• 沒有找到結果。

Capital Market Integration and Exchange Rate Risk Exposure of the Asian Emerging Markets

N/A
N/A
Protected

Academic year: 2021

Share "Capital Market Integration and Exchange Rate Risk Exposure of the Asian Emerging Markets"

Copied!
18
0
0

加載中.... (立即查看全文)

全文

(1)

Capital Mar ket Integr ation and Exchange Risk Exposur e of

the Asian Emer ging Mar kets

Pin-Huang Chou*

Professor, Department of Finance National Central University

Edwar d H. Chow

Professor, Department of Finance National Cheng-Chi University

Gang Shyy

Professor, Department of Finance National Central University

ABSTRACT

This paper investigates the integration relationship between the Asian emerging capital markets (AEMs) and the rest of the world in the context of an international Sharpe-Lintner CAPM during the period 1985-1996. Overall, the results show that the world capital markets, with the inclusion of the AEMs, are not integrated. The rejection can be attributed to exchange risk exposures of the AEMs during the 80's. In addition, the results show that the AEMs become integrated with the world after 1991, the results of which are further confirmed by the fact that no exchange risk exposure is found for the AEMs after the 1991. However, the correlations between the AEMs and the world markets are still low.

JEL Classification: G12.

Keywor ds: CAPM; exchange risk exposure; GMM; SUR; Bootstrap.

*Corresponding author: Pin-Huang Chou. Tel: 886-3-4227151 ext. 6270. Fax:

886-3-4252961. E-mail address: choup@cc.ncu.edu.tw. Comments from John Wei and seminar participants at the 1996 Financial Annual Meeting (Hawaii, USA) are gratefully acknowledged. Financial support from the National Science Council of Taiwan (Grant No: NSC-84-2416-H-008-016) is gratefully appreciated.

(2)

I. Introduction

As cross-border capital flows have been vastly increasing, an interesting question is whether or not the world capital markets are becoming more and more integrated. Conceptually, if two assets of the same risk from two arbitrarily selected capital markets are priced equally, then the markets are said to be integrated. In the canonical equilibrium mean-variance framework, evaluating if risky assets are properly priced relies on the availability of an efficient benchmark portfolio, which is indistinguishable from the test of the underlying asset pricing models. For example, a test of the international CAPM can be viewed as a joint hypothesis that the underlying benchmark portfolio is mean-variance efficient and the capital markets are integrated [see Harvey (1991) and Harvey and Zhou (1993)].

Solnik (1977), Stulz (1981), and Alder and Dumas (1983) prove that an international version of the Sharpe-Lintner CAPM only holds under some restrictive conditions. In particular, the conditions that the world capital markets are fully integrated and that there exist no exchange rate risks are required to ensure the validity of the international CAPM. Otherwise, the markets are segmented, in which case there may not exist a single factor, such as the world market portfolio, that can fully explain the risk-return relationship of assets in different capital markets. As a result, changes in exchange rates may serve as a second factor that explains the cross-sectional differences between assets of different markets [see, e.g., Jorion (1990, 1991)].

There is evidence supporting the significance of exchange rate risk. Dumas and Solnik (1995) find that exchange rate risks are priced in the world's four largest equity markets using a conditional approach that allows for time variation in the rewards for exchange rate risk. Bartov, Bodnar and Kaul (1996) find that part of the increase in the volatility of US multinational firms associated with the breakdown of the Bretton Woods system is related to an increase in the market risk for multinational firms. Chow, Lee and Solt (1977) find that the exchange rate risk exposure of US firms resulting from interest rate and cash flow effect increases significantly with return horizons.

Based on the Sharpe-Lintner CAPM, Harvey and Zhou (1993) apply the tests proposed by Gibbons, Ross, and Shanken (hereafter GRS, 1989) and Zhou (1993) to a sample of 17 developed capital markets (including 16 OECD countries and Hong Kong), and find that the mean-variance efficiency of the MSCI world index (hereafter MSCI WI) cannot be rejected. Using a sample of mutual funds, Cumby and Glen (1990) also cannot reject the efficiency of the MSCI world index. In other words, the developed capital markets are well integrated, and the MSCI world index seems to be a good proxy for the world market portfolio.

Harvey's (1995) study including the Asian emerging capital markets (AEMs) comes to different conclusions. Based on the data from the Emerging Market Data Base (EMDB) of the International Financial Corporation (IFC) for the period from February 1976 to June 1992 Harvey concludes that the MSCI world index fails to explain the risk-return relationship of twenty emerging capital markets around the world. That is, the emerging markets are not fully integrated into the world capital markets. Even after an exchange rate risk factor proxied by an aggregate trade-weighted portfolio of 10 currency deposits is included, the market-plus-exchange-rate-risk two-factor model is rejected.

(3)

In this paper we examine the relationship between capital market integration and exchange rate risk for OECD nations and AEMs. Instead of using Harvey's trade-weighted currency portfolio which includes both the change in the value of the currencies of G-10 countries minus the United States and plus Switzerland and their Eurodeposit rates, we follow Jorion (1990, 1991) and Chow, Lee and Solt (1997) that employ changes in trade-weighted exchange rate indexes to proxy for the exchange rate risk. In specific, we compile a series of trade-weighted exchange rates for each of the countries under consideration. The weights for a particular country's exchange rate index are bilateral trades of the country under investigation with all the other countries in the sample. Thus, each country has a unique series of exchange rates. This is different from Harvey's approach (1995) that basically uses the same currency portfolio for every country in his sample. Our approach allowing the exchange rate index to vary with countries is a more realistic one.

In essence, the two-factor model of Jorion (1991) is extended to an international framework in which each country's equity index is considered as a risky asset. Since different countries' second factor (the exchange rate risk factor) are distinct, the two-factor model can be represented in terms of the seemingly unrelated regression (SUR) model. Zellner's SUR estimators and tests are used that take into account the contemporaneous correlations among different markets.

