International contagion through financial versus non-financial firms

Economic Modelling 59 (2016) 143–163 Contents lists available at ScienceDirect Economic Modelling journal homepage: www.elsevier.com/locate/ecmod I...

Download PDF

3MB Sizes 2 Downloads 80 Views

Report

PDF Reader
Full Text

Economic Modelling 59 (2016) 143–163

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

International contagion through ﬁnancial versus non-ﬁnancial ﬁrms Md Akhtaruzzaman ⁎, Abul Shamsuddin Newcastle Business School, University of Newcastle, Callaghan, NSW 2308, Australia

a r t i c l e

i n f o

Article history: Received 3 March 2016 Received revised form 3 July 2016 Accepted 5 July 2016 Available online xxxx JEL classiﬁcation: G12 G21 G22 G23

a b s t r a c t The role of ﬁnancial ﬁrms in the transmission of ﬁnancial shocks across countries is well recognized in the literature. However, contagion through non-ﬁnancial ﬁrms has not received much attention. This study examines the role of ﬁnancial vis-à-vis non-ﬁnancial ﬁrms in transmitting shocks across countries using a dynamic conditional correlation analysis. We provide empirical evidence from a sample of 49 countries. A novel ﬁnding of our study is that non-ﬁnancial ﬁrms play a more pronounced role in the cross-market transmission of shocks than ﬁnancial ﬁrms. Financial contagion is positively related to the level of equity market development and bilateral trade intensity. It is higher during periods of US economic downturns and ﬁnancial crises. Given that the extent of international contagion varies across economic states and is more prevalent in the non-ﬁnancial than in the ﬁnancial sector, this study has implications for global sector rotation strategies. © 2016 Elsevier B.V. All rights reserved.

Keywords: Financial contagion Non-ﬁnancial ﬁrms Financial ﬁrms Business cycle Dynamic conditional correlation

1. Introduction A large body of literature on ﬁnancial contagion has emerged since the early 1990s. However, there is still no consensus regarding the deﬁnition of contagion.1 The World Bank cites three deﬁnitions of contagion—broad deﬁnition, restrictive deﬁnition, and very restrictive deﬁnition.2 These deﬁnitions are not mutually exclusive. Under the broad deﬁnition, contagion is the cross-country transmission of shocks (Calvo and Reinhart, 1996). In the restrictive deﬁnition, contagion is the cross-country transmission of shocks beyond fundamental economic linkages and beyond common shocks (Bekaert et al., 2005). Under the very restrictive deﬁnition, contagion occurs when crosscountry correlations increase during crisis periods compared to tranquil periods (Forbes and Rigobon, 2002). This study adopts the restrictive deﬁnition of ﬁnancial contagion, which refers to the excess crossmarket correlation that cannot be explained by economic fundamentals. This deﬁnition allows us to avoid erroneously interpreting fundamental economic interdependence as contagion and accommodates the

⁎ Corresponding author. E-mail address: [email protected] (M. Akhtaruzzaman). See Pericoli and Sbracia (2003) and Cheung et al. (2009) for a survey of various deﬁnitions of ﬁnancial contagion. 2 These deﬁnitions of contagion are available at the World Bank website: http://go. worldbank.org/JIBDRK3YC0. 1

http://dx.doi.org/10.1016/j.econmod.2016.07.003 0264-9993/© 2016 Elsevier B.V. All rights reserved.

possibility that contagion may occur not only in a crisis state but also in a tranquil state. We operationalize this measure of contagion within a time-varying conditional correlation model for a sample of 49 countries. This research setting is ﬂexible enough to determine how the extent of cross-market correlation varies over different economic states and across countries with differing degrees of fundamental economic linkages. Thus, our ﬁndings may be of interest to a wider audience regardless what deﬁnition of contagion they use. The unique role of ﬁnancial ﬁrms in the transmission of shocks across markets has been well recognized in the literature (Kaufman, 1994). It has been argued that contagion through ﬁnancial ﬁrms occurs rapidly and destabilizes the ﬁnancial system in the midst of volatile capital ﬂows. World capital ﬂows increased from b7% of world GDP in 1998 to over 20% in 2007 (Milesi-Ferretti and Tille, 2011). However, during the global ﬁnancial crisis (GFC), world capital ﬂows (in US dollars) declined by 44% in December 2008 from its peak in 2007 (Tong and Wei, 2011). Negative capital shocks can lead to an immediate liquidity crunch in ﬁnancial ﬁrms, which can shake market conﬁdence in other ﬁnancial ﬁrms, and induce investors to withdraw money and force those ﬁrms to liquidate assets at a price below their intrinsic value. Thus, liquidity risk may lead to solvency risk and vice versa. Transmission of shocks across ﬁnancial ﬁrms can be intensiﬁed by their linkages within and across national boundaries. Cetorelli and Goldberg (2011) observes that global banks played an important role in the transmission of the GFC to emerging countries, and Kalemli-

144

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

Ozcan et al. (2013) provide similar ﬁndings for developed countries. The role of ﬁnancial ﬁrms in transmitting shocks across countries is also documented in Batten et al. (2013), Kollmann (2013), and Akhtaruzzaman et al. (2014). Therefore, the prevailing notion is that ﬁnancial ﬁrms are the main conduit for ﬁnancial contagion from country to country. To verify this idea, we extend our analysis to examine cross-market contagion through non-ﬁnancial ﬁrms. Domestic non-ﬁnancial ﬁrms are directly interconnected to their foreign counterparts through international trade. Trade linkages lead to business cycle synchronization across countries (see Frankel and Rose, 1998; Kose and Yi, 2001) and increase output co-movement at the sect or level (Giovanni and Levchenko, 2010). This interdependence in real economic activities gives rise to interdependence in non-ﬁnancial stock returns, which has been substantially strengthened by, ﬁrstly, the liberalization of international trade, and, secondly, the rapid rise of multinational corporations. Multinational corporations accounted for about one-third of gross output in many developed countries (Cravino and Levchenko, 2015) and US multinational corporations alone comprised N 10% in several countries (Desai and Foley, 2006). There are a number of interrelated channels through which multinational corporations affect the co-movement of economic activities across countries. First, multinational corporations play an important role in increasing vertical production linkages across countries, which in turn magniﬁes the impact of bilateral trade on output co-movement (Burstein et al., 2008). Second, investment rates and returns of foreign afﬁliates are strongly correlated with those of their parent companies (Desai and Foley, 2006). Third, the sales growth of the headquarter is strongly associated with the sales growth of foreign afﬁliates. Thus, the role of multinational corporations, along with the move towards free trade in recent decades, has facilitated the transmission of demand and supply shocks across countries through nonﬁnancial ﬁrms. This is also echoed in Dungey and Gajurel (2014) ﬁnding that the extent of contagion through ﬁnancial sector stocks was less than that of the aggregate equity market during the GFC, which gives us an empirical motivation to examine the role of non-ﬁnancial ﬁrms in transmitting contagion. This study contributes to the literature in a number of ways. First, we use a large sample of 49 countries to investigate the role of nonﬁnancial ﬁrms in transmitting ﬁnancial shocks from the US to developed, emerging, and frontier economies.3 International contagion through the ﬁnancial sector is also examined to compare with its nonﬁnancial counterpart. Shocks are measured in terms of unexpected sectoral stock returns denominated in local currency. Engle's (2002) dynamic conditional correlations generalized autoregressive conditional heteroskedasticity (hereafter DCC-GARCH) model is used to estimate time-varying correlations. Given the predominant role of the US in the world economy, this study examines the correlation of the US sectoral return with its non-US counterparts. Correlations observed among sectoral returns of non-US stock markets may primarily manifest those markets' correlations to the US stock market (Dimitriou et al., 2013; Kim et al., 2015; Moore and Wang, 2014). Our approach helps avoid capturing spurious cross-market correlations and conduct DCC analysis in a parsimonious framework.4 Our analysis of sector-speciﬁc contagion enables us to test whether contagion occurs primarily

3 As per the FTSE country classiﬁcation, there are three broad categories of countries: developed, emerging and frontier. Financial contagion across these markets has been recently examined by Beirne and Gieck (2014) and (Mollah et al., 2016). However, these studies focus on contagion at the market level, while we focus on contagion through ﬁnancial and non-ﬁnancial sectors. Also, the sample periods are January 2006 to December 2010 in Beirne and Gieck (2014) and July 1998 to June 2011 in (Mollah et al., 2016), which exclude the Asian and Russian ﬁnancial crises and only partly include the European sovereign debt crisis period. Our study uses a larger sample, covering the period January 1990 to March 2014. 4 Under this framework, for our sample of 49 (n) stock markets, we need to estimate 48 rather than 1176 [(n2 −n)/2] correlation coefﬁcients for each month for each country– sector pair.

through the ﬁnancial or the non-ﬁnancial sector.5 Second, we examine how the dynamic conditional correlation is related to the level of market development (developed, emerging and frontier equity markets). More importantly, we use a regression model to explain dynamic conditional correlations in terms of economic state variables, such as ﬁnancial crises, bilateral trade intensity, and business cycles. Finally, we check the robustness of the key ﬁndings to the use of (i) US dollar-denominated returns instead of local-currency-denominated returns, (ii) weekly returns instead of monthly returns, (iii) alternative measures of the business cycle stage, (iv) a multiple structural change model (Bai and Perron, 1998), and (v) a regime-switching model (Hamilton, 1989) of conditational correlations. This study uses monthly data from January 1990 to March 2014 with a sample of 49 countries, comprising 24 developed, 19 emerging, and 6 frontier countries. The major ﬁndings can be summarized as follows. First, conditional correlations are higher for the non-ﬁnancial sector than for the ﬁnancial sector, and this ﬁnding holds for both normal and crisis periods, which suggests that non-ﬁnancial ﬁrms play a more pronounced role in the cross-market transmission of shocks. Second, conditional correlations between the US and developed markets are higher than those between the US and emerging/frontier markets. Moreover, conditional correlations between the US and emerging markets are higher than those between US and frontier markets. These results hold for both ﬁnancial and non-ﬁnancial sectors and suggest that cross-market correlation is positively related to the level of market development. Third, the regression results suggest that conditional correlations generally exhibit counter-cyclical movement and are higher during the GFC, European sovereign debt crisis, and Asian and Russian ﬁnancial crises. Fourth, conditional correlations largely increase as the bilateral trade intensity between the US and a non-US country increases. Finally, these results are robust regardless of whether local currency or US dollar-denominated stock returns are used and whether monthly or weekly returns are used. Moreover, both the Bai and Perron (1998) test and the Markov regime-switching model show that periods of high DCC coincide with a bad economic state, which validates the ﬁndings from regression analysis using predetermined economic state variables. The study proceeds as follows: Section 2 presents a brief review of the related literature, Section 3 describes the methodology and data, Section 4 presents the empirical results, and Section 5 checks robustness. Finally, Section 6 draws conclusions concerning the main themes covered in this paper. 2. Related literature There have been numerous studies on ﬁnancial contagion, but the literature is still overrepresented by studies at the aggregate market level and in the context of developed markets (King and Wadhwani, 1990; Min and Hwang, 2012; Samitas and Tsakalos, 2013). In a seminal study, King and Wadhwani (1990) observe that after the stock market crash of October 1987, cross-market correlations substantially increased between the US, the UK, and Japan. Samitas and Tsakalos (2013) ﬁnd a contagion effect from Greece to European countries during the Greece debt crisis. Min and Hwang (2012) also ﬁnd evidence of contagion spreading from the US to Australia, Japan, Switzerland, and the UK during the GFC. Financial contagion from developed economies to emerging economies has been examined by several recent studies (Aloui et al., 2011; Cai et al., 2016; Chiang et al., 2007; Kenourgios et al., 2011; Shen et al., 2015). Chiang et al. (2007) ﬁnd increased cross-market correlations

