Uncovering frequency domain causality between gold and the stock markets of China and India: Evidence from implied volatility indices

Uncovering frequency domain causality between gold and the stock markets of China and India: Evidence from implied volatility indices

Accepted Manuscript Uncovering frequency domain causality between gold and the stock markets of China and India: Evidence from implied volatility ind...

1MB Sizes 0 Downloads 99 Views

Accepted Manuscript

Uncovering frequency domain causality between gold and the stock markets of China and India: Evidence from implied volatility indices Elie Bouri , David Roubaud , Rania Jammazi , Ata Assaf PII: DOI: Reference:

S1544-6123(17)30002-8 10.1016/j.frl.2017.06.010 FRL 728

To appear in:

Finance Research Letters

Received date: Revised date: Accepted date:

2 January 2017 5 June 2017 7 June 2017

Please cite this article as: Elie Bouri , David Roubaud , Rania Jammazi , Ata Assaf , Uncovering frequency domain causality between gold and the stock markets of China and India: Evidence from implied volatility indices, Finance Research Letters (2017), doi: 10.1016/j.frl.2017.06.010

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Highlights

AC

CE

PT

ED

M

AN US

   

We use implied volatility indices and examine causality dynamics We examine dynamics between gold and the Chinese and Indian stock markets (2011-2016) We uncover some interesting predictability patterns that differ along the spectrum. We report a feedback effect between gold and China’s market in high frequencies Gold predicts India’s market in low frequencies We account for structural breaks and reveal more nuanced dynamics

CR IP T

 

1

ACCEPTED MANUSCRIPT Uncovering frequency domain causality between gold and the stock markets of China and India: Evidence from implied volatility indices Elie Bouri, Holy Spirit University of Kaslik, USEK, Beirut, Lebanon email: [email protected]

CR IP T

David Roubaud,

Montpellier Business School, Montpellier Research in Management, Montpellier, France. email: [email protected] Rania Jammazi

AN US

National schoold of computer science (ENSI), La manouba , Tunisia. email: [email protected] Ata Assaf

Email: [email protected]

ED

Abstract

M

Faculty of Business and Management, University of Balamand, Tripoli, Lebanon.

We use implied volatility indices and examine short-term and long-term causality dynamics

PT

between gold and the Chinese and Indian stock markets from March 2011 to March 2017. We

CE

uncover some interesting predictability patterns that differ along the spectrum. Importantly, we find significant bi-directional effects between gold and the Chinese and Indian stock

AC

markets in both high and low frequencies, suggesting that the safe-haven property of gold is not stable. Our results are robust in the out-of-sample forecasting exercises.

Keywords: Implied volatility; gold; Chinese equities; Indian equities; frequency domain causality JEL classification: C13, G15

2

ACCEPTED MANUSCRIPT 1. Introduction Gold is a strategic resource that is largely used in various national economic activities and in national security. Large fluctuations in the gold market lead to increased price volatility. The latter affects price stabilisation policies and imposes challenges on market participants (producers and consumers), who often try to predict the future price of gold.

CR IP T

Given the valuable role of gold as an investment, increased price volatility of gold also affects the diversification benefits of holding gold in a portfolio and undermines the design of an optimal hedging strategy.

Baur and Lucey (2010) indicate that, in times of market stress, during which equity

AN US

indices plunge, risk-averse investors move away from risky assets to less risky assets, such as gold and gold-related investments. As a result, the gold price increases, and its volatility increases also. Baur (2012) points toward this positive return–volatility in the gold market

M

and relates it to a safe-haven property. According to Baele et al. (2013), flight-to-safety episodes coincide with increases in the stock market volatility index (VIX). Sarwar (2017)

ED

provides evidence that the US VIX Granger causes the gold implied volatility, which is consistent with the flight-to-safety effect.