This paper first examines whether or not the international CAPM holds for 5 AEMs and 17 developed markets. In addition to the F test of GRS (1989) and the GMM test as adopted by Harvey and Zhou (1993) and Harvey (1995), we also employ a bootstrap approach proposed by Chou and Zhou (1996) to test the mean-variance efficiency of a portfolio. The bootstrap approach is also applied to the SUR estimation for the two- factor model. A major advantage of the bootstrap is that no pre-specified distributional assumption is required for the error terms. In particular, under mild regularity conditions, the bootstrap generally yields an approximation to the sampling distribution of an estimator or test statistic that is at least as accurate as the approximation obtained from traditional first-order asymptotic theory. Chou and Zhou (1996) find that the GRS test tends to over-reject the null hypothesis when the error terms significantly deviate from normality, while the bootstrap test does not. However, the GRS test seems to be a little more powerful in rejecting the null than the bootstrap.

Our empirical results confirm Alder and Dumas' (1983) theory that the conditions that the world capital markets are fully integrated and that there exist no exchange rate risks are required to ensure the validity of the international CAPM. We find that rejection of the international CAPM is attributable to exchange rate exposures. During the 1980's the AEMs are segregated from the rest of the world when significant exchange rate exposure is present.

The rest of the paper is organized as follows. Section 2 introduces the data. Section 3 discusses the tests of market integration based on the CAPM and the empirical results of the tests. Section 4 presents the tests of exchange risk exposure. Section 5 concludes the paper.

II. Data

Following Harvey (1995), the monthly index returns for five Asian emerging Markets (Korea, Malaysia, Philippines, Taiwan, and Thailand), constructed by the

(4)

International Financial Corporation (IFC), and the index returns for 17 developed countries (including 16 OECD countries and Hong Kong), compiled by Morgan Stanley Capital International (MSCI), are selected as our samples.1

The 30-day Libor rates are used as the riskfree rates. The sample period starts from February 1985 to May 1996, which contains a total of 130 observations.2 In addition, to examine if structural changes in risk-return relationship occur before and after 1990, the sample is also divided into two sub-samples, 1985:02-1990:12 and 1991:1-1996:5, for further analysis.

Table 1 presents the summery statistics of the returns on the five Asian emerging markets and on the 17 developed capital markets for the sample period. The results indicate that overall the Asian emerging markets have gained higher average returns during the sample period. The Sharpe measures for the AEMs are also higher than those of the developed markets, reflecting the high economic growth of the AEMs since the 80's.

III. Tests of capital mar ket integr ation

It is worth noting that tests of capital market integration in the context of the Sharpe-Lintner CAPM is essentially equivalent to testing the mean-efficiency of the underlying proxy for market portfolio. Several tests of mean-variance efficiency have been proposed for various distributional assumptions. For example, under iid multivariate normality assumptions, Gibbons, Ross, and Shanken (GRS, 1989) show that the mean-variance efficiency of a portfolio, in the case where there exists a risk-free asset, can be tested with a Hotelling's

T

2statistic that has an exact F distribution. Based on the same assumptions, Harvey and Zhou (1991) investigate the Sharpe-Lintner CAPM in the Bayesian framework, and evaluate the hypothesis implied by the CAPM using Monte Carlo numerical integration. However, multivariate normality is not necessary to ensure the validity of the mean-variance framework, upon which the CAPM is derived. The largest class of distributions known thus far that validates the mean-variance framework is the class of elliptical distributions [see, e.g., Ingersoll (1987)]. Noting this fact, Zhou (1993) further proposes an approach based on Monte Carlo simulation to test the CAPM that allows the distribution of the (excess) asset returns and the market model error terms to be elliptical. The members of elliptical distributions include multivariate normal, mixture of multivariate normal, multivariate t, multivariate stable, and so on. Zhou (1993) finds that the GRS test is robust when excess returns follow a market model whose error term is not normally distributed but still retains the property of ellipticity. However, when returns are elliptically but not normally distributed, in which case the market model error term no longer has an elliptical distribution, the GRS test tends to reject the null hypothesis of portfolio efficiency too often. That is, the size of the GRS test is higher than the nominal significance level. When alternative distributional specifications are adopted for the asset returns, the

1

Since IFC data for Indonesia are only available starting from 1989:12, we exclude Indonesia from our sample of Asian emerging markets.

2 To avoid the potential abnormal influence of the global crash in October of 1987, the

(5)

mean-variance efficiency of the CRSP value-weighted index can no longer be rejected. It is worth noting that his analysis is exact in that the statistical inference does not depend on the sample size. However, the inference is legitimate only if the underlying distributional specification is correct and provided that the random number generator for the assumed distribution is available. In addition, the distributions of the asset returns (or error terms) are never known in the real world. Hence, although Zhou's (1993) method extends the test of mean-variance efficiency to the case of elliptical distributions, his test requires that the returns follow a specific distribution.

Table 1: Summar y Statistics

This table presents the sample mean and standard deviation (in parentheses) of returns on the MSCI World Index (denoted `World'), indexes of 17 developed capital markets (including 16 OECD countries and Hong Kong), and 5 Asian emerging markets over the sample period from 1985:03 to 1996:05. The Sharpe measure for each index is also calculated in excess of the 30-day LIBOR rate. The results for two subperiods, 1985:03-1990:12 and 1991:01-1996:05 are also reported.

MacKinlay and Richardson (1991) propose a generalized method of moments (GMM) approach to test portfolio efficiency that requires much weaker distributional assumptions. More specifically, their GMM approach allows the disturbance term to be serially dependent and conditionally heteroscedastic. The