5 An analysis of inter-sector contagion is beyond the scope of this study. Even in our parsimonious framework, an examination of both inter-sector and intra-sector contagion in 49 countries would substantially increase the number of conditional correlations to be estimated. More importantly, inter-sector contagion has been examined by other recent studies (e.g., Baur, 2012). Thus, our study focuses on the extent of contagion through either the ﬁnancial or the non-ﬁnancial sector.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

among Asian markets during the Asian ﬁnancial crisis. They also ﬁnd persistence of higher cross-market correlations in the post-crisis period. They interpret increased correlations during the crisis as contagion and continued higher correlations in the post-crisis period as herding. Similarly, Cai et al. (2016) ﬁnd a signiﬁcant increase in the conditional correlations between eight emerging East Asian countries and the US in the ﬁrst phase of the GFC (contagion) and persistent higher correlations in the second phase of the GFC (herding). Kenourgios et al. (2011) investigate the dynamic correlation between the BRIC (Brazil, Russia, India, and China) and two developed (the US and the UK) stock market indices. They ﬁnd a contagion effect from the crisis country to non-crisis countries during ﬁve recent crises (Asian crisis in 1997, Brazilian crisis in 1997–1998, Russian crisis in 1998, technology bubble collapse in 2000, and Brazilian crisis in 2002). Aloui et al. (2011) demonstrate strong evidence of time-varying dependence between the US and each of the BRIC markets during the GFC. Shen et al. (2015) ﬁnd that the European debt crisis has a limited effect on the Chinese stock market after controlling for the effects of the macro-fundamental factors. However, the above studies examine cross-correlations between aggregate stock market indices. The use of aggregate indices makes it impossible to isolate the changes in the spill-over effect of the individual sectors from those of the market mix, masking ﬁndings of intra-sector contagion. Notable exceptions are Horta et al. (2010), Phylaktis and Xia (2009), and Akhtaruzzaman et al. (2014), who examine contagion effects among sectors. Horta et al. (2010) ﬁnd the contagion effects of the US subprime crisis in ﬁnancial and industrial sectors' indices between the US and the European NYSE Euronext markets. They demonstrate that evidence of contagion appears to be stronger in ﬁnancial than in industrial indices. Phylaktis and Xia (2009) examine whether unexpected shocks from a sector of one country are transmitted to the sectors in other countries. They ﬁnd evidence of the sectoral heterogeneity of international contagion and conclude that the sector heterogeneity of contagion provides international portfolio diversiﬁcation opportunities in times of ﬁnancial crises when the extent of contagion is high at the market level. Akhtaruzzaman et al. (2014) ﬁnd increased correlation between the US and Australian ﬁnancial ﬁrms during the period of the GFC. We extend the literature by comprehensively investigating the extent of contagion through ﬁnancial vis-à-vis nonﬁnancial sectors from the US to developed, emerging and frontier economies.

145

The GARCH models employed in the ﬁrst stage are presented by the following equations: r i;t ¼ α i0 þ α i1 r i;t−1 þ α i2 r us;t þ εi;t

ð1Þ

hi;t ¼ γi0 þ γ i1 ε2i;t−1 þ γ i2 hi;t−1

ð2Þ

ε i;t Ωi;t−1 N 0; hi;t

ð3Þ

r us;t ¼ α us0 þ α us1 rmus;t þ εus;t

ð4Þ

hus;t ¼ γ us0 þ γus1 ε2us;t−1 þ γ us2 hus;t−1

ð5Þ

ε us;t Ωus;t−1 N 0; hus;t

ð6Þ

In these models, subscripts i and us refer to non-US countries and the US, respectively; ri,t and rus,t are sectoral stock returns in non-US and US markets in month t, respectively; rmus,t is the US market return in month t; εi,t is the random error term distributed with a conditional mean zero and conditional variance hi,t, where the conditioning information set is Ωi,t−1, as given in Eq. (3). In Eq. (1) we include US ﬁnancial (non-ﬁnancial) stock returns (rus,t) as an exogenous global factor, and αi2 measures the global systematic risk exposure of the ﬁnancial (non-ﬁnancial) sector of country i. Chiang et al. (2007) use US stock returns as an exogenous global factor to measure the spill-over from the US to other countries at the aggregate level. This is also in line with Elyasiani and Mansur's (2003) speciﬁcation for the domestic and foreign banking sectors. In Eq. (2), γi1 and γi2 represent the effects of past return shock and past conditional variance, respectively, on the current conditional variance. Eqs. (4), (5), and (6) provide the GARCH (1, 1) speciﬁcation for the US. In the second stage, dynamic conditional correlations are estimated from the following variance–covariance matrix: H t ≡ Dt Rt Dt

ð7Þ

In Eq. (7) Ht is a (n × n) conditional covariance matrix, Rt is a (n × n) 1/2 conditional correlation matrix, and Dt = diag(h1/2 11,t… … …hnn,t) is a (n × n) diagonal matrix of conditional standard deviations. Rt ≡ Q −1 Q t Q −1 t t

ð8Þ

3. Methodology and data 3.1. Models for measuring contagion To measure contagion, we use the DCC-GARCH model introduced by Engle (2002). The DCC-GARCH model uses a two-step estimation procedure. In the ﬁrst step, the univariate GARCH models for each return series are estimated. In the second step, standardized stock return residuals from the ﬁrst step are used to estimate the conditional correlation for each market pair (US and a non-US market) for both the ﬁnancial and non-ﬁnancial sectors. In a DCC-GARCH model, the conditional correlation between asset returns varies over time and continuously adjusts to the time-varying return volatility.6

6 The DCC-GARCH model provides a better measure for potential contagion than the volatility-adjusted correlation model of Forbes and Rigobon (2002). Forbes and Rigobon (2002) ﬁnd no evidence of increased cross-market correlations during ﬁnancial crisis once correlation coefﬁcients are corrected for stock return heteroskedasticity. Corsetti et al. (2005) argue that Forbes and Rigobon's (2002) ﬁnding is biased toward the null of interdependence due to the restriction they imposed on the variance of country-speciﬁc noises where the crisis originates to correct for the heteroskedasticity. Engle's (2002) DCCGARCH model does not require any adjustment for heteroskedastic stock returns. Also, the DCC-GARCH model allows other explanatory variables to be included in the estimation to ensure that the correlation coefﬁcient is well controlled for exogenous changes. Moreover, it is a simpler and more parsimonious model than the MGARCH model of Engle and Kroner (1995).

Q t is the conditional variance-covariance matrix of standardized pﬃﬃﬃﬃﬃﬃﬃﬃ residuals: ui;t ¼ εi;t= = hii;t . Q t ¼ ð1−a−bÞQ þ aðut−1 ut−1 ′ Þ þ bQ t−1

ð9Þ

Q is an (n × n) unconditional variance matrix of standardized residuals; a and b are non-negative scalars satisfying the restriction (a + b) b 1. The correlation estimator is: ρij;t ¼ qij;t =

pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ qii;t qjj;t i; j ¼ 1; 2; …n; and i≠j

ð10Þ

In Eq. (10), ρij,t and qij,t represent the conditional correlation and covariance at time t between US and non-US sectoral stock returns, respectively; and qii,t and qjj,t are the conditional variance of sectoral stock returns in the non-US and US markets, respectively. Finally, the following regression model is used to examine whether dynamic conditional correlation is linked to the US business cycle, bilateral trade intensity, and ﬁnancial crises: ρij;t ¼ β0 þ β1 ρij;t−1 þ β2 D1 þ β3 D2 þ β4 D3 þ β5 C j;t þ β6 T ij;t þ eij;t ð11Þ where ρ ij,t is the conditional correlation between US and non-US sectoral stock returns in month t. The lagged dynamic conditional correlation (ρij,t − 1) is added to Eq. (11) to control for serial correlation.

146

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

Indicators of ﬁnancial crises, bilateral trade intensity, and business cycles are used as economic state variables in Eq. (11). Speciﬁcally, the periods of the Asian and Russian ﬁnancial crises (July 1997 to December 1998), the GFC (July 2007 to June 2009), and European debt crisis (January 2010 to August 2012) are proxied by the dummy variables, D1,D2,and D3, respectively.7 The parameter β3 in Eq. (11) provides a direct measure of contagion from the US to a non-US market during the GFC. On the other hand, β2 provides an indirect measure of contagion from the US to a non-US market during the Asian and Russian ﬁnancial crises, while β4 provides an indirect measure of contagion from the US to a non-US market during the European sovereign debt crisis. These are indirect measures of contagion because the US was not the origin of the Asian, Russian, and European sovereign debt crises. Cj,t is the cyclical component of the US monthly leading index. It is estimated by applying the Hodrick–Prescott ﬁlter (Hodrick and Prescott, 1997) to the US leading index, which is obtained from the Economic Cycle Research Institute (ECRI, www.businesscycle.com). The US business cycle is used as an explanatory variable because Ayhan Kose et al. (2008) and Sandra (2007), among others, argue that US macroeconomic shocks signiﬁcantly affect the country's trading partners. A positive (negative) value of Cj,t indicates to a period of economic expansion (contraction). To investigate how the bilateral trade intensity between the US and a non-US country inﬂuences conditional correlations, we use the trade intensity measure of Frankel and Rose (1998). More speciﬁcally, the bilateral trade intensity is measured as follows: T ij;t ¼ X ij;t þ M ij;t = X i;t þ X j;t þ Mi;t þ M j;t ;

ð12Þ

where Tij,t is the bilateral trade intensity between the US (country j) and country i at time t; Xij,t (Mij,t) is the export (import) between the US and country i; Xi,t and Mi,t are the global export and import of country i, respectively; and Xj,t and Mj,t are the global export and import of the US, respectively. In Eq. (11) eij,t is the random error term; β1 is the ﬁrst-order autoregressive coefﬁcient; and β2, β3, and β4 measure the effect of the Asian and Russian ﬁnancial crises, GFC, and European debt crisis, respectively; β5 and β6 measure the sensitivity of the conditional correlation to the US business cycle and bilateral trade intensity of a country with respect to the US. We use heteroscedasticity and autocorrelation consistent (HAC) standard errors (Newey and West, 1987) to make inferences about parameters in Eq. (11). 3.2. Data The monthly data are collected from Datastream and the Economic Cycle Research Institute (ECRI) and cover the period from January 1990 to March 2014. The sample includes all the ﬁnancial and nonﬁnancial ﬁrms in 49 countries. The countries are classiﬁed into developed, emerging, and frontier as described in the FTSE Country Classiﬁcation (www.ftse.com/products/indices/country-classiﬁcation).8 We chose countries for which data are available from January 2001. After using this ﬁlter, the sample includes 24 developed, 19 emerging, and 6 frontier countries. The Global Industry Classiﬁcation Standard (GICS)

7 Since the period of Asian ﬁnancial crisis partly overlaps with the period of the Russian ﬁnancial crisis, we have created a single dummy variable for these crises. Syllignakis and Kouretas (2011) also used a single dummy variable for the Asian and Russian crises to investigate contagion effects among the US, German, Russian and the Central and Eastern European (CEE) stock markets. 8 In 2004, FTSE introduced a formal process for assessing markets against some criteria. Those criteria include market and regulatory environment, custody and settlement, dealing landscape and presence of derivative markets. In addition to these criteria, FTSE considers Gross National Income per capita and market size. Based on each of these criteria, markets are given a grade of Pass, Restricted (partial failure) or Fail. Markets are classiﬁed into developed, advanced emerging, secondary emerging and frontier on the basis of their overall scores (www.ftse.com/products/downloads/FTSE_Country_Classiﬁcation_Paper.pdf).

is used to identify ﬁnancial and non-ﬁnancial ﬁrms. Local-currencydenominated stock returns are calculated from the Datastream total return index using Ri,t = ln (RIi,t/RIi,t − 1) where Ri,t is the return of portfolio i and RIi,t is the return index at time t.9 The value-weighted return on the portfolio of ﬁnancial stocks is calculated from the total return index of individual ﬁnancial stocks. The return on the portfolio of non-ﬁnancial stocks is calculated indirectly from Rm,t = Wf,tRf,t + (1 − Wf,t)Rnf,t. In this formula, Rm,t is the market return; Rf,t and Rnf,t represent returns on ﬁnancial and non-ﬁnancial stock portfolios, respectively; and Wf,t is the weight of the ﬁnancial sector in the domestic market portfolio at time t. We use monthly data because they are less affected by settlement and clearing delays (Baillie and Ramon, 1990). The US leading index is obtained from the ECRI. Monthly data on bilateral trade between US and other countries are obtained from the US Census Bureau. Monthly data on global imports and exports of each country are obtained from Datastream. 4. Empirical results 4.1. Descriptive statistics Descriptive statistics for returns on ﬁnancial, non-ﬁnancial, and market portfolios are presented in Table 1. During the study period, the average stock returns of emerging and frontier markets were much higher than those of their developed counterparts. The standard deviation of stock returns in emerging and frontier markets is higher than that of developed markets. These ﬁndings hold for both ﬁnancial and non-ﬁnancial sectors and indicate that higher returns in emerging and frontier markets are associated with higher risk. The skewness of returns is close to zero, indicating very little asymmetry in returns. The kurtosis is over 3 for all return series, implying a relatively peaked distribution. The Jarque–Bera test also demonstrates that none of the return series are normally distributed. All return series are stationary because the unit root hypothesis is rejected for all of them by the Augmented Dickey-Fuller (ADF) and Phillips–Perron (PP) tests. The null hypothesis of white noise cannot be rejected for most return series by the Box–Pierce–Ljung portmanteau test, which suggests most return series are serially uncorrelated.10 During the 2007–2009 GFC, most developed, emerging and frontier markets experienced negative returns. However, developed market stocks suffered far more losses than emerging/frontier stock markets due to the GFC. During the 1997–1998 Asian ﬁnancial crisis, economies such as Hong Kong, Indonesia, Malaysia, and the Philippines experienced negative returns, while most European countries had negative returns during the European Sovereign Debt Crisis in 2010–2012. In particular, Greece, Portugal, Italy, and Spain were hit very hard by the European Sovereign Debt crisis. The maximum likelihood estimator (Marquardt) is used to estimate the GARCH (1, 1) model. The results are presented in Table 2. Model adequacy is evaluated through diagnostic checks on standardized residuals (εi,t/√hi,t). In most cases, the Box–Pierce–Ljung portmanteau test provides evidence of no autocorrelation in residuals. Also, in most cases, we ﬁnd no remaining ARCH effects in the residuals based on the ARCH-LM test (Engle, 1982) and the Box–Pierce–Ljung portmanteau test of squared standardized residuals. Overall, diagnostic test results

