- Email: [email protected]

PII:

S1059-0560(19)30075-9

DOI:

https://doi.org/10.1016/j.iref.2020.01.001

Reference:

REVECO 1889

To appear in:

International Review of Economics and Finance

Received Date: 22 January 2019 Revised Date:

15 October 2019

Accepted Date: 1 January 2020

Please cite this article as: Xie Z., Chen S.-W. & Wu A.-C., The foreign exchange and stock market nexus:New international evidence, International Review of Economics and Finance (2020), doi: https:// doi.org/10.1016/j.iref.2020.01.001. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. 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. © 2020 Published by Elsevier Inc.

The foreign exchange and stock market nexus: New international evidence Zixiong Xie Department of International Economics and Trade Institute of Resources, Environment, and Sustainable Development College of Economics Jinan University Guangzhou 510632, China E-mail: [email protected] Shyh-Wei Chen1 Department of International Business Tunghai University Taichung 40704, Taiwan, ROC E-mail: [email protected] An-Chi Wu Institute of Economics Academia Sinica Taipei 115, Taiwan, ROC E-mail: [email protected]

1 Corresponding

author. Department of International Business, Tunghai University. No.1727, Sec.4, Taiwan Boulevard, Xitun District, Taichung 40704, Taiwan, ROC. Tel: 886-4-23590121 ext 35310, Fax: 886-423592898 E-mail: [email protected] or [email protected]

Abstract This study attempts to re-examine the exchange rate-stock price nexus for a group of advanced and emerging countries. To this end, we employ the symmetric and asymmetric bootstrap panel Granger non-causality tests, which allows us to untangle the symmetric and asymmetric causations between exchange rates and stock prices under considerations of the cross-sectional dependence and asymmetry of the data at the same time. Among the main results, it is found that, first, the stock prices are helpful for predicting the exchange rates, but not vice versa. Second, the results show weak evidence in support of unidirectional asymmetric causality running from exchange rates to stock prices, and vice versa. Third, the empirical results of the Hatemi-J (2012) symmetric panel Granger non-causality tests show that there is a causation from stock prices to exchange rates, and vice versa. The hypotheses of the Hatemi-J asymmetric panel Granger noncausality from stock prices to exchange rates or from exchange rates to stock prices are not rejected, which are in line with the Emirmahmutoglu and Kose (2011) test. Fourth, when we adopt the raw data to conduct the Emirmahmutoglu and Kose (2011) test, the causal relations are in line with the results using the log of data. Moreover, by using the raw data, the causal relations predicted by using the Hatemi-J (2012) panel Granger non-causality test echo the causal relations predicted by using the Emirmahmutoglu and Kose (2011) test. Keywords: exchange rate, stock price, panel model, Granger non-causality JEL classification: G15, F31

1

Introduction

The causal link between exchange rates and stock prices has long been the focus of research and policy debate in economics. Ajayi and Mougou´e (1996, p. 195) point out that “an understanding of the nature of the intertemporal relation between stock prices and exchange rate movements should enhance the ability of multinational corporations to manage their foreign exchange exposure.” The knowledge of causal relations between the stock and currency markets, and consequently the degree of their integration, will potentially expand the information set available to international investors, multinational corporations and policy makers (Ajayi et al., 1998). Both markets also play crucial roles in the economic growth of an economy (Nieh and Lee, 2001). Theoretically, there are different models for predicting the directions of causation between exchange rates and stock prices.1 Empirically, there could be unidirectional causality running from the exchange rates to stock prices (cf. flow-oriented model) or from the stock prices to exchange rates (cf. stock-oriented model). If a market is subject to the influences of both approaches simultaneously, a feedback loop will prevail with an arbitrary sign of the correlation between the two variables. Therefore, the third and fourth possible outcomes cannot be overlooked, namely, that bidirectional causal (feedback) and neutral relationships (none of causation) exist between exchange rates and stock prices. A myriad of studies have devoted many efforts to dissect the nexus between exchange rates and stock prices for different developed countries and emerging economies. We classify them into two categories. First, many studies use conventional unit root test, cointegration, vector autoregressive model (VAR) and Granger (1969) non-causality test to examine the stock priceexchange rate linkages. For example, a substantial number of studies have examined the stock price-exchange rate linkages, especially in the developed countries (see, for example, Aggarwal, 1981; Soenen and Hennigar, 1988; Bahmani-Oskooe and Sohrabian, 1992; Ajayi and Mougou´e, 1996; Morley and Pentecost, 2000; Nieh and Lee, 2001; Ratanapakorn and Sharma, 2007; Chen and Chen, 2012; Ndako, 2013; Tsagkanos and Siriopoulos, 2013; Caporale, Hunter and Ali, 2014; 1 Readers

are referred to Section 2 of this paper for a brief review.

1

Inci and Lee, 2014). In addition, an equally large number of studies have investigated the stock price-exchange rate relationship for emerging markets (see, for example, Abdalla and Murinde, 1997; Wongbangpo and Sharma, 2002; Muhammad and Rasheed, 2002; Hatemi-J and Irandoust, 2002; Kim, 2003; Smyth and Nandha, 2003; Doong et al., 2005; Hatemi-J and Roca, 2005; Phylaktis ¨ u¨ and Demirci, 2012; and Ravazzolo, 2005; Pan et al., 2007; Yau and Nieh, 2009; Lin, 2012; Ulk Moore and Wang, 2014; Chkili and Nguyen, 2014; Fowowe, 2015; Sui and Sun, 2016; Yang, 2017; Tang and Yao, 2018). Motivated by the statistical power of the advances in panel unit root and panel cointegration tests (Maddala and Wu, 1999; Pedroni, 2000, 2004), Lean et al. (2011), Liang et al. (2013) and Wong (2017) examine the relationship between exchange rates and stock prices by using the panel data approach.2 Second, a growing body of research (see, for example, Yau and Nieh, 2009; Walid et al., 2011; Chen and Chen, 2012; Liu and Wan, 2012; Chkili and Nguyen, 2014; Ho and Huang, 2015; Bahmani-Oskooee and Saha, 2016a, 2016b; Salisu and Ndako, 2018) has turned its attention to the adoption of more sophisticated nonlinear models to examine the nexus between exchange rates and stock prices. This line of research contends that a potential problem with linear model is that if the relationship between exchange rates and stock prices are nonlinear, then conventional unitroot, cointegration and Granger non-causality tests suffer from a loss of power that may lead to bias conclusions. Some studies adopt the conditional volatility model (see, for example, Wong and Li, 2010; Zhao, 2010; Andreou et al., 2013; Moore and Wang, 2014; Caporale et al., 2014; Salisu and Oloko, 2015; Wong, 2017) to examine the volatility spillovers between foreign exchange and stock markets. In addition, Tsai (2012) and Yang et al. (2014) analyze the relationship between stock prices and exchange rates by using the quantile regression. Andrie et al. (2014) and Afshan 2 Lean

et al. (2011) examine the causal relationship between exchange rates and stock prices for eight emerging and

developed Asian markets by using the Westerlund (2006) panel Lagrange multiplier (LM) cointegration test. Instead, Liang et al. (2013) revisit the relationship between the equity market and currency market in ASEAN-5 using the panel cointegration test (Pedroni, 1999, 2004) and panel dynamic ordinary least squares (Kao and Chiang, 2000). Wong (2017) adopts Dumitrescu and Hurlin’s (2012) panel Granger non-causality in heterogeneous panel to test between the timevarying conditional variances of real exchange rate return and real stock price return for Malaysia, the Philippines, Singapore, Korea, Japan, the United Kingdom (UK) and Germany.

2

et al. (2018) examine exchange rate and stock price nexus using the wavelet analysis. An important feature of previous studies is that distinct results based on previous research are due to differences in methodology, approaches and samples and are subject to diverse interpretations, thus making it difficult to reach a corroborative position on the relationship between exchange rates and stock prices. Readers are referred to Wong (2017), Yang (2017) and Salisu and Ndako (2018) for a brief literature review and Bahmani-Oskooee and Saha (2015) for a thorough review. Bahmani-Oskooee and Saha (2015) conclude that any relation that exists between foreign exchange and stock markets is short run. In most studies, no long-run relationship was found between stock prices and exchange rates.3 The aim of this study is to revisit the causal relationship between the exchange rate and the stock market for 20 advanced economies and 6 emerging economies. To this end, we adopt the following approaches. First, we employ Emirmahmutoglu and Kose’s (2011) panel Granger non-causality test to avoid the pre-test bias and thereby mitigate the problem of using the firstdifference of the data. As already outlined, most previous studies adopt linear Granger noncausality test and cointegration approaches to examine the causal relationship between exchange rates and stock prices. However, it is well-known that the standard asymptotic theory is not applicable to hypothesis testing in the level VAR model if the variables are integrated or cointegrated. This is because the usual Wald test statistics for Granger non-causality based on the level VAR model are not only characterized by a non-standard asymptotic distribution, but depend on nuisance parameters in general if the variables are non-stationary (Toda and Phillips, 1993). Therefore, a pre-test is needed to determine the order of integration of variables before estimating the appropriate VAR model from which statistical inferences are derived. However, the Granger non-causality test may suffer from severe pre-test bias. The methodology proposed by Emirmahmutoglu and Kose (2011) can avoid this bias because it is an extension of Toda and Yamamoto’s (1995) test. Toda and Yamamoto (1995) recommend using a modified Wald (MWALD) test in a lag augmented vector autoregression (LA-VAR) which has a conventional asymptotic χ2 distribution 3 For

readers’ information, we summarize the recent contributions to this issue, which are not collected in Table 1 of

Bahmani-Oskooee and Saha (2015), in a table. This table is available from the authors upon request.

3

when VAR(p + dmax) is estimated, where p is the lag order and dmax is the maximal order of integration suspected to occur in the process. The only prior information needed for the LA-VAR approach is the maximum order of integration of the processes. In light of the fact that the pretests for a unit root and cointegrating rank (or taking differences in the data) are not required, the associated pre-test bias and size distortion can be avoided, at least asymptotically (Yamada and Toda, 1998). The simulation study by Emirmahmutoglu and Kose (2011) shows that their test has good power and reasonable size performances even if N and T are small. Second, this study also accounts for asymmetry in the exchange rate-stock price nexus in order to test whether the exchange rate responds asymmetrically to changes in stock price, and vice versa. To this end, based on the idea of Shin et al. (2014), we modify Emirmahmutoglu and Kose’s (2011) approach by incorporating the partial sum process of positive and negative changes in exchange rates and stock prices. The new specification enables us to examine asymmetric causality running from exchange rates to stock prices, and vice versa (please refer to Section 3.1 of this paper for details). To the best of authors’ knowledge, with the exception of Salisu and Ndako (2018), none of the previous studies in the literature have ever examined the causal relationship between exchange rates and stock prices in this way.4 The empirical results allow us to untangle the asymmetric causal relationship between exchange rates and stock prices and help us to discriminate between competing theories (e.g., flow-oriented model or stock-oriented model) for which the hypotheses are applicable to empirical data. As compared to the literature, the incremental contributions of this study are as follows. First, we apply a relatively novel testing methodology in this literature, and we extend the period of examination to a more recent sample period than previous investigations to capture most recent developments in the continued globalization of financial markets. This allows us to derive a stronger inference regarding the relationship between exchange rates and stock prices of, for example, the global financial crisis. Second, we modify the Emirmahmutoglu and Kose (2011) approach in order to extract asymmetric causations between exchange rates and stock prices. The empirical 4 Salisu

and Ndako (2018) employ symmetric and asymmetric panel autoregressive distributed lag (ARDL) model

to examine exchange rate-stock price nexus of 33 OECD countries.

4

results provide an alternative view and new evidence on the causal relationship between exchange rates and stock prices by taking account of the cross-sectional dependence and asymmetry of the data at the same time. Third, in order to know the effect of the global financial crisis (hereafter GFC) on the causal relationship between exchange rates and stock prices, we partition the data into three regimes: the pre-GFC (1998/01/01-2007/07/31), the GFC (2007/08/01-2010/12/31), and the post-GFC (2011/01/01-2019/05/20). Fourth, in order to verify whether the results of the Emirmahmutoglu and Kose (2011) panel Granger non-causality test are indifferent to, for example, using different samples or methods, we redo all of the estimations and tests (i) by dividing the data into the developed economies and emerging economies; (ii) by employing the Hatemi-J (2012) panel Granger non-causality test; and (iii) by using the raw data. The remainder of this paper is organized as follows. Section 2 reviews the theoretical underpinning of the relationship between exchange rates and stock prices. Section 3 briefly introduces the econometric methodology that we employ, and Section 4 describes the data and the empirical test results. Section 5 presents the conclusions that we draw from this research.

2

Theoretical underpinning

Traditionally, there are two competing theoretical perspectives on whether exchange rates Granger cause stock prices or vice-versa. They are the flow-oriented model and stock-oriented model. First, the flow-oriented model of exchange rate determination (Dornbusch and Fisher, 1980), also known as the goods market approach, affirm that currency movements affect international competitiveness and the balance of trade position, and consequently the real output of the country, which in turn affects current and future cash flows of companies and their stock prices. For example, a depreciation of the domestic currency makes the prices of locally produced goods cheaper relative to foreign goods, leading to an increase in exports. This has the effect of increasing the stock prices of such exporting firms. On the contrary, an appreciation of the domestic currency increases prices and this reduces foreign demand, thereby leading to a fall in value and stock prices. From this point of view, assuming direct quotation of exchange rates, there is a positive relationship between

5

exchange rates and stock prices, with causation running from exchange rates to stock prices.5 Conversely, changes in stock prices may also affect exchange rates through the stock-oriented model of exchange rates. In the case of the stock-oriented model, there are two versions, namely the monetary model and portfolio balance model. The monetary model supports a positive relationship and also argues that the relationship is a monetary phenomenon. Equities, being a part of wealth, may affect the behavior of exchange rates through the demand for money according to the monetarist models of exchange rate determination (Gavin, 1989). A rise in stock prices raises its rate of return thus making money less attractive as a store of value. Other things being equal, the fall in demand for domestic money relative to its supply increases the domestic price level, which via purchasing power parity (PPP) increases the exchange rate. This predicts a positive relationship between the two markets as higher stock prices lead to appreciation in exchange rate (Groenewold and Paterson, 2013).6 Similar links can be traced through the portfolio balance model as well (Branson, 1983; Frankel, 1983), which contends that investors are risk averse and they move their investments to countries with higher stock returns, which by implication leads to currency appreciation in countries with higher stock returns and depreciation in countries with lower stock returns. In other words, investors tend to diversify their investment portfolio from countries with lower stock returns to countries with higher stock returns, which, in effect, will lead to higher demand for the currencies of the countries with higher stock return at the expense of the countries with lower stock returns. Hence, countries with higher stock returns are more likely to experience exchange rate deprecia5 The

direct quotation of exchange rate defines the exchange rate as the price of one unit of foreign currency in

domestic currency, such that an increase of the exchange rate implies a depreciation of the domestic currency. 6 Based

L( R, Y ),

on the monetary model of exchange rate determination, suppose we have following three equations:

M∗ P∗

= L( R∗ , Y ∗ ) and E =

P P∗ .

M P

=

The first and second equations are the money market equilibrium conditions of

domestic and foreign countries, respectively. The third equation is the PPP condition. Terms M, P, R, Y are domestic money supply, the price level, the interest rate and the real output, respectively. Terms M∗ , P∗ , R∗ , Y ∗ are foreign money supply, the price level, the interest rate and the real output, respectively. After some manipulations, we have E =

P P∗

=

M L( R,Y ) M∗ L( R∗ ,Y ∗ )

. Therefore, other things being equal, the fall in demand for domestic money relative to its supply

increases the domestic price level, which via purchasing power parity, increases the exchange rate.

6

tion while countries with lower stock returns may be susceptible to exchange rate appreciation.7 Under the assumption of the portfolio balance model, stock prices are expected to lead exchange rates and to be negatively correlated to them. With the exceptions of the flow-oriented and stock-oriented models, an alternative model is called the portfolio rebalancing model, which is proposed by Hau and Rey (2004). This approach also suggests a positive relationship between the two markets, just like the monetary model. The proponents of this approach imagine an internationally-diversified investor who, following an increase in domestic stock prices, finds the portfolio is over-weighted in domestic stocks and so sells domestic and buys foreign equities, which puts pressure on the domestic currency to depreciate, i.e. for the exchange rate to rise (Groenewold and Paterson, 2013). Malliaropulos (1998) also proposed a theoretical model for describing the relationship between stock return differentials and real exchange rate changes. Based on his model, there is a negative and unidirectional causality running from stock return differentials to real exchange rate changes. Readers are referred to Malliaropulos (1998) for theoretical details and Wong and Li (2010) and Moore and Wang (2014) for empirical applications. Finally, the asset market models (Frenkel, 1976) suggest a weak or no association between stock prices and exchange rates. These models treat exchange rates as the price of an asset, the fundamental characteristic of which is that its present value is largely influenced by its expected rate of return. Since developments of stock prices and exchange rates may be driven by different factors, the asset market approach advocates the absence of any linkage between stock prices and exchange rates (Kollias, Mylonidis and Paleologou, 2012, p. 137). For readers’ information, we summarize the predicted causations between the exchange rates and stock prices of the five theoretical models in Table 1. 7 Remember

that we use the direct quotation of an exchange rate, which defines the exchange rate as the price of

one unit of foreign currency in domestic currency. Hence, an uptrend (a downtrend) of the exchange rate implies a depreciation (an appreciation) of the domestic currency.

