- Email: [email protected]
Contagion and impulse response of international stock markets around the 9–11 terrorist attacks
Global Finance Journal 16 (2005) 48 – 68
Contagion and impulse response of international stock markets around the 9–11 terrorist attacks C. Kyung-Chun Mun* Division of Business and Accountancy, Truman State University, Kirksville, MO 63501, United States Received 4 March 2005; received in revised form 9 April 2005; accepted 10 April 2005 Available online 14 June 2005
Abstract This study investigates both return and volatility contagion effects of the 9–11 terrorist attacks across the major markets, and examines the extent to which international stock markets can be destabilized by shocks that arise in the US. Evidence presented in this paper suggests that to the extent that higher correlations with the US market can enhance contagion effects from the US, the attacks brought about volatility contagion (rather than return contagion) from the US to UK and German markets. In contrast, the Japanese market had return contagion (rather than volatility contagion) from the US market. After the attacks, a US shock had a strongly positive effect on the US/Japan return correlation but had little or no effect in response functions of the return correlation for the US/UK and US/German markets. Impulse responses of the volatility correlation to a US shock notably increased after the attacks for the US/UK and US/German markets. An overall analysis of the post-attack period reveals that international market correlations were strengthened through volatility for the US/UK and US/German markets and through return for the US/Japanese market. D 2005 Elsevier Inc. All rights reserved. JEL classification: F21 Keywords: 9–11 terrorist attacks; Contagion; Impulse response
1. Introduction The tragic events of September 11, 2001 exacerbated an already very difficult situation in the US and world stock markets. The shock in the US market driven by the terrorist attacks * Tel.: +1 660 785 4367. E-mail address: [email protected]. 1044-0283/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.gfj.2005.05.002
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
49
Table 1 Summary statistics of two-day stock returns Country
Period
Mean of returns (%)
Standard deviation (%)
US
I 0 II I 0 II I 0 II I 0 II
0.1796 5.0468 0.1396 0.1784 4.9366 0.1615 0.2645 7.1447 0.2006 0.2195 5.8326 0.0548
1.6550 3.0513* 1.5350 1.4101 4.0872* 1.5199 1.7621 5.3320* 2.4053 1.8844 2.7433* 1.8437
UK
Germany
Japan
Period I is the time period before the terrorist attacks; period 0 is the date immediately following the attacks; and period II is the time period after the attacks. *Indicates the values of conditional volatility obtained from the standard deviation of residuals for which each index return is regressed on constant within a univariate GARCH (1,1) process.
gave rise to a synchronous downturn across nearly all major regions of the world. As seen from Table 1, two-day mean returns immediately following the attacks fell by 5.05% for the S & P 500, 4.94% for the FTSE 100 of the UK, 7.14% for the DAX of Germany, and 5.83% for the TOPIX of Japan. At the same time, the market volatility as measured by standard deviation rose drastically across the markets. Apparently, increased international financial linkages have played an important role in the synchronicity of global market downturns. Many people have spoken or written about September 11 and its economic consequences. Enough perspectives exist to assess the impact of the attacks on the global economic situation, i.e., weakened consumer and business confidence, worsening financing conditions for emerging market economies, and rising transactions costs, etc. The attacks, however, have potentially far-reaching implications for stock market behavior and deserve special attention in the finance literature for at least four reasons. First, although there were many incidents that severely affected US stock markets in the past, the 9–11 attacks appeared to be unique in the pattern of US stock market behavior after market crashes.1 As seen from Fig. 1, while there was a gradual market resilience shortly after the crash for other incidents, the 9–11 attacks caused stock markets to plunge to a prolonged bear market after the attacks due to a long-term decline in consumer and investor confidence.2 1
The US stock market crashes for the last three decades as measured by more-than-3% drop in daily S&P 500 market indexes include the Hunt silver crisis of 1980, the Black Monday of 1987, the Iraqi invasion of Kuwait of 1990, the Asian financial crisis of 1997, and the 9–11 terrorist attacks of 2001. 2 One can claim that the prolonged bear market after the 9–11 attacks may be attributable to the business cycle rather than the 9–11 attacks. As reported by the NBER’s Business Cycle Dating Committee (2003; http:// www.nber.org/cycles), a recession began in March 2001 and the economy reached its trough (a turning point) in November 2001, entering a recovery. This observation in the business cycle that the aggregate economic activity has been rising since November 2001 does not seem to be compatible with overall stock market conditions that continued to be bearish for a long period of time even after November 2001. This suggests that a bear market after the 9–11 attacks may not be well explained by the business cycle.
50
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68 180 160 140 120 100 80 60 40 20 0 -200
9/11 Attacks of 2001 Asian Crisis of 1997 Iraqi Invasion of 1990 Black Monday of 1987 Hunt Silver of 1980 -150
-100
-50
0
+50
+100
+150
Fig. 1. S & P 500 index movements around market crashes. S & P index movements are shown here for 200 days before crash date (t = 200 to 1), the date of crash (t = 0), and 200 days after the crash (t = +1 to +200). The values of S & P 500 indecies for t = 200 are set to 100 as the base point.
Second, as we documented in this paper, the long-term bear market after the 9–11 attacks contributed to a structural change in cross-market correlations. Being characterized by a synchronous fall in returns and rise in volatility across international markets, the aftermath of the terrorist attacks implied increased international financial linkages, which in turn contributed to higher cross-market correlations. Third, the fact that the attacks were premeditated and therefore could be repeated has had a significant impact on investor sentiment. Terrorism could now become a part of the fabric of investor sentiment, putting a new risk factor in the equation for international investors. Finally, given the virtually unprecedented shock in nature with global reach, the attacks created a new level of international transmission of financial shocks to the global market. International stock markets now respond more sensitively to the shocks that arise in the US market. Various studies have documented that international stock market correlations increase when markets are volatile (see Campbell, Koedijk, & Kofman ,2002; King, Sentana, & Wadhwani, 1994; Longin & Solnik, 1995; Solnik, Boucrelle, & Fur, 1996). The international market correlation could change because international market volatility changes over time and/or because interdependence across markets changes. If cross-market correlations increase significantly after a crisis, there is a contagion from the US to other markets (see Forbes & Rigobon, 2002).3 Cross-market contagion driven by the attacks could provide an adverse investment environment for risk managers and internationally diversified investors since the changing correlation pattern and 3
There is no general consensus as to the definition of contagion. We will explore the Forbes and Rigobon’s (2002) definition of contagion in the remainder of this paper.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
51
financial disturbances make it more difficult for them to select an ex ante optimal investment strategy. The objective of this paper is twofold. First, we investigate evidence of contagion in returns and volatility of the terrorist attacks across the major markets (UK, Germany, and Japan). We document the evidence by assessing how significant the changes were in cross-market return and volatility correlations after the attacks. Second, in order to obtain additional insights into the contagion of international stock markets, we examine the extent to which shocks to international markets are explained by the US market’s behavior that resulted from the attacks. The attacks could have led other national stock markets to be destabilized through the market contagion. This paper does not intend to investigate the channels through which shocks driven by the attacks are transmitted across countries, i.e., international trade, foreign exchange rates, investor sentiment, or any unobservable fundamental variables. Instead, we focus on the question of whether there was a structural change in international market correlations and in the pattern of impulse responses after the attacks. The dynamic volatility and correlation coefficients across markets are obtained using a bivariate generalized autoregressive conditional heteroskedastic (GARCH) model. Tests for contagion from the US to other countries are conducted by estimating a multivariate GARCH model with constant conditional correlations. The presence of any structural changes in international market correlations around the 9– 11 attacks is tested using a pure structural change model developed by Bai and Perron (1998, 2003). Impulse response functions are obtained by estimating a standard vector autoregression (VAR) model and the analysis is performed over two subsamples (pre-and post-attack). We find that to the extent that significantly higher correlations with the US market after the attacks reflect a contagion from the US, the attacks gave rise to a volatility contagion (rather than a return contagion) from the US to UK and German markets. Although the US/UK and US/German markets showed no return contagion, a strong dinterdependenceT in market returns continued after the attacks.4 In contrast, the Japanese market had a return contagion (rather than a volatility contagion) from the US market and binterdependenceQ in market volatility continued between Japan and the US. Statistical evidence presented in this study indicates that there was an abrupt structural change in the dynamics of international stock market correlations around the 9–11 attacks. We also find that, after the attacks, impulse responses of market correlation to a US shock notably increased through volatility for the US/UK and US/ German markets, whereas the corresponding response of the US/Japanese market increased through returns. The paper is organized as follows: Section 2 presents data and methodology for empirical analysis of international market correlations. Tests for contagion and structural change and impulse response analysis are provided in Section 3. Conclusions are given in Section 4. 4 Forbes and Rigobon (2002) claims that if there is no significant increase in correlation coefficient but there exists a continued high level of correlation after a crisis, there exists an dinterdependenceT, not the contagion, between countries.
52
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
2. Data and methodology 2.1. Data The data set we analyze is based on a newly compiled two-day average series for the period from November 20, 2000 to June 27, 2002, providing a sample size of 203 observations. Following Forbes and Rigobon (2002), two-day average data are obtained from the average of two consecutive daily series with adjustment for weekends and holidays to avoid the problems of non-synchronous trading between the US and other sample countries. The cutoff date for the last observation was selected to avoid a possible confounding effect that may not have been significantly associated with the attacks.5 The first data point was selected to make the time length before and after the attacks balanced in an attempt to make cross-period comparisons of appropriate variables. The data consist of closing stock market indices of four countries: the S & P 500 of US, the FTSE 100 of UK, the German Stock Market Index (DAX) of Germany, and the Tokyo Stock Exchange Price Index (TOPIX) of Japan. These stock market indices are broad market-value-weighted indices and collected from the Global Financial Data base. Two-day index returns are computed as the logarithmic difference of two-day averages of stock index values. To examine time series behavior of the international stock markets before and after the terrorist attacks, we divide the whole sample into two sub-periods: (1) pre-attack period, from November 20, 2000 to September 10, 2001 and (2) post-attack period, from September 20, 2001 to June 27, 2002. The exact date of division of the sample period was chosen to make the sample size of each period balanced. The data points for September 12 and 14 were removed from each of the periods because the New York Stock Exchange (NYSE) was closed for the week of the attacks (September 11– 14, 2001) and reopened on September 17, 2001. The two observations (September 12 and 14, 2001) reflect the immediate impact of the attack on foreign stock markets and the day that the NYSE reopened reflects the immediate impact of the attacks on the US stock market. 2.2. Summary statistics Table 1 presents summary statistics of mean and standard deviation of returns around the attacks for stock indexes of US, UK, Germany, and Japan. Immediately after the attacks, there was a drastic drop in man returns and a concomitant increase in return volatility for all the sample countries. For example, the two-day market returns before the attacks were 0.18% for the US, 0.18% for the UK, 0.26%, and 0.22% for Japan, but immediately after the attacks returns for the corresponding 5
All the sample countries suffered an extreme bear market during July and August of 2002 when the US stock market rapidly fell due largely to the aftermath of financial scandals that have involved accounting irregularities by some major companies. Also, the first half of 2000 should not be included in the sample because the sample markets were exceptionally bullish due to the exuberance in technology-sector stocks.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
53
countries fell to 5.05%, 7.14%, and 5.83%, respectively. On the other hand, the return volatility rose drastically across the sample markets immediately after the attacks: the conditional volatility before the attacks was 1.65% for the US, 1.41% for the UK, 1.76% for Germany, and 1.88% for Japan, but the corresponding volatility instantly after the attacks was 3.05%, 4.09%, 5.33%, and 2.74%, respectively.6 Crossperiod comparisons show that there was an improvement in mean returns (albeit still negative) in the post-attack period across the sample countries. This phenomenon was salient for Japan for which mean returns were improved by 77.3%, from 0.22% before the attacks to 0.05% after the attacks. Yet, data for market volatility show conflicting results across the sample countries. While the market volatility decreased after the attacks for the US and Japan, it increased for the UK and Germany. This suggests that the improvement in mean returns was accompanied by a concomitant increase in market volatility (as measured by the standard deviation) for the UK and Germany but by a decrease in volatility for the US and Japan. To visually observe the immediate impact of the attacks on the market index and volatility of each market, bi-daily movements of the market index and volatility are plotted over the sample period in Fig. 2.7 The plots reveal that while there was no conspicuous change in market indexes and volatility before the attacks, there were an abrupt fall in market indexes and a drastic rise in volatility immediately following the attacks for all the sample countries. This suggests that the attacks brought about a worldwide (albeit temporary) stock market crash, given that the sample countries represent much of the world stock market. Also, it is noticeable that over the entire sample period, the conditional volatility as well as the market indexes were strongly synchronized across the countries, suggesting that international stock markets could be correlated to each other not only through returns but also through volatility. 2.3. International correlations for market returns and volatility Since one of the objectives in this paper is to examine the impact of the attacks on international market correlations, it is necessary to examine how correlations between the US and other sample markets have evolved over time before and after the attacks. Dynamic time series movements of correlation coefficients for returns are obtained using a multivariate GARCH model which permits conditional variances and covariances to change over time. This model is widely applied to an analysis of international correlations for stocks and bonds where correlation coefficients vary over time (see, for example, Bodart & Reding, 1999; Karolyi, 1995; Longin & Solnik, 1995; Ng, 2000; Susmel & Engle, 1994). The format of a multivariate GARCH model has a positive 6
Conditional volatility of each market is obtained from the time path of standard deviation of residuals for which each index return is regressed on constant within a univariate GARCH(1, 1) process. Such a parameterization has been found to be a good representation of many financial series. In our case where movements of time series are examined around the attacks, it is definitely necessary to have a time-varying structure for conditional second moments. 7 In Fig. 2(a), values for the first observation (November 20, 2000) are set to 100 as the base point.
