollar

.02

.04

stock-gold

Jan 1, 2016

0

.04 .02 0 Jan 1, 2014

stock-dollar

.01 Jan 1, 2015 Jan 1, 2016 South Africa

0 .005 .01 .015

stock-gold

Jan 1, 2016

stock-gold

Jan 1, 2016

.005

0 Jan 1, 2015 Russia

Jan 1, 2015 Czech

0

.08 .04 0 Jan 1, 2014

Jan 1, 2014

stock-dollar

.01 .02 .03

stock-gold

Jan 1, 2016

17

0 .005 .01 .015

0

.03

.06

0 .001 .002 .003

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Jan 1, 2015 Malaysia stock-gold

Jan 1, 2016 stock-dollar

Jan 1, 2014

Jan 1, 2015 Thailand stock-gold

Jan 1, 2016 stock-dollar

Fig. 4. The development of the low(in emerging stocks)–high(in gold/the US dollar) tail dependence during Global Commodity Price Slump.

then, the out-of-sample estimates are obtained using the in-sample estimates. Patton (2013) points out that such a method is useful when the model is too computationally intensive to estimate.10 Then, with the forecasts of market correlations, marginal conditional means and variances of asset returns, we forecast the VaR and ES following Siburg et al. (2015): (1) for K = 5000, simulate K observations u(k)T + 1, v(k)T + 1 (k = 1,…,K) from the best fitted copula; (2) convert the u(k)T + 1 and v(k)T + 1 to z(k)T + 1,1 and z(k)T + 1,2, respectively, using the quantile function of the skewed Student-t distribution; (3) transform z(k)T + 1,j into the simulated return R(k)T + 1,j = μT + 1,j + σT + 1,j z(k)T + 1,j, where μT + 1,j and σT + 1,j (j = 1, 2) are the forecasted one-step-ahead conditional mean and volatility values, respectively; (4) compute the simulated portfolio returns as R(k)T + 1,P = wT + 1,1 R(k)T + 1,1 + wT + 1,2 R(k)T + 1,2, where wT + 1,j (j = 1, 2) is determined under the naïve diversification rule and risk parity approach; (5) compute the VaR as the value of the pth quantile in the distribution of the simulated portfolio returns for day T + 1 and the ES as the mean value for situations in which the simulated portfolio returns exceeds the VaR; and (6) update the information set and repeat the preceding steps for day T + 2 and so forth. Table 10 reports the descriptive statistics and hypothesis results of the λ′U from the 270-degree rotated Student-t copula, i.e., the low (in emerging stocks)-high (in gold/US dollars) tail dependence, in the out-of-sample analysis. As the t statistics are all large and positive, both gold and the US dollar can serve as strong safe havens for emerging stock markets. However, in Brazil, Russia, South Africa, and China, the λU′ of paired assets of emerging stocks and gold is higher than that of emerging stocks and the US dollar because negative values of t (US dollar N gold) statistics are observed in these countries. The data shown in Fig. 5 confirm the results. This suggests that, as in the subsample of the global financial crisis, the superiority of the US dollar in hedging infinite extreme risks is weakened in the out-of-sample analysis. Correspondingly, Table 11 shows the out-of-sample evaluations of downside risk gains offered by gold/the US dollar for emerging stock markets. The ERC strategy is still found to be preferred to the EW strategy in most cases. This table again confirms

10 In addition to the out-of-sample estimation used in this paper, a recursive or expanding window in which the forecast for observation t is based on data in the interval [1, t-1] and a rolling window using data only in the interval [t-T, t-1] (T is assumed to be the observations for the in-sample period) are used for the out-of-sample estimation. However, the former easily increases the computational burden and the latter has to “throw away” observations from the start of the in-sample period (Patton, 2013).