7

3 3.1

Methodology Asymmetric panel Granger non-causality test

Emirmahmutoglu and Kose (2011) extend the LA-VAR approach via Meta analysis to test for Granger non-causality between variables in heterogeneous mixed panels. They consider the level VAR model with p + dmax lags in heterogeneous mixed panels: p+dmaxex

exi,t = ci +

∑

p+dmaxex

θi,m exi,t−m +

m =1 p+dmaxsp

spi,t = c˜i +

∑

ϕi,m spi,t−m + ε i,t ,

(1)

ϕ˜ i,m exi,t−m + ei,t ,

(2)

m =1

∑

p+dmaxsp

θ˜i,m spi,t−m +

m =1

∑

m =1

where i (= 1, . . . , N ) denotes the i-th nation and t(= 1, . . . , T ) denotes the time index. The variables ext and spt , respectively, denote the logarithm of effective real exchange rates and stock prices. p is the lag lengths and dmax is the maximal order of integration suspected to occur in the system. sp

Granger non-causality test running from sp to ex is to test for Ho : ϕi,1 = ϕi,2 = · · · = ϕi,p = 0, ∀i and that for the reverse direction running from ex to sp is to test for Hoex : ϕ˜ i,1 = ϕ˜ i,2 = · · · = ϕ˜ i,p = 0, ∀i. As emphasized in the Introduction, the possible asymmetric response of sp on ex and vice versa should be considered in the exchange rate-stock price nexus. Based on the idea of Shin et al. (2014), we decompose the movement of the ext into its negative (depreciation of domestic currency) and positive (appreciation of domestic currency) partial sum as: ext = ex0 + ext+ + ext− , where ext+ and ext− are the partial sum process of positive and negative changes in ext . More precisely: ext+ = ext− =

t

t

j =1

j =1

t

t

∑ ∆ex+j = ∑ max(∆ex j , 0), ∑ ∆ex−j =

j =1

∑ min(∆ex j , 0).

(3) (4)

j =1

Likewise, we decompose the movement of the spt into its negative (decrease in stock price) and − + − positive (increase in stock price) partial sum as: spt = sp0 + sp+ t + spt , where spt and spt are the

8

partial sum process of positive and negative changes in spt . More precisely: sp+ t = sp− t =

t

∑ ∆sp+j =

t

∑ max(∆sp j , 0),

j =1

j =1

t

t

∑ ∆sp−j =

j =1

(5)

∑ min(∆sp j , 0).

(6)

j =1

− + − The two variables sp+ t and spt replace lag order of spt in Eq. (1) and ext and ext replace lag order

of ext in Eq. (2), we have a new heterogeneous mixed panels as follows: p+dmaxex

exi,t = ci +

∑

p+dmaxex

∑

θi,m exi,t−m +

m =1

p+dmaxex

+

m =1

p+dmaxsp

spi,t = c˜i +

+ + ϕi,m spi,t −m

∑

m =1

− − ϕi,m spi,t −m + ε i,t ,

(7)

− − ϕ˜ i,m exi,t −m + ei,t ,

(8)

m =1

p+dmaxsp

θ˜i,m spi,t−m +

∑

∑

p+dmaxsp

+ + ϕ˜ i,m exi,t −m +

m =1

∑

m =1

Once Eqs. (7)–(8) are estimated, we can then judge whether exchange rate (stock price) changes have symmetric or asymmetric effects. If the two partial sums carry the same coefficient in sign and size, the effects are symmetric. Otherwise, they are asymmetric. To test for asymmetric Granger non-causality in this system, considering the following four sp−

sp+

+ + + : ϕi,1 = ϕi,2 = · · · = ϕi,p = 0, ∀i; Ho

hypotheses: Ho

− − − : ϕi,1 = ϕi,2 = · · · = ϕi,p = 0, ∀i;

−

+

+ + + − − − Hoex : ϕ˜ i,1 = ϕ˜ i,2 = · · · = ϕ˜ i,p = 0, ∀i; and Hoex : ϕ˜ i,1 = ϕ˜ i,2 = · · · = ϕ˜ i,p = 0, ∀i. (i) There is one-

way asymmetric Granger causality running from sp+ t (increase in stock prices) to ext (exchange sp+

rates) if H0

−

+

does not hold, but H0ex and H0ex hold. There is one-way asymmetric Granger sp−

causality running from sp− t (decrease in stock prices) to ext (exchange rates) if H0

does not hold,

−

+

but H0ex and H0ex hold. (ii) There is one-way asymmetric Granger causality running from ext+ sp+

+

(appreciation of domestic currency) to spt (stock prices) if H0ex does not hold, but H0

sp−

and H0

hold. There is one-way asymmetric Granger causality running from ext− (depreciation of domestic sp+

−

currency) to spt (stock prices) if H0ex does not hold, but H0

sp−

and H0

hold. (iii) There is two+

−

way Granger causality between nominal exchange rates and stock prices if neither H0ex (H0ex ) sp+

nor H0

sp−

(H0 ) hold. (iv) There is Granger non-causality between nominal exchange rates and +

−

sp+

stock prices if H0ex , H0ex , H0

sp−

and H0

hold.

Emirmahmutoglu and Kose (2011) use the Fisher test statistic proposed by Fisher (1932) in order to test the Granger non-causality hypothesis in heterogeneous panels. Fisher (1932) considered combining several significant levels (p-values) of identical but independent tests. If the test 9

statistics are continuous, the p-values pi (i = 1, ..., N) are independent uniform (0,1) variables. In this case, the Fisher test statistic (λ) is written as follows: N

λ = −2 ∑ ln( pi ),

(9)

i =1

where pi is the p-value corresponding to the Wald statistic of the i-th individual cross-section. This test statistic has a χ2 distribution with 2N degrees of freedom. The test is valid only if N is fixed as T → ∞. However, the limit distribution of the Fisher test statistic is no longer valid in the presence of cross correlations among the cross-sectional units. As a way to deal with such inferential difficulty in panels with cross correlations, Emirmahmutoglu and Kose (2011) use the bootstrap methodology in their Granger non-causality test for cross-sectional dependent panels. In order to accommodate the contemporaneous correlation in panels, we obtain the empirical distribution of the test statistic by using the bootstrap method. We consider the level VAR model with p + dmax lags in heterogeneous mixed panels as shown in Eqs (1) and (2).8

3.2

Tests for cross-sectional dependence

One important issue to be considered in a panel data analysis is testing for cross-sectional dependency across series. For this purpose, we adopt three well-known statistics in the literature to test for cross-sectional dependence. First, we utilize the following Lagrange multiplier statistic for cross-sectional dependence (hereafter, CDBP ) developed by Breusch and Pagan (1980). N −1

CDBP = T

N

∑ ∑

ρˆ 2ij ,

(10)

i =1 j = i +1

where ρˆ ij is the estimated correlation coefficient among the residuals obtained from individual OLS estimations. Under the null hypothesis of no cross-sectional dependence with a fixed N and T → ∞, CDBP is asymptotically distributed as chi-squared with N ( N − 1)/2 degrees of freedom. Pesaran (2004) indicates that the CDBP test has a drawback when N is large, implying that it is not applicable when N → ∞. To overcome this problem, the following Lagrange multiplier 8 About

the steps for conducting the bootstrapping panel Granger non-causality test, readers are referred to

Emirmahmutoglu and Kose (2011) for details.

10

statistic for the cross-sectional dependence (hereafter, CDlm ) developed by Pesaran (2004) can be used. s CDlm =

1 N ( N − 1)

N −1

N

∑ ∑

( T ρˆ2ij − 1).

(11)

i =1 j = i +1

Under the null hypothesis of no cross-sectional dependence with first T → ∞ and then N → ∞, this test statistic is asymptotically distributed as standard normal. However, this test is likely to exhibit substantial size distortions when N is large relative to T. A new test for the cross-sectional dependence (hereafter, CDP ) of Pesaran (2004) can be used where N is large and T is small. The CDP statistic is calculated as follows: s CDP =

2T N ( N − 1)

N −1

!

N

∑ ∑

ρˆ ij

.

(12)

i =1 j = i +1

Under the null hypothesis of no cross-sectional dependence with T → ∞ and N → ∞ in any order, the CDP test is asymptotically distributed as standard normal.

3.3

Tests for slope homogeneity

Determining whether slope coefficients are homogeneous or heterogeneous is also important in a panel non-causality analysis to impose causality restrictions on estimated coefficients. As noted by Granger (2003), imposing the joint restriction for the whole panel is a very strong null hypothesis. Moreover, Breitung (2005) also pointed out that the homogeneity assumption for the parameters is unable to capture heterogeneity due to nation-specific characteristics. By letting the parameter vector b = β of equation spt = α + βext + ε t , the null hypothesis of slope homogeneity (H0 : bi = b, for all i) is tested against the alternative hypothesis of heterogeneity (H1 : bi 6= b) for a non-zero fraction of pair-wise slopes for i 6= j. The standard Wald test is widely used in testing for the null hypothesis of slope homogeneity. However, as pointed out by Pesaran et al. (2008), the test based on the Wald principle is applicable only for cases in which the cross-sectional dimension (N) is relatively small, the time dimension (T) of the panel is large, the explanatory variables are strictly exogenous, and the error variances are homoskedastic. To overcome these problems, Pesaran et al. (2008) developed a standardized 11

version of Swamy’s test (Swamy, 1970) for testing slope homogeneity in large panels. Swamy’s test is valid when ( N, T ) → ∞ without any restrictions on the relative expansion rates of N and T when the error terms are normally distributed. The Swamy test for slope homogeneity is: S˜ =

N

∑

bˆ i − bˆ WFE

i =1

0 x0 M x i τ i ˆ ˆ WFE , b − b i σˆ i2

(13)

where bˆ i is the pooled OLS estimator; bˆ WFE is the weighted fixed effect pooled estimator; Mτ = IT − Zi ( Zi0 Zi )−1 Zi0 and Zi = (τT , xi ), where τT is a T × 1 vector of ones; xi = ext or spt , and σˆ i2 is the estimator of the error variance σi2 . In the case where N is fixed and T → ∞, the S˜ test has an asymptotic chi-squared distribution with k ( N − 1) degrees of freedom where k is the number of explanatory variables.9 The standardized dispersion statistic is as follows: ˜ = ∆

√

N

N −1 S˜ − k √ 2k

.

Under the null hypothesis with the condition of ( N, T ) → ∞, as long as

(14)

√

N/T → ∞ and the error

˜ test has an asymptotic standard normal distribution. The ∆ ˜ terms are normally distributed, the ∆ test can be improved for small samples by using the following bias-adjusted version: ! −1 S˜ − k √ N ˜ adj = N p ∆ . 2k( T − k − 1)/T + 1

4

(15)

Data and results

4.1

Data descriptions

We select 20 developed economies (Australia, Belgium, Canada, Denmark, France, Germany, Greece, Hong Kong, Iceland, Ireland, Italy, Japan, the Netherlands, Norway, Portugal, South Korea, Spain, Taiwan, the United Kingdom (UK) and the United States (US)) and 6 emerging economies (China, India, Indonesia, Malaysia, Philippines, Thailand) with high trade openness or cross-border financial flows, for a total of 26 economies in this study. All of the economies had 9 In

order to save space, we refer to Pesaran et al. (2008) for the details of Swamy’s test and the estimators described

in Eq. (13).

12

listed in WTO membership and MSCI’s market share. Currency linkage tends to affect clusters of countries linked together by international trade flows since the trade linkages are measured through countries’ exports to the common third market or through their bilateral trade. Further, changes in a foreign country’s capital flow can influence home country’s exchange rates though terms of trade. We summarize the trade-to-GDP ratio and capital flows-to-GDP ratio for the period from 1998-2017 of these economies in Table 2 for reference. We use a mixture of developed and emerging markets in this study because, as suggested by Lean et al. (2011, p. 256), “causality may be stronger in countries with developed foreign exchange and stock markets than emerging economies.” In addition, they mentioned using a mixture of developed and emerging economies not only increases the sample size and power of the test, but also allows for heterogeneity among markets. We download the nominal effective exchange rates and the stock prices (stock market benchmark indices) of each economy (please see Table 3 for details) from the Datastream.10 The sample periods are dependent upon the availability of data. For all of the markets the data are daily from the starting date of 1998/01/01 and they all ended on 2019/05/20. All variables are transformed into the logarithm to be implement in the empirical study. As a preliminary analysis, we provide a summary of descriptive statistics of the changes in the log of exchange rate and stock price and categorize them into the full-sample (1998/01/012019/05/20), the pre-GFC (1998/01/01-2007/07/31), the GFC (2007/08/01-2010/12/31), and the post-GFC (2011/01/01-2019/05/20), respectively. The results are given in Tables 4–6, which details the first four moments of each series and presents tests for normality and serial correlation. The results show a clear phenomenon that the average stock return is positive in most countries during the full-sample periods, except for the European countries suffering from their debt crisis, i.e., Greece, Italy, and Portugal (see Table 4(A)). As for the sub-sample periods, on average, during the pre-GFC periods all countries enjoy positive stock returns (see Table 5(A)); but during the GFC periods, only emerging countries have positive stock returns (see Table 5(B)); during post GFC periods, most countries turn out to have positive stock returns except for countries suffering 10 Hatemi-J

and Irandoust (2002) and Bahmani-Oskooee and Saha (2016a; 2016b) also use the nominal effective ex-

change rates to examine the causal relationship with stock prices.

13

from the European debt crisis (see Table 5(C)). The coefficients of skewness of the most of the stock prices of the full-sample and three subsamples are negative, implying that the stock prices of these economies are flatter to the left compared to the normal distribution. The coefficients of excess kurtosis of stock prices are greater than zero, indicating that the empirical distributions of these samples have fat tails. The coefficients of skewness and excess kurtosis reveal non-normality in the data. This is confirmed by the Jarque-Bera normality test as shown in Tables 4–6. The Ljung-Box Q-statistics, LB(24), denote significant autocorrelations of stock prices and exchange rates for all of the economies. We also performed a standard ARCH test for the exchange rate and stock price. The test results indicate that a significant ARCH effect exists for all of the series. The steps for conducting Emirmahmutoglu and Kose’s (2011) panel Granger non-causality analysis is as follows. First of all, we start by testing for the integrated properties of the series by using a variety of panel unit root tests. Next, as highlighted by Bai and Kao (2006), testing for the cross-sectional dependence in a panel non-causality study is crucial for selecting the appropriate estimator. Therefore, in the second step, we test for the cross-sectional dependence of the data and hypothesis of slope homogeneity. Finally, we estimate the LA-VAR model (cf. Eqs. (1) and (2) for symmetric case and Eqs. (7) and (8) for asymmetric case) and conduct the panel Granger noncausality tests with bootstrap critical values from stock prices to exchange rates in Eq. (1) or (7) and from exchange rates to stock prices in Eq. (2) or (8).

4.2

Results of the panel unit root tests

The first step in conducting the Emirmahmutoglu and Kose (2011) test is to investigate the integrated properties of the series for all countries. We employ the panel unit root tests of Im, Pesaran, and Shin (2003) (IPS hereafter) and Maddala and Wu (1999) for detecting the degree of integration of empirical variables in this paper. These tests enable us to recognize that there may be a mixture of stationary and nonstationary processes in the panel under the alternative hypothesis. We present the test results in Table 7. With the exception of ex + , it is found that the null hypothesis of a unit root cannot be rejected at the 5% significance level for the level data of all empirical series

14

based on the IPS test. However, the results of the Maddala and Wu (1999) PP-Fisher χ2 test show that the null hypothesis of a unit root is rejected for the level data of all empirical series at the 5% significance level. Moreover, for the first-difference of the data, the null hypothesis of a unit root must be rejected at the 5% significance level for all series. We do not get a unanimous conclusion from the IPS and PP-Fisher tests. Im et al. (2003) point out that due to the heterogeneous nature of the alternative hypothesis in their test, caution has to be exercised when interpreting results because the null hypothesis of a unit root in each cross section may be rejected when only a fraction of the series in the panel is stationary. An additional concern here is that the presence of cross-sectional dependencies can undermine the asymptotic normality of the IPS test and lead to over-rejection of the null hypothesis of joint non-stationarity (O’Connell, 1998). Third, one of the shortcomings of the panel unit root tests conducted above is that “it does not provide any information about the sources of nonstationarity of the series in question. Such information is required to distinguish between common factors and idiosyncratic components of the series and also to establish whether non-stationarity stems from international or national stochastic trends (Salisu and Ndako, 2018, p. 111).” In order to verify whether the non-stationarity is pervasive, variable-specific or both, we employ the PANICCA – the Panel Analysis of Non-stationarity in Idiosyncratic and Common components (PANIC) with Cross-section Average (CA) test developed by Reese and Westerlund (2016). The PANICCA test is a combined approach of the PANIC test by Bai and Ng (2004, 2010) and the CA augmentation approach of Pesaran (2007) and Pesaran et al. (2013) that exploits the strengths of both tests and overcomes their inherent weaknesses.11 Consider the panel data variable Yi,t , observable for t = 1, ..., T time periods and i = 1, ..., N cross-section units.12 The basic idea in PANIC is to first transform Yi,t by taking first differences. The method of principal components (PC) is then applied to estimate the first-differenced common and idiosyncratic components, which can be cumulated up to levels. The fact that the components 11 We

thank an anonymous referee for raising the PANIC with us.

12 The

exposition of the PANICCA draws heavily from Reese and Westerlund (2016). Readers are referred to Reese

and Westerlund (2016) for technical details.