Full Sample 1985.3-1990.12 1991.1-1996.5 Index Mean(%) Sharpe Mean(%) Sharpe Mean(%) Sharpe World 1.380 (11.057) 0.077 0.422 (11.054) -0.022 1.431 (11.429) 0.067 Korea 1.658 (8.249) 0.137 2.9102 (9.159) 0.245 2.692 (8.741) 0.232 Malaysia 1.534 (7.205) 0.139 0.7994 (8.441) 0.016 1.392 (7.561) 0.096 Philippines 3.539 (10.139) 0.297 3.8424 (11.890) 0.267 4.378 (11.311) 0.328 Taiwan 2.395 (13.062) 0.143 4.1032 (15.490) 0.222 3.519 (14.578) 0.196 Thailand 2.171 (8.070) 0.203 2.4126 (7.552) 0.231 2.553 (7.513) 0.251 Australia 1.638 (12.883) 0.086 0.373 (13.891) -0.021 1.752 (14.169) 0.077 Belgium 1.963 (11.575) 0.124 1.420 (11.538) 0.066 2.581 (12.102) 0.158 Denmark 1.464 (11.409) 0.082 1.158 (11.721) 0.042 1.993 (12.011) 0.106 Germany 1.750 (11.898) 0.103 1.229 (12.959) 0.044 2.283 (12.855) 0.126 France 1.859 (11.443) 0.116 1.399 (12.133) 0.061 2.4427 (11.860) 0.149 U. K. 1.501 (11.340) 0.086 0.788 (11.365) 0.011 1.925 (11.890) 0.106 H. K. 2.301 (12.363) 0.143 0.571 (12.221) -.0075 2.060 (12.720) 0.110 Italy 1.509 (12.128) 0.081 1.197 (12.721) 0.042 2.135 (12.445) 0.118 Japan 1.399 (11.890) 0.073 1.138 (12.388) 0.038 1.833 (12.622) 0.093 Canada 0.990 (11.710) 0.039 -0.264 (11.233) -0.083 0.972 (12.162) 0.026 Netherlands 1.812 (11.225) 0.114 0.655 (11.289) -.0008 1.889 (11.848) 0.104 Norway 1.580 (12.041) 0.087 0.912 (12.217) 0.020 2.074 (12.332) 0.155 Austria 1.842 (12.705) 0.103 2.500 (13.271) 0.138 3.077 (12.671) 0.177 Sweden 2.027 (12.188) 0.123 1.184 (12.714) 0.041 2.295 (13.196) 0.124 Swiss 1.979 (11.759) 0.123 0.796 (12.370) 0.011 1.873 (12.844) 0.094 Spain 1.501 (12.235) 0.079 1.280 (12.453) 0.049 1.965 (13.057) 0.100 USA 1.467 (11.405) 0.082 0.020 (11.052) -0.058 1.242 (11.607) 0.050

(6)

GMM tests are also used by Harvey and Zhou (1993) and Harvey (1995). The GMM test, however, is asymptotic and relies on the availability of large samples.

In this paper, we also adopt a bootstrap approach proposed by Chou and Zhou (1996) to test the mean-variance efficiency that does not assume a specific distribution. Bootstrap method, originally proposed by Efron (1979), is basically a computer-based method that allows one to analyze the distribution of an estimator or test statistic of interest without needing to perform complicated analytical computations. In particular, bootstrap methods usually yield estimates of a higher accuracy than those obtained from classical asymptotic theories. In particular, Chou and Zhou (1996) found that when the distributions of the error terms significantly deviate from normality, the traditional GRS test may suffer from the problem of overrejecting the null.

In the following, we first introduce the model and the various tests of mean-variance efficiency. We then present the empirical results for both the Asian emerging markets and the developed markets during the sample period.

3.1 The model and the tests

Consider the following system of excess-return market model:

,

it mt i i it

r

r

=

α

+

β

+

ε

i

=

1,… ,

N

;

t

=

1

,....

T

,

; (1)

where

=

it

r

the excess return on asset

i

in period t,

=

mt

r

the excess return on the MSCI world index in period t, whose efficiency is being tested

;

=

it

ε

the disturbance term for asset

i

in period t.

N

refers to the number of assets under consideration, and

T

is the number of time series observations. In this paper, returns on risky assets refer to the returns on various country index portfolios.

The above model can be rewritten in a more compact form as follows:

,

t mt

t

r

R

=

α

+

β

+

ε

t

=

1

,...

T

,

(2)

where

R

t is the (

N

×1 ) vector of excess returns;

α

=

(

α

1

,...

α

N

)

;

(

)

=

β

β

N

β

1

,...,

; and

ε

t

=

(

ε

1t

,...,

ε

Nt

)

. The error term

ε

t is assumed to be distributed with unknown distribution function

P

( )

0

,

, whose mean is zero and the covariance matrix is

.

If the Sharpe-Lintner CAPM holds for this set of assets and the benchmark portfolio (i.e., the MSCI WI) is mean-variance efficient, the following restriction on the parameters of equation (1) should hold:

( )

R

t

E

( )

r

mt

E

=

β

(7)

Hence, this restriction implies a testable joint hypothesis [GRS (1989)]:

0

:

0

α

=

H

In the context of Sharpe-Lintner CAPM, test of this hypothesis can be viewed as a joint test of complete integration among capital markets and the mean-variance efficiency of the underlying portfolio. Since the MSCI WI has been shown to well integrated with the developed markets, rejecting the null hypothesis with respect to the AEMs may suggest that the AEMs are not well integrated into the world capital markets.

If the error terms follow an iid multivariate normal distribution, Gibbons, Ross, and Shanken (1989) show that the following GRS statistic has a noncentral F distribution with degrees of freedom

N

and

T

N

−1:

(

)

α

α

, 1

( )

λ

1 1

~

'

2

1

− − ∧ − ∧ ∧ −

Σ

h

F

NT N

T

N

N

T

GRS

, (3) where

 +

=

1

1

m2

T

h

θ

, and m

=

r

m

s

m

θ

is the observed Sharpe measure of the portfolio p;

r

m and

s

m are, respectively, the sample mean and sample standard

deviation

r

mt.3

λ

=

h

−1

α

Σ

−1

α

is the noncentrality parameter of the F distribution. Under the null hypothesis that

α

=0, the distribution reduces to the central F distribution. Hence, critical value and p value of the statistic are easily calculated. Geometrically, test of the zero-intercept hypothesis is equivalent to testing if the inclusion of the benchmark portfolio to the original N risky assets can significantly “span” the efficient frontier. This test is also referred to as intersection test in the literature [see Jobson and Korkie (1989)].