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

18

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 8 Downside risk gains during global financial crisis. VaR (1%) Dec

VaR (0.5%) Dec

VaR (0.1%) Dec

ES (1%) Dec

ES (0.5%) Dec

ES (0.1%) Dec

Panel A: stock index - gold price Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.414 0.432 0.181 0.202 0.334 0.325 0.442 0.520 0.323 0.321 0.331 0.336 0.410 0.412 0.005 0.138 0.382 0.407

0.423 0.443 0.213 0.225 0.333 0.322 0.456 0.533 0.320 0.320 0.330 0.328 0.430 0.435 0.041 0.162 0.402 0.420

0.450 0.471 0.286 0.278 0.352 0.324 0.502 0.543 0.313 0.316 0.333 0.342 0.482 0.493 0.137 0.235 0.460 0.481

0.433 0.453 0.237 0.245 0.342 0.326 0.472 0.538 0.319 0.320 0.332 0.336 0.449 0.456 0.075 0.188 0.422 0.442

0.445 0.465 0.268 0.268 0.347 0.327 0.491 0.546 0.317 0.319 0.333 0.338 0.470 0.480 0.114 0.217 0.446 0.463

0.475 0.494 0.335 0.321 0.371 0.346 0.535 0.577 0.316 0.320 0.342 0.365 0.520 0.535 0.203 0.290 0.500 0.519

Panel B: stock index - US dollar Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.553 0.658 0.397 0.424 0.466 0.471 0.549 0.731 0.520 0.532 0.521 0.946 0.555 0.702 0.483 0.482 0.464 0.864

0.563 0.668 0.396 0.425 0.467 0.468 0.563 0.742 0.523 0.535 0.510 0.941 0.574 0.710 0.467 0.467 0.473 0.871

0.588 0.692 0.401 0.453 0.492 0.504 0.610 0.733 0.534 0.545 0.506 0.866 0.641 0.697 0.519 0.519 0.579 0.831

0.572 0.676 0.402 0.439 0.478 0.484 0.579 0.742 0.527 0.538 0.515 0.923 0.598 0.711 0.491 0.491 0.513 0.860

0.583 0.685 0.405 0.450 0.487 0.494 0.597 0.748 0.531 0.543 0.514 0.907 0.623 0.715 0.504 0.504 0.548 0.855

0.610 0.706 0.421 0.486 0.519 0.541 0.642 0.746 0.541 0.551 0.525 0.858 0.683 0.717 0.564 0.564 0.638 0.843

Note. This table shows the evaluations of downside risk gains provided by gold/the US dollar for emerging stock markets during the global financial crisis. The total number of observations during the period of global financial crisis is 784. VaR (p) Dec. is the averaged VaR decrease between the VaR of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate VaR decrease); ES (p) Dec. is the averaged ES decrease between the ES of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate ES decrease). The numbers written in bold are the highest and lowest reductions in VaR/ES with the US dollar as the safe haven asset and under the ERC strategy.

that both gold and the US dollar can serve as safe havens for emerging stock markets, and that the US dollar is a better safe haven than gold. Using the US dollar as the safe haven asset for stocks and under the ERC strategy, for countries with a longer sample period, Thailand benefits the most, while Chile gains the least. For countries with a shorter sample period, China has attractive downside risk gains; accordingly, the VaR/ES reductions for China, Thailand, and Chile are around 0.89, 0.85, and 0.47 at the most, respectively. In summary, based on the subsample and out-of-sample analysis, the results that both gold and the US dollar can serve as safe havens for emerging stock markets, and that the US dollar is a better safe haven asset in most cases, are robust. However, we should also note that the superiority of the US dollar in hedging infinitely extreme risks is weakened in the subsample of the global financial crisis and in the out-of-sample analysis.

6. Conclusions This paper provides a detailed comparative analysis of the safe haven property of gold and the US dollar for emerging stock markets. It is motivated by the recent slowdown of emerging economies together with world commodity price slumps while an appreciation of the US dollar, and by the issue of whether gold can serve as a safe haven asset for emerging stock markets as it does for developed markets. Using the daily prices of nine emerging stock markets, the world gold spot price, and US dollar rates measured by the corresponding local currency, this paper uses static and GAS dynamic copula functions to characterize the dependence structure Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

19

Table 9 Downside risk gains during Global Commodity Price Slump. VaR (1%) Dec

VaR (0.5%) Dec

VaR (0.1%) Dec

ES (1%) Dec

ES (0.5%) Dec

ES (0.1%) Dec

Panel A: stock index - gold price Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.436 0.447 0.204 0.237 0.418 0.434 0.422 0.422 0.422 0.422 0.437 0.476 0.431 0.430 0.126 0.235 0.357 0.365