PT

In emerging markets, which attract global institutional investors, prior studies provide

CE

evidence of significant linkages between equity and gold prices (e.g., Wang et al., 2010, and Arouri et al., 2015, for China; Jain and Biswal, 2016, for India; Raza et al., 2016, for

AC

emerging markets), and their volatilities (Thuraisamy et al., 2013; Kumar, 2014; Arouri et al., 2015; Raza et al., 2016). Some other studies document hedge and/or safe-haven properties of gold for stock markets (see, among others, Arouri et al., 2015, for China; Gurgun and Unalmis, 2014, for emerging markets). Recently, the link between economic-policy uncertainty and the gold market has been examined, but mixed findings have been reported. For example, Balcilar et al. (2016) find evidence of causality running from economic-policy

3

ACCEPTED MANUSCRIPT uncertainty to the returns and volatility of gold prices, whereas Reboredo and Uddin (2016) use a quantile regression approach and show that neither the US VIX nor policy uncertainty has either a significant co-movement or a causal relation with metal (gold) commodity futures. Using a wavelet approach and a GARCH-based copula methodology, Bekiros et al. (2017) examine the hedging and diversification roles of gold for BRICS stock markets. They

CR IP T

report, among others, evidence of two-way causality linkages, suggesting that gold behaves more like equities.

Based on the above, we notice at least four research gaps. First, prior studies that consider the link between equity and gold markets focus on return and volatility linkages

AN US

using backward looking measures. However, market participants are not only concerned with past volatility but also with future volatility. Historical volatility measures, such as squared returns or GARCH-based volatility, are backward looking measures and, thus, are unlikely to

M

be a good forecast of future realized volatility. The superiority of the implied volatility measure over other measures is well documented, given that it is derived from option prices

ED

and reflects not only historical volatility information but also investors’ expectations for future market conditions (Luo et al., 2016; Maghyereh et al., 2016; Sarwar, 2017). Carr and

PT

Wu (2006) argue that implied volatility is economically more appealing, as it can predict

CE

movements in future realized volatility. Second, while Sarwar (2017) considers the relationships between the implied volatilities of the gold and US stock markets, the author

AC

disregards the relationships in large emerging markets. In this sense, it is not clear whether the flight-to-safety effect is also present in the emerging markets of China and India, which are among the primary forces driving dynamics in the world commodity (gold) market. India is the largest gold consumer, ahead of China, whereas the latter is the largest gold producer. According to 2015 figures from the World Gold Council (2016), China and India consumed 981.5 and 864.3 metric tonnes of gold, respectively. Additionally, China and India officially

4

ACCEPTED MANUSCRIPT hold 1,929.30 and 557.8 metric tonnes of gold, respectively. Furthermore, figures from the World Gold Council indicate a significant increase in the financialization of gold in the last 10 years, especially in China where investors bought more than 30 tonnes of gold in 2016 through the Exchange Traded Funds, an almost six-fold increase in assets under management. These backgrounds might suggest a feedback effect between gold and equity markets. In fact,

CR IP T

using historical volatility measures, Thuraisamy et al. (2013) report evidence of a volatility feedback effect between gold and Chinese equities, whereas Bekiros et al. (2017) indicate two-way causality linkages across frequencies between gold and BRICS stock markets. Third, in using a mean-based Granger causality approach, Sarwar (2017) neglects the

AN US

possibility that the causal relation might vary between the short-run and the long-run. Emerging evidence that causality in volatility varies across frequencies (Chaudhuri and Lo, 2016) implies the suitability of using the frequency domain approach of Breitung and

M

Candelon (2006), which allows causality to be analysed by frequency. Practically, investors often have different investment horizons; long-term fundamentalist traders, such as fund

ED

managers and institutional investors, are more concerned with causalities in low frequencies (i.e., permanent causal relations), whereas short-term speculators, such as noise traders, are

PT

more concerned with causalities in high frequencies (i.e., transitory causal relations).

CE

Recently, Batten et al. (2017) employed the frequency domain causality to examine the relationship between crude oil and natural gas markets. Using the same test, Huang et al.

AC

(2016) study oil, gold, and Chinese equities, but rely on return series. Bekiros et al. (2017) also rely on return data in examining the causal wavelet relation between gold and BRICS stock markets. Finally, we assess the reliability of our findings, using an out-of-sample analysis (Batten et al., 2017). Otherwise, the reliability of the causality estimates might be adversely affected when the underlying entire-sample data have structural breaks.

5

ACCEPTED MANUSCRIPT In view of this, the aim of this paper is to use implied volatility data within the framework of the frequency domain approach to uncover the causality by frequency between the international gold market and the stock markets of China and India, which is an unexplored research area. Our data cover a period after the global financial crisis (GFC), allowing us to examine whether gold retains a flight-to-safety property in the emerging

CR IP T

markets of China and India or, rather, acts as a risky asset (Bekiros et al., 2017).