54
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
(a) Market Indices 120
100
80
60
U.S. U.K. Germany Japan
40
20 Nov-00
Feb-01
May-01
Sep-01
Dec-01
Mar-02
Jul-02
(b) Conditional Volatility 0.06 U.S. U.K. Germany Japan
0.05
0.04
0.03
0.02
0.01
0 Nov-00
Feb-01
Jun-01
Sep-01
Dec-01
Mar-02
Jul-02
Fig. 2. Market indices and conditional volatility.
definite parameterization to ensure that the covariance matrix of residuals is positive definite: yt ¼ m þ et et jXt1
f
N ð0; ht Þ
ht ¼ cVc þ aVet1 eVt1 a þ bVht1 b
ð1Þ
where y t is the 2 1 vector of (R t , R t *)V; R t is the S & P 500 return and R t * is the index return for each individual non-US sample market; m is the 2 1 vector of constants; e t
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
55
is the 2 1 vector of disturbances; X t 1 is the information set available at time t 1; ht is the (2 2) time-dependent conditional covariance matrix; and c, a, and b are the (2 2) parameter matrices. The bivariate GARCH model is estimated separately to obtain the correlation with the US for the non-US sample markets. Hence, we have three market pairs: US/UK, US/Germany, and US/Japan. International stock markets can be correlated not only through their returns but also through their volatilities. The international market correlation through volatility is estimated using the multivariate GARCH model of Eq. (1) by substituting 2 1 vector of (v t , v t*) for y t where v t is the conditional volatility of the S & P 500 return and v t* is the conditional volatility of the index return for each individual non-US sample market. The parameter estimates and diagnostics for model (1) are reported in Table 2. To conserve space, we only report estimates for returns. The constant terms uˆ 1 capture Table 2 GARCH estimates Coefficient
Country
mˆ 1 mˆ 2 cˆ11 cˆ12 cˆ22 aˆ11 aˆ12 aˆ21 aˆ22 bˆ11 bˆ12 bˆ21 bˆ22 LR test: a = b = 0 Q(20): 1st element Q(20): 2nd element Q 2(20): 1st element Q 2(20): 2nd element
UK
Germany
0.0019 (1.91) 0.0017 (2.04)* 0.0066 (1.44) 0.0004 (0.08) 0.0052 (3.17)** 0.1580 (0.43) 0.2027 (0.64) 0.9641 (3.05)** 0.6636 (2.20)* 0.3007 (2.26)* 0.2608 (2.21)* 0.1274 (0.91) 0.5712 (4.54)** 283.66** 23.12 20.35 10.64 21.58
0.0020 0.0029 0.0084 0.0029 0.0046 0.6748 0.1543 0.8900 0.9755 0.1863 0.3390 0.0290 0.5082 31.45** 21.36 19.46 9.52 22.36
Japan (1.87) (2.24)* (1.91) (0.44) (1.35) (1.31) (0.20) (3.03)** (2.36)* (1.18) (2.90)** (0.24) (3.62)**
0.0019 0.0018 0.0055 0.0026 0.0145 0.7848 0.2378 0.1884 0.3840 0.3238 0.5287 0.4389 0.3970 77.84** 21.78 16.37 9.66 15.34
(1.90) (1.39) (2.09)* (0.24) (7.32)** (7.50)** (1.17) (2.67)** (2.38)* (2.29)* (3.49)** (4.88)** (2.34)*
The format of multivariate GARCH model is: yt ¼ m þ et et jXt1
f N ð0; ht Þ
ht ¼ cVc þ aVet1 et1 V a þ bVht1 b where y t is the 2 1 vector of (R t , R*)Vwhere R t is the S & P 500 return and R*t is the index return for each t individual non-US sample market; m is the 2 1 vector of constants; e t is the 2 1 vector of disturbances; X t 1is the information set available at time t 1; ht is the (2 2) time-dependent conditional covariance matrix; and c, a, and b are the (2 2) parameter matrices. The numbers in parentheses are asymptotic t-statistics. * (**) indicates statistical significance at less than 5% (1%) level. The likelihood ratio test statistic for the null of A = B = 0 is v 2(8) that has 95% critical value 15.51. Q(20) is the Ljung-Box statistic for 20th order serial correlation in the residuals; Q 2(20) is a Ljung-Box statistic for residual GARCH effects in the squared standardized residuals. Both Q(20) and Q 2(20) are distributed v 2(20) and have 95% critical value 34.17.
56
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
expected returns for the S & P 500 and uˆ 2 capture expected returns for other sample market indexes over the entire sample period. For example, uˆ 1 for UK is 0.0019 with tstatistics of 1.91, meaning that the average bi-daily return on the S & P 500 is about 0.19% over the whole sample period but statistically insignificant. uˆ 2 for UK is 0.0017 with t-statistics of 2.04, indicating that the average bi-daily return on the FTSE 100 is about 0.17% during the sample period. The multivariate GARCH parameter estimates aˆ11 through bˆ22 are statistically significant for a number of cases, indicating the residual variances and covariances are changing with time and the GARCH estimation procedure should give us better covariance estimates at any point in time than the conventional linear regression method. Also, tests for residual GARCH effects using both Ljung-Box Q2 (20) statistic and likelihood ratio (LR) test of the null: a = b = 0 strongly supports the timevarying variance of disturbances. Table 3 reveals that in the post-attack period there was a modest amount of increase in the unconditional return correlation for the US/UK and US/German markets. The corresponding correlation for the US/Japanese markets showed a relatively large amount of increase after the attacks. For instance, the correlation coefficient for US/UK markets increased from 0.7207 before the attacks to 0.7348 after the attacks, while the corresponding coefficient for US/Japanese markets increased from 0.1792 to 0.2848 after the attacks. To the extent that a significant change in the correlation coefficient is construed as a contagion from one country to the other (see Forbes & Rigobon, 2002), the attacks brought about a contagion in market returns from the US to Japan. In contrast, there was a continued binterdependenceQ in market returns for the US/UK and US/ Germany. The time-varying correlation coefficients are calculated by conditional covariances between returns on the S & P 500 and individual non-US sample markets, and the
Table 3 Correlation coefficients for return and volatility Country
UK
Germany
Japan
Period
I 0 II I 0 II I 0 II
Return correlation with U.S.
Volatility correlation with U.S.
Unconditional
Conditional
Conditional
0.7207 – 0.7348 0.7367 – 0.7724 0.1792 – 0.2848
0.6717 0.7013 0.7077 0.7029 0.7488 0.7465 0.1876 0.7559 0.2935
0.5264 0.9221 0.6589 0.3761 0.9298 0.5591 0.1617 0.2161 –0.0786
(0.1575) (0.1418) (0.0890) (0.0898) (0.1018) (0.1434)
(0.1903) (0.1743) (0.2125) (0.1832) (0.2585) (0.2423)
Note that the figures in parentheses are standard errors. Conditional correlation coefficients for return are calculated by the conditional covariances between returns on the S & P 500 and each individual non-US market index, and the conditional standard deviations of the two return variables obtained by the path of hˆ t in model (1). This table presents mean of the conditional correlation coefficients by period. Period I is the time period before the terrorist attacks; period 0 is the date immediately following the attacks; and period II is the time period after the attacks.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
57
conditional standard deviations of the two return variables are obtained from the time path of hˆ t in model (1). The results for average time-varying correlation coefficients are presented by period in the last two columns of Table 3. In general, the average dynamic return correlation with the S & P 500 increased for all of the sample markets immediately after the attacks, although the extent varied across the markets. For the UK and German markets, the return correlation with the S & P 500 modestly rose immediately following the attacks, while the corresponding return correlation between the US and Japanese markets rose sharply from 0.1876 to 0.7559. This suggests the attacks brought about a modest amount of instant boost to the return correlation for the US/UK and US/German markets, while there was a dramatic rise in the return correlation for the US and Japanese markets. In the post-attack period, the average coefficients for all the sample markets reverted downward but stayed at higher than the pre-attack level. Table 3 also reveals that patterns of the volatility correlation are different from those of the return correlation. For example, immediately following the attacks, there was a drastic increase in the volatility correlation for the US/UK and US/German markets, whereas the corresponding volatility correlation between the US and Japanese market showed a relatively modest increase as compared to the return correlation. Noticeably, immediately following the attacks, the volatility correlation for the US/UK and US/ German markets increased to such a dramatic extent that the volatility correlation overshot the return correlation, while the volatility correlation between the US and Japanese market exhibited a modest increase immediately after the attacks. This suggests that immediately following the attacks, market correlations for the US/UK and US/Germany became stronger through volatility than through returns, but the converse is true for the US/ Japanese stock market. In the post-attack period, volatility correlations across the sample markets stayed at a higher level than the pre-attack level except for Japan, suggesting that market volatility correlations played a more important role in international stock market linkages after the attacks than before. As Campbell et al.(2002) document that cross-market correlations increase in times of bear markets, the increase in correlation in the post-attack period can be attributed to a prolonged period of bear markets after the attacks due to a decline in consumer and investor confidence. Fig. 3 presents the dynamic correlation coefficient between the US and individual non-US sample markets. The solid line represents the volatility correlation coefficients and the dotted line the return correlation coefficients. In panels (a) and (b) of Fig. 3, the magnitude of return correlation coefficients for the UK and Germany before the attacks appears to be larger than that of volatility correlation coefficients. Yet, immediately following the attacks, the volatility correlation coefficient appears to noticeably overshoot the return correlation coefficient for the US/UK and US/German markets. At the same time, these corresponding markets exhibit only a little or no change in return correlations. This is in sharp contrast to the US/Japanese market where return correlations rise considerably but volatility correlations change only a little immediately after the attacks. This reflects that the terrorist attacks had an immediate positive impact on the international return correlation between the US and Japanese market, while there was no noticeable change in the return correlation for the US/UK and US/German
58
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
(a) U.K. 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Nov-00
VolatilityCorr Feb-01
Jun-01
Sep-01
Dec-01
ReturnCorr Mar-02
Jul-02
(b) Germany 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Nov-00
VolatilityCorr Feb-01
Jun-01
Sep-01
Dec-01
ReturnCorr Mar-02
Jul-02
(c) Japan 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Nov-00
VolatilityCorr Feb-01
Jun-01
Sep-01
Dec-01
ReturnCorr Mar-02
Jul-02
Fig. 3. Conditional correlation coefficients with US.
markets. On the contrary, the attacks had a large positive impact on the volatility correlation for the US/UK and US/German markets, but little positive impact for the volatility correlation between the US and Japanese markets. These results reinforce the previous finding that the stock markets of the UK and Germany became more strongly correlated with the US market through volatility than through returns immediately after the attacks, but the contrary is true for the Japanese market. The results are consistent
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
59
with those of recent literature on international stock market correlations in that international correlation between markets rises in periods of high volatility such as the date immediately following the attacks (see, for example, Campbell et al., 2002; Hamao et al., 1990; King & Wadhwani, 1990; Longin & Solnik, 1995; Ramchand and Susmel (1998); Solnik et al., 1996). In sum, evidence presented in this section indicates that while there was a drastic drop in market indexes (or returns) and a concomitant rise in market volatility immediately after the attacks across the sample countries, the risk–return profile was ameliorated for the US and Japanese markets after the attacks. Also, analysis of the correlation coefficient for returns and volatility reveals that, immediately following the attacks, cross-market correlations for the US/UK and US/Germany became stronger through volatility than through returns, whereas the corresponding correlation for the US/Japan became stronger through returns than through volatility. This gives some preliminary evidence that the attacks caused a contagion either through returns or volatility to the non-US sample markets.