If the multivariate normality assumption is violated, the null hypothesis can be tested with Hansen's (1982) generalized method of moments (GMM). Specifically, define the following orthogonality conditions:

(

)

(

)

=

=

T t t T

f

T

g

1

,

1

,

β

α

β

α

(

)

(

)

1 2

,

×





=

N mt t mt mt t t

r

R

r

r

R

f

β

α

β

α

β

α

By maximizing a quadratic function of the following form:

3

The sample standard deviation is not adjusted for

degrees of freedom. Specifically,

(

)

− = 1 2 m mt m T s γ γ

(8)

(

α

,

β

)

g

T

(

α

,

β

)

Wg

T

(

α

,

β

)

,

Q

=

where

W

is a weighting matrix which is symmetric and positive-definite, one obtains the GMM estimates of the model parameters. Under the restriction implied by the null hypothesis, one may define the following moment conditions:

(

)

(

)

=

=

=

=

T t t T

f

T

g

1

,

0

1

,

0

β

α

β

α

Using Hansen's results, the null hypothesis can be tested using the following statistic: 2 1

~

,

0

,

0

T N T

W

g

Tg

α

β

α

β

χ

 =

 =

∧ ∧ − ∧ where ∧

β

and ∧

W

are the restricted GMM estimates. The statistic has a chi-square distribution with N degrees of freedom under the null hypothesis. More specifically, by imposing values of zero on the intercepts, the “restricted” market model can be estimated using GMM and the validity of the restriction can be tested based on a chi-square test with N degrees of freedom [see MacKinlay and Richardson (1991) and Harvey and Zhou (1993) for detail]. In this paper, the Parzen kernel is used to obtain a consistent covariance matrix estimate that is robust against the presence of serial correlation and heteroscedasticity [see Andrews (1991)]. Here it is only required that the returns be stationary and ergodic and that the fourth moments exist for the excess returns.

However, the GMM tests rely on large samples to ensure the convergence of the test statistic to a normal or chi-square distribution. Zhou (1993) proposes a Monte Carlo simulation method that allows the asset returns and the market model disturbance terms to be elliptically distributed. Under a prespecified distribution for the returns or the error terms, he can calculate the exact p value for the test of zero-intercept hypothesis. However, a weakness of his approach is that the parameters for the alternative distributions, such as the degree-of-freedom parameter in the multivariate t distribution and the mixing-probability parameter in the mixture-normal distribution, are unknown, and are obtained by trial and error in his paper.

As an approach complementing the above methods, a bootstrap approach proposed by Chou and Zhou (1996) is also adopted. An advantage of the bootstrap is that no pre-specified distributional assumption is required for the error terms. In particular, under mild regularity conditions, the bootstrap generally yields an approximation to the sampling distribution of an estimator or test statistic that is at least as accurate as the approximation obtained from traditional first-order asymptotic theory. In many instances the sampling distribution of a statistic may not be analytically available, while the bootstrap, on the other hand, obtains the sampling distribution of the statistic via repeatedly resampling from the sample at hand. We briefly introduce the bootstrap approach in the following.

(9)

3.2 The bootstr ap approach

Following Chou and Zhou (1996), we adopt the following quadratic Wald-type statistic to test the null hypothesis of zero intercepts:

∧ − ∧ ∧ −

Σ

=

1

α

1

α

'

h

H

T

,

where ∧

α

and ∧

Σ

are, respectively, the estimates of

α

and

Σ

. The procedure for bootstrap hypothesis test is described below.

(1) Estimate the model using OLS. Let

α

and

ε

denote the OLS intercept estimate and the OLS residual, respectively. Denote

Σ

=

∧ = ∧ ∧

t T t t

T

1

ε

ε

1

. Calculate the following:

∧ − ∧ ∧ −

Σ

=

1

α

1

α

'

h

H

T

Calculate the restricted estimate for

β

:

β

res

=

( )

r

m'

r

m −1

r

m'

R

, where

(

m mT

)

m

r

r

r

1

,...,

'

=

and

R

=

(

R

1

,...,

R

t

)

.

(2) Repeat the following steps a large number of times (we use 100,000 iterations). (a) Draw a bootstrap sample

{ }

ε

t* from

{ }

εˆ

t . Let

R

t*

=

β

res

r

mt

+

ε

t*,

t

=

1

,...,

T

(b) Calculate the OLS estimates for model parameters based on

(

* *

)

1 *

,...,

R

T

R

R

=

and

r

m

'

s

. Denote the estimates

α

ˆ

* and

Σ

ˆ

*. (c) Calculate * 1 * * 1 *

ˆ

ˆ

'

ˆ

α

α

− −

Σ

=

h

H

T

(3)

Calculate the percentage of

H

T*

'

s

that are greater than

H

T, which gives the bootstrap achieved significance level.

The achieved significance level is parallel to the p-value in classical statistical analysis. Based on this test, Chou and Zhou (1996) find that the CRSP value-weighted index is rejected for most of the time from 1926 to 1993, while the GRS F test fails to reject the efficiency for most of the time. Chou and Zhou (1996) also find that both the bootstrap test and the GRS F test generally have reasonable sizes with respect to the nominal significance levels under various distributional assumptions. However, the GRS test tends to over-reject the null hypothesis when the error terms significantly deviate from normality, while the bootstrap test does not suffer from the same problem. However, they also point out that the GRS test seems to be a little more powerful in rejecting the null than the bootstrap test.

3.3 Empir ical results

Table 2 presents the regression results of the market model for each index for the full sample and for two sub-samples. As in Harvey (1995), the results show that the world index generally has very low explanatory power on the variation of the AEMs. Table 2 also shows that the systematic risks

β

'

s

are low for the AEMs. The low correlations between the AEMs and the world capital markets suggest a

(10)

potential benefit of international diversification by including the AEMs into the international portfolio.

Table 2: Regr ession Results

This table presents the systematic risk (

β

i) and

R

2 for each MSCI country index based on the following model:

,

it mt i i it

r

r

=

α

+

β

+

ε

where

r

mt is the excess returns on the MSCI world index (WI). Libor rates are used as the proxy riskfree rate.