15

are estimated from a regression in first differences means that the spurious regression problem is avoided, thereby enabling standard normal inference. However, the use of PC can render PANIC small-sample distorted, especially when N is small (see, for example, Gengenbach et al., 2006, 2010; Pesaran et al., 2013; Westerlund and Larsson, 2009; Westerlund and Urbain, 2015). Reese and Westerlund (2016) suggest using the cross-section average to improve the above problem. The idea underlying the cross-section average augmentation approach, originally put forth by Pesaran (2007) in the context of factor-augmented panel regressions, is to use the cross-section average Y¯t of Yi,t as a proxy for the common component of the data, which is then included in the regression as an additional regressor. The simulation results of Reese and Westerlund (2016) show that the PANICCA outperforms PANIC. The data-generating process (DGP) of Yi,t is assumed to be given the following common factor model: Yi,t = αi0 Dt,p + λi0 Ft + ei,t ,

(16)

where ei,t is a scalar idiosyncratic error, Ft is an r × 1 vector of common factors with λi being the associated (r × 1) vector of loading coefficients, and Dt,p = (1, ..., t p )0 is a ( p + 1) × 1 vector of trends, for which we consider two specifications: (i) a constant ( p = 0) and (ii) a constant and trend ( p = 1). In this paper, Yi,t is considered as the exchange rate or stock price. However, Reese and Westerlund (2016) allow for the presence of an m × 1 vector of additional variables, henceforth denoted Xi,t , whose DGP is given by Xi,t = β0i Dt,p + Λi0 Ft + ui,t .

(17)

0 )0 . In view of equations (16) and (17), the DGP of this variable is easily seen Define Zi,t = (Yi,t , Xi,t

to be given by Zi,t = Bi0 Dt,p + Ci0 Ft + Vi,t .

(18)

0 )0 . Note that the dimention of C is r × ( m + 1). where Bi = (αi , β i ), Ci = (λi , Λi ) and Vi,t = (ei,t , ui,t i

Bai and Ng (2004, 2010) and Reese and Westerlund (2016) proposed three statastics for testig the null hypothesis of a unit root for the idiosyncratic components. They are Pa,p , Pb,p and PMSB p , 16

p = 0 or 1. All three statistics are left-tailed. The appropriate 5% critical value is therefore given by −1.645. In testing for unit root of the common component, Ft , the test to apply is determined by the number of common factors. If only one common factor is to be estimated, then, testing can be carried out using any existing unit root test such as the augmented Dickey-Fuller (ADF) test, which is suggested in Bai and Ng (2004, 2010). However, if at least two factors are estimated, the sequential procedure of Bai and Ng (2004) is applied. The test statistics of this sequential procedure are based on modified versions of the Q f and Qc statistics (denoted by MQ f and MQc , respectively) originally proposed by Stock and Watson (1988). A series is non-stationary if at least one of these two components is non-stationary. If Ft is non-stationary but ei,t is stationary, then Yi,t will be non-stationary due to a pervasive source. If the opposite is verified, i.e., ei,t it is non-stationary but Ft is stationary, then the non-stationarity of Yi,t is due to a series-specific factor that cannot be endorsed in common grounds. The results of the PANICCA test are reported in Table 8. We consider the case of p = 1 because the log of the exchange rates and stock prices exhibit trend. For the level data of ex + and ex − , based on the ADF statistic, the null hypothesis of a unit root of the common factor is significantly rejected at the 5% level, indicating that the common factor of ex + and ex − is a stationary process. The idiosyncratic components of ex + and ex − are non-stationary processes because the three statistics, i.e., Pa,p , Pb,p and PMSB p , are not significant at the 5% level. In the cases of ex, sp, sp+ and sp− , the results of the ADF, Pa,p , Pb,p and PMSB p statistics suggest that both the common factor and idiosyncratic components are non-stationary processes. For the first-difference data of ex, sp, ex + and sp+ , the null hypothesis of a unit root is rejected for both the common factor and idiosyncratic components, implying these variables are I (1) processes. On the other hand, ex − and sp− are I (2) processes. Remember that dmax is the maximal order of integration suspected to occur in the process. The only prior information needed for the LA-VAR approach is the maximum order of integration of the processes. Based on the results of PANICCA, we use dmax = 1 for equations (1) and (2), and dmax = 2 for equations (7) and (8).

17

4.3

Results of the panel Granger non-causality test

Following, for example, Ajayi and Mougou´e (1996), Ajayi et al. (1998), Granger et al. (2000), Lean et al. (2011), Liang et al. (2013) and Wong (2017), we estimate bivariate panel models between exchange rates and stock prices.13 The lag order for the bivariate panel model is determined by the Akaike information criterion (AIC). One important issue to be considered in a panel data analysis is testing for the cross-sectional dependency across nations. Following Kar et al. (2011), we carry out three different tests, i.e., CDBP , CDlm and CDP , to investigate the existence of the cross-sectional dependence. We summarize the results in the left panel of Table 9. It is clear that the null hypothesis of no cross-sectional dependence across the members of the panel is strongly rejected at the 1% significance level of the bivariate panel model, indicating that the bootstrap critical value is required in conducting the panel Granger non-causality test. This finding implies that a shock occurring in one country seems to be transmitted to other countries. Hence, there seems to be evidence of cross-sectional dependence in the data. Table 9 also reports the results of the slope homogeneity test for the bivariate panel model. Both ˜ and ∆ ˜ adj ) reject the null hypothesis of the slope homogeneity at the one percent significance tests (∆ level, supporting the view that the parameters are heterogeneous. This finding simply implies that the panel non-causality analysis by imposing the homogeneity restriction on the variable of interest results in misleading inferences. In this respect, the panel non-causality analysis based on estimating a panel vector autoregression and/or panel vector error correction model by means of the generalized method of moments and the pooled ordinary least squares estimator is not an appropriate approach in detecting causal linkages between exchange rates and stock prices. In order to account for this we use the bootstrapped p-values in conducting the panel Granger 13 Bahmani-Oskooee

and Saha (2015) classify previous empirical studies into two groups. The first includes studies

that rely upon a bivariate model between the stock price index and the exchange rate. The second includes studies that use multivariate models where some additional variables are included in the model. The difference between the bivariate specification and multivariate specification when discussing ’causality’ lies in the direct and indirect causal relations. This paper seeks to investigate the direct linear causal relations between exchange rates and stock prices and, therefore, in line with previous research, we focus only on a specification which consists of two variables.

18

non-causality test. The final step is to perform the LA-VAR approach using mixed panels to test the hypothesis that there is a relationship between exchange rates and stock prices using the level data. We perform the Granger non-causality tests based on the bivariate panel model. An advantage of the LA-VAR procedure is that it is applied irrespective of whether the variables are I (1) or I (0) process. Hence, all variables are entered in level data rather than in first-difference data. The results of the LA-VAR approach are given in the left panel of Table 10. In each case, we perform tests for the null hypothesis of Granger non-causality running from exchange rates to stock prices (Ho : ex 6⇒ sp), and the null hypothesis of Granger non-causality running from stock prices to exchange rates (Ho : sp 6⇒ ex). From Table 10, based on the individual test, in the cases of Australia, Hong Kong, Japan, South Korea, Malaysia, Norway, the Philippines and Portugal, the null hypothesis of the Granger noncausality running from exchange rates to stock prices (Ho : ex 6⇒ sp) is rejected at the 5% significance level. Hence, the empirical results provide strong evidences to show that exchange rates are helpful to predict the future behaviors of stock prices. Table 10 also shows the Fisher test statistic values combined with the p-values to assess an overall hypothesis for the twenty-six countries. This test statistic λ is distributed as χ22N under the cross-section independence assumption.14 As shown in Table 9, the null hypothesis of the cross-section independence assumption is strongly rejected and, therefore, the limit distribution of the Fisher test statistic is no longer valid. In the presence of the cross-section dependence in mixed panels, we apply the bootstrap method to generate the empirical distributions of the Fisher test. The bootstrap distribution of the Fisher test statistics is derived from 10,000 replications. Bootstrap critical values are obtained at the 10%, 5% and 1% levels based on these empirical distributions. The empirical results show that the λ statistics is equal to 139.524 and is insignificant at the 5% level for the null hypothesis of Granger non-causality running from exchange rates to stock prices (Ho : ex 6⇒ sp). Thus, from the perspective of a panel, the empirical test results do not provide strong evidence in support of the 14 We

code the panel Granger non-causality by using the WinRATS software according to the Matlab code provided

by Professor Emirmahmutoglu. We thank him for providing us with his Matlab code for reference.

19

flow-oriented model of exchange rate determination. Table 10 also summarizes the results of Granger non-causality test running from stock prices to exchange rates, i.e., Ho : sp 6⇒ ex, under the bivariate panel model. The cross-sectional dependence test and the slope homogeneity test are also conducted in advanced in Table 9. The testing results reject the nulls, supporting the existence of the cross-sectional dependence and the heterogeneous slope in the system. Next, we turn our attention to test for the null hypothesis of Granger non-causality running from stock prices to exchange rates (Ho : sp 6⇒ ex). The empirical results show that there is unidirectional Granger causality running from stock prices to exchange rates for thirteen economies. They are Belgium, Canada, Denmark, France, Germany, Greece, Iceland, Italy, the Netherlands, Norway, Portugal, Spain, and the US. This implies that it is possible to predict foreign exchange markets from stock markets in these countries. In addition, the λ statistics is equal to 393.9 and is significant at the 5% level, implying that the null hypothesis of Granger non-causality running from stock prices to exchange rates is rejected. As such, the results of the λ statistics are in line with the results of individual Granger non-causality tests. Overall, the empirical evidences show that stock prices are helpful to forecast the future behaviors of exchange rates, which is in support of the stock-oriented model of exchange rate determination (Branson, 1983; Frankel, 1983). The empirical results of this study echo the findings of Lin (2012), Tsai (2012), Liang et al. (2013) and Salisu and Ndako (2018), which demonstrate that the stock-oriented model is important in explaining the relationship between the foreign exchange and the stock markets.

4.4

Results of the asymmetric Granger non-causality test

One main feature of the results reported above is that by assuming symmetric effects of exchange rate changes on stock prices, we are unable to discover significant causal effects. Does the picture change if we shift to nonlinear model (7)–(8)?15 Following the same estimation procedure, we 15 Before

discussing the empirical results of asymmetric Granger non-causality test, we first test for the integrated

order of the partial sum process of positive and negative changes in exchange rates (i.e., ex + and ex − ) and stock prices (i.e., sp+ and sp− ) as shown in Eqs (3)–(6). The test results are given in Tables 7 and 8. We have discussed the results

20

report the results of asymmetric Granger non-causality test on the right panel of Table 10. In order to test for asymmetric Granger non-causality running from ex + to sp, we test the restriction +

+ + + H0ex : ϕ˜ i,1 = ϕ˜ i,2 = · · · = ϕ˜ i,p = 0, ∀i. With the exceptions of East Asian economies, i.e., Hong

Kong, Japan, South Korea, and Malaysia, the null hypothesis of the asymmetric Granger non+

causality running from the positive changes in exchange rates to stock prices (H0ex : ex + 6⇒ sp) is not rejected at the 5% significance level. The estimate of λ equals to 100.979, which is again insignificant at the 5% level for the null hypothesis of Granger non-causality running from ex + (appreciation of domestic currency) to sp (stock prices). For the case of asymmetric non-causality −

− − − running from ex − to sp, we test for the restriction H0ex : ϕ˜ i,1 = ϕ˜ i,2 = · · · = ϕ˜ i,p = 0, ∀i. Except −

for Japan, the null of H0ex : ex − 6⇒ sp cannot be rejected at the 5% significance level. The joint test of λ test either rejects the null that the ex − (depreciation of domestic currency) can not predict to the sp (stock prices). Thus, the results of asymmetric Granger non-causality test echo the results of symmetric Granger non-causality test, indicating that neither depreciation nor appreciation of exchange rates is helpful to forecast the behavior of stock prices. Table 10 also reports the results of asymmetric Granger non-causality test running from stock prices to exchange rates under the bivariate panel model. To test for asymmetric Granger nonsp−

causality running from sp− to ex, we test for the null hypothesis of Ho

− − = ··· = = ϕi,2 : ϕi,1

− ϕi,p = 0, ∀i. We pay our attention to the results of the λ statistics. It is found that the λ statistics is

equal to 167.986, which is significant at the 10% level, but insignificant at the 5% level. Based on the sp−

5% critical value, this result indicates that the null hypothesis of H0

: sp− 6⇒ ex is not rejected.

The uni-directional asymmetric Granger causality running from sp− (decrease in stock price) to ex (exchange rate) is not supportive. In the case of asymmetric Granger non-causality running sp+

from sp+ (increase in stock price) to ex (exchange rate), we test for the following restriction H0

:

previously and, therefore, we omit the detail. Basically, the results of the panel unit root tests suggest using dmax = 2 for these countries in the panel specification. Moreover, the results of the null hypothesis of no cross-sectional dependence across the members of the panel and the null hypothesis of the slope homogeneity are significantly rejected at the 5% level as shown in the right panel of Table 9. There seems to be evidence of cross-sectional dependence in the data and supporting the view that the parameters are heterogeneous.

21

+ + + ϕi,1 = ϕi,2 = · · · = ϕi,p = 0, ∀i. The result of λ is equal to 166.057 and it is significant at the 5% sp+

level, indicating that the null hypothesis of Ho

: sp+ 6⇒ ex is rejected. That is, the increase in

stock price does Granger-cause exchange rate. Overall, the empirical results show that there is an asymmetric (positive) causation running from stock prices to exchange rates, which is in line with the prediction of the monetary model (Gavin, 1989) and the portfolio rebalancing model (Hau and Rey, 2004). Under the assumption of the portfolio rebalancing model, stock prices are expected to lead exchange rates and to be positively correlated to them.

4.5

The effect of the global financial crisis

Are the empirical results sensitive to the global financial crisis (GFC)? The global financial crisis refers to the period of extreme stress in global financial markets and banking systems between mid 2007 and 2010. In the literature, Caporale et al. (2014) have examined the linkages between stock market prices and exchange rates in six advanced economies, namely the US, the UK, Canada, Japan, the euro area, and Switzerland, using data on the banking crisis between 2007 and 2010. They consider two sub-periods: a tranquil or pre-crisis period from August 6, 2003 to August 8, 2007, and a crisis period from August 15, 2007 to December 28, 2011. Salisu and Ndako (2018) divide the data into pre- and post-GFC regimes ranging from 31st May 2004 to 30th September 2007 and from 1st October 2007 to 30th June 2017 respectively to examine stock price-exchange rate nexus in 32 OECD countries. In this study, in order to know the effect of global financial crisis on the causal relationship between exchange rates and stock prices, instead of dividing the sample into the pre-crisis and post-crisis regimes, we partition the data into three regimes as follows: the pre-GFC (1998/01/01-2007/07/31), the GFC (2007/08/01-2010/12/31), and the postGFC (2011/01/01-2019/05/20).16 We redo all of the empirical estimations and tests for the three sub-samples and summarize the results in Tables 11–13.17 16 We

owe this suggestion to an anonymous referee.

17 Owing to the space limitation of the paper, we do not report the results of the panel unit root test, the cross-sectional

dependence test and the slop homogeneity test for the three sub-samples in this paper. These tables are available from the authors upon request.

22

Tables 11–13 report the results of the panel Granger non-causality test for the three sub-samples. We focus on the results of the λ statistics. For the pre-GFC, the GFC and the post-GFC, it is found that the null hypothesis of Granger non-causality running from exchange rates to stock prices is not rejected at the 5% significance level. However, the null hypothesis of Granger non-causality running from stock prices to exchange rates is rejected at the 5% significance level. This result indicates that the global financial crisis did not change the causal relationship between the foreign exchange and stock markets (cf. Table 10). That is, our empirical results are insensitive to the global financial crisis. The results show that, again, the information on the stock market is helpful in predicting the behavior of the foreign exchange market, which lends support to the stock-oriented models such as the monetary model or the portfolio balance model.18 Finally, we report the results of the asymmetric Granger non-causality test for the three subsamples in the right panels of Tables 11 to 13. Again, we focus on the results of the λ statistics. For the pre-GFC (Table 11), the results of the λ statistics show that the null hypotheses of the asymmetric Granger non-causality running from positive and negative changes in exchange rates to stock prices (Hoex

+

: ex + 6⇒ sp and Hoex

−

: ex − 6⇒ sp) are not rejected at the 5% level. For

the null hypothesis of the asymmetric Granger non-causality running from positive changes in sp+

stock prices to exchange rates (Ho

: sp+ 6⇒ ex), it is also not significant at the 5% level. The

null hypothesis of the asymmetric Granger non-causality running from negative changes in stock sp−

prices to exchange rates (Ho

: sp− 6⇒ ex) is also insignificant at the 5% level. On the right panels

of Tables 12 and 13, it is found that in the cases of the GFC and the post-GFC, the results of the λ statistics show that they are all insignificant at the 5% level for the hypothesis of the asymmetric Granger non-causality running from exchange rates to stock prices, and vice versa. 18 A

recent study of Salisu and Ndako (2018) also investigates the causal relationship between exchange rates and

stock prices for 32 OECD countries. Their results lend support the portfolio balance theory for the full OECD, the Euro area, and the non-Euro area, albeit with lesser evidence for the latter. Also, the validity of the theory became more evident after the global financial crisis. Basically, our empirical results are in line with Salisu and Ndako (2018), but slightly different from the impact of the GFC. The distinct results between Salisu and Ndako (2018) and this study are due to differences in methodologies, and samples and are subject to diverse interpretations.