9 Mink (2015) argues that it would be more appropriate to use returns denominated in local currency than common currency returns (e.g. US dollar-denominated returns) as only local currency denominated returns accurately reﬂect price ﬂuctuations in national stock markets. However, returns converted into a common currency reﬂect ﬂuctuations in the exchange rate. Forbes and Rigobon (2002) employ both local currency returns and USD returns to test for a contagion and do not ﬁnd signiﬁcant differences in results. Longin and Solnik (1995) argue that local currency returns should be used when focusing on the correlation across markets rather than across currencies. 10 The results for these tests have not been reported to conserve space. These results are available from the corresponding author on request.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

147

Table 1 Descriptive statistics of monthly stock returns (in decimal form). Financial ﬁrms portfolio

Non-ﬁnancial ﬁrms portfolio

Country

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK USA

0.008 0.004 0.006 0.008 0.008 0.007 0.006 0.006 0.010 0.006 0.007 0.004 −0.001 0.007 0.006 0.008 0.005 0.005 0.006 0.008 0.009 0.007 0.007 0.008

0.038 0.060 0.049 0.041 0.083 0.052 0.055 0.095 0.073 0.059 0.060 0.063 0.057 0.053 0.043 0.065 0.053 0.060 0.087 0.059 0.066 0.045 0.042 0.044

0.010 0.003 0.004 0.010 0.009 0.006 0.003 0.001 0.010 −0.001 0.008 0.003 −0.004 0.005 0.004 0.008 0.001 0.006 −0.001 0.007 0.009 0.005 0.007 0.008

0.044 0.070 0.065 0.046 0.091 0.072 0.067 0.133 0.082 0.137 0.073 0.075 0.075 0.072 0.051 0.089 0.078 0.078 0.107 0.077 0.082 0.067 0.060 0.058

0.007 0.005 0.008 0.007 0.009 0.007 0.006 0.007 0.009 0.005 0.007 0.005 0.001 0.007 0.007 0.008 0.006 0.004 0.007 0.008 0.008 0.008 0.007 0.008

0.042 0.061 0.043 0.045 0.085 0.051 0.055 0.080 0.071 0.057 0.061 0.063 0.056 0.049 0.043 0.066 0.057 0.057 0.088 0.055 0.068 0.040 0.040 0.044

0.656 0.586 0.626 0.733 0.896 0.823 0.742 0.566 0.552 0.659 0.688 0.591 0.789 0.700 0.858 0.871 0.662 0.613 0.800 0.764 0.756 0.749 0.769 0.829

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

0.012 0.014 0.009 0.013 0.005 0.011 0.010 0.013 0.008 0.007 0.017 0.012 0.010 0.009 0.005 0.019 0.013 0.002 0.029

0.074 0.054 0.105 0.063 0.067 0.088 0.087 0.096 0.089 0.070 0.068 0.095 0.057 0.075 0.088 0.115 0.057 0.092 0.135

0.012 0.016 0.013 0.011 0.005 0.011 0.015 0.016 0.003 0.009 0.018 0.016 0.011 0.009 0.007 0.021 0.015 −0.001 0.030

0.080 0.069 0.119 0.066 0.092 0.115 0.120 0.142 0.112 0.091 0.093 0.114 0.060 0.079 0.100 0.183 0.064 0.110 0.149

0.012 0.014 0.007 0.013 0.004 0.010 0.009 0.012 0.009 0.007 0.016 0.012 0.009 0.008 0.004 0.018 0.012 0.003 0.028

0.077 0.056 0.106 0.066 0.069 0.087 0.086 0.093 0.088 0.066 0.068 0.093 0.063 0.079 0.087 0.113 0.061 0.087 0.134

0.767 0.840 0.662 0.581 0.814 0.744 0.796 0.886 0.788 0.771 0.900 0.861 0.736 0.629 0.543 0.911 0.732 0.703 0.614

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

0.011 0.014 0.001 0.010 0.004 0.014

0.093 0.101 0.111 0.116 0.053 0.080

0.010 0.012 −0.000 0.014 −0.003 0.016

0.120 0.124 0.125 0.134 0.079 0.092

0.011 0.013 0.000 0.007 0.005 0.013

0.095 0.113 0.102 0.121 0.053 0.079

0.831 0.674 0.331 0.690 0.928 0.688

Market portfolio

Weight of non-ﬁnancial sector

Note: The weight of the non-ﬁnancial sector is the average percentage of non-ﬁnancial ﬁrms' market value to total market value during the sample period.

suggest that the GARCH (1, 1) model is adequate for describing the return generating process.11

4.2. Spill-over through non-ﬁnancial ﬁrms In this study, the spill-over effects have been assumed to be channeled from the US to developed, emerging, and frontier countries.12

11 To conserve space, diagnostic test results have not been reported, but they are available from the corresponding author on request. 12 A number of studies (Hwang et al., 2013; Kim et al., 2015) ﬁnd spill-over effects from the US to other markets.

We ﬁnd evidence of spill-over effects for both ﬁnancial and nonﬁnancial stock portfolios. From Table 2, it is evident that the coefﬁcient of US ﬁnancial (nonﬁnancial) sector returns is positive and signiﬁcant in the mean equation for the ﬁnancial (non-ﬁnancial) sector of a non-US country. This implies that there is a positive spill-over effect from the US market to developed, emerging, and frontier markets. The magnitude of the spill-over effect ranges from 0.134 (Sri Lanka) to 1.187 (Hungary) for ﬁnancial ﬁrms, and from 0.301 (Sri Lanka) to 1.145 (Finland) for non-ﬁnancial ﬁrms. It is interesting to note that the magnitude of the spill-over effect for the non-ﬁnancial sector is higher than that of the ﬁnancial sector except in Hong Kong, Switzerland, and the UK. These results suggest that nonﬁnancial ﬁrms play a more prominent role in the transmission of information across countries than ﬁnancial ﬁrms do.

148

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

Table 2 GARCH (1, 1) results. Financial ﬁrms portfolio Mean equation

Non-ﬁnancial ﬁrms portfolio Variance equation

Mean equation

Variance equation

Lagged ﬁnancial ﬁrms return

US ﬁnancial ﬁrms return

ARCH

GARCH

Lagged non-ﬁnancial ﬁrms return

US non-ﬁnancial ﬁrms return

ARCH

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK

0.004 0.204*** 0.189*** 0.134*** 0.008 0.148*** −0.009*** 0.121** −0.022 −0.013 0.144** −0.044 0.071 0.098* 0.078 0.000 0.266*** 0.039 −0.011 0.032 −0.013 0.107*** −0.047

0.468*** 0.424*** 0.602*** 0.564*** 0.709*** 0.715*** 0.698*** 0.845*** 0.741*** 0.879*** 0.585*** 0.668*** 0.517*** 0.595*** 0.240*** 0.892*** 0.310*** 0.653*** 0.637*** 0.735*** 0.752*** 0.719*** 0.768***

0.292*** 0.151*** 0.195*** 0.153*** 0.133*** 0.218* 0.131*** 0.143*** 0.224*** 0.232*** 0.129** 0.068** −0.052** 0.272*** 0.083** 0.169*** 0.227*** 0.199*** 0.131*** 0.324*** 0.134*** 0.273*** 0.174**

0.433** 0.809*** 0.678*** 0.754*** 0.860*** 0.361 0.705*** 0.821*** 0.733*** 0.762*** 0.778*** 0.910*** 1.026*** 0.684*** 0.896*** 0.769*** 0.756*** 0.786*** 0.879*** 0.242 0.823*** 0.129 0.727***

−0.007 0.147*** 0.128** 0.078* 0.137*** 0.006 0.032 0.034 0.045 0.147*** 0.016 −0.005 0.100*** 0.093*** −0.020 0.111*** 0.126*** 0.122*** 0.008 0.027 0.007 0.098*** −0.024

0.634*** 0.609*** 0.531*** 0.764*** 1.145*** 0.881*** 0.956*** 0.872*** 0.940*** 0.753*** 0.645*** 0.903*** 0.634*** 0.843*** 0.430*** 0.961*** 0.663*** 0.803*** 0.938*** 0.805*** 0.947*** 0.513*** 0.728***

0.086 0.056** 0.119* 0.092* 0.160*** −0.016** 0.031* 0.096** 0.126** 0.064** 0.075** 0.130*** −0.040** 0.042 −0.028* 0.098*** 0.076** 0.089** 0.121*** 0.045* 0.105** −0.011 0.079**

0.775*** 0.902*** 0.727*** 0.746*** 0.821*** 1.023*** 0.949*** 0.858*** 0.820*** 0.887*** 0.895*** 0.866*** 1.022*** 0.871*** 1.017*** 0.880*** 0.889*** 0.879*** 0.868*** 0.929*** 0.871*** 1.020*** 0.863***

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

0.002 0.141** 0.130** 0.233*** 0.087 0.132 0.120*** 0.082 0.091 0.066 0.085 0.012 0.136 0.118* −0.012 0.187*** −0.051 −0.076 −0.047

0.579*** 0.183*** 0.480*** 0.268*** 0.501*** 0.287*** 1.187*** 0.374*** 0.658*** 0.406*** 0.438*** 0.306* 0.181** 0.497*** 0.724*** 0.966*** 0.435*** 0.588*** 0.737***

0.034 0.218*** 0.135** 0.086*** 0.170*** 0.132* 0.051*** 0.155*** 0.052** 0.249*** 0.125*** 0.042 0.182*** 0.058* 0.061 0.091** 0.186*** 0.073*** 0.067***

0.816 0.769*** 0.813*** 0.894*** 0.770*** 0.672*** 0.925*** 0.785*** 0.941*** 0.752*** 0.864*** 0.856*** 0.713*** 0.916*** 0.894*** 0.868*** 0.776*** 0.918*** 0.926***

0.127** 0.168*** 0.043 0.182*** 0.098* 0.100*** 0.012 0.074 0.154*** 0.038 0.032 −0.012 0.042 0.080 −0.151*** 0.148** −0.024 0.007 −0.144**

0.848*** 0.477*** 0.884*** 0.326*** 0.688*** 0.578*** 1.001*** 0.722*** 0.914*** 0.471*** 0.799*** 0.534*** 0.457*** 0.654*** 0.893*** 0.976*** 0.618*** 0.819*** 1.006***

0.134*** 0.120*** 0.207*** 0.092* 0.074** −0.042** −0.020 0.118*** 0.150*** 0.191*** −0.037*** 0.129** 0.171 −0.022 −0.052*** 0.096* 0.105*** 0.106*** 0.061***

0.853*** 0.831*** 0.759*** 0.815*** 0.861*** 0.904*** 0.504 0.876*** 0.805*** 0.782*** 1.020*** 0.825*** 0.632*** 1.015*** 1.016*** 0.877*** 0.875*** 0.883*** 0.932***