Our main findings provide evidence of a bi-directional causality between the implied volatility of gold and that of equities in China and in India for some short- and long-run horizons. We also report a time variation in the gold-stock causal relationships. Our findings

AN US

suggest that investors and traders in Chinese and Indian markets adjust their equity and gold investments, with a close look at the volatility of both the gold and stock markets. The financialization of gold in the post-GFC period might have driven our results in these large

M

emerging markets, from which comes a large portion of the demand for gold (Bekiros, et al., 2017).

ED

Our findings provide support for policy-makers in China and India, who might consider gold-implied volatility as a tool to contain volatility in their local stock markets. In

PT

particular, uncovering the causality at different frequencies can help policy-makers design

CE

policies to address short- and long-term cycles. Also, traders and investors, whose goals differ across time horizons, can use the frequency-dependent causality results in their investment or

AC

trading strategies to target specific horizons that correspond to their preferences. 2. Causality in the domain Originally introduced by Granger (1969), and widely employed in the literature, the

Granger causality test is used to measure whether one variable x (y) occurs before another variable y (x) and helps predict it. Conventionally conducted within VAR models, the Granger causality considers a single statistical measure to explain predictability at all

6

ACCEPTED MANUSCRIPT frequencies (at an infinite time horizon). However, later studies suggest that the causal influence may change across frequencies (Geweke, 1982; Hosoya, 1991) but point toward the difficulty in estimating the frequency domain causality due to nonlinearities. Accordingly, Breitung and Candelon (2006) impose linear restrictions on the autoregressive parameters in a VAR model allowing for the estimation of the frequency domain approach to causality at

CR IP T

different frequency bands. As such, they differ between short, medium, and long run causality. Under a stationary VAR model, the relation between the implied volatilities of gold and equities in China and India is described as:

AN US

{ (1)

The approach of Breitung and Candelon (2006) has been applied in several recent studies [Huang et al. (2016), in their analysis of the return linkages across oil, gold, and

M

Chinese equities; Batten et al. (2017), in studying the linkages between crude oil and natural

ED

gas; and Bekiros et al. (2017), in examining the causal interactions between the prices of gold and BRICS stock markets]. Importantly, this has been discussed in detail in Ciner (2013).

PT

Therefore, we restrict this paragraph to the presentation of the primary conclusion from Breitung and Candelon (2006), within the context of our study, and indicate that the Granger

CE

causality from chinavix (indiavix) to goldvix at any frequency (ω) can be tested under the

AC

linear restriction given by: H 0 : R(ω )β =0, where β = | (

(2) |′ and )

7

ACCEPTED MANUSCRIPT To test the null hypothesis in the frequency interval ω ϵ (0, π), we follow Breitung and Candelon (2006) by relying on the F-statistics, which are approximately distributed as F(2, T–2p) under the null. 3. Empirical results 3.1 Data and preliminary statistics

CR IP T

The data used in this paper consist of the daily closing price of the implied volatility index of gold, Chinese equities, and Indian equities—denoted goldvix, chinavix, and indiavix, respectively. The Indian volatility index is maintained by the National Stock Exchange of India, and the remaining volatility indices are maintained by the CBOE. Data are collected

AN US

from DataStream for the period March 16, 2011 to March 16, 2017; the start of the period is dictated by data availability.

Table 1. Preliminary statistics chinavix

indiavix

18.610

27.147

18.650

39.950

63.420

37.700

11.120

16.560

11.560

4.443

7.071

4.742

Skewness

1.476

1.638

1.250

Kurtosis

5.883

6.253

4.227

1111.589*

1391.437*

506.042*

ADF

-4.976*

-4.762*

-4.637*

PP

-4.507*

-4.343*

-4.375*

Maximum Minimum

ED

Std. Dev.

M

Mean

goldvix

PT

Jarque–Bera

CE

Observations 1567 1567 1567 Notes: ADF (Augmented-Dickey–Fuller); PP (Phillips–Perron); * denotes statistical significance at 1% level.