3. Contagion and impulse response analysis 3.1. Testing for contagion It is often claimed that there must be a contagion effect from one country to another if stock market correlations increase significantly after the attacks (see Forbes and Rigobon (2002)). The attacks could have affected the extent to which international stock markets are correlated to each other through returns as well as through volatility. In this section, we investigate whether the attacks caused a return/volatility contagion from the US to other countries by estimating a multivariate GARCH model with constant conditional correlations. This model is known to effectively capture the influence of a set of factors for the dynamics of conditional variance and correlations (see Bodart & Reding, 1999; Bollerslev, 1990; Karolyi & Stulz, 1996; Longin & Solnik, 1995; Ng, 2000). Specifically, we estimate Eq. (1) with the following variance–covariance dynamics: hii;t ¼ ci þ ai hii;t1 þ bi e2i;t1 hij;t ¼ ðq þ bDummyt Þ hii;t hjj;t
1=2
ð2Þ
where hii,t and hjj,t are the ii th, jj th, and ij th element in ht ( i = the US market and j = the local market); q is the constant conditional correlation between the US and local market returns; Dummyt is the binary dummy variable that takes a value of 1 on data points in the post-attack period. If the return and volatility story of Table 1 and Fig. 2 is accurate, then b estimates for the return and conditional volatility should be significantly positive. A significantly positive b for the correlation coefficient indicates that the attacks caused a return (volatility) contagion from the US to other sample markets if the dependent variable is a return (conditional volatility) vector.
60
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
The results for b estimates are reported in Table 4. The coefficient estimates for return correlation reveal that the market return correlation with the US increased significantly for Japan in the period following the attacks, while the corresponding correlations for the UK and German markets remained approximately at the same level as the time period before the attacks. Table 4 also shows that the volatility correlation with the US market increased significantly for the UK and German markets but decreased for the Japanese market. These results can be interpreted as evidence that the attacks had a contagion in market returns from the US to Japan, whereas there was no contagion but a continued binterdependenceQ of market returns between the US and the UK and Germany after the attacks. In contrast, it is apparent that the attacks caused a volatility contagion from the US to UK and German markets. In sum, to the extent that higher correlations with the US market can have contagion effects from the US, the attacks caused a volatility contagion from the US to the UK and German markets but a return contagion from the US to the Japanese market. Yet, there was a continued interdependence of market returns (volatility) for the US/UK and US/German (Japanese) markets after the attacks. This suggests that the terrorist attacks gave rise to a structural change in the dynamics of international stock market correlations. Furthermore, these results have important implications for both international asset allocation decisions and risk management. The general increase in correlations as well as the return and
Table 4 Testing for contagion effects bˆ for return bˆ for conditional volatility
UK
Germany
Japan
0.0724 (1.29) 0.1265 (2.14)*
0.0828 (1.42) 0.1509 (3.05)**
0.1230 (2.08)* 0.0755 (0.98)
Testing for contagion effects are conducted by estimating the following regression equation: yt ¼ m þ et et jXt1
f N ð0; ht Þ
V a þ bVht1 b ht ¼ cVc þ aVet1 et1 hii;t ¼ ci þ ai hii;t1 þ bi e2i;t1 hij;t ¼ ðq þ bDummyt Þ hii;t hjj;t
1=2
where y t is the 2 1 vector of (R t , R*)Vor (v t , v*) *) t t where R t (v t ) is the S & P 500 return (volatility) and R*(v t t is the index return (volatility) for each individual non-US sample market; m is the 2 1 vector of constants; e t is the 2 1 vector of disturbances; X t 1 is the information set available at time t 1; ht is the (2 2) time-dependent conditional covariance matrix; and c, a, and b are the (2 2) parameter matrices; hii,t , hjj,t , and hij,t are the ii th, jj th, and ij th element in ht ( i = the US market and j = the local market); q is the constant conditional correlation between the US and local market returns; Dummyt is the binary dummy variable that takes a value of 1 on data points in the post-attack period. The numbers in parentheses are asymptotic t-statistics. * (**) indicates statistical significance at less than 5% (1%) level.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
61
volatility contagion to international markets erodes the expected gains from international diversification. 3.2. Testing for structural change in market correlations8 The presence of a structural change in cross-market correlations around the 9–11 terrorist attacks is tested using the pure structural break model with m breaks developed by Bai and Perron (1998 and 2003). The model is of the following format: y t ¼ dj þ m t
t ¼ Tj1 þ 1; . . . ; Tj for j ¼ 1; . . . ; m þ 1
ð3Þ
where y t is the dependent variable at time t; d j is the constant; m t is the disturbance at time t. By convention T 0 = 0 and T m + 1 = T, the total number of observations. The indices (T 1,. . ., T m) are break points. The number of break points for Eq. (3) is first selected using the Bayesian Information Criterion (BIC) for both return and volatility correlations. Then, as Bai and Perron (2003) suggests, a sequential procedure is used to estimate optimal break points.9 The sequential procedure uses a dynamic programming to find a segment that globally minimizes the overall sum of squared residuals. Statistical test for the breaks is conducted using the following F-statistic defined in Bai and Perron (2003): 1 T ðk þ 1Þ ˆ V V ˆ V 1 ˆ d R RV R FT ðk1 ; . . . ; kk Þ ¼ Rd ð4Þ T k where (T 1,. . ., T k ) is a partition such that Ti = [Tki ](i = 1,. . ., k); (Rd)V= (dV1 dV2,. . ., ˆ = the variance–covariance matrix of dˆ which is heteroskedasticity—and dVk dVk + 1); and V autocorrelation—consistent. It is known that maximizing the F-test Eq. (4), labeled sup F T (k), is asymptotically equivalent to F T (kˆ1,. . .,kˆ k ) minimize the global sum of squared residuals (see Andrews, 1993; Bai & Perron, 1998, 2003). Using the methodology described in Bai and Perron (2003), we first consider the sup F T (k) test of no structural break (m = 0) versus m = k breaks and then conduct a test of the null hypothesis of S breaks against the alternative that an additional break exists, labeled sup F T (S + 1 | S )}. The sup F T (S + 1 | S )} test is tantamount to the application of (S + 1) tests of the null hypothesis of no structural change versus the alternative hypothesis of a single change for each segment containing the observations Tˆi 1 + 1 to Tˆi (i = 1,. . ., S + 1). If there are (S + 1) breaks instead of S breaks in Eq. (3), the overall minimal value of the sum of residuals obtained with (S + 1) breaks should be significantly smaller than that ˆ obtained with S breaks. The parameter estimates, d j , are obtained from applying OLS segment by segment without constraint. The maximum number of structural changes allowed is set to 3. The value of the trimming is set to 0.15 for the construction and critical values of the sup F-type tests. The results are presented in Table 5. The sup F T (k) tests for k = 1 are highly significant for the return correlation for Japan and the volatility correlation for the UK and Germany. For k N 1, the values of the F T (k) 8
We are indebted to an anonymous referee who suggested this structural change analysis. Bai and Perron (2003) claims that the sequential procedure is superior to the procedure based on the information criteria since the latter cannot take into account potential heterogeneity across segments. 9
62
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
Table 5 Bai–Perron tests for structural breaks
A. Correlation in returns UK sup FT ( k ) sup FT ( + 1
Germany
Japan
k= 1
k= 2
k= 3
k= 1
k= 2
k= 3
k= 1
k= 2
k= 3
6.53
3.66
1.12
7.78
4.51
2.99
28.64*
7.31
2.45
)
=1
=2
1.04
0.05
=1 0.98
=2
=1
=2
0.03
0.77
0.01
*Indicates statistical significance at less than 1% level. We use a 5% size for the sequential test sup FT ( + 1 ). B. Correlation in volatility UK sup FT ( k ) sup FT ( + 1
Germany
Japan
k= 1
k= 2
k= 3
k= 1
k =2
k= 3
k= 1
k= 2
k= 3
32.38*
6.23
1.88
44.76*
6.97
2.86
2.43
1.15
0.01
=1
)
1.97
=2
=1
=2
=1
=2
0.01
1.26
0.01
0.33
0.02
*Indicates statistical significance at less than 1% level. We use a 5% size for the sequential test sup FT ( + 1 ). C. Break dates and parameter estimates Correlation in returns
Number of breaks
Correlation in volatility
UK
Germany
Japan
UK
Germany
Sequential
0
0
1
1
1
0
BIC
0 −
0 −
1
1
1
0.68 (0.16) 0.71 (0.15)
0.71 (0.09) 0.75 (0.09)
2001: 09:17 [09:10− 09:21] 0.19 (0.11) 0.30 (0.14)
2001: 09:17 [09:10− 09:19] 0.53 (0.19) 0.66 (0.17)
2001: 09:17 [09:10− 09:19] 0.38 (0.22) 0.56 (0.18)
0 −
Break dates
Parameter estimates
δˆ1 δˆ2
Japan
0.17 (0.26) −0.04 (0.21)
The model is of the following format with m breaks: yt ¼ dj þ mt
t ¼ Tj1 þ 1;. . . ; Tj for j ¼ 1;. . . ; m þ 1
where y t is the dependent variable at time t; d j is the constant; m t is the disturbance at time t. By convention T 0 = 0 and T m + 1 = T, the total number of observations. The first test is the sup F T (k) test of no structural break (m = 0) versus m = k breaks. Asymptotically, sup F T (k) = F T (kˆ 1,. . ., kˆ k ) where kˆ 1,. . ., kˆ k minimize the global sum of squared residuals. The second test has the null hypothesis of S breaks against the alternative that an additional break exists, labeled sup F T (S + 1 | S ). We use a 5% size for the sequential test sup F T (S + 1 | S ). *Indicates statistical significance at less than 1% level.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
63
statistics are insignificant for all countries, suggesting that the model has just one break point (= two segments). This result is further supported by the sup F T (S + 1 | S ) test in that none of the statistics for S z 1 is significant. The BIC selects one break point and so does the sequential procedure. As expected, the break date is estimated at September 17, 2001 and the estimation is precise in the sense that the 95% confidence intervals cover only a few observations before and after. In sum, the statistical results provide evidence that there was an abrupt structural change in the dynamics of international stock market correlations in volatility for the UK and Germany and in returns for Japan around the 9–11 attacks. 3.3. Impulse response analysis To obtain additional insight into the structure of international stock market linkages, we conduct impulse response analysis. The impulse response analysis allows us to gauge to what extent shocks to international markets are explained by the US market, thus making it possible to examine how international stock markets can be destabilized by shocks that arise in the US. To investigate whether or not there was a change in the pattern of impulse responses after the attacks, the impulse response analysis is performed over two subsamples as defined in the previous section: pre-and post-attack period. Impulse response functions are obtained by estimating the following standard vector autoregression (VAR) model: xt ¼ A0 þ A1 xt1 þ A2 xt2 þ . . . . . . þ Ap xtp þ et
ð5Þ
where x t = the (4 1) vector of four market returns included in the VAR, i.e., return on the S & P 500, FTSE 100, DAX, and TOPIX; A 0 = the (4 1) vector of constants; Ai = the (4 4) matrices of coefficients (i = 1, 2,. . ., p); and e t = the error vector. Three lags were chosen in the estimation as this is the lag length that minimizes the Schwartz Baysian Criterion (SBC). A Choleski decomposition is used to orthogonalize the underlying errors using the ordering as defined above. Placed first in the ordering, the exchange rate change is implicitly assumed to be unaffected by the other shocks in the system. Although there is an increased interrelationship and integration among international stock markets, it is consistently documented in the literature that the US market is the price leader and dominates the world market, validating the econometric exogeneity of the US market in the VAR analysis (see, for example, Eun & Shim, 1989; Lin, Engle, and Ito, 1994; Masih & Masih, 2001). We also conduct impulse response analysis of volatility by replacing the four elements of x t by new elements of the four individual conditional volatilities, with the US volatility coming first in the ordering. Impulse response analysis of return (volatility) correlation is conducted by replacing x t by elements of the S & P 500 return (volatility) and three individual correlation coefficients in returns (volatility). Fig. 4 presents the results from the impulse response analysis based on the quadravariate VAR. The upper panel in each response function exhibits the pre-attack responses and the lower panel the post-attack responses. The upper and lower dashed lines in each graph are 95%-confidence bands. Specifically, Fig. 4 traces out impulse response functions from a one standard deviation shock in the US market to each of the non-US markets (UK, Germany, and Japan). The upper panel of Fig. 4(a), for example, reports the
64
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
(a) Responses of Return 0.02
0.02
0.015
0.02
0.015
0.01 0.005
0.015
0.01
0.01
0.005
0.005
0
0
0
-0.005
-0.005
-0.005
-0.01
-0.01
-0.01 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
UK
Germany
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
Japan
0.02
0.02
0.02 0.015
0.015
0.015
0.01
0.01
0.005
0.005
0.01 0.005
0
0
0
-0.005
-0.005
-0.005
-0.01
-0.01
-0.01 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
(b) Responses of Volatility 0.002
0.002
0.002
0.0015
0.0015
0.0015
0.001
0.001
0.001
0.0005
0.0005
0.0005
0
0
0
-0.0005
-0.0005
-0.0005
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
UK
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Germany
Japan
0.002
0.002
0.002
0.0015
0.0015
0.0015
0.001
0.001
0.001
0.0005
0.0005
0.0005
0
0
0
-0.0005
-0.0005
-0.0005
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Fig. 4. Impulse responses of return and volatility.