23

4.6

Robustness analysis

In order to verify whether the results of the Emirmahmutoglu and Kose (2011) panel Granger noncausality test are indifferent to, for example, using different samples or methods, in this section, we consider the following three robustness analyses. First, we redo all of the estimations and tests by dividing the data into the developed economies and emerging economies. Second, in addition to the Emirmahmutoglu and Kose (2011) test, we employ the Hatemi-J (2012) panel Granger noncausality test, which also allows for asymmetry in the causality testing by using the cumulative sums of positive and negative shocks with bootstrapping. Moreover, the estimation method also accounts for the ARCH effects in the data as revealed in Table 4-6. Third, instead of using the log of data in empirical analysis, we repeat the empirical exercise by using the raw data in order to know if there are any differences compared to earlier conclusions.19

4.6.1

Developed and emerging economies

Table 14 reports the results of the λ statistics of the symmetric and asymmetric Granger noncausality tests for the developed and emerging economies, respectively.20 In the case of the developed economies, the λ statistic is 373.577 of the null hypothesis Ho : sp 6⇒ ex and is significantly rejected at the 5% level, indicating that stock price is helpful for predicting the exchange rate. The λ statistic is 110.837 of the null hypothesis Ho : ex 6⇒ sp, favoring a rejection at the 10% level but not at the 5% level. This implies that the exchange rate is not helpful for predicting the stock price based on the 5% significant level. For the null hypotheses of asymmetric Granger non-causality of +

running from positive and negative changes in exchange rates to stock prices (Hoex : ex + 6⇒ sp, Hoex

−

: ex − 6⇒ sp) and running from positive and negative changes in stock prices to exchange sp+

rates (Ho 19 We

sp−

: sp+ 6⇒ ex and Ho

: sp− 6⇒ ex), all of them are insignificant at the 5% level.

thank three anonymous referees for providing us with these suggestions.

20 Due

to space restrictions, we do not present the results of the panel unit root tests, the cross-sectional dependence

of the data and the hypothesis of slope homogeneity for these robustness analysis. These tables are available from the authors upon request.

24

In the case of emerging economies, the symmetric Granger non-causality tests of Ho : sp 6⇒ ex and Ho : ex 6⇒ sp are not significant at the 5% level. The results of asymmetric Granger noncausality tests indicate that positive and negative changes in exchange rates are not helpful for forecasting stock prices because the null hypotheses of Hoex sp+

not rejected at the 5% level. For the hypothesis of Ho

+

: ex + 6⇒ sp, Hoex

−

: ex − 6⇒ sp are

: sp+ 6⇒ ex, the λ statistic is 43.688

and significantly rejected at the 5% level. This indicates that the positive change in stock prices is helpful for predicting the exchange rates. But the hypothesis of asymmetric Granger non-causality sp−

running from negative changes in stock prices to exchange rates (Ho

: sp− 6⇒ ex) is not rejected

at the 5% level. In the developed economies, since the direction of causation is from stock price to exchange rate, it implies that the foreign exchange could be used to hedge investment in the stock market (Hatemi-J and Roca, 2005, p. 544). For the emerging economies, the empirical results show that there is no symmetric causal relationship between foreign exchange market and stock market, indicating that the two financial markets are not close intertwined.21 It implies that investors can use them as effective instruments for portfolio hedging and diversification strategies.

4.6.2

The Hatemi-J (2012) panel Granger non-causality test

Hatemi-J (2012) suggests an approach for implementing asymmetric causality tests within a panel perspective. He shows how cumulative sums for positive and negative shocks can be constructed for this purpose. We can find out, by applying this method, whether the causal impact of positive shocks is different than the causal impact of negative shocks within a panel system. The results of the symmetric and asymmetric Hatemi-J (2012) panel Granger non-causality test are given in 21 Ajayi

et al. (1998), examining seven advanced markets and eight Asian emerging markets, find that no consistent

causal relations exist between stock and exchange rate markets in the case of emerging economies. Ramasamy and Yeung (2005) show that the direction of causality can vary according to the period of study and they also find that stock prices lead exchange rates for Malaysia, Singapore, Thailand, Taiwan and Japan. These studies found that the strength of trade links was helpful in explaining the relationship nature of the asset price linkage.

25

Table 15.22 In the cases of the full-sample (Table 15(A)), the pre-GFC (Table 15(B1)), the GFC (Table 15(B2)), and the post-GFC (Table 15(B3)), for the developed economies (Table 15(C1)) and the emerging economies (Table 15(C2)), the null hypothesis of symmetric Granger non-causality running from exchange rates to stock prices (Ho : ex 6⇒ sp) is significantly rejected at the 5% level. Likewise, the null hypothesis of symmetric Granger non-causality running from stock prices to exchange rates (Ho : sp 6⇒ ex) is also rejected at the 5% level. In other words, there is a bi-directional causal relationship between exchange rates and stock prices, implying that the exchange rate is helpful for predicting stock price, and vice versa. With respect to the results of asymmetric +

Granger non-causality test, the null hypothesis of Hoex : ex + 6⇒ sp for the emerging economies is rejected at the 5% level, implying that the positive changes in exchange rates are helpful for predicting stock prices. For the other cases, i.e., the full-sample, the pre-GFC, the GFC, and the post-GFC, for the developed and emerging economies, the hypotheses of asymmetric Granger non-causality tests from exchange rates to stock prices or from stock prices to exchange rates are not significantly at the 5% level. In sum, with the exception of the hypothesis of Ho : ex 6⇒ sp, basically, the causal relations predicted by the Hatemi-J (2012) panel Granger non-causality test is in line with the causal relations predicted by the Emirmahmutoglu and Kose (2011) test.

4.6.3

The use of nonlogarithmic data

Initially we use the log of exchange rates and stock prices to conduct the empirical estimations and tests. Are the results different from using the raw data? To answer this, we repeat all of the empirical estimations and tests by using the raw data. In sum, the resulting inferences of using raw data to test for the symmetric and asymmetric panel Granger non-causality is fundamentally unchanged as compared to those of the log data. Due to space restrictions, we do not present the whole estimation results of raw data in this paper, but they are available in the Supplement of the paper. 22 Readers

are referred to the appendix of this paper for a brief description of the Hatemi-J (2012) panel Granger

non-causality test.

26

5

Concluding remarks

This paper is directed towards revisiting the direction of causations between exchange rates and stock prices for twenty-six economies, including those of 20 advanced and 6 emerging economies. To this end, we adopt the symmetric and asymmetric panel data approaches. Compared to previous studies, the first contribution of this paper to the literature is the application of a state-of-theart Granger non-causality technique that has recently been developed by Emirmahmutoglu and Kose (2011), and it is based on the estimation of the panel model with the bootstrap critical values. It allows us to untangle the symmetric Granger causal relationship between exchange rates and stock prices. The second contribution of this paper is that we modify Emirmahmutoglu and Kose’s (2011) approach by decomposing the movements of the exchange rate and stock price into their negative and positive partial sums. This modification enables us to test for the asymmetric Granger non-causality between exchange rates and stock prices. Third, in order to know the effect of the global financial crisis on the causal relationship between exchange rates and stock prices, we partition the data into three regimes: the pre-GFC (1998/01/01-2007/07/31), the GFC (2007/08/01-2010/12/31), and the post-GFC (2011/01/01-2019/05/20). Fourth, for the purpose of robustness checking, we redo all of the estimations and tests by (i) dividing the data into the developed economies and emerging economies; (ii) by employing the Hatemi-J (2012) test; and (iii) by using the raw data. For the readers’ information, we summarize all of the results of panel Granger non-causality test in Table 16 and the key findings of this study are as follows. First, the null hypothesis of no cross-sectional dependence across the members of the panel and the null hypothesis of the slope homogeneity are strongly rejected, indicating that the bootstrap critical value is required in conducting the panel Granger non-causality test and supports the view that the parameters are heterogeneous. Therefore, we apply the bootstrap method to generate the empirical distributions of the Fisher test. Second, based on the λ statistic proposed by Emirmahmutoglu and Kose (2011), the null hypothesis of symmetric Granger non-causality running from exchange rates to stock prices cannot be rejected at the 5% significance level. But the null hypothesis of symmetric Granger non-causality running from stock prices to exchange rates can be rejected at the 27

5% significance level. The empirical test results provide evidence that stock prices are helpful for predicting the exchange rates, but not vice versa. Thus, the results do not provide evidence in support of the flow-oriented model of exchange rate determination. Third, the results show weak evidence in support of unidirectional asymmetric causality running from exchange rates to stock prices, and vice versa. Fourth, the empirical results of the Hatemi-J (2012) symmetric panel Granger non-causality tests show that there is a causation from stock prices to exchange rates, and vice versa. The hypotheses of the Hatemi-J asymmetric panel Granger non-causality from stock prices to exchange rates and from exchange rates to stock prices are not rejected, which are in line with the Emirmahmutoglu and Kose (2011) test. Fifth, when we adopt the raw data to conduct the Emirmahmutoglu and Kose (2011) test, the causal relations are in line with the results using the log data. Moreover, by using the raw data, the causal relations predicted by using the Hatemi-J (2012) panel Granger non-causality test echoes the causal relations predicted by using the Emirmahmutoglu and Kose (2011) test.

Acknowledgements We would like to thank the editor, Professor Carl Chen, a review board member of International Review of Economics and Finance, and three anonymous referees of this journal for helpful comments and suggestions. We thank Professor Donald Gotcher for proofreading this article. The usual disclaimer applies.

Appendix: The Hatemi-J (2012) panel Granger non-causality test In this appendix, we briefly describe the model specification and procedure of conducting the Hatemi-J (2012) panel Granger non-causality test. Let yi,t = [exi,t spi,t ]0 , the matrix representation of equations (1) and (2) can be expressed as follows: yi,t = ci + Ai1 yi,t−1 + · · · + Aip yi,t− p + εi,t ,

28

(19)

where country i = 1, . . . , N and time t = 1, . . . , T. Vector ci and matrices Ai1 , . . . , Aip are parameters for country i to be estimated in the p order vector autoregressive (VAR) system. Stacking the equations of (19) for all i yields the specification of panel vector autoregressive seemingly unrelated regression (panel-VAR-SUR) model as follows. y1 Z1 0 · · · 0 β 1 ε1 y2 0 Z2 · · · 0 β 2 ε2 . + . , . = . .. . . .. .. .. . . . .. .. 0 0 · · · ZN yN βN εN {z } | {z } | {z } | {z } | y

Z

β

(20)

ε

where yi = [yi,1 yi,2 · · · yi,T ]0 and ε = [ε 1,t ε i,2 · · · ε 1,T ]0 . Zi contains the explanatory variables of equation (19) and β i is the corresponding coefficient vector. The generalized least square (GLS) estimator of equation (20) is shown as βˆ = [ Z 0 (Σ−1 ⊗ I ) Z ]−1 Z 0 (Σ−1 ⊗ I )y,

(21)

where ⊗ is the Kronecker product operator and I denotes an identity matrix, and Σ is the variancecovariance matrix of the disturbance terms. The null hypothesis of Granger non-causality in the panel-VAR-SUR model is Ho : R βˆ = 0, where R is the matrix indicating the restriction of the model. We can compute the following Wald statistic to conduct the test. WALD = ( R βˆ )0 [ RVar( βˆ ) R0 ]−1 ( R βˆ ). Based on the idea of Toda and Yamamoto (1995), we add additional lags dmax , i.e., the maximum order of integrated of the variables, in the panel-VAR-SUR model, irrespective of the order of integration of variable or cointegration properties among variables. That is, yi,t = ci + Ai1 yi,t−1 + · · · + Aip y p,t− p + Ai( p+1) yi,t−( p+1) + · · · + Ai( p+dmax) yi,t−( p+dmax) + εi,t , (22)

29

A bootstrap method of the modified Wald (MWALD) statistic for the Granger non-causality of the panel-VAR-SUR model is introduced by Hacker and Hatemi-J (2006) and Hatemi-J (2012). Consider a bivariate VAR(p) process for i-th country p

exi,t = ci +

p

∑

θi,m exi,t−m +

∑

θ˜i,m spi,t−m +

m =1 p

spi,t = c˜i +

∑

ϕi,m spi,t−m + ui,t

(23)

∑

ϕ˜ i,m exi,t−m + vi,t

(24)

m =1 p

m =1

m =1

where the lag length p can be determined by, for example, the Akaike information criterion or HJC critrion suggested by Hatemi-J (2003). The steps for running the bootstrap are as follows. Step 1: By using the GLS method in Eq. (21) to run the regressions (23) and (24) for all i under the null hypotheses that sp does not Granger-cause ex and ex does not Granger-cause sp, we can obtain the estimates of coefficients {cˆi , cˆ˜i , θˆi,1 , . . . , θˆi,p , θˆ˜i,1 , . . . , θˆ˜i,p } and residuals uˆ i,t and vˆi,t . Step 2: Let X1i be the matrix that contains the explanatory variables of the restricted regression (23), i.e., all lags of exi,t and Xi be the matrix that contains all variables that determine spi,t in regression (24). Define h1i = diag( X1i ( X1i0 X1i )−1 X1i0 ) h2i = diag( Xi ( Xi0 Xi )−1 Xi0 ) Modify the residuals through the leverage, m uˆ i,t =p

uˆ i,t , 1 − h1i,t

m vˆi,t =p

vˆi,t , 1 − h2i,t

where h1i,t and h2i,t denote i-th and t-th element of h1i and h2i , respectively. The modified residuals are the regression raw residuals modified to have constant variance. m and vˆ m with replacement and subtract from their means, which guarantees the Step 3: Draw uˆ i,t i,t

∗ and vˆ ∗ are zero. means of the bootstrap residuals uˆ i,t i,t

30

∗ and sp∗ such that Step 4: Generate the simulated exi,t i,t p

∗ exi,t

= cˆi +

∑

∗ θˆi,m exi,t−m + ui,t ,

∑

∗ θˆ˜i,m spi,t−m + vi,t .

m =1 p

∗ spi,t = cˆ˜i +

m =1

Step 5: Estimate the panel-VAR-SUR(p + dmax) by GLS estimation p+dmax

p

∗ exi,t

= ci +

∗ spi,t = c˜i +

∑

θi,m exi,t−m +

∑

m =1

m =1

p

p+dmax

∑

θ˜i,m spi,t−m +

m =1

∑

∗ ϕi,m spi,t−m + ui,t

∗ ϕ˜ i,m exi,t−m + vi,t

m =1

and test for the panel Granger non-causality based on the following restrictions Ho :ϕi,1 = ϕi,2 = · · · = ϕi,p = 0, ∀i, Ho : ϕ˜ i,1 = ϕ˜ i,2 = · · · = ϕ˜ i,p = 0, ∀i, and obtain the MWALD statistic of Toda and Yamamoto (1995). Repeating Step 1 to Step 5 for 10,000 times will produce the distribution of the MWALD test statistic. We can find the α-th upper quantile of the distribution of the bootstrap MWALD statistics and obtain the α-level bootstrap critical values.

References Abdalla, I. S. A. and V. Murinde (1997), Exchange rate and stock price interaction is emerging financial markets: Evidence on India, Korea, Pakistan, and Phillips, Applied Financial Economics, 7, 25–35. Afshan, S., Sharif, A. Loganathan, N. and Jammazi, R. (2018), Time frequency causality between stock prices and exchange rates: Further evidences from cointegration and wavelet analysis, Physica A, 495, 225–244. Aggarwal, R. (1981), Exchange rates and stock prices: A study of the U.S. capital markets under floating exchange rates, Akron Business and Economic Review, 12, 7–12. Ajayi, R. A., Friedman, J. and Mehdian, S. M. (1998), On the relationship between stock returns and exchange rate: Tests of Granger causality, Global Finance Journal, 9(2), 241–251. 31

Ajayi, R. A. and Mougou´e, M. (1996), On the dynamic relation between stock prices and exchange rates, The Journal of Financial Research, 19(2), 193–207. Andreou, E., Matis, M. and Savvides, A. (2013), Stock and foreign exchange market linkages in emerging economies, Journal of International Financial Markets, Institutions and Money, 27, 148–268. Andrie, A. M., Ihnatov, I. and Tiwari, A. K. (2014), Analyzing time-frequency relationship between interest rate, stock price and exchange rate through continuous wavelet, Economic Modelling, 41, 227–238. Bahmani-Oskooee, M. and Saha, S. (2015), On the relation between stock prices and exchange rates: A review article, Journal of Economic Studies, 42, 707–732. Bahmani-Oskooee, M. and Saha, S. (2016a), Do exchange rate changes have symmetric or asymmetric effects on stock prices?, Global Finance Journal, 31, 57–72. Bahmani-Oskooee, M. and Saha, S. (2016b), Asymmetry cointegration between the value of the dollar and sectoral stock indices in the U.S., International Review of Economics and Finance, 46, 78–86. Bahmani-Oskooee, M. and Sohrabian, A. (1992), Stock prices and the effective exchange rate of the dollar, Applied Economics, 24, 459–464. Bai, J. and Ng, S. (2004), A PANIC attack on unit roots and cointegration, Econometrica, 72 (4), 1127–1177. Bai, J. and Ng, S. (2010), Panel unit root tests with cross-section dependence: A further investigation, Econometric Theory, 26 (4), 1088–1114. Bai, J. and Kao, C. (2006), On the estimation and inference of a panel cointegration method with cross-sectional dependence, In: Baltagi, B. H. (Ed.), Panel Data Econometrics: Theoretical Contributions and Empirical Applications. Elsevier Science and Technology. Branson, W. H. (1983), Macroeconomic determinants of real exchange rate risk. In R. J. Herring (Ed.), Managing foreign exchange rate risk. Cambridge, MA: Cambridge University Press. Breitung, J. (2005), A parametric approach to the estimation of cointegration vectors in panel data, Econometric Reviews, 24(2), 151–173. Breusch, T. and Pagan, A. (1980), The Lagrange multiplier test and its application to model specifications in econometrics, Reviews of Economic Studies, 47, 239–253. Caporale, G. M., Hunter, J. and Ali, F. M. (2014), On the linkages between stock prices and exchange rates: Evidence from the banking crisis of 2007-2010, International Review of Financial Analysis, 33, 87–103. 32