0.145*** 0.273** 0.164** 0.062 0.133* 0.135**

0.783*** 0.338*** 0.298*** 0.879*** 0.259*** 0.134**

0.113** 0.573*** 0.277*** 0.080*** −0.076*** 0.050*

0.827*** 0.530*** 0.725*** 0.910*** 1.053*** 0.814***

0.051 0.222** 0.083 0.113* 0.330*** 0.130**

0.951*** 0.444*** 0.362*** 1.007*** 0.395*** 0.301***

0.251*** 1.072*** 0.322*** 0.119*** 0.023 −0.004

0.650*** 0.231*** 0.649*** 0.882*** −1.046*** 0.973***

Country

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

GARCH

Notes: This table presents estimates of the parameters of Eqs. (1) to (6): ri,t = αi0 + αi1ri,t−1 + αi2rus,t + εi,t

(1)

hi,t = γi0 + γi1ε2i,t−1 + γi2hi,t−1

(2)

εi,t |Ωi,t−1 ~N (0,hi,t)

(3)

rus,t = αus0 + αus1rmus,t + εus,t

(4)

2 + γus2hus,t−1 hus,t = γus0 + γus1εus,t−1

(5)

εus,t |Ωus,t−1 ~N (0,hus,t)

(6)

In these models, subscripts i and us indicate the non-US country and the US, respectively; ri,t and rus,t are sectoral stock returns in non-US and US markets in month t, respectively; rmus,t is the US market return in month t; εi,t is the random error term distributed with mean 0 and variance hi,t, an information set Ωi,t−1. Eqs. (4), (5) and (6) provide the GARCH (1, 1) speciﬁcation for the US. ⁎ Signiﬁcance at the 0.10 level. ⁎⁎ Signiﬁcance at the 0.05 level. ⁎⁎⁎ Signiﬁcance at the 0.01 level.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

4.3. Dynamic conditional correlations Table 3 presents the average of the conditional correlations over the period January 1990 to March 2014 for each country–sector pair. The null hypothesis of constant correlation is rejected in favor of timevarying correlation based on the Lagrange Multiplier (LM) test13 of Tse (2000), which provides evidence for a time-varying conditional correlation between US sectoral returns and those of a non-US country. The average conditional correlations in the non-ﬁnancial sector are generally greater than those in the ﬁnancial sector. The Kolmogorov– Smirnov test demonstrates that the distribution of DCCs for ﬁnancial stock portfolios signiﬁcantly differs from that of the non-ﬁnancial stock portfolio for all countries except Colombia. In the non-ﬁnancial sector, the average DCC is 0.575 for developed countries, 0.395 for emerging countries, and 0.270 for frontier countries. The respective averages for the ﬁnancial sector are 0.496, 0.289, and 0.200. This ﬁnding of higher-average DCC for non-ﬁnancial ﬁrms is at odds with the notion that ﬁnancial ﬁrms are the main conduit for international transmission shocks. The liberalization of international trade and the rapid growth of multinational corporations may have contributed to the higher conditional return correlation between US and foreign non-ﬁnancial sectors. This ﬁnding corresponds with Claessens et al. (2012) observation that trade linkages rather than ﬁnancial linkages strongly moderated the impact of the GFC on the accounting performance of manufacturing ﬁrms across countries. The conditional return correlation between the US and developed countries is found to be higher than that between the US and emerging/frontier countries. The magnitude of average conditional correlations between US non-ﬁnancial stock returns and those of developed countries ranges from 0.417 (Austria) to 0.744 (UK) while the magnitude of average conditional correlations between US non-ﬁnancial stock returns and those of emerging countries ranges from 0.174 (Pakistan) to 0.571 (Mexico). For both the ﬁnancial and non-ﬁnancial sectors, the conditional return correlations between the US and emerging countries are found to be higher than those between the US and frontier countries. The average conditional correlation between US non-ﬁnancial stock returns and those of frontier countries ranges from 0.169 (Sri Lanka) to 0.456 (Argentina). Overall, these results suggest that the average conditional correlation between US and non-US markets is positively related to the level of development of the non-US market. The time plots of the monthly conditional correlations are depicted in Figs. A1 and A2 in the appendix for ﬁnancial and non-ﬁnancial sectors, respectively. From Panel A of Fig. A1, it is evident that the dynamic conditional correlations between US ﬁnancial stock returns and those of the developed countries ﬂuctuate considerably over time. Though conditional correlations are positive for all months for all developed country pairs, they often ﬂuctuate in a non-synchronized manner over time. The latter creates opportunities for US investors to rebalance their portfolios towards stock markets that are relatively less correlated to the US stock market. Panels B and C of Fig. A1 provide similar ﬁndings for the ﬁnancial sector in emerging and frontier markets, respectively, but episodes of negative conditional correlations are far more prevalent for the US-frontier market pairs than for the US-emerging market pairs. Ten out of 19 emerging markets and ﬁve out of six frontier markets show several episodes of negative conditional correlations with the US market. The time plots of DCCs for the non-ﬁnancial sector are presented in Panels A, B, and C of Fig. A2 for developed, emerging, and frontier markets, respectively. Most developed markets show ﬂuctuations in DCCs around a rising trend while most frontier markets' DCCs ﬂuctuate without any discernible trend. Characteristics of emerging market DCCs fall 13 Tse's (2000) Lagrange Multiplier (LM) test evaluates the null hypothesis, H0: δij = 0 in the following equation: ρij,t = ρij + δijεi,t−1εj,t−1, where εj,t−1 and εi,t−1 are innovations in US (country j) and non-US (country i) sectoral stock returns, respectively.

149

Table 3 Dynamic conditional correlations with respect to the US. Financial ﬁrms portfolio LM test of Tse (2000)

Non-ﬁnancial ﬁrms portfolio Mean

Kolmogorov– Smirnov test

LM test of Tse (2000)

Country

Mean

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK Average DCC-Developed

0.546 0.351 0.560 0.691 0.529 0.584 0.561 0.306 0.438 0.567 0.385 0.426 0.384 0.608 0.433 0.524 0.334 0.426 0.331 0.547 0.578 0.643 0.650 0.496

35.45*** 18.91*** 54.71*** 31.42*** 22.26*** 33.21*** 28.56*** 26.77*** 2.06 110.40*** 14.08*** 63.39*** 43.31*** 12.42*** 87.92*** 27.73*** 55.57*** 7.56*** 12.19*** 75.53*** 26.64*** 17.38*** 13.91***

0.632 0.417 0.574 0.714 0.526 0.691 0.651 0.426 0.583 0.572 0.508 0.558 0.442 0.736 0.480 0.625 0.441 0.608 0.445 0.635 0.596 0.624 0.744 0.575

21.65*** 15.63*** 288.58*** 21.77*** 12.61*** 14.01*** 13.07*** 55.80*** 4.74** 310.86*** 27.66*** 27.99*** 10.69*** 100.09*** 20.18*** 9.55*** 7.49*** 9.27*** 8.88*** 21.36*** 6.92*** 35.38*** 13.82***

0.34*** 0.19*** 0.26*** 0.28*** 0.23*** 0.40*** 0.39*** 0.33*** 0.28*** 0.57*** 0.55*** 0.43*** 0.30*** 1.00*** 0.65*** 0.32*** 0.37*** 0.62*** 0.24*** 0.47*** 0.18*** 0.56*** 0.35***

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey Average DCC-Emerging

0.390 0.254 0.177 0.194 0.341 0.186 0.420 0.127 0.353 0.349 0.303 0.142 0.095 0.341 0.409 0.451 0.441 0.284 0.229 0.289

5.15** 3.34* 9.55*** 13.91*** 91.27*** 13.84*** 46.30*** 24.38*** 550.81*** 3.89** 8.37*** 43.00*** 16.14*** 48.06*** 10.94*** 34.89*** 32.15*** 16.43*** 11.74***

0.555 0.397 0.427 0.197 0.441 0.255 0.473 0.263 0.448 0.438 0.571 0.174 0.322 0.430 0.549 0.485 0.374 0.413 0.297 0.395

4.08** 9.08*** 15.02*** 8.57*** 88.81*** 32.77*** 13.54*** 18.47*** 56.99*** 30.21*** 21.25*** 78.66*** 12.65*** 19.36*** 30.15*** 13.96*** 12.18*** 5.63** 7.37***

0.70*** 0.87*** 0.48*** 0.04 0.64*** 0.42*** 0.20*** 0.28*** 0.79*** 0.49*** 0.47*** 0.32*** 0.55*** 0.67*** 0.56*** 0.26*** 0.30*** 0.47*** 0.28***

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka Average DCC-Frontier

0.352 0.191 0.211 0.257 0.110 0.077 0.200

27.95*** 7.44*** 54.38*** 10.85*** 199.87*** 75.37***

0.456 0.262 0.159 0.271 0.302 0.169 0.270

22.89*** 29.88*** 65.53*** 41.59*** 227.13*** 83.72***

0.34*** 0.44*** 0.16*** 0.19*** 0.67*** 0.87***

Notes: a) This table presents the estimates of the parameters of Eq. (10): pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ρij,t = qij;t = qii;t qjj;t i; j ¼ 1; 2; …n; and i≠j

(10)

In Eq. (10) ρij,t and qij,t denote the conditional correlation and covariance at time t between US and non-US sectoral stock returns, respectively. qii,t and qjj,t are the conditional variance of sectoral stock returns in the non-US and US markets, respectively. b) Tse's (2000) Lagrange Multiplier (LM) test evaluates the null hypothesis, H0: δij = 0 using the following equation: ρij,t = ρij + δijεi,t−1εj,t−1, where εj,t−1 and εi,t−1 are innovations in US (country j) and non-US country i sectoral stock returns, respectively. c) The two-sample Kolmogorov–Smirnov test is used to test the null hypothesis that there is no difference between the distributions of DCC for ﬁnancial ﬁrms and non-ﬁnancial ﬁrms with respect to the US. D-statistics for the two-sample Kolmogorov–Smirnov test are reported in the last column. ⁎ Signiﬁcance at the 0.10 level. ⁎⁎ Signiﬁcance at the 0.05 level. ⁎⁎⁎ Signiﬁcance at the 0.01 level.

150

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

somewhere between those of the other two types of markets. These patterns, along with evidence of greater DCCs in the non-ﬁnancial than the ﬁnancial sector, suggest that emerging/frontier market stocks can be viewed as a different asset class, providing US investors opportunities to reduce their portfolio risk by adopting sector rotation strategies across countries. The shaded areas in Figs. A1 and A2 represent recession periods as determined by the National Bureau of Economic Research (NBER). For most country–sector pairs, DCC is higher in times of recession. It is more so during the GFC-induced US recession of 2007–2009. 4.4. Determinants of dynamic conditional correlations Conditional correlation is regressed on its own lag, the US business cycle, bilateral trade intensity between the US and a non-US country, and ﬁnancial crises, based on Eq. (11). The OLS regression results are presented in Table 4. With the exception of three countries, the coefﬁcient of the lagged conditional correlation is large and statistically signiﬁcant for both sectors, implying a high degree of persistence in conditional correlations. This is expected since fundamental economic links between countries are normally persistent. The model adequacy for Eq. (11) has been performed through diagnostic checks of standardized residuals. The Box–Pierce–Ljung portmanteau test provides evidence of no autocorrelation in residuals in any of the non-ﬁnancial stock portfolios. For ﬁnancial stocks, we ﬁnd evidence of autocorrelation in only two out of 48 portfolios. The ARCH-LM test (Engle, 1982) and the Box–Pierce–Ljung portmanteau test of squared standardized residuals show evidence of remaining ARCH effects in only one ﬁnancial stock portfolio and ﬁve non-ﬁnancial stock portfolios. Hence, Eq. (11) is an adequate representation of the determinants of dynamic conditional correlations. The extent of contagion during ﬁnancial crises is measured by the coefﬁcients of ﬁnancial crisis dummies. Table 4 shows that the coefﬁcient of the dummy variable for the GFC is positive and statistically signiﬁcant for several countries, providing evidence of contagion. Table 4 demonstrates that contagion occurs through ﬁnancial ﬁrms for ten developed countries (Australia, Austria, Canada, France, Greece, Ireland, Italy, Japan, New Zealand, and Switzerland), two emerging countries (Malaysia and Philippines), and one frontier country (Argentina) during the GFC. Contagion occurs through non-ﬁnancial ﬁrms for 12 developed countries (Australia, Austria, Finland, Greece, Italy, Japan, the Netherlands, New Zealand, Singapore, Spain, Switzerland, and the UK), eight emerging countries (China, Czech Republic, Peru, Philippines, Poland, Russia, South Africa, and Turkey), and ﬁve frontier countries (Argentina, Bulgaria, Cyprus, Romania, and Sri Lanka) during the GFC. Finally, contagion occurs through both ﬁnancial and nonﬁnancial ﬁrms for seven developed countries (Australia, Austria, Greece, Italy, Japan, New Zealand, and Switzerland), one emerging country (Philippines), and one frontier country (Argentina) during the GFC. These results reinforce the contention that the GFC's effect on conditional correlation is more prevalent in the non-ﬁnancial sector and pervasive across countries, regardless of their development level. Evidence of the GFC's positive effect on contagion at the sectoral level is consistent with ﬁndings at the aggregate market level, provided by Kim and Kim (2013), Syllignakis and Kouretas (2011), and Hwang et al. (2013). Horta et al. (2016) ﬁnd the transmission of ﬁnancial contagion from the US to Belgium, France, the Netherlands, and Portugal during the GFC. They show that the largest markets correlate most with the US market with France having the highest correlation, followed by the Netherlands, Belgium, and Portugal. For this group of countries, we ﬁnd similar results whereby the extent of correlation is highest for the Netherlands, followed by France, Belgium, and Portugal.14 The 14 This discrepancy may be due to methodological differences. We use the DCC-GARCH model, while Horta et al. (2016) uses the copula theory and the maximum likelihood approach. Also, our sample period differs from theirs.