impulse response function of non-US market returns before the attacks. In the pre-attack period, a shock in the US market has a short-lived positive effect with a relatively small impulse response of the non-US market returns. In the post-attack period, the UK and German markets exhibit little or no change in the response of market returns as compared to the pre-attack period. On the contrary, a US shock in the post-attack period has a rapid and large short-run response from the Japanese market. This suggests that in the period following the attacks, the short run dependence of market returns to shocks that arise in the US market appears to have greatly increased in the Japanese market but remained nearly unchanged in the UK and German markets. The upper panel of Fig. 4(b) reveals that in the pre-attack period there is a modest amount of positive response of the non-US market volatility to a shock in the US market, with the most rapid response early on and later responses tapering off. By contrast, in the post-attack period, the non-US sample markets show a large amount of persistence in
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
65
volatility responses to a shock in the US market. Responses of the German and UK market volatility are especially dramatic after the attacks, in that they are nearly twice the magnitude of the pre-attack response and appear to persist for a long time. In sum, an overall market response of volatility to a shock in the US market increased after the attacks, with the most dramatic increase in response from the German and UK markets. Fig. 5(a) shows impulse response functions from a one standard deviation shock in the US market to each of the non-US marketsT return correlation with the US market. The upper panel of Fig. 5(a) shows the pre-attack impulse response functions of return correlation for each of the non-US markets. In the pre-attack period, a shock in the US market provides a temporary boost to the US/UK and US/Germany return correlations while the corresponding shock to the Japanese market seems to have a temporary turbulent impact on the Japan/US return correlation in terms of switching responses between positive and negative. In the post-attack period, the US shock has a strongly positive effect on the US/Japan return correlation but the response functions of the US/UK and the US/
(a) Responses of Return Correlation 0.06
0.06
0.05
0.05
0.04
0.04
0.03
0.03
0.02
0.02
0.06 0.05 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.05
0.01
0.01 0
0
-0.01
-0.01
-0.02
-0.02
-0.03
-0.03
-0.04
-0.04
-0.05
-0.05 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
UK
Germany
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Japan
0.06
0.06
0.06
0.05
0.05
0.05
0.04
0.04
0.04
0.03
0.03
0.03
0.02
0.02
0.02
0.01
0.01
0
0
0
-0.01
-0.01
-0.01
-0.02
-0.02
-0.02
-0.03
-0.03
-0.03
-0.04
-0.04
-0.05
-0.05
0.01
-0.04 -0.05 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
(b) Responses of Volatility Correlation 0.035
0.035
0.035 0.03
0.03
0.03
0.025
0.025
0.025
0.02
0.02
0.02
0.015
0.015
0.015
0.01
0.01
0.01
0.005
0.005
0.005
0
0
0
-0.005
-0.005 1 2 3 4 5 6 7 8 9 101112131415161718192021222324
-0.005 1 2 3 4 5 6 7 8 9 101112131415161718192021222324
UK
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Germany
Japan
0.035
0.035
0.03
0.03
0.03
0.025
0.025
0.025
0.02
0.02
0.02
0.015
0.015
0.015
0.035
0.01
0.01
0.01
0.005
0.005
0.005
0
0
0
-0.005
-0.005
-0.005 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Fig. 5. Impulse responses of correlation for return and volatility.
66
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
Germany return correlation show little or no change over the two sub-periods. This implies that a US shock appears to have strengthened the return correlation between the US and Japanese markets after the attacks than before but had no noticeable change in response for the US/UK and US/German markets. The lower panel of Fig. 5(b) exhibits the pre-attack impulse responses of volatility correlation to shocks in the US market. In general, the US shock in the pre-attack period has a sufficiently positive impact on the volatility correlation between the US and non-US markets. Overall patterns of the post-attack responses of the UK and German markets are distinguishable from those of the pre-attack responses in that post-attack responses of volatility correlations are more strongly positive than the pre-attack responses. In sharp contrast to the case of the response function of return correlation, a US shock notably strengthened the volatility correlation for the US/UK and US/German markets after the attacks than before but had no noticeable change in response for the US/Japanese markets. Impulse response analysis conducted here reveals that overall impulse responses of the sample markets to a US shock were not conspicuously highly-positive for any of the variables before the attacks. In the period following the attacks, however, responses of market volatility to a US shock increased noticeably for the UK and Germany after the attacks but had little or no change for the Japanese market. The converse was true for responses of market returns. Responses of correlation with the US reveal that after the attacks, the UK and German markets were more strongly correlated through volatilities with the US market but the Japanese market was more strongly correlated through returns with the US market. These results reinforce the previous finding based on the analysis of dynamic time series behavior of correlation coefficients.
4. Conclusions In this paper we investigate evidence of both return and volatility contagion effects of the terrorist attacks across the major markets, and examine the extent to which international stock markets can be destabilized by shocks that arise in the US. To this end, the whole sample of four national stock market indexes (the S & P 500 of US, the FTSE 100 of UK, the DAX of Germany, and the TOPIX of Japan) is divided into two subperiods: the pre-attack period and post-attack period. We find that there was a drastic drop in market returns and a concomitant rise in market volatility immediately after the attacks. After the attacks, the overall market condition for the US and Japan was ameliorated from the risk and return perspective. Yet, the situation for the UK and Germany with regard to the risk profile was aggravated in the post-attack period, with a higher level of standard deviation than its pre-attack level. International stock markets can be correlated not only through their returns but also through their volatilities. Evidence presented in this paper indicates that, after the attacks, international stock markets were more strongly positively correlated through volatilities rather than through returns for the US/UK and US/German markets, while a market correlation between the US and Japan was strengthened through returns rather than through volatilities. This suggests that to the extent that higher correlations with the US market can enhance contagion effects from the US to other markets, the terrorist attacks
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
67
brought about a volatility contagion (rather than a return contagion) from the US to UK and German markets. In contrast, the Japanese market had a return contagion (rather than a volatility contagion) from the US market, suggesting that the terrorist attacks gave rise to a structural change in the dynamics of international stock market correlations. This is further supported by statistical evidence that there was an abrupt structural break in international stock market correlations in volatility for the UK and Germany and in returns for Japan around the 9–11 attacks. Impulse response analysis conducted in this paper shows that post-attack responses of the sample markets to a US shock were short-lived and positive in return and return correlations, whereas the corresponding responses of volatility and volatility correlations were fairly persistent. The non-US sample market returns showed a positive response to a shock in the US market, with the most intense response of the Japanese market returns after the attacks. An overall response of market volatility to a US shock increased after the attacks, with the most dramatic increase in the UK and German markets. The US shock had a strongly positive effect on the US/Japan return correlation but had little or no change over the two sub-periods for response functions of the US/UK and US/Germany return correlations. Impulse responses of the volatility correlation to a US shock notably increased after the attacks for the US/UK and US/Germany volatility correlations. An overall analysis of the post-attack period reveals that international market correlations were strengthened through volatility for the US/UK and US/German markets and through returns for the US/Japanese market.
Acknowledgement The author thanks Manuchehr Shahrokhi (the editor) and an anonymous referee for their helpful comments and suggestions. The author also benefited from comments of participants at the 2003 International Applied Business Research Conference, Acapulco, Mexico. The author is grateful to Steve Smith and Sandy Fleak for their comments/suggestions, and Brandon Duede for research assistance. The author gratefully acknowledges financial support from the Division of Business and Accountancy, Truman State University.
References Andrews, D. (1993). Tests for parameter instability and structural change with unknown change point. Econometrica, 61, 821 – 856. Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66, 47 – 78. Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change models. Journal of Applied Econometrics, 18, 1 – 22. Bodart, V., & Reding, P. (1999). Exchange rate regime, volatility and international correlations on bond and stock markets. Journal of International Money and Finance, 18, 133 – 151. Bollerslev, T. (1990). Modelling the coherence in short-run nominal exchange rates: A multivariate generalized ARCH model. Review of Economics and Statistics, , 498 – 505. Campbell, R., Koedijk, K., & Kofman, P. (2002). Increased correlation in bear markets. Financial Analysts Journal, , 87 – 94.
68
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
Eun, C., & Shim, S. (1989). International transmission of stock market movements. Journal of Financial and Quantitative Analysis, 24, 241 – 256. Forbes, K. J., & Rigobon, R. (2002). No contagion, only interdependence: Measuring stock market comovements. Journal of Finance, 57, 2223 – 2261. Hamao, Y., Masulis, R. W., & Ng, V. K. (1990). Correlation in price changes and volatility across international stock markets. Review of Financial Studies, 3, 281 – 307. Karolyi, G. (1995). A multivariate GARCH model of international transmissions of stock returns and volatility: The case of the United States and Canada. Journal of Business and Economic Statistics, 13, 11 – 25. Karolyi, G., & Stulz, R. (1996). Why do markets move together? An investigation of US–Japan stock return comovements using ADRs. Journal of Finance, 51, 951 – 986. King, M. A., Sentana, E., & Wadhwani, S. (1994). Volatility and links between national stock markets. Econometrica, 62, 901 – 933. King, M. A., & Wadhwani, S. (1990). Transmission of volatility between stock markets. Review of Financial Studies, 3, 5 – 33. Lin, W. L., Engle, R. F., & Ito, T. (1994). Do bulls and bears move across borders? International transmission of stock returns and volatility. Review of Financial Studies, 7, 507 – 539. Longin, F., & Solnik, B. (1995). Is the correlation in international equity returns constant: 1960–1990? Journal of International Money and Finance, 14, 3 – 26. Masih, R., & Masih, A. M. (2001). Long and short term dynamic causal transmission amongst international stock markets. Journal of International Money and Finance, 20, 563 – 587. Ng, A. (2000). Volatility spillover effects from Japan and the US to the Pacific-Basin. Journal of International Money and Finance, 19, 207 – 233. Ramchand, L., & Susmel, R. (1998). Volatility and cross correlation across major stock markets. Journal of Empirical Finance, 5, 397 – 416. Solnik, B., Boucrelle, C., & Fur, Y. (1996). International market correlation and volatility. Financial Analysts Journal, , 17 – 34. Susmel, R., & Engle, R. (1994). Hourly volatility spillovers between international equity markets. Journal of International Money and Finance, 13, 3 – 25.