AC

For the entire period, the summary statistics and unit root tests of the level series are presented in Table 1. Interestingly, goldvix has the lowest mean and standard deviation. All series are positively skewed and more peaked than a normal distribution. This finding is also confirmed by the non-normality test of Jarque–Bera. The results from the AugmentedDickey–Fuller and Phillips–Perron unit root tests show that all series are stationary at levels. Accordingly, we can use level series to conduct a cointegration analysis in line with Johansen, in addition to time and frequency domain Granger causalities. For the full sample

8

ACCEPTED MANUSCRIPT period, unreported results from a VAR cointegration analysis point toward the presence of two cointegration vectors for each pair of variables under study. This result suggests that the implied volatility of the international gold market is closely linked to the implied volatility of the Chinese (Indian) stock market. For the full sample period, we conduct the conventional time domain causality test within the VAR framework (see Table 2), where the number of

CR IP T

lagged variables was selected based on SIC criteria. The results show a unidirectional causality from chinavix to goldvix and no evidence of a feedback effect. However, we find that goldvix Granger causes indiavix, and we report evidence of a feedback effect at the 10%

AN US

significance level.

Table 2. VAR Granger causality test Sample Full sample

ED

M

Null hypothesis Gold-China goldvix ≠› chinavix chinavix ≠› goldvix Gold-India Full sample goldvix ≠› indiavix Indiavix ≠› goldvix Gold-India 1 goldvix ≠› indiavix Indiavix ≠›goldvix goldvix ≠› indiavix Gold-India 2 Indiavix ≠›goldvix goldvix ≠› indiavix Gold-India 3 Indiavix ≠›goldvix Notes: This table tests the null hypothesis of no Granger causality.

Chi-sq P values 1.006 0.315 4.998 0.025 12.782 0.000 3.279 0.070 19.742 0.000 0.638 0.408 0.127 0.721 0.157 0.691 0.152 0.696 4.726 0.029 of freedom are determined by

PT

SIC.

df 1 1 1 1 1 1 1 1 1 1 The degrees

However, it is well documented that the presence of structural breaks may affect the

CE

relationships between the examined variables. Using Bai and Perron’s1 (2003) sequential and

AC

repartition tests on the chinavix equation, which includes a constant and four lags each of chinavix and goldvix, we could not detect any break. As for the indiavix equation, which also comprises a constant and four lags each of indiavix and goldvix, the structural breaks were found at June 7, 2012 and 16 May, 2014. It is worth noting the intuitive economic meaning of these break dates. The first corresponds to extreme fear territory encountered by global 1

The detailed results for the break dates are available upon request. It is also worth noting that the choice of the lag length as four eliminates all correlation present in the residuals.

9

ACCEPTED MANUSCRIPT markets regarding solvency risks in Europe and slowing global growth; during that time of monetary stimulus in the US, the UK, and Europe, global equity prices decreased along with commodity prices, including gold. As for the second, it coincides with the Indian election, which triggered high levels of volatility following a historic win by the main opposition alliance.2 Table 2 also reports the results for the time domain causality between goldvix and

CR IP T

indiavix in each sub-period. Evidence reveals a unidirectional causality running from goldvix to indiavix in the first sub-sample at the 1% significance level and no evidence of any causality in the second sub-sample. As for the third sub-sample, we report a unidirectional causal effect from goldvix to indiavix at the 5% level.

AN US

3.2 Frequency domain analysis

Unlike conventional time domain causality, which describes the relation between variables by estimating a fixed coefficient that is constant at all frequencies, the spectral-

M

causality approach of Breitung and Candelon (2006) assumes that the sensitivity of a variable (i.e., goldvix) to transitory—high frequency—shocks in another variable (i.e., stock index

ED

implied volatility) is not equal to permanent—low frequency—shocks. This approach allows

PT

us to decompose the causality test statistic into different frequencies. Figures 1–2 provide the results for the full period. The frequency parameter (ω) on the horizontal axis is used to

CE

calculate the length of the period T measured in days, which corresponds to a cycle where T = 2π /ω. Interestingly, the results from the frequency domain causality test differ from those

AC

reported in Table 2 and offer a broader view on the direction and strength of causality in different frequencies that has not been presented elsewhere. This confirms the need for uncovering short and long run causality.

2

https://www.bloomberg.com/news/articles/2014-05-16/india-s-nifty-stock-index-futures-drop-before-electionresults

10

ACCEPTED MANUSCRIPT

Figure 1. Frequency domain causality between goldvix and chinavix—full sample period 8

CR IP T

goldvix ≠> chinavix

7

chinavix ≠> goldvix

6

5% critical value

5 4 3

AN US

2 1 0 0

0.5

1

1.5

2

2.5

3

M

Note: The frequencies (Omega) are on the x-axis, whereas the y-axis shows the F statistics testing the null hypothesis of no Granger-causality.