coefﬁcients of the dummy variable for the European sovereign debt crisis and the Asian and Russian ﬁnancial crises are positive and statistically signiﬁcant for several country pairs. Unlike the GFC effect, the contagion effects of these two regional crises are more prevalent in the ﬁnancial than in the non-ﬁnancial sector. Furthermore, the contagion effect of the European sovereign debt crisis is less pervasive than those of the GFC and Asian and Russian ﬁnancial crises. Thus, the contagion effect of a ﬁnancial crisis depends on its nature and differs by sector. Even in times of a ﬁnancial crisis, a country's domestic sector may not exhibit excess correlation with the corresponding US sector, which opens up portfolio diversiﬁcation opportunities for global investors. In general, conditional correlation between US sectoral stock returns and those of the non-US country increases (decreases) during US economic downturns (upturns). In the ﬁnancial sector, 12 out of 23 developed countries, 13 out of 19 emerging countries, and 4 out of 6 frontier countries experienced statistically signiﬁcant increases (decreases) in DCC during US economic downturns (upturns). In the non-ﬁnancial sector, we ﬁnd similar results for US-developed market pairs, but a lesser prevalence of the business cycle effect for US-emerging and USfrontier market pairs. These results imply that the potential beneﬁts of international portfolio diversiﬁcation may vary across countries and sectors, depending on the phase of the US business cycle. Our results complement the study by Akhtaruzzaman et al. (2014) which demonstrates that conditional correlation between the US and Australian ﬁnancial stock returns moves counter-cyclically. Ferreira and Gama (2010) also ﬁnd a higher correlation of global industry portfolios with the global market portfolio during a recession in the US. In general, an increase in trade intensity of a country with the US increases the corresponding conditional correlation. In the nonﬁnancial sector, 10 out of 23 developed countries and 3 out of 19 emerging countries experienced statistically signiﬁcant increases in DCC as the bilateral trade intensity increases. These results are interesting when we relate them to the largest trading partners of the US. Eight out of 15 US top trading partners experienced increases in DCC for non-ﬁnancial ﬁrms as the bilateral trade intensity increases.15 Also, in case of frontier countries, the bilateral trade intensity does not have any impact on DCC for both ﬁnancial and non-ﬁnancial sectors. This result may be due to the fact that compared to developed and emerging countries, frontier countries have weaker trade linkages with the US (www.census.gov/ foreign-trade/balance/index.html). However, in the ﬁnancial sector, only six developed countries and three emerging countries experienced increases in DCC as the bilateral trade intensity increases. Hence, the effect of trade intensity on contagion is more prevalent in the nonﬁnancial sector than in the ﬁnancial sector. 5. Robustness checks We performed the following robustness checks for our results. First, we examine whether the ﬁndings for returns denominated in local currency remain valid for US dollar-denominated returns. Second, we use weekly data to check the robustness of the results from monthly data. Third, we use NBER business cycle dates for US recessions instead of the cyclical components of the ECRI leading index in Eq. (11) to determine whether the business cycle-contagion nexus is robust for an alternative business cycle measure. Fourth, Eq. (11) identiﬁes structural breaks in DCC using dummy variables for the ﬁnancial crises, conditional on other variables. As an alternative, the incidence of structural breaks is examined for individual country–sector pairs using a data-driven approach proposed by Bai and Perron (1998).16 Finally, a Markov regime-switching model (Hamilton, 1989) is used to identify whether

15 The US top 15 trading partners as at March 2014 has been obtained from the US international trade data by the US Census Bureau (https://www.census.gov/foreign-trade/ statistics/highlights/top/top1412yr.html). 16 Details of the Bai and Perron (1998) test procedure can be found in (see IHS Global Inc., 2015).

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

151

Table 4 Determinants of dynamic conditional correlations with respect to the US. Panel A: Financial ﬁrms portfolio Country

Constant

Lagged conditional correlation

Dummy for global ﬁnancial crisis

Dummy for European sovereign debt crisis

Dummy for Asian and Russian crises

US business cycle

Trade intensity

Diagnostic statistics Q (12)

Q2(12)

ARCH LM test of Engle (1982)

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK

0.075*** 0.002 0.053*** 0.050* 0.027*** 0.004 0.017** 0.026** 0.002 0.097*** −0.001 −0.009 0.023* 0.022** 0.076*** 0.048*** 0.019** 0.012 −0.014 0.104*** 0.028** 0.012 0.017

0.843*** 0.959*** 0.892*** 0.910*** 0.929*** 0.955*** 0.924*** 0.918*** 0.977*** 0.828*** 0.944*** 0.934*** 0.936*** 0.953*** 0.824*** 0.869*** 0.907*** 0.970*** 0.957*** 0.819*** 0.921*** 0.977*** 0.964***

0.023*** 0.013* 0.010 0.025** 0.007 0.011** 0.002 0.016* −0.004 0.001*** 0.003 0.023** 0.014*** 0.002 0.005*** −0.005 0.004 0.009 0.006 0.010 0.006 0.006*** 0.006

0.016* 0.015*** 0.002 0.010 0.008** 0.004 0.005 0.012* 0.005* 0.001 0.006 0.010 0.009*** 0.003 0.001 0.015** 0.005 0.010 0.013** 0.004 0.014** 0.001 0.006**

0.008 0.029** 0.005 0.025 0.024** 0.018*** 0.037*** n/a −0.004 0.001*** 0.014** 0.017 0.009 0.003 0.008 0.036* 0.017** 0.006 0.004 0.024** 0.029*** 0.013*** 0.001

−0.002*** −0.001 −0.001 −0.001 0.001 −0.001 −0.002* −0.002** −0.002** −0.001 −0.001*** −0.002* −0.001* −0.001 −0.001 −0.002 −0.002*** −0.002** −0.003* −0.003** −0.002* 0.001 −0.001

0.389 1.298 0.316 0.024 1.505 0.698* 0.201*** 6.028 0.484 0.484 0.865*** 1.401** −0.010 0.244 −0.172 2.562* 3.382** −0.025 0.570 −2.196 1.177 0.099 0.152

4.71 16.5 13.7 18.2 16.1 10.9 14.2 10.8 8.23 8.79 21.3** 24.7** 7.18 10.6 5.86 12.9 6.22 12.6 13.6 5.89 10.5 11.4 12.8

2.38 2.03 18.4 8.25 1.10 8.71 7.23 10.9 4.98 1.49 3.61 9.95 0.55 6.47 0.17 12.3 9.85 6.28 18.1 6.22 2.97 5.08 4.63

0.63 0.03 0.01 0.01 0.01 0.65 0.15 0.29 0.06 0.13 0.21 0.53 0.01 0.85 0.02 0.01 0.35 0.73 0.16 0.01 0.03 3.30* 0.02

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

0.008 0.022** −0.016** 0.001 0.064*** 0.006 0.018 −0.010** 0.339*** 0.009 −0.003 −0.002 −0.016 0.019** 0.013** 0.082*** 0.044*** −0.003 −0.002

0.953*** 0.914*** 0.930*** 0.910*** 0.802*** 0.955*** 0.845*** 0.951*** 0.022 0.964*** 0.958*** 0.814*** 0.914*** 0.922*** 0.944*** 0.792*** 0.879*** 0.924*** 0.982***

0.001 −0.000 −0.006 0.008 0.005 0.004 0.017 0.006 0.001 0.007* −0.004 −0.011** 0.009 0.017** 0.010 0.001 0.002 0.000 0.002

0.001 0.001 0.014* 0.004 0.005** 0.005 0.037** 0.004 0.007 0.006 0.001 0.001 0.022** 0.006 0.014*** 0.011* 0.009 0.014** −0.005

0.007 n/a 0.007 0.002 0.008 −0.002 0.082*** 0.011*** 0.004 −0.000 0.037** 0.017 n/a 0.019** 0.009 0.058** 0.033* 0.018 0.014

−0.001** −0.000 −0.002** −0.002 −0.001** −0.001** −0.006*** −0.001 −0.001*** −0.001 −0.003*** −0.002*** −0.003* −0.001 −0.002*** −0.003*** −0.002*** −0.003** −0.001

0.299** 0.126 0.218*** 0.901 0.455 0.197 11.694** 0.829 0.328 0.089 0.086 6.938 1.872 0.296 2.415 0.456 0.696 0.512 0.926

7.31 11.1 8.86 12.2 11.8 16.0 11.6 7.13 12.6 7.40 14.5 10.0 17.7 12.2 11.3 8.21 9.73 3.74 2.04

6.32 9.09 9.09 8.39 1.71 15.5 9.04 1.62 6.46 13.07 4.26 11.0 15.3 7.54 0.88 0.99 3.72 16.3 1.19

0.02 0.60 0.63 0.22 0.01 0.50 0.47 0.19 0.31 0.10 0.01 0.69 1.07 0.21 0.01 0.01 0.13 0.01 0.01

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

0.020 0.052 0.013* −0.004 0.082 0.011**

0.908*** 0.944*** 0.924*** 0.951*** 0.260*** 0.788***

0.024** 0.009 0.015* 0.019** 0.076** 0.002

0.039** −0.001 −0.009 0.017 n/a 0.016**

−0.002 −0.001* −0.002** −0.003* −0.008* −0.001

0.393 −16.459 3.702 8.513 −32.997 1.871

8.85 6.43 3.13 15.4 6.09 4.23

7.25 6.23 8.85 16.2 9.01 3.78

0.11 1.05 0.36 0.06 0.49 0.17

Trade intensity

Diagnostic statistics

0.029* 0.008 0.013 0.005 0.063 0.006

Panel B: Non-ﬁnancial ﬁrms portfolio Country

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel

Constant

0.046** 0.002 0.535*** 0.025* 0.003 0.004 0.003 0.110*** 0.005 0.219*** 0.061***

Lagged conditional correlation

Dummy for global ﬁnancial crisis

Dummy for European sovereign debt crisis

Dummy for Asian and Russian Crises

US business cycle

0.903*** 0.950*** 0.023 0.961*** 0.980*** 0.964*** 0.985*** 0.746*** 0.977*** 0.620*** 0.882***

0.017** 0.019*** 0.013 0.001 0.007* 0.007 0.003 0.059*** 0.005 −0.003* 0.012

0.011 0.014*** −0.008 0.001 0.003 0.005*** −0.002 0.025** 0.004* −0.001 0.012**

0.006 0.019*** 0.018** 0.005 0.015*** 0.005** 0.013** n/a −0.002 −0.005 0.000

−0.002** −0.001* −0.001 −0.001 −0.001 −0.001 −0.000 −0.004** −0.001** 0.001** −0.001

0.709 2.409** 1.222* 0.009 1.370* 0.694* 0.060* 9.103 0.465 −3.689 −0.147

Q (12)

Q2 (12)

ARCH LM test of Engle (1982)

13.0 15.5 13.5 10.5 10.7 8.76 18.1 12.9 9.39 6.16 12.3

3.21 2.53 17.2 5.78 5.81 17.1 6.31 7.85 6.25 0.45 4.37

0.66 0.01 0.29 0.27 0.51 0.01 0.42 1.88 0.56 0.02 0.01

(continued on next page)

152

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

Table 4 (continued) Panel B: Non-ﬁnancial ﬁrms portfolio Country

Constant

Lagged conditional correlation

Dummy for global ﬁnancial crisis

Dummy for European sovereign debt crisis

Dummy for Asian and Russian Crises

US business cycle

Trade intensity

Diagnostic statistics Q (12)