Contagion and impulse response of international stock markets around the 9–11 terrorist attacks C. Kyung-Chun Mun* Division of Business and Accountancy, Truman State University, Kirksville, MO 63501, United States Received 4 March 2005; received in revised form 9 April 2005; accepted 10 April 2005 Available online 14 June 2005
Abstract This study investigates both return and volatility contagion effects of the 9–11 terrorist attacks across the major markets, and examines the extent to which international stock markets can be destabilized by shocks that arise in the US. Evidence presented in this paper suggests that to the extent that higher correlations with the US market can enhance contagion effects from the US, the attacks brought about volatility contagion (rather than return contagion) from the US to UK and German markets. In contrast, the Japanese market had return contagion (rather than volatility contagion) from the US market. After the attacks, a US shock had a strongly positive effect on the US/Japan return correlation but had little or no effect in response functions of the return correlation for the US/UK and US/German markets. Impulse responses of the volatility correlation to a US shock notably increased after the attacks for the US/UK and US/German markets. An overall analysis of the post-attack period reveals that international market correlations were strengthened through volatility for the US/UK and US/German markets and through return for the US/Japanese market. D 2005 Elsevier Inc. All rights reserved. JEL classification: F21 Keywords: 9–11 terrorist attacks; Contagion; Impulse response
1. Introduction The tragic events of September 11, 2001 exacerbated an already very difficult situation in the US and world stock markets. The shock in the US market driven by the terrorist attacks * Tel.: +1 660 785 4367. E-mail address: [email protected]. 1044-0283/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.gfj.2005.05.002
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
49
Table 1 Summary statistics of two-day stock returns Country
Period
Mean of returns (%)
Standard deviation (%)
US
I 0 II I 0 II I 0 II I 0 II
0.1796 5.0468 0.1396 0.1784 4.9366 0.1615 0.2645 7.1447 0.2006 0.2195 5.8326 0.0548
1.6550 3.0513* 1.5350 1.4101 4.0872* 1.5199 1.7621 5.3320* 2.4053 1.8844 2.7433* 1.8437
UK
Germany
Japan
Period I is the time period before the terrorist attacks; period 0 is the date immediately following the attacks; and period II is the time period after the attacks. *Indicates the values of conditional volatility obtained from the standard deviation of residuals for which each index return is regressed on constant within a univariate GARCH (1,1) process.
gave rise to a synchronous downturn across nearly all major regions of the world. As seen from Table 1, two-day mean returns immediately following the attacks fell by 5.05% for the S & P 500, 4.94% for the FTSE 100 of the UK, 7.14% for the DAX of Germany, and 5.83% for the TOPIX of Japan. At the same time, the market volatility as measured by standard deviation rose drastically across the markets. Apparently, increased international financial linkages have played an important role in the synchronicity of global market downturns. Many people have spoken or written about September 11 and its economic consequences. Enough perspectives exist to assess the impact of the attacks on the global economic situation, i.e., weakened consumer and business confidence, worsening financing conditions for emerging market economies, and rising transactions costs, etc. The attacks, however, have potentially far-reaching implications for stock market behavior and deserve special attention in the finance literature for at least four reasons. First, although there were many incidents that severely affected US stock markets in the past, the 9–11 attacks appeared to be unique in the pattern of US stock market behavior after market crashes.1 As seen from Fig. 1, while there was a gradual market resilience shortly after the crash for other incidents, the 9–11 attacks caused stock markets to plunge to a prolonged bear market after the attacks due to a long-term decline in consumer and investor confidence.2 1
The US stock market crashes for the last three decades as measured by more-than-3% drop in daily S&P 500 market indexes include the Hunt silver crisis of 1980, the Black Monday of 1987, the Iraqi invasion of Kuwait of 1990, the Asian financial crisis of 1997, and the 9–11 terrorist attacks of 2001. 2 One can claim that the prolonged bear market after the 9–11 attacks may be attributable to the business cycle rather than the 9–11 attacks. As reported by the NBER’s Business Cycle Dating Committee (2003; http:// www.nber.org/cycles), a recession began in March 2001 and the economy reached its trough (a turning point) in November 2001, entering a recovery. This observation in the business cycle that the aggregate economic activity has been rising since November 2001 does not seem to be compatible with overall stock market conditions that continued to be bearish for a long period of time even after November 2001. This suggests that a bear market after the 9–11 attacks may not be well explained by the business cycle.
50
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68 180 160 140 120 100 80 60 40 20 0 -200
9/11 Attacks of 2001 Asian Crisis of 1997 Iraqi Invasion of 1990 Black Monday of 1987 Hunt Silver of 1980 -150
-100
-50
0
+50
+100
+150
Fig. 1. S & P 500 index movements around market crashes. S & P index movements are shown here for 200 days before crash date (t = 200 to 1), the date of crash (t = 0), and 200 days after the crash (t = +1 to +200). The values of S & P 500 indecies for t = 200 are set to 100 as the base point.
Second, as we documented in this paper, the long-term bear market after the 9–11 attacks contributed to a structural change in cross-market correlations. Being characterized by a synchronous fall in returns and rise in volatility across international markets, the aftermath of the terrorist attacks implied increased international financial linkages, which in turn contributed to higher cross-market correlations. Third, the fact that the attacks were premeditated and therefore could be repeated has had a significant impact on investor sentiment. Terrorism could now become a part of the fabric of investor sentiment, putting a new risk factor in the equation for international investors. Finally, given the virtually unprecedented shock in nature with global reach, the attacks created a new level of international transmission of financial shocks to the global market. International stock markets now respond more sensitively to the shocks that arise in the US market. Various studies have documented that international stock market correlations increase when markets are volatile (see Campbell, Koedijk, & Kofman ,2002; King, Sentana, & Wadhwani, 1994; Longin & Solnik, 1995; Solnik, Boucrelle, & Fur, 1996). The international market correlation could change because international market volatility changes over time and/or because interdependence across markets changes. If cross-market correlations increase significantly after a crisis, there is a contagion from the US to other markets (see Forbes & Rigobon, 2002).3 Cross-market contagion driven by the attacks could provide an adverse investment environment for risk managers and internationally diversified investors since the changing correlation pattern and 3
There is no general consensus as to the definition of contagion. We will explore the Forbes and Rigobon’s (2002) definition of contagion in the remainder of this paper.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
51
financial disturbances make it more difficult for them to select an ex ante optimal investment strategy. The objective of this paper is twofold. First, we investigate evidence of contagion in returns and volatility of the terrorist attacks across the major markets (UK, Germany, and Japan). We document the evidence by assessing how significant the changes were in cross-market return and volatility correlations after the attacks. Second, in order to obtain additional insights into the contagion of international stock markets, we examine the extent to which shocks to international markets are explained by the US market’s behavior that resulted from the attacks. The attacks could have led other national stock markets to be destabilized through the market contagion. This paper does not intend to investigate the channels through which shocks driven by the attacks are transmitted across countries, i.e., international trade, foreign exchange rates, investor sentiment, or any unobservable fundamental variables. Instead, we focus on the question of whether there was a structural change in international market correlations and in the pattern of impulse responses after the attacks. The dynamic volatility and correlation coefficients across markets are obtained using a bivariate generalized autoregressive conditional heteroskedastic (GARCH) model. Tests for contagion from the US to other countries are conducted by estimating a multivariate GARCH model with constant conditional correlations. The presence of any structural changes in international market correlations around the 9– 11 attacks is tested using a pure structural change model developed by Bai and Perron (1998, 2003). Impulse response functions are obtained by estimating a standard vector autoregression (VAR) model and the analysis is performed over two subsamples (pre-and post-attack). We find that to the extent that significantly higher correlations with the US market after the attacks reflect a contagion from the US, the attacks gave rise to a volatility contagion (rather than a return contagion) from the US to UK and German markets. Although the US/UK and US/German markets showed no return contagion, a strong dinterdependenceT in market returns continued after the attacks.4 In contrast, the Japanese market had a return contagion (rather than a volatility contagion) from the US market and binterdependenceQ in market volatility continued between Japan and the US. Statistical evidence presented in this study indicates that there was an abrupt structural change in the dynamics of international stock market correlations around the 9–11 attacks. We also find that, after the attacks, impulse responses of market correlation to a US shock notably increased through volatility for the US/UK and US/ German markets, whereas the corresponding response of the US/Japanese market increased through returns. The paper is organized as follows: Section 2 presents data and methodology for empirical analysis of international market correlations. Tests for contagion and structural change and impulse response analysis are provided in Section 3. Conclusions are given in Section 4. 4 Forbes and Rigobon (2002) claims that if there is no significant increase in correlation coefficient but there exists a continued high level of correlation after a crisis, there exists an dinterdependenceT, not the contagion, between countries.
52
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
2. Data and methodology 2.1. Data The data set we analyze is based on a newly compiled two-day average series for the period from November 20, 2000 to June 27, 2002, providing a sample size of 203 observations. Following Forbes and Rigobon (2002), two-day average data are obtained from the average of two consecutive daily series with adjustment for weekends and holidays to avoid the problems of non-synchronous trading between the US and other sample countries. The cutoff date for the last observation was selected to avoid a possible confounding effect that may not have been significantly associated with the attacks.5 The first data point was selected to make the time length before and after the attacks balanced in an attempt to make cross-period comparisons of appropriate variables. The data consist of closing stock market indices of four countries: the S & P 500 of US, the FTSE 100 of UK, the German Stock Market Index (DAX) of Germany, and the Tokyo Stock Exchange Price Index (TOPIX) of Japan. These stock market indices are broad market-value-weighted indices and collected from the Global Financial Data base. Two-day index returns are computed as the logarithmic difference of two-day averages of stock index values. To examine time series behavior of the international stock markets before and after the terrorist attacks, we divide the whole sample into two sub-periods: (1) pre-attack period, from November 20, 2000 to September 10, 2001 and (2) post-attack period, from September 20, 2001 to June 27, 2002. The exact date of division of the sample period was chosen to make the sample size of each period balanced. The data points for September 12 and 14 were removed from each of the periods because the New York Stock Exchange (NYSE) was closed for the week of the attacks (September 11– 14, 2001) and reopened on September 17, 2001. The two observations (September 12 and 14, 2001) reflect the immediate impact of the attack on foreign stock markets and the day that the NYSE reopened reflects the immediate impact of the attacks on the US stock market. 2.2. Summary statistics Table 1 presents summary statistics of mean and standard deviation of returns around the attacks for stock indexes of US, UK, Germany, and Japan. Immediately after the attacks, there was a drastic drop in man returns and a concomitant increase in return volatility for all the sample countries. For example, the two-day market returns before the attacks were 0.18% for the US, 0.18% for the UK, 0.26%, and 0.22% for Japan, but immediately after the attacks returns for the corresponding 5
All the sample countries suffered an extreme bear market during July and August of 2002 when the US stock market rapidly fell due largely to the aftermath of financial scandals that have involved accounting irregularities by some major companies. Also, the first half of 2000 should not be included in the sample because the sample markets were exceptionally bullish due to the exuberance in technology-sector stocks.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
53
countries fell to 5.05%, 7.14%, and 5.83%, respectively. On the other hand, the return volatility rose drastically across the sample markets immediately after the attacks: the conditional volatility before the attacks was 1.65% for the US, 1.41% for the UK, 1.76% for Germany, and 1.88% for Japan, but the corresponding volatility instantly after the attacks was 3.05%, 4.09%, 5.33%, and 2.74%, respectively.6 Crossperiod comparisons show that there was an improvement in mean returns (albeit still negative) in the post-attack period across the sample countries. This phenomenon was salient for Japan for which mean returns were improved by 77.3%, from 0.22% before the attacks to 0.05% after the attacks. Yet, data for market volatility show conflicting results across the sample countries. While the market volatility decreased after the attacks for the US and Japan, it increased for the UK and Germany. This suggests that the improvement in mean returns was accompanied by a concomitant increase in market volatility (as measured by the standard deviation) for the UK and Germany but by a decrease in volatility for the US and Japan. To visually observe the immediate impact of the attacks on the market index and volatility of each market, bi-daily movements of the market index and volatility are plotted over the sample period in Fig. 2.7 The plots reveal that while there was no conspicuous change in market indexes and volatility before the attacks, there were an abrupt fall in market indexes and a drastic rise in volatility immediately following the attacks for all the sample countries. This suggests that the attacks brought about a worldwide (albeit temporary) stock market crash, given that the sample countries represent much of the world stock market. Also, it is noticeable that over the entire sample period, the conditional volatility as well as the market indexes were strongly synchronized across the countries, suggesting that international stock markets could be correlated to each other not only through returns but also through volatility. 2.3. International correlations for market returns and volatility Since one of the objectives in this paper is to examine the impact of the attacks on international market correlations, it is necessary to examine how correlations between the US and other sample markets have evolved over time before and after the attacks. Dynamic time series movements of correlation coefficients for returns are obtained using a multivariate GARCH model which permits conditional variances and covariances to change over time. This model is widely applied to an analysis of international correlations for stocks and bonds where correlation coefficients vary over time (see, for example, Bodart & Reding, 1999; Karolyi, 1995; Longin & Solnik, 1995; Ng, 2000; Susmel & Engle, 1994). The format of a multivariate GARCH model has a positive 6
Conditional volatility of each market is obtained from the time path of standard deviation of residuals for which each index return is regressed on constant within a univariate GARCH(1, 1) process. Such a parameterization has been found to be a good representation of many financial series. In our case where movements of time series are examined around the attacks, it is definitely necessary to have a time-varying structure for conditional second moments. 7 In Fig. 2(a), values for the first observation (November 20, 2000) are set to 100 as the base point.