Table 3 presents the empirical results of zero-intercept hypothesis of the market model regression based on GRS F-test, GMM test, and the bootstrap test of Chou and Zhou (1996). We test the mean-variance efficiency of the MSCI world index with respect to the 5 Asian emerging markets (denoted ‘+AEMs’), the 17 developed markets (denoted ‘+ OECD17’), and the combination of all markets (22 markets, denoted ‘+AEMs+OECD17’). Full Sample 1985.3-1990.12 1991.1-1996.5 Index

β

2

R

β

R

2

β

R

2 Korea 0.00099 0.0000 0.05503 0.0048 0.00116 0.0000 Malaysia 0.03444 0.0027 0.04443 0.0040 0.03784 0.0036 Philippines 0.12364 0.0180 0.12010 0.0130 0.13509 0.0317 Taiwan 0.05347 0.0020 0.15377 0.0132 0.03081 0.0009 Thailand 0.17290 0.0548 0.21843 0.0973 0.11561 0.0213 Australia 1.06818 0.8605 1.11462 0.8072 1.02782 0.9247 Belgium 0.97990 0.9067 0.98122 0.8765 0.99861 0.9386 Denmark 0.96597 0.8728 0.97938 0.8311 0.96687 0.9162 Germany 0.98053 0.8463 1.02082 0.7957 0.97885 0.9248 France 0.96821 0.8817 0.97005 0.8392 0.99480 0.9314 U. K. 0.97734 0.9264 0.97628 0.8982 0.98154 0.9515 H. K. 0.96727 0.7622 0.93857 0.7342 0.97936 0.7696 Italy 0.95775 0.7577 0.97561 0.7525 0.96299 0.7543 Japan 1.02771 0.8862 1.08127 0.9063 0.96467 0.8596 Canada 1.00244 0.9281 0.97469 0.9086 1.01952 0.9423 Netherlands 0.97663 0.9558 0.98432 0.9476 0.96984 0.9627 Norway 0.96487 0.8052 0.94618 0.7538 1.02492 0.8679 Austria 0.94905 0.7009 0.92979 0.5776 1.02387 0.8995 Sweden 1.02244 0.8720 1.09289 0.8852 0.95479 0.8556 Swiss 1.00046 0.9043 1.06069 0.8946 0.95113 0.9201 Spain 1.00980 0.8607 1.03613 0.8312 1.00038 0.8312 0.8938 USA 0.99891 0.9526 0.95715 0.9311 1.03231 0.9722

(11)

Table 3: Test of Por tfolio Efficiency and Integr ation Relationship

This table presents empirical results on test of zero-intercept hypothesis for the following model based on the GRS (1989) F-test, GMM test, and bootstrap test of Chou and Zhou (1996): , it mt i i it r r = α + β + ε

where

r

mt is the excess returns on the MSCI world index (WI). Libor rates are used as the proxy riskfree rate. ‘OECD17’ refers to the set of 16 OECD countries and Hong Kong. ‘AEMs’ refers to the 5 Asian emerging markets. The p-values of the bootstrap test are calculated based on 100,000 iterations.

Rt θ*a GRSa GMMc Bootstrapd

Panel A: Full sample

WI 0.0767 +AEMs 0.3130 0.0446 0.1055 0.02012 +OECD17 0.4772 0.1113 0.3432 0.00584 +OECD17+AEMs 0.5665 0.0709 0.3249 0.00036 Panel B: 1985:03-1990:12 WI -0.0218 +AEMs 0.4327 0.0582 0.1399 0.01114 +OECD17 0.6893 0.2025 0.5838 0.00078 +OECD17+AEMs 0.8598 0.1452 0.4918 0.00081 Panel C: 1901:01-1996:05 WI 0.0670 +AEMs 0.2945 0.5117 0.5227 0.38404 +OECD17 0.8285 0.0440 0.2244 0.00000 +OECD17+AEMs 0.4446 0.0681 0.3524 0.00000 a

The Sharpe measure of the underlying portfolio or the maximum attainable Sharpe measure for a set of assets.

b The p-value for the GRS F-test of zero-intercept hypothesis; c The p-value based on the GMM test, distributed as

χ

2

with the degrees of freedom equal to the number if assets (country indexs);

d

The p-value (achieved significance level) based on the bootstrap test of Chou and Zhou (1996).

For the full sample, the results of GRS F-test and the GMM test show that efficiency of the MSCI WI cannot be rejected with respect to the developed markets, as in Cumby and Glen (1990) and Zhou and Harvey (1993), implying that the developed markets are well integrated with the world. However, the GRS test shows that the null hypothesis is rejected with respect to the AEMs with a p-value of 0.0446, and is marginally rejected with respect to all markets (i.e. the developed and the Asian emerging markets) with a p-value of 0.0709, suggesting that the segregation of world capital markets is mainly due to the segregation of the AEMs. Without a specific distributional assumption, on the contrary, the bootstrap test

(12)

shows that the mean-variance efficiency of the MSCI world index is rejected for both the developed and the emerging markets. The efficiency with respect to the developed markets is even rejected with a much lower p-value (0.00584), comparing with the p-value of 0.02012 with respect to the AEMs. Since assets returns are known to deviate from normality, the results of bootstrap tests suggest that the world capital markets are in fact not fully integrated.

To further investigate the structural stability of the integration relationship before and after 1990, tests are also applied to two sub-periods, the results of which are presented in Panels B and C of Table 3. For the first sub-period (from 1985:03 to 1990:12), the GRS F test cannot reject the efficiency of the MSCI WI for either set of markets. The efficiency with respect to the AEMs is marginally rejected with a p-value of 0.0582. The bootstrap test, again, rejects the efficiency for each of the scenarios.

Interestingly, however, for the second sub-period starting from 1991, the GRS test shows that the efficiency of the MSCI WI cannot be rejected with respect to the AEMs, whereas the efficiency with respect to the developed markets is rejected with a p-value of 0.0440. As a result, the efficiency of the index with respect to the set of all markets is also rejected marginally. The result suggests that the AEMs become integrated with the world capital markets in the second sub-period. The (marginal) deviation of the world integration is thus attributable to segregation of the developed markets. The evidence of the integration of the AEMs into the world is further strengthened by the result of bootstrap test, which shows that for the second sub-period the p-value of the efficiency of the world index with respect to the AEMs is 0.38404. Again, the results of the bootstrap test show that the efficiency with respect to the developed markets and the full markets is rejected.