Q2 (12)

ARCH LM test of Engle (1982)

18.4

22.7**

0.09

Developed Italy

−0.028

0.949***

0.015*

0.010**

0.009

−0.001*

Developed Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK

0.015 0.100*** 0.047** 0.015* 0.006 0.020 −0.010 0.034*** 0.006 0.038*** −0.023**

0.967*** 0.862*** 0.906*** 0.967*** 0.977*** 0.960*** 0.955*** 0.955*** 0.970*** 0.936*** 0.988***

0.012** 0.002* 0.017** 0.008 0.007 0.010** 0.011 0.006* 0.004 0.003** 0.004*

0.006** −0.001 0.005 0.006** 0.004 0.004 0.008 0.001 0.004* −0.001 0.002

0.001 0.001 0.003 0.008 0.006 −0.001 −0.007 0.004 0.007 0.004* 0.010*

−0.001** −0.001* −0.000 −0.001 −0.001 −0.000 −0.001* −0.001** −0.001 −0.001* −0.001

0.001 0.069 −0.830 0.824 1.273 0.127 0.637** −1.799 1.021* 0.070* 0.631***

17.3 3.03 5.96 7.70 15.8 6.60 10.6 8.47 5.64 9.63 9.88

2.12 6.75 6.26 1.10 26.7*** 16.6 15.5 4.99 14.9 2.06 1.05

0.89 2.22 0.02 0.25 3.56* 0.14 0.02 0.02 9.1*** 0.09 0.21

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

0.010 0.036*** 0.005 0.006 0.084*** 0.010 0.017** −0.004 0.068*** 0.058*** 0.018 0.027*** 0.014 0.027*** 0.069*** 0.040** 0.002 −0.023 −0.001

0.969*** 0.916*** 0.934*** 0.915*** 0.778*** 0.916*** 0.951*** 0.973*** 0.844*** 0.848*** 0.968*** 0.801*** 0.918*** 0.934*** 0.861*** 0.915*** 0.975*** 0.985*** 0.985***

−0.002 0.004 0.016** 0.030 0.036*** 0.012 0.010 0.003 0.003 0.009 0.010 −0.007** 0.030* 0.002** 0.018** 0.020* 0.012* 0.008 0.010*

0.006 0.000 0.018** 0.017 0.004 0.004 0.016* −0.006 0.003 0.012 0.006 −0.000 0.023 0.001 0.017** 0.023*** 0.007 0.005* −0.002

0.017** n/a 0.001 −0.008 0.033* −0.003 0.024* 0.004 0.002 0.008 0.020 0.003 n/a 0.001 0.013 0.008 0.019 0.007* 0.016

−0.001** −0.001* −0.001 −0.001 −0.004*** −0.002** −0.002* −0.001 −0.001** −0.002** −0.002** −0.000 −0.002 −0.000 −0.003*** −0.002** −0.001 −0.001* −0.001

0.188 −0.001 0.147 0.327 2.606 1.591 0.864 0.658** 0.119 0.205 −0.015 1.885** 0.271 0.072 0.758 −0.375 0.600 0.665* 0.560

8.14 5.05 15.7 16.4 13.0 7.78 8.54 6.37 6.14 5.37 8.45 15.4 10.8 5.92 12.7 9.71 6.57 14.2 15.9

2.60 2.45 14.5 6.58 5.29 10.7 3.31 15.7 5.46 1.80 2.47 14.3 6.55 3.99 13.2 14.6 0.60 12.6 5.07

1.74 0.30 4.67** 0.01 0.05 0.29 0.41 2.09 2.28 0.18 0.77 0.62 0.12 0.14 0.20 0.01 0.02 0.17 0.04

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

0.030** 0.109 0.006 −0.024 0.272* 0.026***

0.910*** 0.858*** 0.873*** 0.811*** 0.106 0.830***

0.019** 0.031** 0.050* 0.088* 0.032 0.010***

0.001 0.010* 0.025* 0.075 0.029 0.003

0.033** −0.001 −0.034 0.094 n/a 0.016**

−0.001 −0.000 −0.003 −0.011** −0.007** −0.001

9.85 15.7 4.54 11.7 14.0 15.1

2.99 6.79 4.36 6.42 10.2 4.04

0.16 0.74 0.08 0.89 0.27 0.03

2.259**

0.717 −7.872 35.518 27.898 −15.911 0.270

Notes: Presents the OLS estimates of the parameters of Eq. (11): ρij,t = β0 + β1ρij,t−1 + β2D1 + β3D2 + β4D3 + β5Cj,t + β6Tij,t + eij,t,

(11)

where ρij,t is the conditional correlation between US and non-US sectoral stock returns in month t. The periods of the Asian and Russian ﬁnancial crises, the global ﬁnancial crisis, and European debt crisis are proxied by the dummy variables, D1,D2,and D3, respectively. Cj,t is the cyclical component of the leading index of the US. Tij,t is the bilateral trade intensity between US and country j. pﬃﬃﬃ Diagnostic statistics are calculated from standardized residuals (eij;t = hij;t ). Engle's ARCH LM test (Engle, 1982) examines whether there are remaining ARCH effects. Q (12) and Q2 (12) are Box–Pierce–Ljung Q test statistics for autocorrelation in standardized residuals and their squares of order 12, respectively. ⁎ Signiﬁcance at the 0.10 level. ⁎⁎ Signiﬁcance at the 0.05 level. ⁎⁎⁎ Signiﬁcance at the 0.01 level.

high DCC regimes coincide with particular economic states, which helps to validate the ﬁndings from the regression model in Eq. (11).

signiﬁcance of individual coefﬁcients slightly differ from those obtained from local-currency-denominated return, the key ﬁndings regarding the effects of the ﬁnancial crises and the US business cycle remain robust.

5.1. US dollar-denominated returns 5.2. Use of weekly data A DCC analysis based on stock returns denominated in local currency ignores the effects of foreign exchange rate ﬂuctuations on estimates of conditional correlations. Therefore, the DCC analysis is replicated using stock returns denominated in US dollars for all country-sector pairs. Eq. (11) is re-estimated using DCC estimated from US dollardenominated stock returns.17 Though the magnitudes and statistical 17

The results are available from the corresponding author on request.

Eq. (11) is re-estimated using DCCs obtained from weekly data. The results are presented in Table A1 in the Appendix.18 Though the magnitudes and statistical signiﬁcance of individual coefﬁcients vary slightly from those obtained from monthly data, the main ﬁndings remain

18 In Table A1 the bilateral trade intensity is not used as a regressor since trade data are not available in a weekly frequency.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

153

robust. Similar to results obtained from monthly data, non-ﬁnancial ﬁrms played a more prominent role in transmitting contagion than ﬁnancial ﬁrms do. For example, conditional correlations between US non-ﬁnancial stock returns and those of 13 developed, nine emerging and ﬁve frontier countries increased signiﬁcantly during the GFC. In case of ﬁnancial ﬁrms, a GFC-induced increase in contagion is for 11 developed, 5 emerging, and 1 frontier countries. 5.3. Alternative measure of business cycle In Section 3, the cyclical component of macroeconomic activity was estimated by applying the Hodrick–Prescott ﬁlter (Hodrick and Prescott, 1997) to the ECRI leading index for the US and then used as a determinant of DCC. As an alternative measure of the business cycle, a dummy variable for NBER business cycle dates for the US recession is used in Eq. (11).19 Overall, the results conﬁrm that DCC is higher during the recession, which may be attributed to the fact that the cyclical component of the US leading index from ECRI coincides well with NBER business cycle dates. 5.4. The Bai and Perron (1998) test of structural breaks Since contagion is often accompanied by an upward structural break in dynamic conditional correlations, the Bai and Perron (1998) test for structural breaks is employed. For example, Chiang et al. (2007) identify two distinct phases of crisis transmission during the Asian ﬁnancial crisis. The ﬁrst phase, the contagion phase, is characterized by an increased DCC. During this phase, shocks spread from crisis-hit countries to other countries, thus increasing stock return volatility. The second phase is characterized by continued high correlation, displaying herding behavior. In the second phase, return correlations remain consistently high as the public becomes more aware of the crisis. To empirically identify these phases, Table A2 shows the results of structural break tests. While we set the maximum number of structural breaks at 5 in the Bai and Perron (1998) test, Table A2 shows the incidence of structural breaks in DCC around the Asian and Russian ﬁnancial crises, GFC, and European debt crisis.20 In this table, the symbols 0, +, and +⁎ are used to indicate no structural break, a structural break, and a structural break with conditional correlation exceeding one standard above the mean, respectively. In the ﬁnancial sector, 40 out 48 country pairs show statistically signiﬁcant structural breaks in DCC during the GFC. Speciﬁcally, in most cases, DCC increases by one standard deviation above the mean during the GFC in the ﬁnancial sector. For example, the DCC for the UK jumps from 0.6930 to 0.8260; for China it changes from 0.1962 to 0.4576. A very similar change in structural breaks in DCC during the GFC is found in the nonﬁnancial sector. Overall, the results demonstrate that the contagion phase of the GFC starts with these breaks, and this is followed by a phase in which DCCs remain persistently high for a certain period of time (herding). However, DCC does not experience such a structural shift during the European debt crisis or the Asian and Russian ﬁnancial crises. Therefore, the beneﬁts of international diversiﬁcation decrease more during a global crisis than a regional crisis. This ﬁnding holds for both sectors and agrees with the results of Eq. (11), where potential structural breaks are predetermined by using dummy variables for ﬁnancial crises. 5.5. Markov regime-switching model In Eq. (11), predetermined economic state variables were used to determine their role in driving conditional correlations. Here, we resort to a data-driven approach to determine contagion regime—a two-state Markov switching model (Hamilton, 1989). In general, we ﬁnd that

Fig. 1. Dynamic conditional correlations and ﬁltered regime probabilities from Markov regime-switching models: January 1990–March 2014. Notes: Regimes 1 and 2 have been derived from the Markov regime-switching model. Following Hamilton (1989), we use a Markov regime-switching model to calculate the regime probabilities of high DCC regimes and low DCC regimes. The ﬁrst-order Markov assumption requires that the probability of being in a regime depends on the previous state, so that P ðSt ¼ jjSt−1 ¼ iÞ ¼ pij ðt Þ We may write these probabilities in a transition matrix as follows: 2 pðt Þ ¼ 4

p11 ðt Þ … … … pM1 ðt Þ …

3 p1M ðt Þ … 5 pMM ðt Þ

where the ijth element represents the probability of transitioning from regime i in period t− 1 to regime j in period t. The vertical axis of each graph represents DCCs and ﬁltered probabilities. Shaded areas represent US recession periods as identiﬁed by the National Bureau of Economic Research (NBER) and the periods of global ﬁnancial crisis, Asian and Russian ﬁnancial crises, and European debt crisis.

the high DCC regime coincides with periods of bad economic states (GFC and US economic downturns). From our numerous results for the Markov switching model, Fig. 1 presents regime probabilities along with DCCs for two country–sector pairs: the US and Italian ﬁnancial sectors, and the US and Chinese non-ﬁnancial sectors. For the US– Italy pair, Fig. 1 clearly shows that regime 1 often corresponds to a low DCC scenario accompanied by a good economic state (unshaded area), while regime 2 often corresponds to a high DCC scenario with a bad economic state (shaded area). For the US–China pair, we ﬁnd such evidence only after the outbreak of the GFC. These ﬁndings validate the results of the regression model of DCC in Eq. (11). 6. Conclusion

19

To conserve space these results have not been reported but they are available from the corresponding author. 20 The complete set of results is available from the corresponding author on request.