54
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
(a) Market Indices 120
100
80
60
U.S. U.K. Germany Japan
40
20 Nov-00
Feb-01
May-01
Sep-01
Dec-01
Mar-02
Jul-02
(b) Conditional Volatility 0.06 U.S. U.K. Germany Japan
0.05
0.04
0.03
0.02
0.01
0 Nov-00
Feb-01
Jun-01
Sep-01
Dec-01
Mar-02
Jul-02
Fig. 2. Market indices and conditional volatility.
definite parameterization to ensure that the covariance matrix of residuals is positive definite: yt ¼ m þ et et jXt1
f
N ð0; ht Þ
ht ¼ cVc þ aVet1 eVt1 a þ bVht1 b
ð1Þ
where y t is the 2 1 vector of (R t , R t *)V; R t is the S & P 500 return and R t * is the index return for each individual non-US sample market; m is the 2 1 vector of constants; e t
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
55
is the 2 1 vector of disturbances; X t 1 is the information set available at time t 1; ht is the (2 2) time-dependent conditional covariance matrix; and c, a, and b are the (2 2) parameter matrices. The bivariate GARCH model is estimated separately to obtain the correlation with the US for the non-US sample markets. Hence, we have three market pairs: US/UK, US/Germany, and US/Japan. International stock markets can be correlated not only through their returns but also through their volatilities. The international market correlation through volatility is estimated using the multivariate GARCH model of Eq. (1) by substituting 2 1 vector of (v t , v t*) for y t where v t is the conditional volatility of the S & P 500 return and v t* is the conditional volatility of the index return for each individual non-US sample market. The parameter estimates and diagnostics for model (1) are reported in Table 2. To conserve space, we only report estimates for returns. The constant terms uˆ 1 capture Table 2 GARCH estimates Coefficient
Country
mˆ 1 mˆ 2 cˆ11 cˆ12 cˆ22 aˆ11 aˆ12 aˆ21 aˆ22 bˆ11 bˆ12 bˆ21 bˆ22 LR test: a = b = 0 Q(20): 1st element Q(20): 2nd element Q 2(20): 1st element Q 2(20): 2nd element
UK
Germany
0.0019 (1.91) 0.0017 (2.04)* 0.0066 (1.44) 0.0004 (0.08) 0.0052 (3.17)** 0.1580 (0.43) 0.2027 (0.64) 0.9641 (3.05)** 0.6636 (2.20)* 0.3007 (2.26)* 0.2608 (2.21)* 0.1274 (0.91) 0.5712 (4.54)** 283.66** 23.12 20.35 10.64 21.58
0.0020 0.0029 0.0084 0.0029 0.0046 0.6748 0.1543 0.8900 0.9755 0.1863 0.3390 0.0290 0.5082 31.45** 21.36 19.46 9.52 22.36
Japan (1.87) (2.24)* (1.91) (0.44) (1.35) (1.31) (0.20) (3.03)** (2.36)* (1.18) (2.90)** (0.24) (3.62)**
0.0019 0.0018 0.0055 0.0026 0.0145 0.7848 0.2378 0.1884 0.3840 0.3238 0.5287 0.4389 0.3970 77.84** 21.78 16.37 9.66 15.34
(1.90) (1.39) (2.09)* (0.24) (7.32)** (7.50)** (1.17) (2.67)** (2.38)* (2.29)* (3.49)** (4.88)** (2.34)*
The format of multivariate GARCH model is: yt ¼ m þ et et jXt1
f N ð0; ht Þ
ht ¼ cVc þ aVet1 et1 V a þ bVht1 b where y t is the 2 1 vector of (R t , R*)Vwhere R t is the S & P 500 return and R*t is the index return for each t individual non-US sample market; m is the 2 1 vector of constants; e t is the 2 1 vector of disturbances; X t 1is the information set available at time t 1; ht is the (2 2) time-dependent conditional covariance matrix; and c, a, and b are the (2 2) parameter matrices. The numbers in parentheses are asymptotic t-statistics. * (**) indicates statistical significance at less than 5% (1%) level. The likelihood ratio test statistic for the null of A = B = 0 is v 2(8) that has 95% critical value 15.51. Q(20) is the Ljung-Box statistic for 20th order serial correlation in the residuals; Q 2(20) is a Ljung-Box statistic for residual GARCH effects in the squared standardized residuals. Both Q(20) and Q 2(20) are distributed v 2(20) and have 95% critical value 34.17.
56
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
expected returns for the S & P 500 and uˆ 2 capture expected returns for other sample market indexes over the entire sample period. For example, uˆ 1 for UK is 0.0019 with tstatistics of 1.91, meaning that the average bi-daily return on the S & P 500 is about 0.19% over the whole sample period but statistically insignificant. uˆ 2 for UK is 0.0017 with t-statistics of 2.04, indicating that the average bi-daily return on the FTSE 100 is about 0.17% during the sample period. The multivariate GARCH parameter estimates aˆ11 through bˆ22 are statistically significant for a number of cases, indicating the residual variances and covariances are changing with time and the GARCH estimation procedure should give us better covariance estimates at any point in time than the conventional linear regression method. Also, tests for residual GARCH effects using both Ljung-Box Q2 (20) statistic and likelihood ratio (LR) test of the null: a = b = 0 strongly supports the timevarying variance of disturbances. Table 3 reveals that in the post-attack period there was a modest amount of increase in the unconditional return correlation for the US/UK and US/German markets. The corresponding correlation for the US/Japanese markets showed a relatively large amount of increase after the attacks. For instance, the correlation coefficient for US/UK markets increased from 0.7207 before the attacks to 0.7348 after the attacks, while the corresponding coefficient for US/Japanese markets increased from 0.1792 to 0.2848 after the attacks. To the extent that a significant change in the correlation coefficient is construed as a contagion from one country to the other (see Forbes & Rigobon, 2002), the attacks brought about a contagion in market returns from the US to Japan. In contrast, there was a continued binterdependenceQ in market returns for the US/UK and US/ Germany. The time-varying correlation coefficients are calculated by conditional covariances between returns on the S & P 500 and individual non-US sample markets, and the
Table 3 Correlation coefficients for return and volatility Country
UK
Germany
Japan
Period
I 0 II I 0 II I 0 II
Return correlation with U.S.
Volatility correlation with U.S.
Unconditional
Conditional
Conditional
0.7207 – 0.7348 0.7367 – 0.7724 0.1792 – 0.2848
0.6717 0.7013 0.7077 0.7029 0.7488 0.7465 0.1876 0.7559 0.2935
0.5264 0.9221 0.6589 0.3761 0.9298 0.5591 0.1617 0.2161 –0.0786
(0.1575) (0.1418) (0.0890) (0.0898) (0.1018) (0.1434)
(0.1903) (0.1743) (0.2125) (0.1832) (0.2585) (0.2423)
Note that the figures in parentheses are standard errors. Conditional correlation coefficients for return are calculated by the conditional covariances between returns on the S & P 500 and each individual non-US market index, and the conditional standard deviations of the two return variables obtained by the path of hˆ t in model (1). This table presents mean of the conditional correlation coefficients by period. Period I is the time period before the terrorist attacks; period 0 is the date immediately following the attacks; and period II is the time period after the attacks.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
57
conditional standard deviations of the two return variables are obtained from the time path of hˆ t in model (1). The results for average time-varying correlation coefficients are presented by period in the last two columns of Table 3. In general, the average dynamic return correlation with the S & P 500 increased for all of the sample markets immediately after the attacks, although the extent varied across the markets. For the UK and German markets, the return correlation with the S & P 500 modestly rose immediately following the attacks, while the corresponding return correlation between the US and Japanese markets rose sharply from 0.1876 to 0.7559. This suggests the attacks brought about a modest amount of instant boost to the return correlation for the US/UK and US/German markets, while there was a dramatic rise in the return correlation for the US and Japanese markets. In the post-attack period, the average coefficients for all the sample markets reverted downward but stayed at higher than the pre-attack level. Table 3 also reveals that patterns of the volatility correlation are different from those of the return correlation. For example, immediately following the attacks, there was a drastic increase in the volatility correlation for the US/UK and US/German markets, whereas the corresponding volatility correlation between the US and Japanese market showed a relatively modest increase as compared to the return correlation. Noticeably, immediately following the attacks, the volatility correlation for the US/UK and US/ German markets increased to such a dramatic extent that the volatility correlation overshot the return correlation, while the volatility correlation between the US and Japanese market exhibited a modest increase immediately after the attacks. This suggests that immediately following the attacks, market correlations for the US/UK and US/Germany became stronger through volatility than through returns, but the converse is true for the US/ Japanese stock market. In the post-attack period, volatility correlations across the sample markets stayed at a higher level than the pre-attack level except for Japan, suggesting that market volatility correlations played a more important role in international stock market linkages after the attacks than before. As Campbell et al.(2002) document that cross-market correlations increase in times of bear markets, the increase in correlation in the post-attack period can be attributed to a prolonged period of bear markets after the attacks due to a decline in consumer and investor confidence. Fig. 3 presents the dynamic correlation coefficient between the US and individual non-US sample markets. The solid line represents the volatility correlation coefficients and the dotted line the return correlation coefficients. In panels (a) and (b) of Fig. 3, the magnitude of return correlation coefficients for the UK and Germany before the attacks appears to be larger than that of volatility correlation coefficients. Yet, immediately following the attacks, the volatility correlation coefficient appears to noticeably overshoot the return correlation coefficient for the US/UK and US/German markets. At the same time, these corresponding markets exhibit only a little or no change in return correlations. This is in sharp contrast to the US/Japanese market where return correlations rise considerably but volatility correlations change only a little immediately after the attacks. This reflects that the terrorist attacks had an immediate positive impact on the international return correlation between the US and Japanese market, while there was no noticeable change in the return correlation for the US/UK and US/German
58
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
(a) U.K. 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Nov-00
VolatilityCorr Feb-01
Jun-01
Sep-01
Dec-01
ReturnCorr Mar-02
Jul-02
(b) Germany 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Nov-00
VolatilityCorr Feb-01
Jun-01
Sep-01
Dec-01
ReturnCorr Mar-02
Jul-02
(c) Japan 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Nov-00
VolatilityCorr Feb-01
Jun-01
Sep-01
Dec-01
ReturnCorr Mar-02
Jul-02
Fig. 3. Conditional correlation coefficients with US.
markets. On the contrary, the attacks had a large positive impact on the volatility correlation for the US/UK and US/German markets, but little positive impact for the volatility correlation between the US and Japanese markets. These results reinforce the previous finding that the stock markets of the UK and Germany became more strongly correlated with the US market through volatility than through returns immediately after the attacks, but the contrary is true for the Japanese market. The results are consistent
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
59
with those of recent literature on international stock market correlations in that international correlation between markets rises in periods of high volatility such as the date immediately following the attacks (see, for example, Campbell et al., 2002; Hamao et al., 1990; King & Wadhwani, 1990; Longin & Solnik, 1995; Ramchand and Susmel (1998); Solnik et al., 1996). In sum, evidence presented in this section indicates that while there was a drastic drop in market indexes (or returns) and a concomitant rise in market volatility immediately after the attacks across the sample countries, the risk–return profile was ameliorated for the US and Japanese markets after the attacks. Also, analysis of the correlation coefficient for returns and volatility reveals that, immediately following the attacks, cross-market correlations for the US/UK and US/Germany became stronger through volatility than through returns, whereas the corresponding correlation for the US/Japan became stronger through returns than through volatility. This gives some preliminary evidence that the attacks caused a contagion either through returns or volatility to the non-US sample markets.