Overall, our empirical results show that the global integration relationship between the AEMs and the developed capital markets does not hold for the full sample. A further examination reveals that the rejection of the global integration is mainly attributablt to the failure of the MSCI world index to account for the performances of the developed countries, especially after 1990. Although the AEMs are also not fully integrated into the world markets for the full sample, we find that they become well integrated into the world capital markets after 1990. The integration relationship may be attributed to several liberalization and internalization actions taken by the AEMs since the 1990's.

In next section, we investigate if exchange rate risk can account for the failure of the single-factor pricing model. Since we only have 130 observations for the full sample and 65 observations of each of the sub-samples, the GMM test fails to reject the null hypothesis in all scenarios. The results of the GMM test may not be as credible with respect to the GRS and bootstrap tests.

3.4 Exchange r isk exposure of the AEMs vs. the developed mar kets

We have rejected the use of the MSCI world index as the sole factor explaining the risk-return behavior of the world capital markets. A natural investigation along this line is to see if the rejection can be attributed to an omitted factor, namely the exchange rates. Hence, following the idea of Jorion (1990, 1991), we investigate the following model:

(13)

it it i mt i i it

a

b

R

X

e

R

=

+

+

γ

+

i

,

t (4)

,

where

R

it and

R

mt are raw returns on asset

i

and the world index at time

t

, respectively;

X

it is the rate of change in a trade-weighted exchange rate for country

i

,

expressed in terms of the corresponding local currency. For each country, the trade-weighted exchange rate is calculated as the weighted average of its exchange rates with respect to the remaining 21 countries. The model differs from Jorion (1991) mainly in that here we investigate the exchange risk exposure in a country level, rather than in a firm level as in Jorion (1990, 1991). In addition, the “market factor” in the model is the MSCI world index, rather than a country index. Our model also differs from Harvey (1995) in that the second factor of Harvey's two-factor model is a trade-weighted portfolio of 10 currency deposits. Since each country has different trades with other countries, a unanimous exchange rate factor may not well capture the exchange risk exposure. This may also explain why Harvey (1995) rejects both the single- and two-factor models.

If there is no exchange risk exposure, one expects the coefficient

γ

ito be zero. Hence, in addition to testing the significance of the individual coefficients, we can also test the following joint hypotheses for different set of markets:

0

:

0 i

=

H

γ

i

∈AEMs;

0

:

0 i

=

H

γ

i

∈OECD and Hong Kong;

0

:

0 i

=

H

γ

i

∈AEMs, OECD and Hong Kong.

Since the currency rates

X

it

'

s

differ for different countries, we estimate the model (4) by Zellner's seemingly unrelated regression (SUR) estimation approach. We estimate the model for the full sample and each of the sub-samples. The joint tests are then tested based on an F statistics with degrees of freedom

N

and

N

(

T

2

)

, where

N

refers to the number of countries in the system and

T

is the sample size [see, e.g., Greene (1993, page 491)]. The empirical results are reported in Table 4.

For the full sample, the results show that most AEMs do not experience exchange risk exposures, except that the exposure coefficient of Taiwan is significantly negative at 5% level. However, the joint test that the AEMs do not experience exchange risk exposures cannot be rejected. The null hypothesis of no exchange rate exposure is rejected for bboth the developed markets an the set of all markets. The exposure coefficient is significantly negative with a p-value of 0.0001 for five of the developed countries: Denmark, United Kingdom, Japan, Canada, and the United States. Based on firm-level data, Jorion (1990) also documents a negative relationship between stock returns and exchange rate changes in the US market, though the relationship is not significant. He, Ng, and Wu (1995) also find that stock returns are negatively correlated to contemporaneous exchange rate changes in the Japanese market. The results explain, at least in part, why the mean-variance efficiency of the world index is rejected with respect to the developed markets and the set of all markets for the full sample.

(14)

Table 4: Exchange Risk Exposur e of the Asian Emer ging Mar kets vs. the Developed Mar kets

The table presents the empirical results of the following model: , ,t i e X R b a Rit = i + i mt + γ i it + it

where

X

it is the rate of change in a trade-weighted exchange rate for country i. The model is estimated based on Zellner’s SUR estimator, and the joint hypotheses are tested using F test. ‘OECD17’ refers to the set of 16 OECD countries and Hong Kong. ‘AEMs’ refers to the 5 Asian emerging markets.

Full Sample 1985.3-1990.12 1991.1-1996.5 Index Mean(%) (P-value) Mean(%) (P-value) Mean(%) (P-value)

Korea 0.0303 (0.9197) -0.1495 (0.7281) 0.4284 (0.2409) Malaysia 0.2647 (0.2230) 0.0196 (0.9476) 0.3804 (0.0934) Philippines -0.0761 (0.7794) 0.1598 (0.6949) -0.2218 (0.3461) Taiwan -1.2287 (0.0149) -2.9873 (0.0001) -0.2085 (0.6812) Thailand 0.1322 (0.5433) 0.0560 (0.8125) 0.3499 (0.1976) Australia -0.1859 (0.0945) -0.5869 (0.0048) -0.1088 (0.1901) Belgium -0.0363 (0.4016) -0.0518 (0.2943) 0.3838 (0.0053) Denmark -0.8123 (0.0001) -1.5032 (0.0001) -0.3276 (0.0519) Germany 0.0888 (0.3927) 0.0335 (0.8855) 0.2474 (0.0015) France 0.1879 (0.1227) 0.0975 (0.6466) 0.2400 (0.0142) U. K. -0.5167 (0.0001) -0.5860 (0.0009) -0.4793 (0.0001) H. K. 0.2798 (0.2890) 0.7340 (0.0995) 0.3223 (0.2508) Italy -0.2952 (0.0995) 0.0299 (0.8961) -0.6531 (0.0077) Japan -0.6963 (0.0001) -0.9206 (0.0001) -0.5606 (0.0149) Canada -0.7139 (0.0001) -1.2436 (0.0001) -0.6265 (0.0001) Netherlands 0.1136 (0.0930) 0.0790 (0.4537) 0.3119 (0.0001) Norway -0.1949 (0.1154) 0.0115 (0.9576) -0.0764 (0.5172) Austria -0.2446 (0.5609) 0.3631 (0.7063) -0.3185 (0.2297) Sweden -0.2673 (0.0742) -0.1250 (0.5810) -0.2636 (0.1238) Swiss 0.1664 (0.1160) 0.2951 (0.0698) 0.2573 (0.0420) Spain -0.1474 (0.2906) -0.1071 (0.5911) -0.3739 0.8312 (0.0070) USA -0.3224 (0.0001) -0.4548 (0.0001) -0.1920 (0.0023)