The contagion literature is overrepresented by studies at the aggregate market level, often for a small number of countries. At the sectoral

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

154

level, the key focus of the extant literature is contagion through ﬁnancial ﬁrms. This study examines contagion through non-ﬁnancial ﬁrms using a sample of 49 countries, including developed, emerging, and frontier countries. The DCC-GARCH models show that conditional correlation between the US sectoral return and that of another country varies over time. The magnitude of conditional correlation is positively related to the level of equity market development of the respective non-US country. For most country–sector pairs, conditional correlations ﬂuctuate considerably over time, but the values remain positive. However, conditional correlations switch between positive and negative values for several frontier and emerging markets. These ﬁndings highlight the need for US investors to rebalance frequently their international equity portfolio in response to ﬂuctuating DCCs. A novel ﬁnding of our study is that non-ﬁnancial ﬁrms play a more pronounced role in the cross-country transmission of shocks than ﬁnancial ﬁrms. The average conditional correlation over 25 years (1990–2014) for the non-ﬁnancial sector is greater than that of the ﬁnancial sector in 44 out of 48 country pairs. Regressing DCCs on the

economic state variables, we ﬁnd that the GFC's effect on conditional correlations is more prevalent in the non-ﬁnancial sector and pervasive across countries, regardless of their level of market development. Sectoral contagion across countries was least pervasive during the European sovereign debt crisis. The contagion effect of the Asian and Russian ﬁnancial crises lies somewhere in between the two other crises examined in this study. As expected, conditional correlation is positively related to the bilateral trade intensity. These ﬁndings from OLS regressions are robust to the use of a data-driven approach to determining structural shifts in DCCs (Bai and Perron, 1998) and contagion regime (Hamilton, 1989). This study suggests that the potential beneﬁts of international diversiﬁcation depend on the time-varying properties of conditional correlations, which in turn vary with US business cycle, bilateral trade intensity, ﬁnancial crises, and the type of sector. Consequently, understanding dynamic conditional correlations is critical for international investors and corporate policymakers if they are to develop strategies for coping better with foreign shocks. Regulators may also ﬁnd this information useful in determining the nature and timing of policy interventions to contain contagion.

Appendix A Table A1 Determinants of dynamic conditional correlations with respect to the US using weekly data. Panel A: Financial ﬁrms portfolio Country

Constant

Lagged conditional correlation

Dummy for global ﬁnancial crisis

Dummy for European sovereign debt crisis

Dummy for Asian and Russian crises

US business cycle

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK

0.006*** 0.003** 0.002 0.018*** 0.002* 0.002 0.002 0.0007 0.027*** 0.095*** 0.001* 0.005** 0.006*** 0.004 0.057*** 0.0018 0.006*** 0.003** 0.0011 0.015*** 0.009*** 0.015*** 0.002*

0.983*** 0.989*** 0.997*** 0.970*** 0.995*** 0.996*** 0.997*** 0.996*** 0.930*** 0.791*** 0.995*** 0.989*** 0.979*** 0.994*** 0.739*** 0.994*** 0.967*** 0.990*** 0.995*** 0.969*** 0.980*** 0.975*** 0.996***

0.002* 0.004* 0.001 0.005** 0.002 0.001* 0.0001 0.003* 0.0010 0.007** 0.003* 0.0022 0.002*** 0.0002 0.007* 0.0018 0.002** 0.0020 0.0003 0.003 0.003 0.001* 0.0020

0.004** 0.005** 0.0001 0.003 0.002* 0.001 0.001 −0.0005 0.006* 0.0033 0.002* 0.0017 0.001** 0.0013 −0.0013 0.002* −0.00004 0.0012 0.0016 0.002 0.005* 0.003 0.001*

0.001 0.002 −0.0001 0.003 0.004* 0.003** 0.003* 0.003*** 0.0008 0.007** 0.002** 0.0029 0.000 0.0011 0.012** 0.0003* −0.001 0.001* 0.001** 0.003* 0.003 0.001* 0.0009

−0.001*** −0.001 −0.0001 −0.0003 −0.0002 −0.0001 −0.0002 −0.0001 −0.002*** 0.0001 −0.0000 −0.0002 −0.0001* −0.0000 −0.0005* −0.0009* −0.0001* −0.0004*** −0.0004* −0.0003* −0.001** −0.0002 −0.0000

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

0.004*** 0.0010 0.002** 0.001** 0.001* 0.001** 0.002*** 0.001 0.028*** 0.003*** 0.003** 0.002*** 0.0003 0.003*** 0.001 0.002** 0.002* 0.001 0.003

0.989*** 0.992*** 0.987*** 0.993*** 0.994*** 0.985*** 0.989*** 0.997*** 0.812*** 0.985*** 0.990*** 0.958*** 0.965*** 0.984*** 0.995*** 0.992*** 0.993*** 0.997*** 0.976***

0.003 0.0024 0.003* 0.0019 0.002* 0.003** 0.002 0.001 0.007 0.0010 0.0032 0.0004 0.006*** 0.003* 0.001 0.001 0.001 0.0004 0.0040

0.002 0.0020 0.003* 0.0007 0.001 0.003** 0.003* 0.001 0.007*** 0.0008 0.0013 0.001** 0.008** 0.002** 0.002* 0.003* 0.001 0.001* 0.0050

−0.001 0.0014 −0.001 0.0001 0.001 −0.0012 0.005* 0.0004* 0.003 0.0004 0.0018 0.0004 n/a 0.005** 0.003* 0.0001 0.003* 0.002 0.006*

−0.0002* −0.0001 −0.0003* −0.0002* −0.0001** −0.0002* −0.001** −0.0002* −0.0002 −0.0001 −0.0004** 0.000 −0.0002 −0.000 −0.0003*** −0.0002* −0.0004*** −0.0002** −0.0002

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

155

Table A1 (continued) Panel A: Financial ﬁrms portfolio Country

Constant

Lagged conditional correlation

Dummy for global ﬁnancial crisis

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

0.038*** 0.010*** 0.002* 0.003*** 0.004** 0.001***

0.891*** 0.909*** 0.989*** 0.985*** 0.972*** 0.961***

0.002* 0.005 0.002 0.001 0.005 0.002

Dummy for European sovereign debt crisis 0.005** 0.003 0.002* 0.002* 0.003* 0.001

Dummy for Asian and Russian crises

US business cycle

0.007** n/a 0.0001 −0.002 n/a 0.003*

−0.001** −0.0003 −0.0003* −0.0002 −0.0001 −0.0001

Panel B: Non-ﬁnancial ﬁrms portfolio Country

Constant

Lagged conditional correlation

Dummy for global ﬁnancial crisis

Dummy for European sovereign debt crisis

Dummy for Asian and Russian crises

US business cycle

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK

0.006*** 0.006*** 0.004** 0.066*** 0.002 0.002 0.002 0.002** 0.003** 0.007*** 0.009*** 0.0028 0.005*** 0.004 0.013*** −0.0210 0.002*** 0.001 0.002* 0.002 0.001 0.006* 0.003

0.988*** 0.985*** 0.991*** 0.905*** 0.997*** 0.997*** 0.997*** 0.990*** 0.993*** 0.985*** 0.979*** 0.994*** 0.981*** 0.994*** 0.957*** 0.700*** 0.977*** 0.996*** 0.995*** 0.996*** 0.998*** 0.989*** 0.995***

0.003* 0.004* 0.003 0.0007 0.001 0.001** 0.0008 0.003** 0.0020 0.0019 0.0030 0.002** 0.001*** 0.002 0.007 0.112* 0.003** 0.001* 0.002** 0.001* 0.001* 0.001** 0.0012

0.002 0.005** 0.001 0.0031 0.0013 0.0011 0.0006 0.0011 0.0014 0.004** 0.0023 0.0019 0.001** 0.001 0.005** 0.101*** −0.00004 0.001 0.0021 0.001 0.000 0.0013 0.0010

0.001 0.001* 0.001 0.0024 0.003*** 0.002* 0.003* 0.004** 0.0007 0.0036 0.0017 0.0026 0.0003 0.001 0.007*** −0.0637 −0.0006 0.002 0.0011 0.001 0.001 0.002* 0.004*

−0.001* 0.0001 −0.000 −0.0004 −0.0001 0.0001 0.0001 −0.0001* −0.0003*** −0.0001 −0.0001 0.0002 −0.0001* 0.0001 −0.000 0.0002 −0.0001* −0.0001* −0.0001** −0.00003 −0.0001 −0.00001* −0.00003

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

0.005*** 0.009*** 0.003* 0.002* 0.016*** 0.003*** 0.005*** 0.001 0.002*** 0.005*** 0.002 0.003*** 0.004** 0.005*** 0.003* 0.004** 0.002* 0.001 0.002

0.989*** 0.966*** 0.987*** 0.987*** 0.933*** 0.979*** 0.983*** 0.997*** 0.987*** 0.981*** 0.996*** 0.968*** 0.975*** 0.979*** 0.990*** 0.986*** 0.992*** 0.995*** 0.986***

0.003 0.009 0.004* 0.003 0.009** 0.004* 0.002 0.001 0.003 0.0010 0.001 0.0004 0.007** 0.004** 0.002 0.003* 0.003* 0.004** 0.005*

0.002 0.007* 0.004* 0.003 0.004 0.002 0.002 0.001 0.003** 0.0015 0.000 0.002*** 0.005* 0.002** 0.002* 0.003* 0.002 0.002 0.006

0.001 0.009*** 0.002 0.002 0.008* −0.0003 0.004*** 0.001 0.003* 0.0003 0.002 0.0004 n/a 0.003** 0.004* −0.001 0.005** 0.004** 0.006*

0.00003 −0.001** −0.0004** −0.001** −0.001** −0.0002 −0.0003* −0.0001 −0.00003 −0.0001* −0.0002* 0.00003 −0.0002 −0.00002 −0.0002* −0.0001** 0.00001 −0.0001* −0.0002

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

0.045*** −0.0005 0.003*** 0.001 0.001* 0.004***

0.879*** 0.990*** 0.978*** 0.985*** 0.989*** 0.799***

0.003* 0.004** 0.006* 0.008*** 0.004* 0.007

0.007* n/a −0.002 0.001 n/a 0.009*

−0.001 −0.00002 −0.001* −0.0001 −0.0002* −0.0002

0.004 0.0008 0.002 0.005** −0.0001 0.004

Notes: Presents the OLS estimates of the parameters of Eq. (11): ρij,t = β0 + β1ρij,t−1 + β2D1 + β3D2 + β4D3 + β5Cj,t + Tij,t + eij,t,

(11)

where ρij,t is the conditional correlation between US and non-US sectoral stock returns in month t. The periods of the Asian and Russian ﬁnancial crises, the Global ﬁnancial crisis and European debt crisis are proxied by the dummy variables, D1,D2,and D3, respectively. Cj,t is the cyclical component of the leading index of the US. Tij,t is the bilateral trade intensity between US and country j. ⁎ Signiﬁcance at the 0.10. ⁎⁎ Signiﬁcance at the 0.05. ⁎⁎⁎ Signiﬁcance at the 0.01 level.

156

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

Table A2 Structural breaks in dynamic conditional correlations of ﬁnancial and non-ﬁnancial ﬁrms using the Bai–Perron method. Country

Structural breaks in DCC of financial firms

Structural breaks in DCC of non-ﬁnancial firms

Asian and Russian ﬁnancial crises

Global ﬁnancial crisis

European debt crisis

Asian and Russian ﬁnancial crises

Global ﬁnancial crisis

European debt crisis

Developed Australia Austria Belgium Canada Finland France Germany Greece Hong Kong Ireland Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore South Korea Spain Sweden Switzerland UK

+ + 0 0 0 + + 0 0 0 0 0 + 0 + 0 0 + + 0 0 + 0

+⁎ +⁎ + +⁎ +⁎ +⁎ +⁎ +⁎

+ 0 0 0 0 0 0 0 + + 0 0 0 0 0 + 0 0 0 + 0 0 0

+ 0 0 0 0 + + 0 0 0 + 0 0 0 0 + + + 0 + + + 0

+⁎ +⁎ 0 0 + +⁎ +⁎

+ 0 0 0 0 0 0 + 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Emerging Brazil Chile China Colombia Czech Republic Egypt Hungary India Indonesia Malaysia Mexico Pakistan Peru Philippines Poland Russia South Africa Taiwan Turkey

+ + + 0 0 0 0 + 0 0 + 0 0 + + + + + +

+⁎ +⁎ +⁎ +⁎ +⁎ +⁎

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

+ 0 0 + 0 0 0 + 0 + +⁎

+⁎ + +⁎ +⁎

0 0 + + + + + +

0 +⁎ + +⁎ +⁎ +⁎ +⁎ +⁎

Frontier Argentina Bulgaria Cyprus Romania Slovenia Sri Lanka

+ 0 0 + 0 0

+⁎ +⁎ +⁎ +⁎

0 0 0 0 0 0

0 0 0 + 0 0

+⁎ +⁎ +⁎

0 0 +⁎ +⁎ +⁎ +⁎ + 0 +⁎ +⁎ +⁎ 0 +⁎ +⁎ +⁎

0 +⁎ 0 +⁎ +⁎ 0 +⁎ +⁎ +⁎ +⁎ +⁎ +⁎ +⁎

0 +

0 +⁎ 0 +⁎ +⁎ +⁎ +⁎ + +⁎ +⁎ +⁎ +⁎ +⁎ +⁎ +⁎ +⁎

0 +⁎ +⁎ 0 +⁎ +⁎ +⁎

0 0 +

0 + 0 0 0 + 0 0 0 0 0 0 + 0 + 0 0 0 0

+ + + 0 0 0

Notes: (a) Bai and Perron (1998) test is used to ﬁnd if there are structural breaks in DCC using Bayesian Information Criterion (BIC). (b) + indicates a structural break in the period of respective ﬁnancial crisis while 0 indicates no structural break in the period of respective ﬁnancial crisis. +⁎ indicates a structural break with the dynamic conditional correlation coefﬁcient exceeding one standard above the mean.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

157

A) Developed countries

Fig. A1. Dynamic conditional correlations (Financial ﬁrms) with respect to the US. Notes: This ﬁgure illustrates conditional correlation between US ﬁnancial stock returns and those of pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ developed, emerging, and frontier countries. Conditional correlations are calculated from the DCC-GARCH model in Eq.(10): ρij;t ¼ qij;t = qii;t qjj;t i; j ¼ 1; 2; …n; and i≠j (10). In Eq. (10), ρij,t and qij,t represent the conditional correlation and covariance at time t between US and non-US sectoral stock returns, respectively. qii,t and qjj,t are the conditional variance of sectoral stock returns in the non-US and US markets, respectively. Shaded areas represent US recession periods as identiﬁed by the National Bureau of Economic Research (NBER). The vertical axis for Panels A, B, and C indicates the coefﬁcient of DCC.