Figure 1 reveals that chinavix Granger causes goldvix over the short- and long-runs

ED

within (0.28, 1.32) and (2.15, 2.65) frequency bands, corresponding to wave lengths of five to 22 days and two to three days, respectively. This suggests that the implied volatility of the

PT

Chinese stock market predicts the implied volatility of the international gold market in both

CE

shorter and longer cycle lengths. However, the reciprocal causal effect is depicted from goldvix to chinavix within the lower and middle frequency bands of (0.44, 1.46), matching a

AC

wave length period of 13 to four days. This latest finding implies that goldvix has a short- to medium-term predictive power of the Chinese implied volatility. The above-mentioned results provide evidence on the existence of significant bidirectional causalities between the two volatility measures not detected through the conventional causality test. Furthermore, the frequency-domain causality from chinavix to goldvix is more noticeable than is the frequency-domain causality from goldvix to chinavix, especially in the short-run. This result suggests that short-term shocks in the Chinese stock market can spread quickly to the gold 11

ACCEPTED MANUSCRIPT market. As such, sharp declines in equity prices could lead to higher gold prices, to the extent that short-term stock traders, such as speculators, move part of their funds from equities to gold to reduce the risk of holding equities. In contrast, long-term fundamentalist traders, such as fund managers and institutional investors, in Chinese equities can design their investment strategies independently from the gold market volatility.

CR IP T

Figure 2. Frequency domain causality between goldvix and indiavix—full sample period 7

goldvix ≠> indiavix indiavix ≠> goldvix 5% critical value

6 5 4

AN US

3 2 1 0 0

0.5

1

1.5

2

2.5

3

M

Note: The frequencies (Omega) are on the x-axis, whereas the y-axis shows the F statistics testing the null hypothesis of no Granger-causality.

ED

A bi-directional causality is also reported for the case of India. Figure 2 indicates that indiavix Granger causes goldvix within the middle and long frequency intervals of (1, 1.25)

PT

and (2.16, 2.23), matching periods of five to six days and three days, respectively. Evidence

CE

of a reverse causality is also reported at the short/middle/long frequency ranges, suggesting a more persistent causality effect than the one detected from indiavix to goldvix. This result

AC

implies that both long-term investors and short-term traders should design their investment strategies, while having a closer look at the gold market volatility; whereas long-term investors in Indian equities should design their investment strategies by relying on the gold market volatility. This finding is not surprising, given that India is the largest gold consumer. Overall, our results for China and India are in line with Bekiros et al. (2017) who report a quite similar conclusion when they use return series in examining the time-scale

12

ACCEPTED MANUSCRIPT causality between gold and BRICS stock markets. Bekiros et al. (2017) argued that the financialization of commodity markets, including the gold market, during and after the GFC, has made investments in gold much easier and, thus, might also make the gold asset behave increasingly like stocks (Bekiros, 2017). This explanation is very relevant to our study, given that our sample covers the post-GFC period. It seems that Chinese and Indian investors adopt

CR IP T

a quite homogenous trend in rebalancing their portfolios and in switching between gold and equities. Our reported evidence on causal interactions corroborates evidence of the timevarying volatility transmission between gold and stock markets (Arouri et al., 2015), which may undermine the safe-haven property of gold.

AN US

3.3 Out-of-sample rolling causality analysis

To assess whether the bi-directional causality findings are also significant in out-ofsample forecasting exercises, we apply a rolling causality analysis in the frequency domain.

M

According to Batten et al. (2017), the out-of-sample forecasting exercises tend to be more robust to structural changes3 than is in-sample causality analysis. We consider a fixed

ED

window size of 250 observations4 and report results that are consistent with the bi-directional effect reported in the in-sample analysis, although they are richer and more nuanced. Figures

PT

3 and 4 depict the time-varying causality patterns between chinavix—goldvix, and indiavix—

CE

goldvix, respectively. Notably, short-tern causality tests are conducted at frequency 2.5, corresponding to a wave length of two to three days. Long-term causality tests are conducted

AC

at frequency 0.5, corresponding to a wave length of 12 to 13 days (Batten et al., 2017).