0.435 0.443 0.224 0.256 0.416 0.436 0.410 0.412 0.434 0.433 0.453 0.498 0.446 0.444 0.146 0.233 0.364 0.377

0.430 0.421 0.276 0.303 0.410 0.446 0.247 0.061 0.459 0.461 0.509 0.555 0.407 0.384 0.225 0.246 0.385 0.411

0.434 0.438 0.241 0.271 0.416 0.441 0.364 0.320 0.442 0.442 0.475 0.519 0.438 0.430 0.174 0.243 0.375 0.391

0.433 0.432 0.262 0.291 0.414 0.445 0.323 0.246 0.453 0.454 0.497 0.542 0.436 0.423 0.203 0.249 0.385 0.405

0.432 0.420 0.309 0.336 0.419 0.457 0.217 0.030 0.480 0.482 0.554 0.595 0.420 0.394 0.273 0.289 0.418 0.441

Panel B: stock index - US dollar Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.476 0.498 0.392 0.418 0.456 0.559 0.479 0.471 0.473 0.494 0.479 0.890 0.568 0.747 0.417 0.422 0.490 0.759

0.464 0.459 0.401 0.419 0.452 0.548 0.476 0.465 0.480 0.498 0.484 0.853 0.581 0.756 0.410 0.414 0.478 0.768

0.433 0.349 0.427 0.430 0.424 0.534 0.279 0.132 0.495 0.501 0.556 0.812 0.409 0.775 0.413 0.413 0.434 0.800

0.456 0.433 0.410 0.424 0.448 0.548 0.424 0.378 0.485 0.500 0.517 0.852 0.538 0.763 0.415 0.418 0.473 0.780

0.444 0.392 0.420 0.429 0.441 0.543 0.383 0.310 0.492 0.503 0.543 0.840 0.511 0.772 0.416 0.417 0.463 0.792

0.420 0.336 0.447 0.447 0.446 0.553 0.261 0.125 0.510 0.510 0.617 0.831 0.439 0.799 0.432 0.432 0.484 0.821

Note. This table shows the evaluations of downside risk gains offered by gold/the US dollar for emerging stock markets during the recent global commodity price slump. The total number of observations during the recent global commodity price slump is 656. VaR (p) Dec. is the averaged VaR decrease between the VaR of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate VaR decrease); ES (p) Dec. is the averaged ES decrease between the ES of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate ES decrease). The numbers written in bold are the highest and lowest reductions in VaR/ES with the US dollar as the safe haven asset and under the ERC strategy.

between emerging stock markets and gold prices/the US dollar rates. This study rotates some commonly used copulas by 270 degrees/90 degrees to examine how decreases in emerging stock markets co-move with increases (appreciations) in gold prices (US dollar values) under extreme circumstances. Considering that tails offer information about extreme risks at a limit, we also calculate the downside risk gains provided by gold/the US dollar at various confidence levels. The subsample and out-of-sample analyses are also done to ensure that the results are robust. We find that both gold and the US dollar can serve as safe haven assets for emerging stock markets; that the US dollar is superior to gold in most cases, while its superiority of hedging infinitely extreme risks is weakened in the subsample of the global financial crisis and in the out-of-sample analysis; and that the downside risk gains provided by the US dollar are the most attractive for China and Thailand, followed by India and Brazil, while offering relatively less attractive gains for Chile, South Africa, and Russia. The results imply that despite the prominent role of gold as a safe haven asset, for emerging countries, the potential of the US dollar for hedging against extreme negative shocks should be further explored, especially for China and Thailand. India and Brazil should also pay greater attention to the US dollar. Reverse market movements in extreme cases (with the confidence levels being infinitely close to one to capture the extreme cases as much as possible) are an important angle for detecting the safe haven property, and downside risk gains at various confidence levels should be another important dimension. In this way, direct and concrete portfolio implications are provided. The period of the global financial crisis and out-of-sample analysis should be specifically analyzed, as they may alter some of the results. Future studies can be extended to seek more potential assets as safe havens for emerging countries. As indicated by Liu et al. (2016), the econometric framework with multivariate non-normal distribution should be developed when examining alternative safe haven assets. Finally, various portfolio strategies for risk hedging, such as dynamic portfolio strategies, also merit consideration. Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