Hypothesis F stat. P-value F stat. P-value F stat. P-value

0 : i = O H γ ∈ ∀

i

AEMs 1.7388 (0.1223) 3.8121 (0.0020) 1.3083 (0.2578) 0 : i = Ho γ ∈ ∀

i

OECD17 8.4222 (0.0001) 8.4470 (0.0001) 8.6883 (0.0001) 0 : i = Ho γ ∈ ∀

i

AEMs and OECD17 6.9253 (0.0001) 7.4539 (0.0001) 7.0271 (0.0001)

(15)

Applying the test to the first sub-sample shows that the null hypothesis of no exposure is rejected for either set of markets. For the AEMs, the exposure coefficient of Taiwan is highly significantly with a p-value of 0.0001. For the developed markets, six countries (Australia, Denmark, U.K., Japan, Canada, and the US) have significantly negative exposure coefficients.

For the second sub-period starting from 1991, however, the results show that the null hypothesis of no exposure for the AEMs cannot be rejected with a p-value of 0.2578. The exposure coefficient of Taiwan becomes insignificant, and none of the five Asian emerging markets has a significant exposure coefficient. The results are nicely consistent with the result in the last section which shows that the AEMs become integrated with the world markets.

The Taiwanese market has been considered one of the most restrictive emerging markets in the 1980's because foreign institutional investors must meet tough registration requirements in order to invest in Taiwanese market “Unqualified” institutional investors and individuals may only invest in Taiwanese market via government-approved mutual funds. As the Taiwan government has adopted several liberalization and internationalization actions since the 90's, our empirical results seem to have conformed to the changes in Taiwan's market during the past decade.

For the developed markets over the subperiod starting from 1991, however, we still reject the null hypothesis of no eexposure. The results are also consistent with the results of efficiency test that the developed markets are not integrated with the world markets. Interestingly, though, there are six countries (U.K., Italy, Japan, Canada, Spain, and the US.) that have significant negative exposures, and three countries (Belgium, Germany, and Swiss) exhibit significant positive exposure.

Overall we found exchange rate exposures for the AEMs for the period from 1985 to 1990, during which the AEMs are not integrated with the world market. After 1990, no exchange rate exposures are found for the AEMs, and the AEMs become fully integrated with the world capital markets. On the other hand, we found significant exposures (mostly negative) for the developed markets, and the developed markets are not fully integrated with the world capital markets, especially after 1990.

3.5 Conclusion

This paper investigates the integration relationship between the Asian emerging capital markets and 17 developed markets in the context of an international asset-pricing model for the period from 1985 to 1996. The results show that overall the hypothesis of global integration is rejected. A further investigation of the samples reveals that, however, the rejection of the hypothesis can be partly attributed to the segregation of the AEMs from the rest of the world during the 80's, during which we also found significant exchange rate exposures for the AEMs. After 1990, though, the AEMs seem to have been well integrated with the world capital markets, and no exchange rate exposure is found.

It should be noted, however, that the two models used in this paper are not either complements or substitutes to each other. Rejection of the integration relationship does not necessarily imply that the exchange rate risk will be priced. Although we found that rejection of the international CAPM seems to be attributable to exchange rate exposures, we have not tested if the exposure is priced

(16)

in the context of a two-factor model. One reason we cannot test the pricing of exchange rate exposures in the form of, say, Jorion (1991) is that here the factor premiums may differ from country to country. Hence, an international version of the two-factor pricing model, with the inclusion of a proper exchange-rate related factor, should be a topic that merits further research.

References

Adler, M & B. Dumas, 1983, International portfolio selection and corporation finance: A synthesis, Journal of Finance,38: 925-984.

Andrews, D. W. K., 1991, Heteroscedasticity and autocorrelation consistent covariance matrix, Economctrica, 59: 817-858.

Bailey, W., & R. M. Stulz, 1990, Benefits of international diversification: The case of Pacific-basin stock markets, Journal of Portfolio Management: 57-61. Bailey, W., R. M. Stulz & S. Yen, 1990, Properties of daily stock returns from the

Pacific basin stock markets: Evidence and implications. in: S. G. Rhee and R. P. Chang, eds, Pacific-Basin Capital Markets Research, 1: 155-171.

Bakaert, G., & C. R. Harvey, 1995, Time-varying world market integration, Journal of Finance, 50: 403-444.

Bartov, E., G. M. Bodnar & A. Kaul, 1996, Exchange Rate Variability and the Riskiness of U.S. Multinational Firms : Evidence from the Breakdown of the Bretton Woods System, Journal of Financial Economics, 42:105-132.

Brown, S. J. & T. Otsuki, 1993, Risk premia in Pacific-basin capital markets, working paper, New York University, USA.

Cambell, J. Y. & Y. Hamao, 1992, Predictable stock returns in the United States and Japan: A study of long-term market integration,Journal of Finance, 47: 43-69. Chan, K. C., A. Karolyi & R. M. Stulz, 1993, Global financial markets and the risk

premium on U. S. equity, Journal of Financial Economics, 32: 137-168.

Cheung, Y-W, J. He & L. Ng, 1994, Pacific-basin stock markets and real activity, Pacific-Basin Finance Journal, 2: 349-373.

Cho, D. C., S. Eun, & L. Senbet, 1986, International asset pricing theory: An empirical investigation, Journal of Finance, 41: 313-330.

Chou, P. H. & Guofu Zhou, 1996, Using bootstrap to test mean-variance efficiency of a given portfolio, working paper, Department of Finance, National Central University, Taiwan.