3

The results provided by the Bai and Perron (2003) test, which indicate the presence of structural breaks, legitimate the choice of the rolling causality method to conduct the out-of-sample evaluations in a more flexible way. 4 We also conducted a rolling causality analysis using a window size of 500 observations, and the overall results were quite the same. These results are available from the authors upon request.

13

ACCEPTED MANUSCRIPT

Figure 3a. Rolling-window frequency domain causality gold—China (chinavix ≠> goldvix) 14.00 12.00

CR IP T

10.00 8.00 6.00 4.00 2.00

AN US

0.00 29-02-2012 05-11-2012 13-07-2013 20-03-2014 25-11-2014 02-08-2015 08-04-2016 14-12-2016

Notes: The date is given on the x-axis, whereas the y-axis shows the F statistics testing the null hypothesis of no Granger-causality. The horizontal solid line indicates the 5% critical values. The dotted line represents the shortrun causality (ω=2.5), whereas the dashed line represents the long-run causality (ω=0.5).

M

As shown in Figure 3a, and regardless of the frequency, a significant unidirectional

ED

causality is detected from chinavix to goldvix. Indeed, the two (short and long) frequency curves commonly exhibit a dominant peak during 09/2012, which is much more pronounced

PT

in the short term (highest amplitude). Furthermore, and despite the fact that the short-term fluctuations depict another significant peak from July 2013 to February 2014, the long-run

CE

causality fluctuations are typically governed by three successive short-lived peaks (June

AC

2014, August 2015, and January 2017).

14

ACCEPTED MANUSCRIPT

Figure 3b. Rolling-window frequency domain causality gold—China (goldvix ≠> chinavix) 10.00 9.00 8.00 7.00

CR IP T

6.00 5.00 4.00 3.00 2.00 1.00

AN US

0.00 29-02-2012 05-11-2012 13-07-2013 20-03-2014 25-11-2014 02-08-2015 08-04-2016 14-12-2016

Notes: The date is given on the x-axis, whereas the y-axis shows the F statistics testing the null hypothesis of no Granger-causality. The horizontal solid line indicates the 5% critical values. The dotted line represents the shortrun causality (ω=2.5), whereas the dashed line represents the long-run causality (ω=0.5).

M

A careful look at the plots in Figure 3b validates the hypothesis of feedback effects,

ED

i.e. the existence of a significant bi-directional causality between chinavix and goldvix. In fact, the short-run and long-run curves follow a quite homogenous trend, showing at least

PT

three peaks (March 2012, July 2013, and February 2017).

Furthermore, the short-run

causality exhibits a peak in March and April 2013.

AC

CE

Figure 4a. Rolling-window frequency domain causality gold—India (indiavix ≠> goldvix)

15

ACCEPTED MANUSCRIPT

8.00 7.00 6.00 5.00 4.00 3.00

CR IP T

2.00 1.00

0.00 29-02-2012 05-11-2012 13-07-2013 20-03-2014 25-11-2014 02-08-2015 08-04-2016 14-12-2016

AN US

Notes: The date is given on the x-axis, whereas the y-axis shows the F statistics testing the null hypothesis of no Granger-causality. The horizontal solid line indicates the 5% critical values. The dotted line represents the shortrun causality (ω=2.5), whereas the dashed line represents the long-run causality (ω=0.5).

For the case of India, Figure 4a shows that indiavix can predict goldvix at particular periods across both the short- and long-run causalities. More precisely, the highest fluctuations in the causality curves are shown twice during November 2012 and July–August

M

2013. We also observed two long-run causality peaks in April–May 2014 and December

ED

2016.

PT

Figure 4b. Rolling-window frequency domain causality gold—India (goldvix ≠> indiavix)

10.00

AC

8.00

CE

12.00

6.00 4.00 2.00

0.00 29-02-2012 05-11-2012 13-07-2013 20-03-2014 25-11-2014 02-08-2015 08-04-2016 14-12-2016

16

ACCEPTED MANUSCRIPT Notes: The date is given on the x-axis, whereas the y-axis shows the F statistics testing the null hypothesis of no Granger-causality. The horizontal solid line indicates the 5% critical values. The dotted line represents the shortrun causality (ω=2.5), whereas the dashed line represents the long-run causality (ω=0.5).