3. Contagion and impulse response analysis 3.1. Testing for contagion It is often claimed that there must be a contagion effect from one country to another if stock market correlations increase significantly after the attacks (see Forbes and Rigobon (2002)). The attacks could have affected the extent to which international stock markets are correlated to each other through returns as well as through volatility. In this section, we investigate whether the attacks caused a return/volatility contagion from the US to other countries by estimating a multivariate GARCH model with constant conditional correlations. This model is known to effectively capture the influence of a set of factors for the dynamics of conditional variance and correlations (see Bodart & Reding, 1999; Bollerslev, 1990; Karolyi & Stulz, 1996; Longin & Solnik, 1995; Ng, 2000). Specifically, we estimate Eq. (1) with the following variance–covariance dynamics: hii;t ¼ ci þ ai hii;t1 þ bi e2i;t1 hij;t ¼ ðq þ bDummyt Þ hii;t hjj;t
1=2
ð2Þ
where hii,t and hjj,t are the ii th, jj th, and ij th element in ht ( i = the US market and j = the local market); q is the constant conditional correlation between the US and local market returns; Dummyt is the binary dummy variable that takes a value of 1 on data points in the post-attack period. If the return and volatility story of Table 1 and Fig. 2 is accurate, then b estimates for the return and conditional volatility should be significantly positive. A significantly positive b for the correlation coefficient indicates that the attacks caused a return (volatility) contagion from the US to other sample markets if the dependent variable is a return (conditional volatility) vector.
60
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
The results for b estimates are reported in Table 4. The coefficient estimates for return correlation reveal that the market return correlation with the US increased significantly for Japan in the period following the attacks, while the corresponding correlations for the UK and German markets remained approximately at the same level as the time period before the attacks. Table 4 also shows that the volatility correlation with the US market increased significantly for the UK and German markets but decreased for the Japanese market. These results can be interpreted as evidence that the attacks had a contagion in market returns from the US to Japan, whereas there was no contagion but a continued binterdependenceQ of market returns between the US and the UK and Germany after the attacks. In contrast, it is apparent that the attacks caused a volatility contagion from the US to UK and German markets. In sum, to the extent that higher correlations with the US market can have contagion effects from the US, the attacks caused a volatility contagion from the US to the UK and German markets but a return contagion from the US to the Japanese market. Yet, there was a continued interdependence of market returns (volatility) for the US/UK and US/German (Japanese) markets after the attacks. This suggests that the terrorist attacks gave rise to a structural change in the dynamics of international stock market correlations. Furthermore, these results have important implications for both international asset allocation decisions and risk management. The general increase in correlations as well as the return and
Table 4 Testing for contagion effects bˆ for return bˆ for conditional volatility
UK
Germany
Japan
0.0724 (1.29) 0.1265 (2.14)*
0.0828 (1.42) 0.1509 (3.05)**
0.1230 (2.08)* 0.0755 (0.98)
Testing for contagion effects are conducted by estimating the following regression equation: yt ¼ m þ et et jXt1
f N ð0; ht Þ
V a þ bVht1 b ht ¼ cVc þ aVet1 et1 hii;t ¼ ci þ ai hii;t1 þ bi e2i;t1 hij;t ¼ ðq þ bDummyt Þ hii;t hjj;t
1=2
where y t is the 2 1 vector of (R t , R*)Vor (v t , v*) *) t t where R t (v t ) is the S & P 500 return (volatility) and R*(v t t is the index return (volatility) for each individual non-US sample market; m is the 2 1 vector of constants; e t is the 2 1 vector of disturbances; X t 1 is the information set available at time t 1; ht is the (2 2) time-dependent conditional covariance matrix; and c, a, and b are the (2 2) parameter matrices; hii,t , hjj,t , and hij,t are the ii th, jj th, and ij th element in ht ( i = the US market and j = the local market); q is the constant conditional correlation between the US and local market returns; Dummyt is the binary dummy variable that takes a value of 1 on data points in the post-attack period. The numbers in parentheses are asymptotic t-statistics. * (**) indicates statistical significance at less than 5% (1%) level.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
61
volatility contagion to international markets erodes the expected gains from international diversification. 3.2. Testing for structural change in market correlations8 The presence of a structural change in cross-market correlations around the 9–11 terrorist attacks is tested using the pure structural break model with m breaks developed by Bai and Perron (1998 and 2003). The model is of the following format: y t ¼ dj þ m t
t ¼ Tj1 þ 1; . . . ; Tj for j ¼ 1; . . . ; m þ 1
ð3Þ
where y t is the dependent variable at time t; d j is the constant; m t is the disturbance at time t. By convention T 0 = 0 and T m + 1 = T, the total number of observations. The indices (T 1,. . ., T m) are break points. The number of break points for Eq. (3) is first selected using the Bayesian Information Criterion (BIC) for both return and volatility correlations. Then, as Bai and Perron (2003) suggests, a sequential procedure is used to estimate optimal break points.9 The sequential procedure uses a dynamic programming to find a segment that globally minimizes the overall sum of squared residuals. Statistical test for the breaks is conducted using the following F-statistic defined in Bai and Perron (2003): 1 T ðk þ 1Þ ˆ V V ˆ V 1 ˆ d R RV R FT ðk1 ; . . . ; kk Þ ¼ Rd ð4Þ T k where (T 1,. . ., T k ) is a partition such that Ti = [Tki ](i = 1,. . ., k); (Rd)V= (dV1 dV2,. . ., ˆ = the variance–covariance matrix of dˆ which is heteroskedasticity—and dVk dVk + 1); and V autocorrelation—consistent. It is known that maximizing the F-test Eq. (4), labeled sup F T (k), is asymptotically equivalent to F T (kˆ1,. . .,kˆ k ) minimize the global sum of squared residuals (see Andrews, 1993; Bai & Perron, 1998, 2003). Using the methodology described in Bai and Perron (2003), we first consider the sup F T (k) test of no structural break (m = 0) versus m = k breaks and then conduct a test of the null hypothesis of S breaks against the alternative that an additional break exists, labeled sup F T (S + 1 | S )}. The sup F T (S + 1 | S )} test is tantamount to the application of (S + 1) tests of the null hypothesis of no structural change versus the alternative hypothesis of a single change for each segment containing the observations Tˆi 1 + 1 to Tˆi (i = 1,. . ., S + 1). If there are (S + 1) breaks instead of S breaks in Eq. (3), the overall minimal value of the sum of residuals obtained with (S + 1) breaks should be significantly smaller than that ˆ obtained with S breaks. The parameter estimates, d j , are obtained from applying OLS segment by segment without constraint. The maximum number of structural changes allowed is set to 3. The value of the trimming is set to 0.15 for the construction and critical values of the sup F-type tests. The results are presented in Table 5. The sup F T (k) tests for k = 1 are highly significant for the return correlation for Japan and the volatility correlation for the UK and Germany. For k N 1, the values of the F T (k) 8
We are indebted to an anonymous referee who suggested this structural change analysis. Bai and Perron (2003) claims that the sequential procedure is superior to the procedure based on the information criteria since the latter cannot take into account potential heterogeneity across segments. 9
62
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
Table 5 Bai–Perron tests for structural breaks
A. Correlation in returns UK sup FT ( k ) sup FT ( + 1
Germany
Japan
k= 1
k= 2
k= 3
k= 1
k= 2
k= 3
k= 1
k= 2
k= 3
6.53
3.66
1.12
7.78
4.51
2.99
28.64*
7.31
2.45
)
=1
=2
1.04
0.05
=1 0.98
=2
=1
=2
0.03
0.77
0.01
*Indicates statistical significance at less than 1% level. We use a 5% size for the sequential test sup FT ( + 1 ). B. Correlation in volatility UK sup FT ( k ) sup FT ( + 1
Germany
Japan
k= 1
k= 2
k= 3
k= 1
k =2
k= 3
k= 1
k= 2
k= 3
32.38*
6.23
1.88
44.76*
6.97
2.86
2.43
1.15
0.01
=1
)
1.97
=2
=1
=2
=1
=2
0.01
1.26
0.01
0.33
0.02
*Indicates statistical significance at less than 1% level. We use a 5% size for the sequential test sup FT ( + 1 ). C. Break dates and parameter estimates Correlation in returns
Number of breaks
Correlation in volatility
UK
Germany
Japan
UK
Germany
Sequential
0
0
1
1
1
0
BIC
0 −
0 −
1
1
1
0.68 (0.16) 0.71 (0.15)
0.71 (0.09) 0.75 (0.09)
2001: 09:17 [09:10− 09:21] 0.19 (0.11) 0.30 (0.14)
2001: 09:17 [09:10− 09:19] 0.53 (0.19) 0.66 (0.17)
2001: 09:17 [09:10− 09:19] 0.38 (0.22) 0.56 (0.18)
0 −
Break dates
Parameter estimates
δˆ1 δˆ2
Japan
0.17 (0.26) −0.04 (0.21)
The model is of the following format with m breaks: yt ¼ dj þ mt
t ¼ Tj1 þ 1;. . . ; Tj for j ¼ 1;. . . ; m þ 1
where y t is the dependent variable at time t; d j is the constant; m t is the disturbance at time t. By convention T 0 = 0 and T m + 1 = T, the total number of observations. The first test is the sup F T (k) test of no structural break (m = 0) versus m = k breaks. Asymptotically, sup F T (k) = F T (kˆ 1,. . ., kˆ k ) where kˆ 1,. . ., kˆ k minimize the global sum of squared residuals. The second test has the null hypothesis of S breaks against the alternative that an additional break exists, labeled sup F T (S + 1 | S ). We use a 5% size for the sequential test sup F T (S + 1 | S ). *Indicates statistical significance at less than 1% level.