Chow, E. H. , W. Y. Lee & M. Solt, 1997, The Exchange Rate Risk Exposure of Asset Returns, Journal of Business, 70: 105-123.

Cumby, R. E. & J. D. Glen, 1990, Evaluating the performance of international mutual funds,

Journal of Finance, 45: 497-522.

Efron, B., 1979, Bootstrap methods: Another look at the jacknife, Annals of Statistics, 7: 1-26.

Ferson, W. F. & C. R. Harvey, 1994, Sources of risk and expected returns in global equity markets, Journal of Banking and Finance, 18: 775-803.

Gibbons, Michael, Stephen A. Ross, & Jay Shanken. 1989, Testing the efficiency of a given portfolio, Econometrica, 57: 1121-1152.

Greene, W., 1993, Econometric Analysis, New York: Mcmillan.

(17)

capital segmentation: The evidence from the Japanese and American stock markets, Journal of Finance, 44: 849-869.

Hamao, Y., R. Masulis, & V. Ng, 1990, Correlations in price changes and volatility across international atock markets, Review of Financial Studies, 3: 281-308. He, Jia, Lilian K. Ng, & Xueping Wu, 1995, Foreign exchange exposure, risk, and

the Japan market, working paper, City University of Hong Kong, Hong Kong. Hietala, P., 1989, Asset pricing in partially segmented markets: Evidence from the

Finnish markets, Journal of Finance, 44: 697-718.

Harvey, C. R., 1991, The world price of covariance risk,Journal of Finance, 46: 111-157.

Harvey, C. R., 1995, Predictable risk and returns in emerging markets, Review of Financial Studies, 8: 773-816.

Harvey, C. R. & G. Zhou, 1990, Bayesian inference in asset pricing tests, Journal of Financial Economics, 26: 221-254.

Harvey, C. R. & G. Zhou, 1993, International asset pricing with alternative distributional specifications, Journal of Empirical Finance, 1: 107-131.

Hawawini, G., 1994, Equity price behavior: Some evidence from markets around the world, Journal of Banking and Finance, 18: 603-620.

Ingersoll, J. E. Jr., 1987, Theory of financial decision making,. Maryland: Rowman & Littlefield.

Jorion, P., 1991, The pricing of exchange rate risk in the stock market, Journal of Financial and Quantitative Analysis, 26: 363-376.

Jorion, P., 1990, The exchange-rate exposure of U.S. multinationals, Journal of Business, 63: 331-345.

Lintner, J., 1965, The valuation of risk assets and the securities if risky investments in stock portfolios and capital budgets,Review of Economics and Statistics, 47: 13-37.

Korajczyk, R. & C. Viallet, 1989, An empirical investigation of international asset pricing, Review of Financial Studies, 2: 553-585.

MacKinlay, A. Craig, 1987, On multivariate tests of the CAPM, Journal of Financial Economics, 18: 341-371.

MacKinlay, A. Craig & Matthew P. Richardson, 1991, Using generalized method of moments to test mean-variance efficiency, Journal of Finance, 46: 511-527. Richardson, Matthew, and Tom Smith, 1993, A test for multivariate normality in

stock returns. Journal of Business, 66: 295-321.

Ross, S. A., 1976, The arbitrage theory of capital asset pricing, Journal of Economic Theory, 13: 341-360.

Sharpe, W., 1964, Capital asset prices: A theory of market equilibrium under conditions of risk, Journal of Finance, 19: 425-442.

Solnik, B., 1977, Testing international asset pricing: Some pessimistic views, Journal of Finance, 32: 500-524.

Stulz, R., 1981, A model of international asset pricing, Journal of Financial Economics, 9: 383-406.

Wheatley, S., 1988, Some tests of international equity integration, Journal of Financial Economics, 21: 177-122.

Zhou, Guofu, 1993, Asset pricing tests under alternative distributions. Journal of Finance, 48: 1925-1942.

(18)

亞洲新興市場的整合與匯率風險暴露

周賓凰

國立中央大學財務金融學系教授

周行一

國立政治大學財務管理學系教授

史綱

國立中央大學財務金融學系教授

摘要

本文以 Sharpe-Lintner CAPM 模型為架構,研究 1985-1996 年期間亞洲新 興股市與開發國家股市的整合關係。整體而言,我們發現在引進亞洲新興市場 的考量後,世界資本市場並不存在一整合的關係,而此非整合關係可部分歸諸 於亞洲新興市場在 1980 年代的匯率風險。此外,我們亦發現亞洲新興股市在 1991 年後變得與世界存在整合關係,進一步分析發現亞洲股市在這一段期間在 考慮系統風險後,並無顯著的匯率風險暴露。不過,我們發現,亞洲新興股市 與世界其他股市的相關性仍相當低。 關鍵字: 資本資產定價理論、一般化動差法、匯率風險、拔靴複製法

數據

Table 1: Summar y Statistics
Table 2: Regr ession Results
Table 3: Test of Por tfolio Efficiency and Integr ation Relationship
Table  4:  Exchange  Risk  Exposur e  of  the  Asian  Emer ging  Mar kets  vs.  the  Developed Mar kets

參考文獻

相關文件

• The  ArrayList class is an example of a  collection class. • Starting with version 5.0, Java has added a  new kind of for loop called a for each

 Promote project learning, mathematical modeling, and problem-based learning to strengthen the ability to integrate and apply knowledge and skills, and make. calculated

Wang, Solving pseudomonotone variational inequalities and pseudocon- vex optimization problems using the projection neural network, IEEE Transactions on Neural Networks 17

Define instead the imaginary.. potential, magnetic field, lattice…) Dirac-BdG Hamiltonian:. with small, and matrix

Monopolies in synchronous distributed systems (Peleg 1998; Peleg

Based on the reformulation, a semi-smooth Levenberg–Marquardt method was developed, and the superlinear (quadratic) rate of convergence was established under the strict

Corollary 13.3. For, if C is simple and lies in D, the function f is analytic at each point interior to and on C; so we apply the Cauchy-Goursat theorem directly. On the other hand,

Corollary 13.3. For, if C is simple and lies in D, the function f is analytic at each point interior to and on C; so we apply the Cauchy-Goursat theorem directly. On the other hand,