As in the case of China, there is evidence of a reverse causality from goldvix to indiavix. Figure 4b indicates that the goldvix has a predictive power against indiavix. The causality magnitudes climb sharply four times, regardless of the frequencies. In particular,

CR IP T

they are marked by a major peak during July–October 2012, July–August 2013, May 2015– February 2016, and December 2016. However, the long-run causality continued for the first quarter of 2017.

Overall, the out-of-sample exercises give more insights in the time-variation in the

AN US

short- and long-runs’ causality and capture spikes not necessarily shown by Bai and Perron (2003) test of structural breaks. Importantly, we confirm the robustness of the bi-directional causality effects reported in the in-sample analysis, and argue that the causality from goldvix to stock market VIXs is characterized by higher amplitudes with longer duration causalities,

M

which is not necessarily the case in the opposite direction.

ED

Our analysis extended the findings of Srawar (2017) and highlighted the importance of using the spectral causality in studying the causal relation between the implied volatility of

PT

gold and equity markets. In particular, we reported evidence of a dynamical causal

CE

relationship that has experienced some breaks due to factors such as the heightened political and geopolitical risks in Europe, the US quantitative easing, and the uncertainty surrounding

AC

the Donald Trump election with the Federal Reserve raising interest rates in December. These factors resulted in some dynamic behaviour between the implied volatilities of gold and equity markets, where the market of equity (gold) has been playing a time-varying dynamics in its predictive power of the gold (equity) market volatility as investors shift their focus in and out of gold. 4. Conclusion

17

ACCEPTED MANUSCRIPT This study extends the prior literature on the interactions between gold and the stock markets of China and India, while shifting attention to the spectral causal interactions among implied volatility indices in the post-GFC period, during which the financialization of gold increased significantly. Our findings show that the implied volatility of stock markets in China and India Granger causes the implied volatility of gold at different frequencies, but

CR IP T

there is also evidence of a feedback effect, implying that gold may sometimes behave like stocks (Bekiros et al., 2017). While prior studies show that investors rebalance their portfolios toward less risky assets, such as gold with a close look at the US VIX fluctuations (Sarwar, 2017), our results suggest that, in China and India, investors and traders adjust their

AN US

equity (gold) investments, with a close look at gold (stock) implied volatility. References

Arouri, M.E.H., Lahiani, A. and Nguyen, D.K., 2015. World gold prices and stock returns in

M

China: insights for hedging and diversification strategies. Econ. Model. 44, 273-282.

ED

Baele, L., Bekaert, G., Inghelbrecht, K. and Min, W. 2013. Flights to safety (No. w19095). National Bureau of Economic Research.

PT

Balcilar, M., Gupta, R., Pierdzioch, C. 2016. Does uncertainty move the gold price? New evidence from a nonparametric causality-in-quantiles test. Resour. Policy. 49, 74–80.

CE

Bai, J., Perron, P., 2003. Computation and analysis of multiple structural change models. J.

AC

Appl. Econometr., 18(1), 1–22.doi:10.1002/jae.659. Batten, J.A., Ciner, C., Lucey, B.M., 2017. The dynamic linkages between crude oil and natural gas markets. Energ. Econ. 62, 150–170. doi:10.1016/j.eneco.2016.10.019.

Baur, D.G., Lucey, B.M., 2010. Is gold a hedge or a safe haven? An analysis of stocks, bonds and gold. Fin. Rev., 45(2), 217–229.doi:10.1111/j.1540-6288.2010.00244.x Baur, D.G. 2012. Asymmetric volatility in the gold market. Journal of Alternative Investments, 14 (4), 26–38.

18

ACCEPTED MANUSCRIPT 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 Fin. 73, 317–334. doi:10.1016/j.jimonfin.2017.02.010. Bouri, E., Jain, A., Biswal, P.C., Roubaud, D., 2017. Cointegration and nonlinear causality amongst gold, oil, and the Indian stock market: Evidence from implied volatility

CR IP T

indices. Resour. Policy 52, 201-206. doi:10.1016/j.resourpol.2017.03.003.

Breitung, J., Candelon, B., 2006. Testing for short- and long-run causality: a frequencydomain approach. J. Econ. 132, 363–378. doi:10.1016/j.jeconom.2005.02.004.

Chaudhuri, S.E., Lo, A.W., 2016. Spectral portfolio theory. http://alo.mit.edu/wp-

AN US

content/uploads/2016/06/spectral_14_standard.pdf.