20

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

Table 10 Descriptive statistics of the low-high tail dependence in out-of-sample. Panel A: stock index - gold price

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

0.0116 0.0001 0.0018 0.0028 0.0128 0.0002 0.0012 0.0097 0.0004

130.8 94.5 165.6 75.8 98.1 55.2 113.3 278.3 130.0

0.0034 0.0000 0.0004 0.0014 0.0050 0.0001 0.0004 0.0011 0.0001

0.0055 0.0000 0.0010 0.0009 0.0047 0.0000 0.0005 0.0062 0.0002

0.0209 0.0001 0.0033 0.0066 0.0271 0.0011 0.0025 0.0151 0.0008

Panel B: Stock Index - US dollar

Brazil Chile Czech Russia South Africa China India Malaysia Thailand

Mean

t

Std

Min

Max

t (US dollar N gold)

0.0069 0.0234 0.0024 0.0013 0.0080 0.0000 0.0266 0.0252 0.0282

140.0

0.0019 0 0.0010 0.0004 0.0040 0.0000 0.0028 0.0029 0.0036

0.0036 0.0234 0.0005 0.0005 0.0010 0.0000 0.0195 0.0178 0.0184

0.0109 0.0234 0.0045 0.0023 0.0213 0.0000 0.0364 0.0308 0.0381

−55.5

89.5 125.2 76.4 46.9 363.1 265.1 298.2

23.3 −39.1 −30.6 −55.1 351.1 160.3 292.5

Jan 1, 2013 Jan 2, 2015 Brazil stock-dollar

Jan 1, 2011

.004 .002 0 stock-dollar

stock-gold

stock-gold

stock-dollar

Jan 1, 2011

stock-gold

stock-dollar

Jan 1, 2011

stock-dollar

Jan 1, 2013 Jan 2, 2015 China stock-gold

stock-dollar

.02

.02 0

Jan 1, 2013 Jan 2, 2015 India

Jan 1, 2013 Jan 2, 2015 Czech

0 .0004.0008 Jan 1, 2013 Jan 2, 2015 South Africa

.005.01.015.02.025.03

.04

stock-dollar

Jan 1, 2011

.04

Jan 1, 2013 Jan 2, 2015 Russia stock-gold

Jan 1, 2011

stock-gold

0

.004 0 Jan 1, 2011

Jan 1, 2013 Jan 2, 2015 Chile

.01 .02 .03

.008

stock-gold

Jan 1, 2011

0

Jan 1, 2011

0

0

.01

.01

.02

.02

Note. This table provides summary statistics for λU′ from the 270-degree rotated Student-t copula (i.e., the low (in emerging stocks)-high (in gold/the US dollar) tail dependence), including the mean, standard deviation, minimum, maximum of λU′ and t statistic for testing λU′ N 0; t (US dollar N gold) is the t statistic for the hypothesis that the λU′ of emerging stock-the US dollar pair is larger than that of emerging stock–gold pair.

Jan 1, 2013 Jan 2, 2015 Malaysia stock-gold

stock-dollar

Jan 1, 2011

Jan 1, 2013 Jan 2, 2015 Thailand stock-gold

stock-dollar

Fig. 5. The development of the low(in emerging stocks)–high(in gold/the US dollar) tail dependence in out-of-sample.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

21

Table 11 Downside risk gains in out-of-sample. VaR (1%) Dec

VaR (0.5%) Dec

VaR (0.1%) Dec

ES (1%) Dec

ES (0.5%) Dec

ES (0.1%) Dec

Panel A: stock index - gold price Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.416 0.417 0.084 0.154 0.378 0.378 0.451 0.451 0.303 0.303 0.283 0.296 0.380 0.384 −0.078 0.089 0.423 0.420

0.413 0.413 0.081 0.156 0.370 0.372 0.444 0.448 0.294 0.302 0.274 0.273 0.379 0.385 −0.082 0.072 0.422 0.417