Chen, S.-W. and Chen, T.-C. (2012), Untangling the non-linear causal nexus between exchange rates and stock prices: New evidence from the OECD countries, Journal of Economic Studies, 39(2), 231–259. Chkili, W. and Nguyen, D. K. (2014), Exchange rate movements and stock market returns in a regime-switching environment: Evidence for BRICS countries, Research in International Business and Finance, 31, 46–56. Dickey, D. A. and Fuller, W. A. (1981), Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root, Econometrica, 49, 1057–1072. Doong, S. C, Yang, S. Y. and Wang, A. T. (2005), The dynamic relationship and pricing of stocks and exchange rates: Empirical evidence from Asian emerging markets, Journal of American Academy of Business 7(1), 118–123. Dornbusch, R. and Fisher, S. (1980), Exchange rates and the current account, American Economic Review, 70, 960–971. Dumitrescu, E. I. and Hurlin, C. (2012), Testing for Granger non-causality in heterogeneous panels, Economic Modelling, 29(4), 1450–1460. Emirmahmutoglu, F. and Kose, N. (2011), Testing for Granger causality in heterogeneous mixed panels, Economic Modelling, 28, 870–876. Fisher, R. A. (1932), Statistical Methods for Research Workers, 4th edition. Oliver and Boyd, Edinburgh. Fowowe, B. (2015), The relationship between stock prices and exchange rates in South Africa and Nigeria: structural breaks analysis, International Review of Applied Economics, 29, 1–14. Frenkel, J. A. (1976), A monetary approach to the exchange rate: Doctrinal aspects and empirical evidence, Scandinavian Journal of Economics, 78, 200–224. Frankel, J. A. (1983), Monetary and portfolio balance models of exchange rate determination. In J. S. Bhandari, & B. H. Putnam (Eds.), Economic interdependence and flexible exchange rates, 84–115, Cambridge, MA: MIT Press. Gavin, M., (1989), The stock market and exchange rate dynamics, Journal of International Money and Finance, 8(2), 181–200. Gengenbach, C., Palm, F. C. and Urbain, J.-P. (2006), Cointegration testing in panels with common factors, Oxford Bulletin of Economics and Statistics, 68, 683–719. Supplement. Gengenbach, C., Palm, F. C. and Urbain, J. P. (2010), Panel unit root tests in the presence of crosssectional dependencies: Comparison and implications for modelling, Econometric Reviews, 29, 111–145. 33

Granger, C. W. J. (1969), Investigating causal relations by econometric models and cross-spectral methods, Econometrica, 37, 424–438. Granger, C. W. J. (2003), Some aspects of causal relationships, Journal of Econometrics, 112, 69–71. Granger, C. W. J., Huang, B.-N. and Yang, C.-W. (2000), A bivariate causality between stock prices and exchange rates: evidence from recent Asian flu, The Quarterly Review of Economics and Finance, 40, 337–354. Groenewold, N. and Paterson, J. E. (2013), Stock prices and exchange rates in Australia: Are commodity prices the missing link? Australian Economic Papers, 52, 159–170. Hacker, R. S. and Hatemi-J A. (2006), Tests for causality between integrated variables using asymptotic and bootstrap distributions: Theory and application, Applied Economics, 38(13), 1489–1500. Hatemi-J, A. (2003), A new method to choose optimal lag order in stable and unstable VAR models, Applied Economics Letters, 10(3), 135–137. Hatemi-J, A. (2012), Asymmetric panel causality tests with an application to the impact of fiscal policy on economic performance in Scandinavia, MPRA Paper No 55527. Munich: University Library of Munich. Hatemi-J, A. and Irandoust, M. (2002), On the causality between exchange rates and stock prices: A note, Bulletin of Economic Research, 54(2), 197–203. Hatemi-J, A. and Roca, E. (2005), Exchange rates and stock prices interaction during good and bad times: Evidence from the ASEAN4 countries, Applied Financial Economics, 15(8), 539–546. Hau, H. and Rey, H. (2004), Can portfolio rebalancing approach explain the dynamics of equity returns, equity flows, and exchange rates?, American Economic Review, 94(2), 126–133. Ho, L.-C. and Huang, C.-H. (2015), The nonlinear relationships between stock indexes and exchange rates, Japan and the World Economy, 33, 20–27. Im, K., Pesaran, M. H. and Shin, Y. (2003), Testing for unit roots in heterogeneous panels, Journal of Econometrics, 115, 53–74. Inci, A.C. and Lee, B.S. (2014), Dynamic relations between stock returns and exchange rate changes, European Financial Management, 20 (1), 71–106. Kao, C. and Chiang, M. H. (2000), On the estimation and inference of a cointegrated regression in panel data, Advances in Econometrics, 15, 7–51. Kar, M., Nazlioglu, S. and Agir, H. (2011), Financial development and economic growth nexus in the MEAN countries: Bootstrap panel Granger causality analysis, Economic Modelling, 28, 685–693. 34

Kim, K. H. (2003), Dollar exchange rate and stock price: Evidence from multivariate cointegration and error correction model, Review of Financial Economics, 12, 301–313. Kollias, C., Mylonidis, N., and Paleologou, S.-M. (2012), The nexus between exchange rates and stock markets: Evidence from the euro-dollar rate and composite European stock indices using rolling analysis, Journal of Economics and Finance, 36, 136–147. Lean, H.H., Narayan, P. and Smyth, R. (2011), Exchange rate and stock price interaction in major Asian markets: Evidence for industrial countries and panels allowing for structural breaks, The Singapore Economic Review, 56, 255–277. Liang, C.-C., Lin, J.-B, and Hsu, H.-C. (2013), Reexamining the relationships between stock prices and exchange rates in ASEAN-5 using panel Granger causality approach, Economic Modelling, 32, 560–563. Lin, C. H. (2012), The comovement between exchange rates and stock prices in the Asian emerging markets, International Review of Economics and Finance, 22(1), 161–172. Liu, l. and Wan, J. (2012), The relationships between Shanghai stock market and CNY/USD exchange rate: New evidence based on cross-correlation analysis, structural cointegration and nonlinear causality test, Physica A, 391, 6051–6059. Maddala, G. S. and Wu, S. (1999), A comparative study of unit root tests with panel data and a new simple test, Oxford Bulletin of Economics and Statistics, 61, 631–652. Malliaropulos, D. (1998), International stock return differentials and real exchange rate changes, Journal of International Money and Finance, 17, 493–511. Morley, B. and Pentecost, E. J. (2000), Common trends and cycles in G-7 countries exchange rates and stock prices, Applied Economics Letters, 7, 7–10. Moore, T. and Wang, P. (2014), Dynamic linkage between real exchange rate and stock prices: Evidence from developed and emerging Asian markets, International Review of Economics and Finance, 29, 1–11. Muhammad, N. and Rasheed, A. (2002), Islamabad stock prices and exchange rates: Are they related? Evidence from South Asian countries, The Pakistan Development Review, 41(4), 535– 550. Ndako, U.B. (2013), Dynamics of stock prices and exchange rates relationship: evidence from five sub-Saharan African financial markets, Journal of African Business, 14 (1), 47–57. Nieh, C. C. and Lee, C. F. (2001), Dynamic relationship between stock prices and exchange rates for G-7 countries, Quarterly Review of Economics and Finance, 41(4), 477–490. O’Connell, P. J. (1998), The overvaluation of purchasing power parity, Journal of International Economics, 44, 1–19. 35

Pan, M.S., Fok, C.W. and Liu, Y. (2007), Dynamic linkages between exchange rates and stock prices: Evidence from East Asian markets, International Review of Economics and Finance, 16, 503–520. Pedroni, P. (1999), Critical values for cointegration tests in heterogeneous panels with multiple regressors, Oxford Bulletin of Economics and Statistics, 61, 653–670. Pedroni, P. (2000), Fully Modified OLS for Heterogeneous Cointegrated Panels, Advances in Econometrics, 15, 93–130. Pedroni, P. (2004), Panel cointegration; asymptotic and finite sample properties of pooled time series tests with an application to the PPP hypothesis, Econometric Theory, 20, 597–625. Pesaran, M. H. (2004), General diagnostic tests for cross section dependence in panels, CESifo Working Paper 1229, IZA Discussion Paper 1240. Pesaran, M. H. (2007), A simple panel unit root test in presence of cross-section dependence, Journal of Applied Econometrics, 22, 265–312. Pesaran, M. H., Ullah, A., Yamagata, T. (2008), A bias-adjusted LM test of error cross-section independence, Econometrics Journal, 11, 105–127. Pesaran, M. H., Smith, L. V., and Yamagata, T. (2013), Panel unit root test in the presence of a multifactor error structure, Journal of Econometrics, 175, 94–115. Phylaktis, K. and Ravazzolo, F. (2005), Stock prices and exchange rate dynamics, Journal of International Money and Finance, 24, 1031–1053. Ramasamy, B., and Yeung, M. C. (2005), The causality between stock returns and exchange rates: Revisited, Australian Economic Papers, 44(2), 162–169. Ratanapakorn, O. and Sharma, S. C. (2007), Dynamic analysis between the US stock returns and the macroeconomic variables, Applied Financial Economics, 17, 369–377. Reese, S. and Westerlund, J. (2016), PANICCA - PANIC on cross-section averages, Journal of Applied Econometrics, 31 (6), 961–981. Salisu, A. A. and Ndako, U. B. (2018), Modelling stock price-exchange rate nexus in OECD countries: A new perspective, Economic Modelling, 74, 105–123. Salisu, A. A. and Oloko, T. F. (2015), Modelling spillovers between stock market and FX market: Evidence for Nigeria, Jornal of African Business, 16, 84–108. Shin, Y., Yu, B. C., and Greenwood-Nimmo, M. (2014), Modelling asymmetric cointegration and dynamic multipliers in a nonlinear ARDL framework. In R. Sickels, and W. Horrace (Eds.), Festschrift in honor of Peter Schmidt: Econometric methods and applications (pp. 281–314). Springer. 36

Smyth, R. and Nandha, M. (2003), Bivariate causality between exchange rates and stock prices in South Asia, Applied Economics Letters, 10, 699–704. Soenen, L. A. and Hennigar, E. S. (1988), An analysis of exchange-rates and stock-prices-the united-states experience between 1980 and 1986, Akron Business and Economic Review, 19(4), 7–16. Stock, J. H. and Watson, M. W. (1988), Testing for common trends, Journal of American Statistical Association, 83, 1097–1107. Swamy, P. A. V. B. (1970), Efficient inference in a random coefficient regression model. Econometrica, 38, 311–323. Sui, L. and Sun, L. (2016), Spillover effects between exchange rates and stock prices:Evidence from BRICS around the recent global financial crisis, Research in International Business and Finance, 36, 459–471. Tang, X. and Yao, X. (2018), Do financial structures affect exchange rate and stock price interaction? Evidence from emerging markets, Emerging Markets Review, 34, 64–76. Toda, H. Y. and Phillips, P. C. B. (1993), Vector autoregression and causality, Econometrica, 59, 229–255. Toda, H. Y. and Yamamoto, T. (1995), Statistical inference in vector autoregressions with possibly integrated processes, Journal of Econometrics, 66, 225–250. Tsagkanos, A., Siriopoulos, C. (2013), A long-run relationship between stock price index and exchange rate: A structural nonparametric cointegrating regression approach, Journal of International Financial Markets, Institutions and Money, 25, 106–118. Tsai, I.-C. (2012), The relationship between stock price index and exchange rate in Asian markets: A quantile regression approach, Journal of International Financial Markets, Institutions and Money, 22, 609–621. ¨ u, ¨ N. and Demirci, E. (2012), Joint dynamics of foreign exchange and stock markets in emergUlk ing Europe, Journal of International Financial Markets, Institutions and Money, 22, 55–86. Walid, C. Chaker, A., Masood, O. and Fry, J. (2011), Stock market volatility and exchange rates in emerging countries: A Markov-state switching approach, Emerging Markets Review, 12, 272–292. Westerlund, J. (2006), Testing for panel cointegration with multiple structural breaks, Oxford Bulletin of Economics and Statistics, 68, 101–132. Westerlund, J. and Larsson, R. (2009), A note on the pooling of individual PANIC unit root tests, Econometric Theory, 25, 1851–1868. 37

Westerlund, J. and Urbain, J. P. (2015), Cross-sectional averages versus principal components, Journal of Econometrics, 185, 372–377. Wong, H. T. (2017), Real exchange rate returns and real stock price returns, International Review of Economics and Finance, 49, 340–352. Wong, D. K. T. and Li, K.-W. (2010), Comparing the performance of relative stock return differential and real exchange rate in two financial crises, Applied Financial Economics, 20, 137–150. Wongbangpo, P. and Sharma, S. C. (2002), Stock market and macroeconomic fundamental dynamic interactions: ASEAN-5 countries, Journal of Asian Economics, 13, 27–51. Yamada, H. and Toda, H. Y. (1998), Inference in possibly integrated vector autoregressive models: Some finite sample evidence, Journal of Econometrics, 86, 55–95. Yang, S.-P. (2017), Exchange rate dynamics and stock prices in small open economies: Evidence from Asia-Pacific countries, Pacific-Basin Finance Journal, 46, 337–354. Yang, Z., Tu, A. H. and Zeng, Y. (2014), Dynamic linkages between Asian stock prices and exchange rates: new evidence from causality in quantiles, Applied Economics, 46(11),1184–1201. Yau, H.-Y. and Nieh, C.-C. (2009), Testing for cointegration with threshold effect between stock prices and exchange rates in Japan and Taiwan, Japan and the World Economy, 21, 292–300. Zhao, H. (2010), Dynamic relationship between exchange rate and stock price: Evidence from China, Research in International Business and Finance, 24, 103–112.

38

Table 1: Summary of theoretical models Hypothesis

Proponents

Theoretical prediction

Flow-oriented model

Dornbusch and Fischer (1980)

Positive causation from exchange rate to stock price

Stock-oriented model Monetary model Portfolio balance model

Gavin (1989) Branson (1983), Frankel (1983)

Positive causation from stock price to exchange rate Negative causation from stock price to exchange rate

Portfolio rebalancing model

Hau and Rey (2004)

Positive causation from stock price to exchange rate

Asset market model

Frenkel (1976)

No association between stock prices and exchange rates

39

40

42.0 154.0 68.7 43.1 91.9 54.7 72.5 51.9 342.9 84.2 38.2 54.9 174.6 51.7 27.8 81.9 178.0 126.1 69.8 71.5 68.6 55.3 116.2 119.4 54.1 25.5

1998-2017 42.3 143.8 74.0 43.7 83.7 53.7 63.1 53.8 286.9 72.7 33.1 66.1 143.2 49.1 23.7 69.6 205.3 115.5 70.7 77.6 64.7 55.5 106.3 120.9 51.0 23.9

1998-2007

Trade-to-GDP ratio 42.7 160.1 63.1 45.6 101.0 60.0 82.9 60.1 398.9 96.9 49.1 43.2 184.0 55.0 32.7 96.5 148.1 146.6 69.0 58.7 75.7 59.1 137.3 130.3 59.0 28.6

2008-2017 14.2 58.4 13.0 5.4 17.7 16.4 13.5 12.5 83.6 59.5 4.1 2.9 151.8 10.9 7.3 36.1 10.4 62.8 21.5 5.1 19.9 17.1 8.7 6.8 23.6 9.5

1998-2017 16.6 57.9 13.4 6.0 24.3 25.3 19.4 15.6 71.3 69.2 3.3 2.5 237.8 14.1 6.5 30.7 8.6 80.0 27.4 5.2 21.7 25.2 8.1 6.2 31.9 10.9

1998-2007

Capital flows-to-GDP ratio 14.2 58.6 14.4 4.7 13.9 11.3 10.1 14.2 96.0 71.4 4.8 3.8 126.0 8.9 8.8 42.8 13.9 64.4 20.4 4.7 21.9 13.4 11.5 8.4 18.6 9.1

2008-2017

Note: The numbers 1, 2, 3 denote that the country is the member of the OECD, APEC, or EU, respectively. The trade-to-GDP ratio is the sum of imports and exports normalized by GDP. The capital flows-to-GDP ratio is the sum of FDI, portfolio equity, debt securities, as a gross capital flows as the percentage of GDP. We obtain the data from the World Bank and IMF Database.