158

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

B) Emerging countries

Fig. A1 (continued).

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

C) Frontier countries

Fig. A1 (continued).

159

160

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

A) Developed countries

Fig. A2. Dynamic conditional correlations (non-ﬁnancial ﬁrms) with respect to the US. Notes: This ﬁgure illustrates conditional correlation between US ﬁnancial stock returns and those of pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ developed, emerging, and frontier countries. Conditional correlations are calculated from the DCC-GARCH model in Eq. (10): ρij;t ¼ qij;t = qii;t qjj;t i; j ¼ 1; 2; …n; and i≠j (10). In Eq. (10), ρij,t and qij,t represent the conditional correlation and covariance at time t between US and non-US sectoral stock returns, respectively. qii,t and qjj,t are the conditional variance of sectoral stock returns in the non-US and US markets, respectively. Shaded areas represent US recession periods as identiﬁed by the National Bureau of Economic Research (NBER). The vertical axis of each graph represents DCCs.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

B) Emerging countries

Fig. A2 (continued).

161

162

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163

C) Frontier countries

Fig. A2 (continued).

References Akhtaruzzaman, M., Shamsuddin, A., Easton, S., 2014. Dynamic correlation analysis of spill-over effects of interest rate risk and return on Australian and US ﬁnancial ﬁrms. J. Int. Financ. Mark. Inst. Money 31, 378–396. Aloui, R., Aïssa, M.S.B., Nguyen, D.K., 2011. Global ﬁnancial crisis, extreme interdependences, and contagion effects: the role of economic structure? J. Bank. Financ. 35, 130–141. Ayhan Kose, M., Otrok, C., Whiteman, C.H., 2008. Understanding the evolution of world business cycles. J. Int. Econ. 75, 110–130. Bai, J., Perron, P., 1998. Estimating and testing linear models with multiple structural changes. Econometrica 66, 47–78. Baillie, R.T., Ramon, P.D., 1990. Stock returns and volatility. J. Financ. Quant. Anal. 25, 203–214. Batten, J., Loncarski, I., Szilagyi, P., 2013. Recent Cross-Border Banking within Europe: International Balance Sheet Perspectives. Vol. 14. Emerald Group Publishing Limited, pp. 15–30 (2013). Baur, D.G., 2012. Financial contagion and the real economy. J. Bank. Financ. 36, 2680–2692.

Beirne, J., Gieck, J., 2014. Interdependence and contagion in global asset markets. Rev. Int. Econ. 22, 639–659. Bekaert, G., Harvey, C.R., Ng, A., 2005. Market integration and contagion. J. Bus. 78, 39–70. Burstein, A., Kurz, C., Tesar, L., 2008. Trade, production sharing, and the international transmission of business cycles. J. Monet. Econ. 55, 775–795. Cai, X.J., Tian, S., Hamori, S., 2016. Dynamic correlation and equicorrelation analysis of global ﬁnancial turmoil: evidence from emerging East Asian stock markets. Appl. Econ. 1–15. Calvo, S., Reinhart, C.M., 1996. Capital ﬂows to Latin America: is there evidence of contagion effects? World Bank Policy Research Working Paper Cetorelli, N., Goldberg, L.S., 2011. Global banks and international shock transmission: evidence from the crisis. IMF Econ. Rev. 59, 41–76. Cheung, L., Tam, C.-S., Szeto, J., 2009. Contagion of ﬁnancial crises: a literature review of theoretical and empirical frameworks. Hong Kong Monetary Authority Research Paper 02. Chiang, T.C., Jeon, B.N., Li, H., 2007. Dynamic correlation analysis of ﬁnancial contagion: evidence from Asian markets. J. Int. Money Financ. 26, 1206–1228. Claessens, S., Tong, H., Wei, S.-J., 2012. From the ﬁnancial crisis to the real economy: using ﬁrm-level data to identify transmission channels. J. Int. Econ. 88, 375–387.

M. Akhtaruzzaman, A. Shamsuddin / Economic Modelling 59 (2016) 143–163 Corsetti, G., Pericoli, M., Sbracia, M., 2005. ‘Some contagion, some interdependence’: more pitfalls in tests of ﬁnancial contagion. J. Int. Money Financ. 24, 1177–1199. Cravino, J., Levchenko, A.A., 2015. Multinational ﬁrms and international business cycle transmission. University of Michigan Working Papers. Vols. 1-58. Desai, M.A., Foley, C.F., 2006. The comovement of returns and investment within the multinational ﬁrm. NBER International Seminar on Macroeconomics 2004, NBER International Seminar on Macroeconomics 2004. The MIT Press, pp. 197–240 (2006). Dimitriou, D., Kenourgios, D., Simos, T., 2013. Global ﬁnancial crisis and emerging stock market contagion: a multivariate FIAPARCH–DCC approach. Int. Rev. Financ. Anal. 30, 46–56. Dungey, M., Gajurel, D., 2014. Equity market contagion during the global ﬁnancial crisis: evidence from the world's eight largest economies. Econ. Syst. 38, 161–177. Elyasiani, E., Mansur, I., 2003. International spillover of risk and return among major banking institutions: a bivariate GARCH model. J. Account. Audit. Finance 18, 303–330. Engle, R.F., 1982. Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inﬂation. Econometrica 50, 987–1007. Engle, R.F., 2002. Dynamic conditional correlation: a simple class of multivariate generalized autoregressive conditional heteroskedasticity models. J. Bus. Econ. Stat. 20, 339. Engle, R.F., Kroner, K.F., 1995. Multivariate simultaneous generalized ARCH. Econ. Theory 11, 122–150. Ferreira, M.A., Gama, P.M., 2010. Correlation dynamics of global industry portfolios. J. Multinational Financ. Manag. 20, 35–47. Forbes, K.J., Rigobon, R., 2002. No contagion, only interdependence: measuring stock market comovements. J. Financ. 57, 2223–2261. Frankel, J.A., Rose, A.K., 1998. The endogenity of the optimum currency area criteria. Econ. J. 108, 1009–1025. Giovanni, D.J., Levchenko, A.A., 2010. Putting the parts together: trade, vertical linkages, and business cycle comovement. Am. Econ. J. Macroecon. 2, 95–124. Hamilton, J.D., 1989. A new approach to the economic analysis of nonstationary time series and the business cycle. Econ. J. Econ. Soc. 57, 357–384. Hodrick, R.J., Prescott, E.C., 1997. Postwar U.S. business cycles: an empirical investigation. J. Money Credit Bank. 29, 1–16. Horta, P., Lagoa, S., Martins, L., 2016. Unveiling investor-induced channels of ﬁnancial contagion in the 2008 ﬁnancial crisis using copulas. Quant. Finan. 16, 1–13. Horta, P., Mendes, C., Vieira, I., 2010. Contagion effects of the subprime crisis in the European NYSE Euronext markets. Port. Econ. J. 9, 115–140. Hwang, E., Min, H.-G., Kim, B.-H., Kim, H., 2013. Determinants of stock market comovements among US and emerging economies during the US ﬁnancial crisis. Econ. Model. 35, 338–348. IHS Global Inc., 2015. EViews 9 User's Guide II. IHS Global Inc. Kalemli-Ozcan, S., Papaioannou, E., Perri, F., 2013. Global banks and crisis transmission. J. Int. Econ. 89, 495–510. Kaufman, G.G., 1994. Bank contagion: a review of the theory and evidence. J. Financ. Serv. Res. 8, 123–150.

163

Kenourgios, D., Samitas, A., Paltalidis, N., 2011. Financial crises and stock market contagion in a multivariate time-varying asymmetric framework. J. Int. Financ. Mark. Inst. Money 21, 92–106. Kim, B.-H., Kim, H., Lee, B.-S., 2015. Spillover effects of the U.S. ﬁnancial crisis on ﬁnancial markets in emerging Asian countries. Int. Rev. Econ. Financ. 39, 192–210. Kim, B.-H., Kim, S., 2013. Transmission of the global ﬁnancial crisis to Korea. J. Policy Model 35, 339–353. King, M.A., Wadhwani, S., 1990. Transmission of volatility between stock markets. Rev. Financ. Stud. 3, 5–33. Kollmann, R., 2013. Global banks, ﬁnancial shocks, and international business cycles: evidence from an estimated model. J. Money Credit Bank. 45, 159–195. Kose, M.A., Yi, K.-M., 2001. International trade and business cycles: is vertical specialization the missing link? Am. Econ. Rev. 91, 371–375. Longin, F., Solnik, B., 1995. Is the correlation in international equity returns constant: 1960-1990? J. Int. Money Financ. 14, 3–26. Milesi-Ferretti, G.-M., Tille, C., 2011. The great retrenchment: international capital ﬂows during the global ﬁnancial crisis. Econ. Policy 26, 289–346. Min, H.-G., Hwang, Y.-S., 2012. Dynamic correlation analysis of US ﬁnancial crisis and contagion: evidence from four OECD countries. Appl. Financ. Econ. 22, 2063–2074. Mink, M., 2015. Measuring stock market contagion: local or common currency returns? Emerg. Mark. Rev. 22, 18–24. Mollah, S., Quoreshi, A.S., Zaﬁrov, G., 2016. Equity market contagion during global ﬁnancial and Eurozone crises: evidence from a dynamic correlation analysis. J. Int. Financ. Mark. Inst. Money 41, 151–167. Moore, T., Wang, P., 2014. Dynamic linkage between real exchange rates and stock prices: evidence from developed and emerging Asian markets. Int. Rev. Econ. Financ. 29, 1–11. Newey, W.K., West, K.D., 1987. A simple, positive semi-deﬁnite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55, 703–708. Pericoli, M., Sbracia, M., 2003. A primer on ﬁnancial contagion. J. Econ. Surv. 17, 571–608. Phylaktis, K., Xia, L., 2009. Equity market comovement and contagion: a sectoral perspective. Financ. Manag. 38, 381–409. Samitas, A., Tsakalos, I., 2013. How can a small country affect the European economy? The Greek contagion phenomenon. J. Int. Financ. Mark. Inst. Money 25, 18–32. Sandra, E., 2007. Business cycle transmission from the US to Germany – a structural factor approach. Eur. Econ. Rev. 51, 521–551. Shen, P.-L., Li, W., Wang, X.-T., Su, C.-W., 2015. Contagion effect of the European ﬁnancial crisis on China's stock markets: interdependence and pure contagion. Econ. Model. 50, 193–199. Syllignakis, M.N., Kouretas, G.P., 2011. Dynamic correlation analysis of ﬁnancial contagion: evidence from the Central and Eastern European markets. Int. Rev. Econ. Financ. 20, 717–732. Tong, H., Wei, S.-J., 2011. The composition matters: capital inﬂows and liquidity crunch during a global economic crisis. Rev. Financ. Stud. 24, 2023–2052. Tse, Y.K., 2000. A test for constant correlations in a multivariate GARCH model. J. Econ. 98, 107–127.

International contagion through financial versus non-financial firms

International contagion through financial versus non-financial firms

Recommend Documents