0.380 0.366 −0.095 0.093 0.315 0.316 0.320 0.322 0.250 0.286 0.170 0.070 0.306 0.317 −0.376 0.079 0.399 0.385

0.402 0.398 0.028 0.136 0.356 0.357 0.411 0.413 0.281 0.297 0.248 0.223 0.359 0.367 −0.155 0.087 0.421 0.413

0.392 0.384 −0.017 0.120 0.340 0.342 0.383 0.386 0.267 0.292 0.223 0.173 0.343 0.353 −0.211 0.091 0.419 0.409

0.352 0.332 −0.204 0.039 0.299 0.297 0.316 0.305 0.224 0.272 0.160 0.063 0.288 0.309 −0.360 0.139 0.426 0.394

Panel B: stock index - US dollar Brazil EW ERC Chile EW ERC Czech EW ERC Russia EW ERC South Africa EW ERC China EW ERC India EW ERC Malaysia EW ERC Thailand EW ERC

0.568 0.613 0.403 0.473 0.504 0.563 0.572 0.593 0.482 0.492 0.492 0.886 0.599 0.712 0.434 0.432 0.590 0.792

0.568 0.604 0.408 0.473 0.502 0.567 0.577 0.596 0.486 0.494 0.507 0.871 0.610 0.711 0.424 0.412 0.596 0.801

0.574 0.566 0.422 0.466 0.483 0.580 0.481 0.528 0.497 0.498 0.542 0.833 0.569 0.673 0.449 0.439 0.566 0.832

0.572 0.595 0.414 0.473 0.500 0.573 0.549 0.577 0.490 0.496 0.522 0.866 0.597 0.703 0.441 0.431 0.592 0.812

0.574 0.583 0.420 0.472 0.496 0.579 0.530 0.563 0.495 0.498 0.538 0.856 0.591 0.696 0.449 0.437 0.591 0.824

0.587 0.575 0.437 0.474 0.501 0.596 0.432 0.502 0.506 0.502 0.595 0.828 0.547 0.681 0.495 0.488 0.598 0.850

Note. This table shows the out-of-sample evaluations of downside risk gains provided by gold/the US dollar for emerging stock markets. For China and Malaysia (other countries), the total number of observations in the out-of-the-sample is 917 (1438). VaR (p) Dec. is the averaged VaR decrease between the VaR of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate VaR decrease); ES (p) Dec. is the averaged ES decrease between the ES of the emerging stock and that of portfolio EW/ERC at the (1 − p) confidence level (positive values indicate ES decrease). The numbers written in bold are the highest and lowest reductions in VaR/ES with the US dollar as the safe haven asset and under the ERC strategy.

Acknowledgements The authors would like to thank the editor, Professor Jonathan Batten, and the two anonymous reviewers for their helpful comments. Financial support offered by the National Natural Science Foundation of China [no. 71601157] is gratefully acknowledged. Appendix A. The skewed-t distribution The density function of the skewed-t distribution is 8    −ηþ1=2 > 1 bz þ a 2 a > > ; zb− < bc 1 þ η−2 1−ϕ b skewed−t ðzjη; ϕÞ ¼   −ηþ1=2  > 1 bz þ a 2 a > > : bc 1 þ ; z ≥− η−2 1 þ ϕ b

ðA:1Þ

The values of a, b, and c are defined as. a ≡ 4ϕc

η−2 Γðη þ 1=2Þ 2 2 ; b ≡ 1 þ 3ϕ −a ; c ≡ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; η−1 πðη−2ÞΓðη=2Þ