Canada1,2 China2 Denmark1,3 France1,3 Germany1,3 Greece1,3 Hong Kong2 Iceland1,3 India Indonesia2 Ireland1 Italy1,3 Japan1,2 South Korea1,2 Malaysia2 Netherlands1,3 Norway1 Philippines2 Portugal1,3 Spain1,3 Taiwan2 Thailand2 United Kingdom1,3 United States1,2

Belgium1,3

Australia1,2,3

Country

Table 2: The trade-to-GDP ratio and capital flows-to-GDP ratio of each country

Table 3: Description of stock market benchmark indices of each country Country

Market / Region

Currency name

Stock market index

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand UK US

DM / Pacific DM / Europe DM / American EM / Pacific DM / Europe DM / Europe DM / Europe DM / Europe DM / Pacific DM / Europe EM / Pacific EM / Pacific DM / Europe DM / Europe DM / Pacific DM / Pacific EM / Pacific DM / Europe DM / Europe EM / Pacific DM / Europe DM / Europe DM / Pacific EM / Pacific DM / Europe DM / American

Australian dollar Euro Canadian dollar Renminbi Danish krone Euro Euro Euro Hong Kong dollar Euro Indian rupee Indonesian rupiah Euro Euro Japanese yen South Korean won Malaysian ringgit Euro Norwegian krone Philippine peso Euro Euro New Taiwan dollar Thai baht Pound sterling United States dollar

ASX 200 price index BEL 20 price index TSX composite price index SSE Shanghai composite index OMX Copenhagen price index CAC 40 price index DAX 30 performance price index Athex composite price index HSI hang seng index OMX Iceland all share price index BSE Sensex price index IDX composite price index ISEQ price index FTSE MIB price index Nikkei 225 price index KOSPI composite index FTSE KLSE price index AEX price index OSEAX price index Philippine se composite index PSI all share price index IBEX 35 price index TAIEX price index Bangkok SET price index FTSE 100 price index S&P 500 composite price index

Note: We use benchmark indices of the countries as representatives of their respective stock markets. All indices are capitalization-weighted index. Besides, all currencies are the most traded currencies in the world. Terms DM and EM denote the developed markets and emerging markets, respectively.

41

Table 4: Descriptive Statistics of the changes in the log of stock price (sp) and exchange rate (ex), full-sample periods (1998/1/1∼2019/05/20) (A) The change in the log of sp, full-sample periods (1998/1/1∼2019/05/20)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

0.016 0.007 0.016 0.016 0.028 0.011 0.019 −0.013 0.017 0.007 0.047 0.048 0.008 −0.003 0.006 0.030 0.018 0.005 0.032 0.025 −0.010 0.004 0.004 0.026 0.006 0.019

0.950 1.210 1.062 1.519 1.238 1.410 1.465 1.865 1.522 1.789 1.458 1.465 1.325 1.515 1.455 1.631 1.153 1.393 1.326 1.338 1.192 1.454 1.337 1.423 1.156 1.180

Min (%)

−8.704 −8.319 −9.788 −9.256 −11.723 −9.472 −8.875 −17.713 −13.582 −109.601 −12.885 −12.732 −13.964 −13.331 −12.111 −12.805 −24.153 −9.590 −9.709 −13.089 −10.379 −13.185 −9.936 −16.063 −9.266 −9.470

Max (%) 5.628 9.334 9.370 9.401 9.496 10.595 10.797 13.431 13.407 5.135 15.034 13.128 9.733 10.877 13.235 11.284 20.817 10.028 9.186 16.178 10.196 13.484 8.520 11.350 9.384 10.957

SK

−0.464 −0.026 −0.667 −0.332 −0.284 −0.058 −0.096 −0.292 0.088 −42.982 −0.489 −0.210 −0.626 −0.201 −0.358 −0.189 0.635 −0.137 −0.590 0.268 −0.359 −0.135 −0.188 −0.011 −0.149 −0.232

EK 5.699 5.986 9.802 5.390 5.321 4.997 4.413 6.112 8.214 2558.036 7.867 9.054 8.430 4.824 6.450 6.259 87.294 6.366 6.345 12.802 6.421 5.981 3.989 10.379 6.057 8.417

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.557 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.002∗∗∗ 0.000∗∗∗ 0.109 0.000∗∗∗ 0.024∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.416 0.007∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.005∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.002∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 1.000 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

(B) The change in the log of ex, full-sample periods (1998/1/1∼2019/05/20)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

0.000 0.001 0.002 0.004 0.002 0.001 0.002 0.001 −0.001 −0.009 −0.010 −0.022 0.001 0.002 0.003 0.006 −0.002 0.001 −0.002 −0.006 0.001 0.001 0.000 0.007 −0.004 0.002

0.633 0.171 0.508 0.295 0.185 0.177 0.213 0.224 0.249 0.800 0.392 1.265 0.276 0.183 0.647 0.663 0.436 0.202 0.434 0.481 0.092 0.148 0.266 0.441 0.429 0.324

Min (%)

−7.085 −1.342 −3.031 −2.317 −1.555 −1.400 −1.699 −10.424 −1.847 −23.647 −4.153 −20.852 −2.073 −1.428 −3.952 −6.683 −6.352 −1.602 −3.941 −7.308 −0.689 −1.214 −2.779 −4.157 −6.184 −2.279

Max (%) 6.764 1.251 3.875 1.574 1.375 1.340 1.657 1.393 1.683 18.665 3.115 23.265 1.953 1.412 6.466 10.161 5.974 1.585 3.259 13.955 0.682 1.030 2.743 5.760 2.154 2.020

SK

−0.522 0.017 −0.105 −0.201 0.018 0.021 0.022 −17.978 −0.146 −1.087 −0.207 −1.484 0.005 0.041 0.563 0.934 0.074 0.054 −0.246 3.933 0.063 −0.020 −0.041 0.779 −0.992 0.067

EK 10.046 3.831 3.601 3.630 4.233 3.856 4.193 836.336 3.112 224.217 7.040 86.041 3.579 3.990 6.533 33.602 41.518 3.893 6.280 149.007 3.801 4.100 9.398 29.896 12.132 3.418

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.317 0.087∗ 0.002∗∗∗ 0.518 0.237 0.121 0.137 0.000∗∗∗ 0.088∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.062∗ 0.111 0.152 0.000∗∗∗ 0.000∗∗∗ 0.054∗ 0.197 0.000∗∗∗ 0.168 0.152 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.039∗∗

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.994∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

Note: Symbols *, **, *** denote significance at the 10%, 5% and 1%, respectively. Mean and S.D. refer to the mean and standard deviation, respectively. SK is the skewness coefficient. EK is the excess kurtosis coefficient. JB is the p-value corresponding to the Jarque-Bera statistic. LB(24) is the p-value corresponding to the Ljung-Box Q statistic calculated with twenty-four lags. ARCH(4) is the p-value corresponding to the ARCH test calculated with four lags.

42

Table 5: Descriptive Statistics of the change in the log of stock price (sp), sub-sample periods (A) The change in the log of sp, pre global financial crisis periods (1998/1/1∼2007/07/31)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

0.035 0.024 0.029 0.053 0.035 0.026 0.023 0.047 0.031 0.083 0.069 0.071 0.030 0.020 0.005 0.066 0.034 0.010 0.049 0.025 0.017 0.029 0.005 0.033 0.009 0.016

0.757 1.143 1.012 1.429 1.139 1.396 1.571 1.598 1.518 0.767 1.618 1.666 1.065 1.340 1.365 2.031 1.537 1.464 1.197 1.482 1.042 1.366 1.542 1.697 1.129 1.109

Min (%)

−5.550 −5.610 −8.465 −9.256 −6.258 −7.678 −8.875 −9.692 −9.285 −5.619 −12.885 −12.732 −6.124 −7.867 −7.234 −12.805 −24.153 −7.531 −5.899 −8.692 −9.590 −7.339 −9.936 −16.063 −5.885 −7.044

Max (%)

SK

EK

3.445 9.334 4.684 9.401 4.970 7.002 7.553 7.620 13.395 5.135 7.695 13.128 5.835 7.626 7.222 10.024 20.817 9.517 7.340 16.178 5.395 6.323 8.520 11.350 5.903 5.573

−0.440 0.223 −0.598 0.069 −0.338 −0.117 −0.173 −0.102 0.172 −0.526 −0.693 0.028 −0.472 −0.155 −0.056 −0.092 0.710 −0.088 −0.505 0.916 −0.656 −0.198 −0.052 0.239 −0.181 −0.026

2.839 5.345 5.198 5.850 2.227 2.882 2.771 4.357 6.579 6.478 4.852 7.886 3.411 3.396 2.031 3.375 59.974 4.257 3.256 13.622 6.640 2.930 2.942 8.534 2.830 3.027

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.308 0.000∗∗∗ 0.064∗ 0.004∗∗∗ 0.012∗∗ 0.001∗∗∗ 0.006∗∗∗ 0.000∗∗∗ 0.004∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.704 0.245 0.000∗∗∗ 0.000∗∗∗ 0.073∗ 0.000∗∗∗ 0.000∗∗∗ 0.004∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.018∗∗

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

(B) The change in the log of sp, global financial crisis periods (2007/08/01∼2010/12/31)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

−0.029 −0.059 −0.003 −0.052 −0.010 −0.046 −0.010 −0.140 −0.001 −0.294 0.030 0.051 −0.122 −0.077 −0.059 0.007 0.011 −0.046 −0.018 0.020 −0.064 −0.046 −0.004 0.021 −0.008 −0.016

1.519 1.704 1.675 2.075 1.773 1.864 1.735 2.098 2.232 4.096 1.965 1.792 2.231 1.897 1.977 1.743 0.971 1.887 2.090 1.578 1.550 1.915 1.607 1.595 1.665 1.811

Min (%)

−8.704 −8.319 −9.788 −8.044 −11.723 −9.472 −7.433 −10.214 −13.582 −109.601 −11.592 −10.954 −13.964 −8.598 −12.111 −11.172 −9.979 −9.590 −9.709 −13.089 −10.379 −9.586 −6.735 −11.090 −9.266 −9.470

Max (%) 5.628 9.221 9.370 9.034 9.496 10.595 10.797 9.114 13.407 5.063 15.034 7.623 9.733 10.877 13.235 11.284 4.259 10.028 9.186 9.365 10.196 13.484 6.525 7.549 9.384 10.957

(Continued on next page)

43

SK

−0.349 −0.030 −0.596 −0.181 −0.125 0.217 0.258 −0.102 0.134 −22.221 −0.087 −0.496 −0.328 0.194 −0.354 −0.554 −1.326 −0.055 −0.539 −0.729 0.094 0.291 −0.284 −0.713 −0.049 −0.177

EK 3.139 4.548 6.311 2.164 4.931 5.496 6.540 2.322 5.554 576.628 6.892 5.646 3.606 5.052 6.857 7.042 14.380 5.832 3.684 8.782 7.790 6.261 2.077 6.431 5.518 6.273

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.600 0.022∗∗ 0.000∗∗∗ 0.013∗∗ 0.025∗∗ 0.002∗∗∗ 0.022∗∗ 0.017∗∗ 0.092∗ 0.715 0.021∗∗ 0.000∗∗∗ 0.002∗∗∗ 0.000∗∗∗ 0.107 0.641 0.292 0.004∗∗∗ 0.043∗∗ 0.004∗∗∗ 0.101 0.078∗ 0.022∗∗ 0.045∗∗ 0.000∗∗∗ 0.000∗∗∗

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 1.000 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

Table 5 (cont.): Descriptive Statistics of the change in the log of stock price (sp), sub-sample periods (C) The change in the log of sp, post global financial crisis periods (2011/01/01∼2019/05/20)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

0.013 0.014 0.009 0.001 0.035 0.016 0.026 −0.030 0.009 0.044 0.029 0.021 0.036 0.002 0.033 0.000 0.003 0.021 0.033 0.027 −0.018 −0.003 0.007 0.020 0.010 0.038

0.841 1.026 0.749 1.338 1.070 1.194 1.194 2.037 1.115 0.777 0.922 0.999 1.070 1.526 1.295 0.911 0.558 1.023 1.032 1.021 1.183 1.328 0.890 0.914 0.905 0.899

Min (%)

−4.176 −6.613 −4.123 −8.873 −6.781 −8.384 −7.067 −17.713 −6.018 −3.842 −6.947 −9.300 −10.416 −13.331 −11.153 −6.420 −3.237 −5.873 −5.759 −6.989 −7.247 −13.185 −6.521 −5.812 −4.779 −6.896

Max (%)

SK

EK

3.587 5.354 3.941 5.604 5.143 6.089 5.210 13.431 5.519 4.763 3.348 4.649 4.666 6.386 7.426 4.900 3.322 4.426 4.291 5.542 4.604 5.884 4.459 5.752 3.943 4.840

−0.342 −0.282 −0.369 −0.986 −0.353 −0.300 −0.303 −0.426 −0.342 0.230 −0.396 −0.820 −0.794 −0.488 −0.628 −0.458 −0.398 −0.314 −0.302 −0.553 −0.454 −0.514 −0.621 −0.362 −0.222 −0.539

2.088 3.241 2.946 7.121 3.081 3.699 2.870 7.706 2.913 3.143 2.486 6.950 7.056 4.253 6.560 5.095 3.247 2.823 3.046 4.410 2.247 6.144 4.811 5.357 2.688 5.416

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.017∗∗ 0.001∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.558 0.066∗ 0.144 0.000∗∗∗ 0.159 0.042∗∗ 0.047∗∗ 0.000∗∗∗ 0.004∗∗∗ 0.015∗∗ 0.037∗∗ 0.000∗∗∗ 0.071∗ 0.155 0.240 0.002∗∗∗ 0.000∗∗∗ 0.002∗∗∗ 0.000∗∗∗ 0.130 0.096∗ 0.006∗∗∗

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

Note: Symbols *, **, *** denote significance at the 10%, 5% and 1%, respectively. Mean and S.D. refer to the mean and standard deviation, respectively. SK is the skewness coefficient. EK is the excess kurtosis coefficient. JB is the p-value corresponding to the Jarque-Bera statistic. LB(24) is the p-value corresponding to the Ljung-Box Q statistic calculated with twenty-four lags. ARCH(4) is the p-value corresponding to the ARCH test calculated with four lags.

44

Table 6: Descriptive Statistics of the change in the log of exchange rate (ex), sub-sample periods (A) The change in the log of ex, pre global financial crisis periods (1998/1/1∼2007/07/31)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

0.006 0.003 0.011 −0.001 0.003 0.003 0.004 0.000 −0.005 0.000 −0.004 −0.033 0.002 0.004 0.000 0.022 0.000 0.003 0.003 −0.009 0.001 0.002 −0.004 0.014 0.003 −0.004

0.564 0.171 0.431 0.295 0.187 0.180 0.218 0.287 0.242 0.571 0.334 1.803 0.288 0.187 0.624 0.626 0.505 0.213 0.364 0.615 0.097 0.140 0.290 0.570 0.342 0.284

Min (%)

−2.918 −0.759 −2.265 −2.317 −0.854 −0.790 −0.926 −10.424 −1.703 −7.214 −1.984 −20.852 −1.493 −0.812 −3.952 −5.187 −6.352 −0.921 −3.941 −7.308 −0.447 −0.612 −2.779 −4.149 −1.771 −2.122

Max (%) 3.552 1.251 1.849 1.518 1.315 1.340 1.657 1.393 1.270 3.392 2.364 23.265 1.953 1.412 6.466 10.161 5.974 1.585 2.648 13.955 0.682 0.984 2.743 5.760 1.263 1.538

SK

−0.319 0.370 −0.122 −0.274 0.350 0.363 0.380 −19.068 −0.220 −0.987 −0.118 −1.115 0.254 0.397 0.905 2.126 0.002 0.338 −0.777 4.210 0.349 0.357 −0.202 1.057 −0.266 −0.181

EK 2.697 2.682 1.281 3.488 2.434 2.624 2.728 693.328 2.557 15.649 3.416 44.951 2.507 2.761 9.295 42.531 49.102 2.563 8.915 122.825 2.588 2.357 11.913 21.001 1.459 2.722

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.299 0.198 0.424 0.154 0.329 0.241 0.142 0.008∗∗∗ 0.621 0.000∗∗∗ 0.002∗∗∗ 0.000∗∗∗ 0.226 0.178 0.034∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.225 0.003∗∗∗ 0.000∗∗∗ 0.272 0.290 0.000∗∗∗ 0.000∗∗∗ 0.020∗∗ 0.443

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.001∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.999 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

(B) The change in the log of ex, global financial crisis periods (2007/08/01∼2010/12/31)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

0.012 −0.001 0.005 0.010 −0.002 −0.002 −0.003 0.000 −0.009 −0.071 −0.015 −0.008 0.000 −0.002 0.039 −0.034 0.005 −0.003 0.000 −0.006 0.000 −0.001 0.002 −0.012 −0.032 −0.005

0.992 0.185 0.797 0.364 0.211 0.200 0.236 0.179 0.313 1.603 0.471 0.657 0.314 0.201 0.881 1.102 0.348 0.230 0.581 0.448 0.096 0.173 0.280 0.407 0.615 0.442

Min (%)

−7.085 −1.298 −3.012 −2.169 −1.555 −1.400 −1.699 −1.204 −1.847 −23.647 −2.084 −6.893 −2.073 −1.428 −3.305 −6.683 −1.582 −1.602 −2.889 −1.793 −0.674 −1.214 −1.161 −4.157 −3.448 −2.279

Max (%)

SK

EK

6.764 1.124 3.875 1.574 1.375 1.178 1.450 0.972 1.683 18.665 3.115 5.046 1.930 1.177 4.724 9.666 1.601 1.419 3.259 2.294 0.596 1.030 1.670 1.989 2.154 2.020

−0.639 −0.203 −0.062 −0.199 −0.235 −0.205 −0.224 −0.119 −0.197 −0.620 0.252 −0.685 −0.157 −0.191 0.269 0.258 0.034 −0.180 0.131 0.055 −0.178 −0.173 0.268 −2.098 −0.548 0.168

8.134 5.179 1.597 2.797 6.643 5.081 5.773 4.573 3.191 84.416 4.104 20.846 4.622 5.360 2.187 13.399 2.303 5.251 3.860 1.862 5.604 5.299 3.163 19.555 3.157 2.474

(Continued on next page)