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
63
statistics are insignificant for all countries, suggesting that the model has just one break point (= two segments). This result is further supported by the sup F T (S + 1 | S ) test in that none of the statistics for S z 1 is significant. The BIC selects one break point and so does the sequential procedure. As expected, the break date is estimated at September 17, 2001 and the estimation is precise in the sense that the 95% confidence intervals cover only a few observations before and after. In sum, the statistical results provide evidence that there was an abrupt structural change in the dynamics of international stock market correlations in volatility for the UK and Germany and in returns for Japan around the 9–11 attacks. 3.3. Impulse response analysis To obtain additional insight into the structure of international stock market linkages, we conduct impulse response analysis. The impulse response analysis allows us to gauge to what extent shocks to international markets are explained by the US market, thus making it possible to examine how international stock markets can be destabilized by shocks that arise in the US. To investigate whether or not there was a change in the pattern of impulse responses after the attacks, the impulse response analysis is performed over two subsamples as defined in the previous section: pre-and post-attack period. Impulse response functions are obtained by estimating the following standard vector autoregression (VAR) model: xt ¼ A0 þ A1 xt1 þ A2 xt2 þ . . . . . . þ Ap xtp þ et
ð5Þ
where x t = the (4 1) vector of four market returns included in the VAR, i.e., return on the S & P 500, FTSE 100, DAX, and TOPIX; A 0 = the (4 1) vector of constants; Ai = the (4 4) matrices of coefficients (i = 1, 2,. . ., p); and e t = the error vector. Three lags were chosen in the estimation as this is the lag length that minimizes the Schwartz Baysian Criterion (SBC). A Choleski decomposition is used to orthogonalize the underlying errors using the ordering as defined above. Placed first in the ordering, the exchange rate change is implicitly assumed to be unaffected by the other shocks in the system. Although there is an increased interrelationship and integration among international stock markets, it is consistently documented in the literature that the US market is the price leader and dominates the world market, validating the econometric exogeneity of the US market in the VAR analysis (see, for example, Eun & Shim, 1989; Lin, Engle, and Ito, 1994; Masih & Masih, 2001). We also conduct impulse response analysis of volatility by replacing the four elements of x t by new elements of the four individual conditional volatilities, with the US volatility coming first in the ordering. Impulse response analysis of return (volatility) correlation is conducted by replacing x t by elements of the S & P 500 return (volatility) and three individual correlation coefficients in returns (volatility). Fig. 4 presents the results from the impulse response analysis based on the quadravariate VAR. The upper panel in each response function exhibits the pre-attack responses and the lower panel the post-attack responses. The upper and lower dashed lines in each graph are 95%-confidence bands. Specifically, Fig. 4 traces out impulse response functions from a one standard deviation shock in the US market to each of the non-US markets (UK, Germany, and Japan). The upper panel of Fig. 4(a), for example, reports the
64
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
(a) Responses of Return 0.02
0.02
0.015
0.02
0.015
0.01 0.005
0.015
0.01
0.01
0.005
0.005
0
0
0
-0.005
-0.005
-0.005
-0.01
-0.01
-0.01 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
UK
Germany
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
Japan
0.02
0.02
0.02 0.015
0.015
0.015
0.01
0.01
0.005
0.005
0.01 0.005
0
0
0
-0.005
-0.005
-0.005
-0.01
-0.01
-0.01 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
(b) Responses of Volatility 0.002
0.002
0.002
0.0015
0.0015
0.0015
0.001
0.001
0.001
0.0005
0.0005
0.0005
0
0
0
-0.0005
-0.0005
-0.0005
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
UK
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Germany
Japan
0.002
0.002
0.002
0.0015
0.0015
0.0015
0.001
0.001
0.001
0.0005
0.0005
0.0005
0
0
0
-0.0005
-0.0005
-0.0005
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Fig. 4. Impulse responses of return and volatility.
impulse response function of non-US market returns before the attacks. In the pre-attack period, a shock in the US market has a short-lived positive effect with a relatively small impulse response of the non-US market returns. In the post-attack period, the UK and German markets exhibit little or no change in the response of market returns as compared to the pre-attack period. On the contrary, a US shock in the post-attack period has a rapid and large short-run response from the Japanese market. This suggests that in the period following the attacks, the short run dependence of market returns to shocks that arise in the US market appears to have greatly increased in the Japanese market but remained nearly unchanged in the UK and German markets. The upper panel of Fig. 4(b) reveals that in the pre-attack period there is a modest amount of positive response of the non-US market volatility to a shock in the US market, with the most rapid response early on and later responses tapering off. By contrast, in the post-attack period, the non-US sample markets show a large amount of persistence in
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
65
volatility responses to a shock in the US market. Responses of the German and UK market volatility are especially dramatic after the attacks, in that they are nearly twice the magnitude of the pre-attack response and appear to persist for a long time. In sum, an overall market response of volatility to a shock in the US market increased after the attacks, with the most dramatic increase in response from the German and UK markets. Fig. 5(a) shows impulse response functions from a one standard deviation shock in the US market to each of the non-US marketsT return correlation with the US market. The upper panel of Fig. 5(a) shows the pre-attack impulse response functions of return correlation for each of the non-US markets. In the pre-attack period, a shock in the US market provides a temporary boost to the US/UK and US/Germany return correlations while the corresponding shock to the Japanese market seems to have a temporary turbulent impact on the Japan/US return correlation in terms of switching responses between positive and negative. In the post-attack period, the US shock has a strongly positive effect on the US/Japan return correlation but the response functions of the US/UK and the US/
(a) Responses of Return Correlation 0.06
0.06
0.05
0.05
0.04
0.04
0.03
0.03
0.02
0.02
0.06 0.05 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.05
0.01
0.01 0
0
-0.01
-0.01
-0.02
-0.02
-0.03
-0.03
-0.04
-0.04
-0.05
-0.05 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
UK
Germany
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Japan
0.06
0.06
0.06
0.05
0.05
0.05
0.04
0.04
0.04
0.03
0.03
0.03
0.02
0.02
0.02
0.01
0.01
0
0
0
-0.01
-0.01
-0.01
-0.02
-0.02
-0.02
-0.03
-0.03
-0.03
-0.04
-0.04
-0.05
-0.05
0.01
-0.04 -0.05 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
(b) Responses of Volatility Correlation 0.035
0.035
0.035 0.03
0.03
0.03
0.025
0.025
0.025
0.02
0.02
0.02
0.015
0.015
0.015
0.01
0.01
0.01
0.005
0.005
0.005
0
0
0
-0.005
-0.005 1 2 3 4 5 6 7 8 9 101112131415161718192021222324
-0.005 1 2 3 4 5 6 7 8 9 101112131415161718192021222324
UK
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Germany
Japan
0.035
0.035
0.03
0.03
0.03
0.025
0.025
0.025
0.02
0.02
0.02
0.015
0.015
0.015
0.035
0.01
0.01
0.01
0.005
0.005
0.005
0
0
0
-0.005
-0.005
-0.005 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1 2 3 4 5 6 7 8 9 101112131415161718192021 22 23 24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Fig. 5. Impulse responses of correlation for return and volatility.
66
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
Germany return correlation show little or no change over the two sub-periods. This implies that a US shock appears to have strengthened the return correlation between the US and Japanese markets after the attacks than before but had no noticeable change in response for the US/UK and US/German markets. The lower panel of Fig. 5(b) exhibits the pre-attack impulse responses of volatility correlation to shocks in the US market. In general, the US shock in the pre-attack period has a sufficiently positive impact on the volatility correlation between the US and non-US markets. Overall patterns of the post-attack responses of the UK and German markets are distinguishable from those of the pre-attack responses in that post-attack responses of volatility correlations are more strongly positive than the pre-attack responses. In sharp contrast to the case of the response function of return correlation, a US shock notably strengthened the volatility correlation for the US/UK and US/German markets after the attacks than before but had no noticeable change in response for the US/Japanese markets. Impulse response analysis conducted here reveals that overall impulse responses of the sample markets to a US shock were not conspicuously highly-positive for any of the variables before the attacks. In the period following the attacks, however, responses of market volatility to a US shock increased noticeably for the UK and Germany after the attacks but had little or no change for the Japanese market. The converse was true for responses of market returns. Responses of correlation with the US reveal that after the attacks, the UK and German markets were more strongly correlated through volatilities with the US market but the Japanese market was more strongly correlated through returns with the US market. These results reinforce the previous finding based on the analysis of dynamic time series behavior of correlation coefficients.
4. Conclusions In this paper we investigate evidence of both return and volatility contagion effects of the terrorist attacks across the major markets, and examine the extent to which international stock markets can be destabilized by shocks that arise in the US. To this end, the whole sample of four national stock market indexes (the S & P 500 of US, the FTSE 100 of UK, the DAX of Germany, and the TOPIX of Japan) is divided into two subperiods: the pre-attack period and post-attack period. We find that there was a drastic drop in market returns and a concomitant rise in market volatility immediately after the attacks. After the attacks, the overall market condition for the US and Japan was ameliorated from the risk and return perspective. Yet, the situation for the UK and Germany with regard to the risk profile was aggravated in the post-attack period, with a higher level of standard deviation than its pre-attack level. International stock markets can be correlated not only through their returns but also through their volatilities. Evidence presented in this paper indicates that, after the attacks, international stock markets were more strongly positively correlated through volatilities rather than through returns for the US/UK and US/German markets, while a market correlation between the US and Japan was strengthened through returns rather than through volatilities. This suggests that to the extent that higher correlations with the US market can enhance contagion effects from the US to other markets, the terrorist attacks
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
67
brought about a volatility contagion (rather than a return contagion) from the US to UK and German markets. In contrast, the Japanese market had a return contagion (rather than a volatility contagion) from the US market, suggesting that the terrorist attacks gave rise to a structural change in the dynamics of international stock market correlations. This is further supported by statistical evidence that there was an abrupt structural break in international stock market correlations in volatility for the UK and Germany and in returns for Japan around the 9–11 attacks. Impulse response analysis conducted in this paper shows that post-attack responses of the sample markets to a US shock were short-lived and positive in return and return correlations, whereas the corresponding responses of volatility and volatility correlations were fairly persistent. The non-US sample market returns showed a positive response to a shock in the US market, with the most intense response of the Japanese market returns after the attacks. An overall response of market volatility to a US shock increased after the attacks, with the most dramatic increase in the UK and German markets. The US shock had a strongly positive effect on the US/Japan return correlation but had little or no change over the two sub-periods for response functions of the US/UK and US/Germany return correlations. Impulse responses of the volatility correlation to a US shock notably increased after the attacks for the US/UK and US/Germany volatility correlations. An overall analysis of the post-attack period reveals that international market correlations were strengthened through volatility for the US/UK and US/German markets and through returns for the US/Japanese market.
Acknowledgement The author thanks Manuchehr Shahrokhi (the editor) and an anonymous referee for their helpful comments and suggestions. The author also benefited from comments of participants at the 2003 International Applied Business Research Conference, Acapulco, Mexico. The author is grateful to Steve Smith and Sandy Fleak for their comments/suggestions, and Brandon Duede for research assistance. The author gratefully acknowledges financial support from the Division of Business and Accountancy, Truman State University.
References Andrews, D. (1993). Tests for parameter instability and structural change with unknown change point. Econometrica, 61, 821 – 856. Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66, 47 – 78. Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change models. Journal of Applied Econometrics, 18, 1 – 22. Bodart, V., & Reding, P. (1999). Exchange rate regime, volatility and international correlations on bond and stock markets. Journal of International Money and Finance, 18, 133 – 151. Bollerslev, T. (1990). Modelling the coherence in short-run nominal exchange rates: A multivariate generalized ARCH model. Review of Economics and Statistics, , 498 – 505. Campbell, R., Koedijk, K., & Kofman, P. (2002). Increased correlation in bear markets. Financial Analysts Journal, , 87 – 94.
68
C.K.-C. Mun / Global Finance Journal 16 (2005) 48–68
Eun, C., & Shim, S. (1989). International transmission of stock market movements. Journal of Financial and Quantitative Analysis, 24, 241 – 256. Forbes, K. J., & Rigobon, R. (2002). No contagion, only interdependence: Measuring stock market comovements. Journal of Finance, 57, 2223 – 2261. Hamao, Y., Masulis, R. W., & Ng, V. K. (1990). Correlation in price changes and volatility across international stock markets. Review of Financial Studies, 3, 281 – 307. Karolyi, G. (1995). A multivariate GARCH model of international transmissions of stock returns and volatility: The case of the United States and Canada. Journal of Business and Economic Statistics, 13, 11 – 25. Karolyi, G., & Stulz, R. (1996). Why do markets move together? An investigation of US–Japan stock return comovements using ADRs. Journal of Finance, 51, 951 – 986. King, M. A., Sentana, E., & Wadhwani, S. (1994). Volatility and links between national stock markets. Econometrica, 62, 901 – 933. King, M. A., & Wadhwani, S. (1990). Transmission of volatility between stock markets. Review of Financial Studies, 3, 5 – 33. Lin, W. L., Engle, R. F., & Ito, T. (1994). Do bulls and bears move across borders? International transmission of stock returns and volatility. Review of Financial Studies, 7, 507 – 539. Longin, F., & Solnik, B. (1995). Is the correlation in international equity returns constant: 1960–1990? Journal of International Money and Finance, 14, 3 – 26. Masih, R., & Masih, A. M. (2001). Long and short term dynamic causal transmission amongst international stock markets. Journal of International Money and Finance, 20, 563 – 587. Ng, A. (2000). Volatility spillover effects from Japan and the US to the Pacific-Basin. Journal of International Money and Finance, 19, 207 – 233. Ramchand, L., & Susmel, R. (1998). Volatility and cross correlation across major stock markets. Journal of Empirical Finance, 5, 397 – 416. Solnik, B., Boucrelle, C., & Fur, Y. (1996). International market correlation and volatility. Financial Analysts Journal, , 17 – 34. Susmel, R., & Engle, R. (1994). Hourly volatility spillovers between international equity markets. Journal of International Money and Finance, 13, 3 – 25.