ðA:2Þ

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006

22

X. Wen, H. Cheng / Emerging Markets Review xxx (2018) xxx–xxx

where η is the kurtosis parameter and ϕ is the asymmetry parameter. These are restricted to 2 b η b ∞ and −1 b ϕ b 1. Like the Student's t distribution, it is well defined only for η N 2; skewness exists for η N 3 and kurtosis exists only if η N 4. In the case of ϕ N 0 (ϕ b 0), the mode of the density is to the left (right) of 0 and the variable is skewed to the right (left). References Adams, Z., Glück, T., 2015. Financialization in commodity markets: a passing trend or the new normal? J. Bank. Financ. 60, 93–111. Akram, Q.F., 2009. Commodity prices, interest rates and the dollar. Energy Econ. 31, 838–851. Aloui, R., Aϊssa, M.S.B., Nguyen, D.K., 2013. Conditional dependence structure between oil prices and exchange rates: a copula-GARCH approach. J. Int. Money Financ. 32, 719–738. Anderson, R.M., Bianchi, S.W., Goldberg, L.R., 2012. Will my risk parity strategy outperform? Financ. Anal. J. 68, 75–93. Arouri, M.E.H., Lahiani, A., Nguyen, D.K., 2015. World gold prices and stock returns in China: insights for hedging and diversification strategies. Econ. Model. 44, 273–282. Avdulaj, K., Barunik, J., 2015. Are benefits from oil-stock diversification gone? New evidence from a dynamic copula and high frequency data? Energy Econ. 51, 31–44. Batten, J.A., Ciner, C., Lucey, B.M., 2013. On the Economic Determinants of the Gold-inflation Relation (Working Paper). Batten, J.A., Ciner, C., Lucey, B.M., 2015. Are Precious Metals Really a Homogenous Asset Class? (Working Paper) Baur, D.G., Lucey, B.M., 2010. Is gold a hedge or a safe haven? An analysis of stocks, bonds and gold. Financ. Rev. 45, 217–229. Baur, D.G., McDermott, T.K., 2010. Is gold a safe haven? International evidence. J. Bank. Financ. 34, 1886–1898. Baur, D.G., McDermott, T.K., 2016. Why is gold a safe haven? J. Behav. Exp. Financ. 10, 63–71. Beckmann, J., Berger, T., Czudaj, R., 2015. Does gold act as a hedge or a safe haven for stocks? A smooth transition approach. Econ. Model. 48, 16–24. Bekiros, S., Boubaker, S., Nguyen, D.K., Uddin, G.S., 2017. Black swan events and safe havens: the role of gold in globally integrated emerging markets. J. Int. Money Financ. 73, 317–334. Bessler, W., Wolff, D., 2015. Do commodities add value in multi-asset portfolios? An out-of-sample analysis for different investment strategies. J. Bank. Financ. 60, 1–20. Bredin, D., Conlon, T., Potì, V., 2015. Does gold glitter in the long-run? Gold as a hedge and safe haven across time and investment horizon. Int. Rev. Financ. Anal. 41, 320–328. Brooks, C., 2008. Introductory Econometrics for Finance. Cambridge University Press, New York. Caballero, R.J., Krishnamurthy, A., 2008. Collective risk management in a flight to quality episode. J. Financ. 63, 2195–2230. Christoffersen, P.F., 1998. Evaluating interval forecasts. Int. Econ. Rev. 39, 841–862. Ciner, C., Gurdgiev, C., Lucey, B.M., 2013. Hedges and safe havens: an examination of stocks, bonds, gold, oil and exchange rates. Int. Rev. Financ. Anal. 29, 202–211. Commodity Market Review from World Economic Outlook, October 2013. International Monetary Fund. Creal, D., Koopman, S.J., Lucas, A., 2013. Generalized autoregressive score models with applications. J. Appl. Econ. 28, 777–795. Daskalaki, C., Skiadopoulos, G.S., 2011. Should investors include commodities in their portfolios after all? New evidence. J. Bank. Financ. 35, 2606–2626. Delatte, A.L., Lopez, C., 2013. Commodity and equity markets: some stylized facts from a copula approach. J. Bank. Financ. 