45

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.501 0.091∗ 0.114 0.580 0.095∗ 0.086∗ 0.145 0.157 0.553 0.000∗∗∗ 0.046∗∗ 0.002∗∗∗ 0.088∗ 0.088∗ 0.270 0.000∗∗∗ 0.064∗ 0.081∗ 0.040∗∗ 0.190 0.089∗ 0.101 0.002∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.262

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

Table 6 (cont.): Descriptive Statistics of the change in the log of exchange rate (ex), sub-sample periods (C) The change in the log of ex, post global financial crisis periods (2011/01/01∼2019/05/20)

Australia Belgium Canada China Denmark France Germany Greece Hong Kong Iceland India Indonesia Ireland Italy Japan South Korea Malaysia Netherlands Norway Philippines Portugal Spain Taiwan Thailand United Kingdom United States

Mean (%)

S.D. (%)

Min (%)

Max (%)

SK

EK

−0.011 0.001 −0.010 0.007 0.002 0.001 0.001 0.002 0.006 0.006 −0.013 −0.014 −0.001 0.002 −0.008 0.005 −0.007 0.001 −0.009 −0.001 0.001 0.002 0.004 0.006 0.000 0.011

0.506 0.166 0.432 0.262 0.171 0.163 0.196 0.145 0.226 0.457 0.416 0.435 0.245 0.172 0.553 0.419 0.379 0.176 0.438 0.273 0.085 0.146 0.228 0.238 0.425 0.310

−2.862 −1.342 −3.031 −1.750 −1.410 −1.316 −1.647 −1.198 −1.344 −3.692 −4.153 −4.320 −1.923 −1.394 −3.159 −2.384 −2.651 −1.421 −3.322 −1.460 −0.689 −1.175 −1.450 −1.048 −6.184 −1.883

2.235 0.878 2.152 1.490 0.851 0.846 1.045 0.861 1.139 3.723 3.022 4.271 1.137 0.934 4.654 2.460 3.026 0.923 3.005 1.387 0.471 0.783 1.760 1.491 2.086 1.804

−0.228 −0.298 −0.158 −0.078 −0.275 −0.331 −0.361 −0.268 0.067 0.067 −0.500 −0.041 −0.322 −0.328 0.255 −0.053 0.257 −0.313 −0.220 −0.260 −0.285 −0.287 0.137 0.033 −1.657 0.175

1.708 4.341 2.718 3.530 4.350 4.331 4.915 4.846 2.268 9.009 9.331 13.495 3.828 4.601 6.165 2.952 6.504 4.276 4.764 1.708 4.618 4.353 5.142 2.526 23.813 2.750

JB 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

LB(24) 0.326 0.677 0.757 0.645 0.595 0.757 0.713 0.797 0.518 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.662 0.769 0.625 0.213 0.034∗∗ 0.664 0.591 0.577 0.632 0.759 0.001∗∗∗ 0.044∗∗ 0.068∗ 0.423

ARCH(4) 0.000∗∗∗ 0.000∗∗∗ 0.007∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.001∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.001∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗

Note: Symbols *, **, *** denote significance at the 10%, 5% and 1%, respectively. Mean and S.D. refer to the mean and standard deviation, respectively. SK is the skewness coefficient. EK is the excess kurtosis coefficient. JB is the p-value corresponding to the Jarque-Bera statistic. LB(24) is the p-value corresponding to the Ljung-Box Q statistic calculated with twenty-four lags. ARCH(4) is the p-value corresponding to the ARCH test calculated with four lags.

46

Table 7: Results of the panel unit root tests for the full-sample periods (1998/1/1∼2019/05/20) Level

ex sp ex + ex − sp+ sp−

I (d)

1st Diff

IPS

ADF-Fisher

PP-Fisher

IPS

ADF-Fisher

PP-Fisher

−1.369∗ [0.086] −0.840 [0.200] −2.697∗∗∗ [0.004] −0.701 [0.242] 8.173 [1.000] 6.041 [1.000]

74.978∗∗ [0.020]

11395.300∗∗∗ [0.000]

–

–

–

0

46.673 [0.683]

13224.000∗∗∗ [0.000]

–

1

319.394∗∗∗ [0.000]

2502.240∗∗∗ [0.000]

–

–

0

224.764∗∗∗ [0.000]

8478.130∗∗∗ [0.000]

–

–

1

94.698∗∗∗ [0.000]

9805.490∗∗∗ [0.000]

–

–

1

72.099∗∗ [0.034]

2371.910∗∗∗ [0.000]

–

–

1

−194.325∗∗∗ [0.000] – −128.908∗∗∗ [0.000] −132.525∗∗∗ [0.000] −99.865∗∗∗ [0.000]

70.242∗∗ [0.047]

Note: Number in brackets is p-value. Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

47

Table 8: Results of the PANICCA for the full-sample periods (1998/1/1∼2019/05/20) Level common factor

1st Diff

idiosyncratic components

common factor

ADF

Pa

Pb

PMSB

ex

18.960 [1.000]

−0.578 [0.282]

−0.534 [0.297]

−0.493 [0.311]

sp

22.951 [1.000]

−0.394 [0.347]

−0.379 [0.353]

−0.363 [0.358]

ex +

−67.526∗∗∗ [0.000]

2.777 [0.997]

12.129 [1.000]

52.978 [1.000]

ex −

−60.713∗∗∗ [0.000]

2.146 [0.984]

9.147 [1.000]

38.979 [1.000]

sp+

55.641 [1.000]

4.706 [1.000]

25.204 [1.000]

134.993 [1.000]

sp−

38.840 [1.000]

4.295 [1.000]

21.398 [1.000]

106.620 [1.000]

2nd Diff

idiosyncratic components

I (d)

idiosyncratic components

Pa

Pb

−73.878∗∗∗ [0.000]

−73.404∗∗∗ [0.000]

−10.875∗∗∗ [0.000]

−1.571∗ [0.058]

–

1

−74.103∗∗∗ [0.000]

−10445.729∗∗∗ [0.000]

−166.854∗∗∗ [0.000]

−1.675∗∗ [0.047]

–

1

–

−7704.376∗∗∗ [0.000]

−183.331∗∗∗ [0.000]

−3.235∗∗∗ [0.001]

–

−14.069∗∗∗ [0.000]

−4.235∗∗∗ [0.000]

−74.082∗∗∗ [0.000]

−11395.235∗∗∗ [0.000]

−237.941∗∗∗ [0.000]

−3.479∗∗∗ [0.000]

−72.946∗∗∗ [0.000]

−14276.322∗∗∗ [0.000]

−167.796∗∗∗ [0.000]

−1.244 [0.107]

ADF

PMSB

−1.260 [0.104]

PMSB

1

−1.538∗ [0.062] –

1

−2.371∗∗∗ [0.009]

2

Note: Number in brackets is p-value. Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

48

2

Table 9: Results of the cross-sectional dependence and homogeneity test for the full-sample periods (1998/1/1∼2019/05/20) Regressions for Symmetric Granger non-causality test Ho : ex ; sp

Ho : sp ; ex

Regressions for Asymmetric Granger non- causality test Ho : ex + ; sp

Ho : ex − ; sp

Ho : sp+ ; ex

Ho : sp− ; ex

The results of the cross-sectional dependence CDBP

281093.420∗∗∗

326624.435∗∗∗

278162.101∗∗∗

322914.901∗∗∗

CDlm

11012.643∗∗∗

12798.517∗∗∗

10897.668∗∗∗

12653.017∗∗∗

447.514∗∗∗

60.022∗∗∗

446.129∗∗∗

59.627∗∗∗

CDP

The results of slope homogeneity ˜ ∆

4526.706∗∗∗

4340.530∗∗∗

7159.715∗∗∗

5290.380∗∗∗

˜ adj ∆

4527.924∗∗∗

4341.698∗∗∗

7162.283∗∗∗

5292.277∗∗∗

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

49

50

24

24

10

18

24

2

4

24

24

7

9

Malaysia

Netherlands

Norway

Philippines

Portugal

Spain

Taiwan

Thailand

United Kingdom

United States

0.046

6.172∗∗

171.004

260.534

1%

0.022

7.631∗∗

0.063 0.002

22.865∗ 21.055∗∗∗

0.000

101.636∗∗∗

260.534

171.004

132.744

393.900∗∗∗

0.522

0.084

34.055∗ 6.157

0.122

0.042

9.923∗∗ 32.218

0.009

0.271

0.000

9.389∗∗∗

27.734

51.404∗∗∗

0.140 0.001

28.925∗∗∗

0.300

31.506

27.092

0.096

0.055

36.014∗

22.460∗

0.474

15.704

0.000

0.000

37.943∗∗∗

62.694∗∗∗

0.000

25.035∗∗∗

0.058

0.001

31.278∗∗∗

34.515∗

0.470

0.000

133.099∗∗∗ 9.673

0.367 0.025

4.301

p-value

11.121∗∗

Wald Satistics

Ho : sp ; ex

11

24

24

23

4

16

24

18

20

24

24

24

20

24

24

20

24

23

15

20

20

20

24

24

20

8

Lags

271.918

180.878

140.610

100.979

18.556∗

24.769

0.070

0.418

0.738

0.099 19.265

0.990 32.044∗

0.286

0.276

0.148

0.431

0.023

0.000

0.010

0.695

0.501

0.389

0.502

0.992

0.000

0.768

0.350

0.459

0.272

0.110

0.316

0.535

0.801

p-value

0.293

18.677

27.637

24.234

20.431

39.659∗∗

61.054∗∗∗

42.990∗∗∗

16.348

23.324

25.303

19.303

10.565

60.572∗∗∗

10.785

21.818

19.982

23.347

32.739

26.750

18.799

4.580

Wald Satistics

Ho : ex + ; sp

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

132.744

5%

139.524∗

0.171

0.054

12.819

0.387

13.825∗

0.620

25.353

21.320

0.768

0.016

1.825

0.035

41.263∗∗

0.005

45.766∗∗∗

30.250∗∗

0.002

48.425∗∗∗ 0.252

0.000

74.067∗∗∗

12.518

0.456

0.889

0.392

0.102

5.716

8.001

25.250

23.478

0.930

0.001

49.996∗∗∗

13.105

0.646

0.552

0.845

0.268

0.520

0.714

0.873

2.099

2.701

12.253

9.125

7.995

0.702

0.018

11.887∗∗ 2.186

p-value

Wald Satistics

10%

Bootstrap critical value

λ statistic

15

South Korea

22

Iceland

Japan

23

Hong Kong

6

2

Greece

Italy

3

Germany

14

6

France

Ireland

10

Denmark

24

10

China

Indonesia

11

Canada

16

4

Belgium

India

4

Australia

Lags

Ho : ex ; sp

Symmetric Granger non-causality test

296.075

195.652

152.128

61.275

8.017

21.330

22.794

18.513

1.899

17.557

30.231

21.641

21.284

20.931

34.494∗

41.420∗∗

19.374

23.496

0.712

0.619

0.532

0.729

0.754

0.350

0.177

0.248

0.381

0.643

0.076

0.015

0.498

0.491

0.118

0.072

32.357

0.931

29.893∗

0.487

0.077

0.886

0.635

0.288

0.129

0.341

0.797

0.218

p-value

14.646

22.563

23.358∗

12.806

17.277

23.026

31.921

26.242

14.631

10.717

Wald Satistics

Ho : ex − ; sp

243.581

162.092

126.072

166.057∗∗

50.352∗∗∗

42.792∗∗

37.373∗∗

34.868∗

7.902∗

16.377

42.105∗∗

21.312

26.673

36.296∗

42.966∗∗∗

23.013

22.478

35.341∗

43.267∗∗∗

40.719∗∗∗

23.683

38.439∗∗

21.839

31.176∗

26.467

27.874

30.238

54.197∗∗∗

20.587

6.862

Wald Satistics

0.000

0.011

0.040

0.054

0.095

0.427

0.013

0.264

0.145

0.051

0.010

0.519

0.315

0.064

0.009

0.004

0.480

0.023

0.112

0.053

0.151

0.112

0.177

0.000

0.422

0.552

p-value

Ho : sp+ ; ex

Asymmetric Granger non-causality test

Table 10: Results of the EK panel Granger non-causality test of full-sample periods

316.573

205.266

157.975

167.986∗

43.588∗∗∗

29.791

23.588

33.869∗

2.283

11.630

21.609

38.375∗∗∗

20.492

34.134∗

21.995

21.553

15.575

24.538

23.681

20.102

105.366∗∗∗

32.655∗

11.282

24.322

21.413

21.103

20.679

88.696∗∗∗

16.151

4.757

Wald Satistics

0.000

0.192

0.485

0.067

0.684

0.769

0.603

0.003

0.428

0.082

0.580

0.606

0.743

0.431

0.480

0.452

0.000

0.087

0.732

0.229

0.373

0.391

0.658

0.000

0.707

0.783

p-value

Ho : sp− ; ex

51

23

8

Hong Kong

Iceland

10.171

6

14

24

2

2

24

24

9

7

Malaysia

Netherlands

Norway

Philippines

Portugal

Spain

Taiwan

Thailand

United Kingdom

United States

183.435

278.379

1%

0.337

0.548

0.463

0.375

0.000

27.777∗∗∗

0.036 0.000

17.957∗∗ 28.050∗∗∗

238.355

158.896

123.780

226.401∗∗

0.195

29.696

0.127

0.001

14.655∗∗∗ 32.002

0.182

3.405

0.283 0.009

43.395∗∗∗

0.009

17.011∗∗∗ 16.517

0.126

0.088

33.831∗ 32.039

0.001 0.081

24.083∗∗∗ 19.325∗

0.640

0.047

36.660∗∗ 3.388

0.667

0.000

28.177∗∗∗ 12.154

0.164

29.496

0.188

0.000

11.255

0.021

20.659∗∗∗

0.000

36.388∗∗∗ 9.693∗∗

0.004

40.945∗∗∗ 0.207

0.029

17.067∗∗

5.897

p-value

Wald Satistics

Ho : sp ; ex

15

23

24

24

19

16

24

14

20

24

24

15

6

24

24

20

24

23

16

20

20

20

10

7

20

15

Lags

304.579

203.301

158.733

79.456

23.684∗

9.868

23.068

29.219

25.204

23.315

26.904

16.818

22.956

39.122∗∗

41.449∗∗

16.479

6.959

24.245

0.071

0.992

0.516

0.212

0.154

0.106

0.309

0.266

0.291

0.027

0.015

0.351

0.325

0.448

0.422

0.011

24.702

0.555

37.297∗∗

0.324

0.726

0.384

0.297

0.708

0.677

0.160

0.194

0.093

p-value

22.406

25.518

12.255

21.216

22.834

16.146

7.510

10.547

25.208

22.611∗

Wald Satistics

Ho : ex + ; sp

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

142.257

5%

93.998

7.948

7.866

23.970

25.580

0.138

0.001

14.621∗∗∗ 3.967

0.111

0.023

26.391∗∗

32.700

0.780

3.225

0.298 0.034

27.134

0.386

0.296

0.999

0.501

38.095∗∗

12.765

10%

Bootstrap critical value

λ statistic

24

24

South Korea

6

12

Italy

Japan

23.326

0.809

0.645

0.001

49.913∗∗∗ 6.017

0.001

25.654∗∗∗

7.275

8

Greece

0.620

0.738

0.177

0.229

0.534

0.557

0.350

p-value

0.957

0.202

2

Germany

1.986

5

4

France

4.936

Ireland

3

Denmark

5.628

24

4

China

6.048

Indonesia

7

Canada

18.467

15

20

Belgium

8.915

Wald Satistics

India

8

Australia

Lags

Ho : ex ; sp

Symmetric Granger non-causality test

366.269

243.332

189.614

105.752

14.058

0.521

0.426

0.071 23.610

0.547 34.817∗

0.577

0.053

0.494

0.009

0.300

0.565

0.303

0.993

0.146

0.297

0.109

0.068

0.375

0.081

0.000

0.784

0.766

0.202

0.165

0.124

0.516

0.265

p-value

22.534

17.185

26.062∗

23.438

29.479∗∗∗

22.768

22.235

27.022

4.864

9.533

27.159

32.761

30.136∗

25.572

33.024∗

65.744∗∗∗

14.870

15.187

24.978

14.190

11.347

19.083

17.958

Wald Satistics

Ho : ex − ; sp

249.825

166.795

130.568

124.543

35.715∗∗∗

21.349

39.086∗∗

26.327

36.653∗∗∗

6.988

49.434∗∗∗

16.296

22.059

36.209∗

24.327

16.282

11.845∗

17.572

41.905∗∗

34.057∗∗

30.336

36.103∗∗

19.050

31.699∗∗

31.863∗∗

26.799

9.306

11.597

25.336

23.716∗

Wald Satistics

0.002

0.560

0.027

0.337

0.009

0.973

0.002

0.296

0.337

0.052

0.443

0.364

0.066

0.823

0.013

0.026

0.174

0.040

0.266

0.047

0.045

0.141

0.503

0.115

0.189

0.070

p-value

Ho : sp+ ; ex

Asymmetric Granger non-causality test

252.493

167.549

131.105

107.579

36.344∗∗∗

35.919∗∗

0.002

0.042

0.426

0.044

36.933∗∗ 24.636

0.156

0.257

0.110

0.622

0.164

0.002

0.447

0.773

0.047

0.779

0.356

0.651

0.002

0.350

0.684

0.292

0.577

0.658

0.101

0.007

0.273

0.167

p-value

25.146

19.236

32.709

11.809

26.045

48.929∗∗∗

24.252

10.701

12.745∗∗

18.474

25.945

17.023

49.715∗∗∗

25.002

12.843

22.938

18.155

16.917

15.965

19.334∗∗∗

23.330

20.137

Wald Satistics

Ho : sp− ; ex

Table 11: Results of the Emirmahmutoglu and Kose (2011) panel Granger non-causality test for the pre-GFC periods