Ciner, C., 2013. Oil and stock returns: Frequency domain evidence. J. Int. Finan. Markets, Inst. Money., 23, pp.1-11. doi:10.1016/j.intfin.2012.09.002

M

Geweke, J., 1982. Measurement of linear dependence and feedback between multiple time series. J. Am. Stat. Assoc. 77, 304–313. doi:10.1080/01621459.1982.10477803.

ED

Granger, C. W. J. 1969. Investigating causal relations by econometric models and cross-

PT

spectral methods. Econometrica, 37, 424−438. Gupta, R., Gil-Alana, L.A., Yaya, O.S., 2015. Do sunspot numbers cause global

CE

temperatures? Evidence from a frequency domain causality test. Appl. Econ. 47, 798– 808. doi:10.1080/00036846.2014.980575.

AC

Gurgun, G., Unalmis, I., 2014. Is gold a safe haven against equity market investment in emerging

and

developing

countries.

Financ.

Res.

Lett.

11,

341–

348.doi:10.1016/j.frl.2014.07.003

Hosoya, Y., 1991. The decomposition and measurement of the interdependency between second-order stationary processes. Probab. Th. Rel. Fields 88, 429–444. doi:10.1007/BF01192551.

19

ACCEPTED MANUSCRIPT Huang, S., An, H., Gao, X., Huang, X., 2016. Time–frequency featured co-movement between the stock and prices of crude oil and gold. Physica 444, 985–995. doi:10.1016/j.physa.2015.10.080. Jain, A., Biswal, P.C., 2016. Dynamic linkages among oil price, gold price, exchange rate, stock

market

in

India.

doi:10.1016/j.resourpol.2016.06.001.

Resour.

Policy

49,

179–185.

CR IP T

and

Kumar, D., 2014. Return and volatility transmission between gold and stock sectors: application of portfolio management and hedging effectiveness. IIMB Manage. Rev. 26, 5–16. doi:10.1016/j.iimb.2013.12.002.

AN US

Luo, X., Qin, S., Ye, Z., 2016. The information content of implied volatility and jumps in forecasting volatility: evidence from the Shanghai gold futures market. Fin. Res. Lett. 19, 105–111. doi:10.1016/j.frl.2016.06.012.

M

Maghyereh, A.I., Awartani, B., Bouri, E., 2016. The directional volatility connectedness between crude oil and equity markets: new evidence from implied volatility indexes.

ED

Energ. Econ. 57, 78–93. doi:10.1016/j.eneco.2016.04.010.

PT

Mensi, W., Hammoudeh, S., Reboredo, J.C., Nguyen, D.K., 2014. Do global factors impact BRICS stock markets? A quantile regression approach. Emerg. Markets Rev. 19, 1–

CE

17. doi:10.1016/j.ememar.2014.04.002. Raza, N., Shahzad, S. J. H., Tiwari, A. K., Shahbaz, M. 2016. Asymmetric impact of gold, oil

AC

prices and their volatilities on stock prices of emerging markets. Resour. Policy 49, 290–301.

Reboredo, J. C., Uddin, G. S. 2016. Do financial stress and policy uncertainty have an impact on the energy and metals markets? A quantile regression approach. Int. Rev. Econ. Fin. 43, 284–298.

20

ACCEPTED MANUSCRIPT Sarwar, G., 2017. Examining the flight-to-safety with the implied volatilities. Fin. Res. Lett. 20, 118–124. doi:10.1016/j.frl.2016.09.015. Thuraisamy, K.S., Sharma, S.S., Ali Ahmed, H.J., 2013. The relationship between Asian equity

and

commodity

futures

markets.

J.

Asian

Econ.

28,

67–75.

doi:10.1016/j.asieco.2013.04.003.

analysis for India. Econ. Bull. 34, 663–680.

CR IP T

Tiwari, A., Arouri, M., Teulon, F., 2014. Oil prices and trade balance: a frequency domain

Wang, M.L., Wang, C.P., Huang, T.Y., 2010. Relationships among oil price, gold price, exchange rate and international stock markets. Int. Res. J. Fin. Econ. 47, 82–91.

AN US

World Gold Council, 2016. Gold Demand Trends Q1 2016. http://www.gold.org/supply-anddemand/gold-demand-trends/back-issues/gold-demand-trends-q1-2016.(accessed

AC

CE

PT

ED

M

16.17.09).

21