37, 5346–5356. Engle, R., 2002. Dynamic conditional correlation: a simple class of multi-variate generalized autoregressive conditional heteroskedasticity models. J. Bus. Econ. Stat. 20, 339–350. Erb, C.B., Harvey, C.R., 2006. The strategic and tactical value of commodity futures. Financ. Anal. J. 62, 69–97. Flavin, T.J., Morley, C.E., Panopoulou, E., 2014. Identifying safe haven assets for equity investors through an analysis of the stability of shock transmission. J. Int. Financ. Mark. Inst. Money 33, 137–154. Gorton, G.B., Rouwenhorst, G.K., 2006. Facts and fantasies about commodity futures. Financ. Anal. J. 62, 47–68. Hammoudeh, S., Nguyen, D.K., Sousa, R.M., 2015. US monetary policy and sectoral commodity prices. J. Int. Money Financ. 57, 61–85. Hansen, B.E., 1994. Autoregressive conditional density estimation. Int. Econ. Rev. 35, 705–730. Huberman, G., Jiang, W., 2006. Offering vs. choice in 401(k) plans: equity exposure and number of funds. J. Financ. 61, 763–801. Joe, H., Xu, J.J., 1996. The estimation method of inference functions for the margins for multivariate models. Technical Report 166. Department of Statistics, University of British Columbia. Liu, C.S., Chang, M.S., Wu, X., Chui, C.M., 2016. Hedges or safe havens-revisit the role of gold and USD against stock: a multivariate extended skew-t copula approach. Quant. Financ. https://doi.org/10.1080/14697688.2016.1176238. Lucey, B.M., Li, S., 2015. What precious metals act as safe haven, and when? Some US evidence. Appl. Econ. Lett. 22, 35–45. Maillard, S., Roncalli, T., Teïletche, J., 2010. The properties of equally weighted risk contribution portfolios. J. Portf. Manag. 36 (4), 60–67. Mensi, W., Hammoudeh, S., Reboredo, J.C., Nguyen, D.K., 2015. Are Sharia stocks, gold and U.S. Treasury hedges and/or safe havens for the oil-based GCC markets? Emerg. Mark. Rev. 24, 101–121. Nazlioglu, S., Soytas, U., 2012. Oil price, agricultural commodity prices, and the dollar: a panel cointegration and causality analysis. Energy Econ. 34, 1098–1104. Ning, C., 2010. Dependence structure between the equity market and the foreign exchange market – a copula approach. J. Int. Money Financ. 29, 743–759. Oh, D.H., Patton, A.J., 2017. Time-varying systemic risk: Evidence from a dynamic copula model of CDS spreads. J. Bus. Econ. Stat. 1, 1–15. Patton, A.J., 2006. Modeling asymmetric exchange rate dependence. Int. Econ. Rev. 47, 527–556. Patton, A.J., 2013. Copula methods for forecasting multivariate time series. Handbook of Economic Forecasting. Vol. 2. Elsevier North-Holland, Amsterdam (Chapter 16). Pukthuanthong, K., Roll, R., 2011. Gold and the dollar (and the euro, pound, and yen). J. Bank. Financ. 35, 2070–2083. Reboredo, J.C., 2013. Is gold a safe haven or a hedge for the US dollar? Implications for risk management. J. Bank. Financ. 37, 2665–2676. Reboredo, J.C., Ugolini, A., 2015. Systemic risk in European sovereign debt markets: a CoVaR-copula approach. J. Int. Money Financ. 51, 214–244. Reboredo, J.C., Rivera-Castro, M.A., Ugolini, A., 2016. Downside and upside risk spillovers between exchange rates and stock prices. J. Bank. Financ. 62, 76–96. Siburg, K.F., Stoimenov, P., Weiß, G.N.F., 2015. Forecasting portfolio value-at-risk with nonparametric lower tail dependence estimates. J. Bank. Financ. 54, 129–140. Tang, K., Xiong, W., 2012. Index investment and financialization of commodities. Financ. Anal. J. 68 (6), 54–74. Wang, Y., Wu, J.L., Lai, Y.H., 2013. A revisit to the dependence structure between the stock and foreign exchange markets: a dependence-switching copula approach. J. Bank. Financ. 37, 1706–1719. Wen, X., Wei, Y., Huang, D., 2012. Measuring contagion between energy market and stock market during financial crisis: a copula approach. Energy Econ. 34, 1435–1446.

Please cite this article as: Wen, X., Cheng, H., Which is the safe haven for emerging stock markets, gold or the US dollar?, Emerg. Mark. Rev. (2018), https://doi.org/10.1016/j.ememar.2017.12.006