52

14

4

14

22

18

1

1

2

15

5

4

South Korea

Malaysia

Netherlands

Norway

Philippines

Portugal

Spain

Taiwan

Thailand

United Kingdom

United States

178.080

270.983

1%

0.154

0.477

0.517 0.000

76.486∗∗∗

331.858

213.311

165.192

213.524∗∗

0.463

0.669

0.323

0.640

0.642

4.630

12.131

2.262

0.219

0.217

0.479

0.004

43.316∗∗∗ 17.653

0.403

14.644

0.061 0.080

8.326∗

0.007

0.252

0.285

0.156

22.957∗

15.941∗∗∗

1.315

1.141

20.423

0.124

0.000

51.559∗∗∗ 5.762

0.802

0.063

0.551

0.073

0.355

0.494

3.218∗

0.062

3.490∗ 1.411

0.824

0.000

65.203∗∗∗ 0.049

0.335

0.096

p-value

0.931

21.229∗

Wald Satistics

Ho : sp ; ex

4

23

18

23

19

20

23

22

20

16

20

15

20

21

24

16

21

4

19

19

20

19

1

11

19

14

Lags

346.505

232.522

182.549

157.527

0.376

0.031 4.227

0.369 37.160∗∗

0.001

51.087∗∗∗ 19.373

0.555

0.706 17.518

16.167

0.124 0.016

29.788

0.417

0.336

0.012

0.000

0.216

0.374

0.163

0.237

0.607

39.823∗∗

20.673

17.802

36.793∗∗

62.962∗∗∗

24.637

22.442

30.690

19.635

18.662

0.000

0.015

34.646∗∗ 35.953∗∗∗

0.689

0.390

15.526

21.130

0.480

0.014

18.650

0.679

6.075∗∗

0.811

0.025

p-value

8.382

13.512

26.169∗∗

Wald Satistics

Ho : ex + ; sp

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

137.823

5%

168.438∗

6.669

4.520

14.108

0.192

0.078

3.298

0.521

3.100∗

0.101

0.164

0.413

25.959

28.372

0.062

38.554∗∗∗

22.872∗

0.000

90.849∗∗∗ 0.316

0.000

3.858∗∗

4.734

0.979 0.050

0.001

0.695

0.051

7.783∗

11.790

0.986

0.044

4.051∗∗ 8.024

0.395

0.440

0.141

0.432

0.314

0.372

0.724

0.597

3.922

0.618

1.014

7.570

0.369

0.043

24.270∗∗ 0.807

p-value

Wald Satistics

10%

Bootstrap critical value

λ statistic

5

19

Iceland

Japan

1

Hong Kong

1

1

Greece

1

1

Germany

Italy

2

France

Ireland

1

Denmark

15

1

China

Indonesia

7

Canada

3

1

Belgium

India

14

Australia

Lags

Ho : ex ; sp

Symmetric Granger non-causality test

0.036

346.879

233.083

183.266

138.351

0.578

26.103

17.348

37.344∗∗

32.803∗∗

15.985

32.052∗

32.432∗

30.851∗

17.497

46.660∗∗∗

46.203∗∗∗

0.965

0.296

0.499

0.030

0.025

0.718

0.099

0.070

0.057

0.354

0.001

0.000

0.060

34.056∗∗ 30.640∗

0.112

0.016

0.558

0.273

0.106

32.646

30.410∗∗

19.427

5.144

26.962

0.048

0.014

36.420∗∗ 30.300∗∗

0.231

0.137

0.215

23.145

2.213

14.343

0.141 0.066

19.652

p-value

29.019∗

Wald Satistics

Ho : ex − ; sp

291.889

197.230

154.997

149.804

22.109∗∗∗

42.151∗∗∗

21.787

42.468∗∗∗

0.000

0.009

0.242

0.008

0.181

0.003

41.531∗∗∗ 24.402

0.182

0.002

0.161

0.079

0.013

0.673

0.155

28.943

46.190∗∗∗

26.135

24.490∗

36.736∗∗

12.075

26.330

0.060

0.000 31.885∗

0.008 61.046∗∗∗

0.786

0.673

0.270

0.397

0.301

0.164

0.974

0.014

0.390

0.841

p-value

32.581∗∗∗

15.703

2.342

22.297

19.962

22.754

24.898

0.001

23.798∗∗

20.079

8.850

Wald Satistics

Ho : sp+ ; ex

Asymmetric Granger non-causality test

14.656

458.670

293.774

225.271

199.944

33.987∗∗∗

40.579∗∗

11.397

42.342∗∗∗

20.173

18.666

40.635∗∗

45.903∗∗∗

20.525

29.119∗∗

22.998

27.465∗∗

18.102

34.532∗∗

41.943∗∗

19.080

94.094∗∗∗

2.675

20.011

22.291

23.214

10.392

0.100

51.429∗∗∗

20.376

0.000

0.013

0.877

0.008

0.384

0.544

0.013

0.002

0.426

0.023

0.289

0.025

0.581

0.032

0.013

0.265

0.000

0.614

0.394

0.270

0.278

0.943

0.751

0.000

0.372

0.402

p-value

Ho : sp− ; ex Wald Satistics

Table 12: Results of the Emirmahmutoglu and Kose (2011) panel Granger non-causality test for the GFC periods

53

3

8

4

3

2

10

5

10

8

24

6

8

Japan

South Korea

Malaysia

Netherlands

Norway

Philippines

Portugal

Spain

Taiwan

Thailand

United Kingdom

United States

0.021 0.047

14.856∗∗

15.722∗∗

160.796

246.619

1%

0.000

26.941∗∗∗ 8.123∗∗

244.208

160.128

123.949

290.784∗∗∗

74.227∗∗∗

0.000

0.083

0.058

35.749∗ 11.184∗

0.458 0.024

9.802 17.618∗∗

0.032

12.195∗∗

0.000

30.653∗∗∗ 0.264

0.000

18.763∗∗∗ 12.324

0.162

0.436

0.044

6.542

7.978

0.147

0.072

14.397∗ 5.371

0.250

0.424

0.722

9.036

0.638

1.329

0.226

0.002

14.852∗∗∗ 6.932

0.101

10.626

0.155

0.000

51.281∗∗∗ 0.000

0.033

13.754∗∗ 2.024

0.053

19.469∗

33.779∗∗∗

p-value

Wald Satistics

Ho : sp ; ex

8

6

24

8

15

12

10

6

5

7

10

7

10

5

18

11

24

3

19

10

11

13

24

9

6

11

Lags

6.782

0.237

261.015

171.484

132.913

85.232

12.668

7.024

17.305

7.777

12.178

0.124

0.319

0.835

0.456

0.666

0.700

0.029 9.033

0.182

20.007∗∗

0.364

0.067

0.323

0.003

0.778

8.860

5.448

13.216∗

11.456

21.964∗∗∗

6.428

0.539

0.020

22.668∗∗ 16.770

0.792

0.000

0.814

0.484

0.682

0.539

0.896

0.300

0.559

0.398

p-value

18.215

23.568∗∗∗

13.463

9.515

8.350

11.864

15.776

10.663

4.885

11.549

Wald Satistics

Ho : ex + ; sp

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

124.367

5%

103.818

0.554

0.103

0.834

0.927

0.246

0.343

22.430

13.271

5.774

1.380

12.619

2.143

0.540

0.005

14.922∗∗∗ 2.157

0.000 0.096

13.484∗

0.363

31.144∗∗∗

10%

Bootstrap critical value

λ statistic

0.744

5.458

0.107

1.239

13.152

5

0.750

0.530

3

4.251

0.394

Ireland

1

Iceland

0.000

19.551∗∗∗

Italy

3

Hong Kong

0.424

4.931

0.628

8

5

Greece

1.741

0.774

Indonesia

3

Germany

3.269

0.818

0.654

0.607

0.553

0.433

p-value

7

6

France

2.221

0.201

0.999

4.928

11.123

Wald Satistics

India

1

5

2

Canada

Denmark

6

Belgium

China

11

Australia

Lags

Ho : ex ; sp

Symmetric Granger non-causality test

264.504

173.552

133.830

90.294

3.915

21.654∗∗∗

20.948

6.731

15.243

9.335

12.275

15.062∗∗

3.988

4.666

26.283∗∗∗

29.622∗∗∗

9.873

4.552

29.388∗∗

8.834

25.552

3.594

26.209

5.725

6.562

9.418

31.365

10.469

2.706

15.832

Wald Satistics

0.865

0.001

0.642

0.566

0.434

0.674

0.267

0.020

0.551

0.701

0.003

0.000

0.452

0.473

0.044

0.637

0.376

0.309

0.124

0.838

0.833

0.741

0.144

0.314

0.845

0.148

p-value

Ho : ex − ; sp

254.799

167.462

130.003

121.638

19.347∗∗

8.615

22.792

8.384

16.500

17.651

6.925

25.337∗∗∗

16.887∗∗∗

8.113

17.316∗

7.848

13.559

14.056∗∗

16.947

6.690

30.366

1.660

24.815

19.227∗∗

17.253

26.935∗∗

31.713

22.999∗∗∗

14.306∗∗

6.677

Wald Satistics

0.013

0.196

0.532

0.397

0.350

0.127

0.733

0.000

0.005

0.323

0.068

0.346

0.194

0.015

0.527

0.824

0.173

0.646

0.167

0.037

0.101

0.013

0.134

0.006

0.026

0.825

p-value

Ho : sp+ ; ex

Asymmetric Granger non-causality test

256.595

170.742

132.200

124.817

38.643∗∗∗

4.065

28.686

14.946∗

10.147

12.584

14.336

19.327∗∗∗

2.052

9.415

4.483

17.282∗∗

3.004

7.167

17.028

6.574

30.364

3.375

0.000

0.668

0.232

0.060

0.810

0.400

0.158

0.004

0.842

0.224

0.923

0.016

0.981

0.209

0.521

0.832

0.173

0.337

0.601 0.006

8.285

0.948

0.107

0.243

0.000

0.910

0.008

p-value

37.764∗∗∗

4.622

19.565

28.409

39.851∗∗∗

2.099

25.363∗∗∗

Wald Satistics

Ho : sp− ; ex

Table 13: Results of the Emirmahmutoglu and Kose (2011) panel Granger non-causality test for the post-GFC periods

Table 14: Results of the Emirmahmutoglu and Kose (2011) panel Granger non-causality test for developed and emerging economies Symmetric Granger causality test Ho : ex ; sp

Ho : sp ; ex

Asymmetric Granger causality test Ho : ex + ; sp

Ho : ex − ; sp

Ho : sp+ ; ex

Ho : sp− ; ex

83.440

42.845

118.743∗

154.749∗

Developed economies λ statistic

110.837∗

373.577∗∗∗

10% c.v.

97.197

105.674

102.745

116.561

97.096

133.520

5% c.v.

125.808

136.134

131.905

150.435

124.477

173.039

1% c.v.

192.230

207.818

200.070

230.356

188.552

266.304

Emerging economies λ statistic

28.687

20.323

17.940

17.460

43.688∗∗

11.405

10% c.v.

32.676

27.993

37.522

34.972

29.416

29.516

5% c.v.

42.137

36.294

47.799

44.887

37.697

37.562

1% c.v.

64.149

54.703

71.715

67.556

56.908

55.705

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

54

Table 15: Results of the Hatemi-J (2012) panel Granger non-causality test Symmetric Granger non-causality test Ho : ex ; sp

Ho : sp ; ex

Asymmetric Granger non-causality test Ho : ex + ; sp

Ho : ex − ; sp

Ho : sp+ ; ex

Ho : sp− ; ex

(A) Full sample periods (1998/1/1∼2019/05/20) MWALD

356.494∗∗∗

570.325∗∗∗

256.304

104.771

176.911

264.212

10% c.v.

87.343

60.110

629.685

674.726

1209.704

1147.206

5% c.v.

93.495

65.139

647.435

693.587

1258.717

1188.188

1% c.v.

108.871

75.874

681.109

726.117

1346.713

1271.332

161.730

67.604

80.430

120.688

(B1) Pre global financial crisis periods (1998/1/1∼2007/07/31) MWALD

211.244∗∗∗

219.805∗∗∗

10% c.v.

64.549

57.585

223.571

207.255

935.744

866.423

5% c.v.

69.953

62.467

233.810

216.639

977.843

900.652

1% c.v.

81.810

72.394

253.260

236.443

1052.754

972.967

97.557

115.858

127.040

286.867

(B2) Global financial crisis periods (2007/08/01∼2010/12/31) MWALD

169.567∗∗∗

560.836∗∗∗

10% c.v.

61.868

56.767

269.677

248.212

353.442

505.017

5% c.v.

67.088

61.939

283.714

260.719

384.061

536.989

1% c.v.

77.512

72.812

310.216

285.434

448.951

598.988

104.511

93.389

88.183

128.538

(B3) Post global financial crisis periods (2011/01/01∼2019/05/20) MWALD

145.902∗∗∗

268.620∗∗∗

10% c.v.

57.230

43.400

252.626

318.222

453.965

492.415

5% c.v.

62.445

47.435

263.558

330.637

475.958

513.361

1% c.v.

72.685

56.165

285.600

354.074

528.277

555.678

138.253

52.525

128.489

256.395

(C1) Developed economies MWALD

189.163∗∗∗

495.210∗∗∗

10% c.v.

59.346

50.867

332.852

393.456

602.504

660.446

5% c.v.

64.511

55.761

345.913

407.039

636.119

694.726

1% c.v.

75.988

65.132

370.240

432.048

703.744

765.080

18.513∗∗

137.059∗∗∗

34.836

22.491

9.868

(C2) Emerging economies MWALD

171.351∗∗∗

10% c.v.

37.932

13.677

111.269

100.594

130.577

111.995

5% c.v.

42.717

15.861

118.923

107.526

138.687

118.997

1% c.v.

52.149

20.693

132.439

122.166

156.103

133.144

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

55

Table 16: Summary of the panel Granger non-causality test Symmetric Granger causality test Ho : ex ; sp

Ho : sp ; ex

Asymmetric Granger causality test Ho : ex + ; sp

Ho : ex − ; sp

Ho : sp+ ; ex

Ho : sp− ; ex

166.057∗∗

167.986∗

The Emirmahmutoglu and Kose (2011) λ statistic using the log of data 139.524∗

393.900∗∗∗

100.979

61.275

93.998

226.401∗∗

79.456

105.752

124.543

107.579

168.438∗

213.524∗∗

157.527

138.351

149.804

199.944

The Post-GFC

103.818

290.784∗∗∗

85.232

90.294

121.638

124.817

Developed economies

110.837∗

373.577∗∗∗

83.440

42.845

118.743∗

154.749∗

Emerging economies

28.687

17.940

17.460

Full-sample The Pre-GFC The GFC

20.323

43.688∗∗

11.405

The Hatemi-J (2012) MWALD test using the log of data Full-sample

356.494∗∗∗

570.325∗∗∗

256.304

104.771

176.911

264.212

The Pre-GFC

211.244∗∗∗

219.805∗∗∗

161.730

67.604

80.430

120.688

The GFC

169.567∗∗∗

560.836∗∗∗

97.557

115.858

127.040

286.867

The Post-GFC

145.902∗∗∗

268.620∗∗∗

104.511

93.389

88.183

128.538

Developed economies

189.163∗∗∗

495.210∗∗∗

138.253

52.525

128.489

256.395

Emerging economies

171.351∗∗∗

18.513∗∗

137.059∗∗∗

34.836

22.491

9.868

The Emirmahmutoglu and Kose (2011) λ statistic using raw data Full-sample

145.708∗

459.109∗∗∗

93.071

59.604

163.438∗∗

154.475∗

The Pre-GFC

78.309

256.476∗∗∗

64.908

63.808

118.784

136.162∗

The GFC

159.828∗∗

200.084∗∗

135.780∗

104.286

111.477

142.185

The Post-GFC

111.870

296.516∗∗∗

83.650

80.333

139.727∗

115.429

Developed economies

110.928∗

436.963∗∗∗

76.884

40.650

129.703∗∗

143.136∗∗

Emerging economies

33.321∗

13.970

18.288

35.296∗∗

11.215

22.616

The Hatemi-J (2012) MWALD test using raw data Full-sample

175.976∗∗∗

471.564∗∗∗

122.945

66.737

158.380

187.214

The Pre-GFC

140.612∗∗∗

235.531∗∗∗

115.352

44.604

76.784

125.078

The GFC

143.424∗∗∗

505.320∗∗∗

65.128

84.663

173.272

141.483

The Post-GFC

117.516∗∗∗

175.952∗∗∗

87.131

75.284

71.202

85.708

Developed economies

120.188∗∗∗

435.228∗∗∗

82.305

39.705

128.893

205.347

Emerging economies

57.242∗∗∗

43.452

18.454

15.073

4.034

11.026

Note: Symbols ***, **, and * denote rejection at the significance levels of 1%, 5%, and 10%, respectively.

56

► This paper examines re-examine the exchange rate-stock price nexus for a group of advanced and emerging countries ►We adopt the symmetric and asymmetric bootstrap panel Granger non-causality tests. ► Empirical results show that the stock prices are helpful for predicting the exchange rates, but not vice versa. ► The results show weak evidence in support of unidirectional asymmetric causality running from exchange rates to stock prices, and vice versa. ► The empirical results of the Hatemi-J (2012) symmetric panel Granger non-causality tests show that there is a causation from stock prices to exchange rates, and vice versa.