Are Developing Markets the Same as Developed Markets? Lu Liu∗ 2013 June

Abstract This paper uses a dynamic panel-data gravity model to explain the correlations between 40 markets from 1996 to 2010 using four types of market linkages: information capacity, nancial integration, economic integration, and similarity in industrial structure. The mechanism of interdependence of developed markets and that of developing markets are heterogeneous: 1)information capacity and industrial structure similarity have signicant impact on the correlations of a developed market with other markets; 2)economic integration drives the correlations of a developing market with other markets; 3)nancial integration is important for interdependence among developed markets and that among developing markets, but not for that between developed and developing markets. The EMU has a signicant positive impact on stock market integration from 1996 to 2002. This impact increases after the inauguration of the EMU in 1999 but does not increase further after the monetary transition being accomplished in 2002.

Keywords:

stock market interdependence; market linkages; heterogeneity;

gravity model; dynamic panel-data; information capacity

JEL Classication:

G15; C33

Department of Economics, Lund University, P.O. Box 7082, S-220 07 Lund, Sweden. Email: [email protected]; Tel.: +46 46 222 4290. The Bankforskningsinstitutet is gratefully acknowledged for funding this research. ∗

1

1

Introduction

Global stock markets are undergoing ever-increasing integration.

It is essential

for investors who pursue international diversication strategies to understand the driving forces behind stock market interdependence in order to evaluate the potential benets and risks of diversication.

This is also important for policy

makers that aim at stabilizing nancial system and reducing nancial contagion. There is an important literature documenting stock market synchronization being driven by linkages between economies, but there is hardly a consensus among economists over the importance of these linkages. For instance, Wälti (2011) nds that monetary integration leads to stronger stock market synchronization, but Roll (1992) nds similarity in industrial structure to be the most important driving factor.

Forbes and Chinn (2004) nd bilateral trade to be the primary channel

through which the largest nancial markets aect other markets, while Flavin et al. (2002) argue for the impact of geographic location. The diversity of these conclusions, as stated by Beine and Candelon (2011), may be attributable to the heterogeneous characteristics of the markets. The present paper studies the impact of several bilateral market linkages on the pairwise correlations between the returns of 40 national stock market indexes. It pays attention to the overall magnitudes of the impacts, as well as to their heterogeneity across markets. The investigated linkages are information capacity, economic integration, nancial integration, and industrial similarity. Specically, it classies the pairwise market correlations into three groups:

among developed

markets, among developing markets, and between developing & developed markets. And thereby one can distinguish the impacts of the linkages on dierent groups of

2

stock market correlations. The present paper attempts to answer three questions. First, which linkages drive stock market comovement?

Second, how do the

mechanisms of interdependence (i.e., the eects of market linkages on stock market correlations) dier in dierent groups of markets?

For example, does economic

integration drive the correlations among developed markets to the same extent as it drives the correlations among developing markets? Third, when the selected bilateral linkages are taken into account, does the implementation of the European Economic and Monetary Union (EMU) still matter for stock market integration? This paper relies on the gravity model with a dynamic panel specication. The gravity model in economics, mimicking the gravitational interaction in Newton's law of gravity, explains the relationship between two economies based on their masses and the distance (or closeness) between them. The gravity model is widely used in the empirical study of international trade, and has become popular in the study of capital market synchronization (see Flavin et al., 2002, Beine et al., 2010, and Beine and Candelon, 2011 for example). In our study, the relationship between two economies is the correlation between their primary national stock markets, while the sizes of the markets are regarded as the masses, and the cross-market linkages, as the distances. One merit of the gravity model approach, particularly for our study, is its exibility in describing cross-market heterogeneity. This paper diers from the existing literature in several respects.

First, it

provides a comprehensive view of a large sample of national stock markets while distinguishing the impact of the linkages with respect to specic types of markets. There exist studies of the interdependence of developing or developed markets only (see for example Pretorius, 2002, and Beine and Candelon, 2011), while others investigate a combination of developed and developing countries without allowing

3

for potential heterogeneity. However, it is implausible that the stock markets in developed economies are linked via the same mechanism as are developing markets. Therefore, investors should implement dierent strategies of diversication in developing markets from those in developed markets. The present paper provides new insight for understanding stock market interdependence. Second, this paper addresses the important role of information capacity in explaining stock market comovement. According to Sims (2006), the information capacity of a country includes its wiring capacity and internal human capacity. Wiring capacity refers to the availability of communication technologies that allow investors to access information, whereas internal human capacity refers to investor's capability and eciency of using the information.

Stock market

correlations are expected to increase as information capacity increases, since a large information capacity implies easier access to information, which in turn reduces information asymmetries and fosters cross-country investment in equities. Moreover, a market with a large information capacity may have a fast information diusion process, and therefore may respond faster to external shocks, whereas a market with a very small information capacity tends to be isolated from other markets. Third, this paper adds to the literature that studies the EMU eect on stock market integration.

The existing literature (e.g., Yang et al., 2003) nds

that the EMU has signicantly strengthened stock market integration among its member countries; however, the increase in integration may also be attributable to other factors such as larger volumes of bilateral trade and faster information transmission.

The present paper examines whether joint EMU participation

matters after controlling for the selected bilateral linkages dened above and how

4

the EMU eect on stock market integration changes along the implementation of EMU. Fourth, our dynamic panel specication is able to capture the dynamics of stock market correlations, which is ignored by the existing related studies (e.g., Wälti, 2011, and Flavin et al., 2002). Empirical evidence (see Kim et al., 2005 for example) shows that stock market integration is a persistent process and one of its main determinant is the existing level of integration. Estimation of the impacts of bilateral factors may thus be spurious if autocorrelation is not taken into account. The remainder of this paper is organized as follows.

Section 2 presents the

dynamic panel gravity model. Section 3 presents the selected variables and data. Section 4 contains the estimation results and Section 5 concludes.

2

Econometric modelling

The econometric model is a gravity model with dynamic panel specication allowing for both cross-section xed eects and time-specic xed eects.

To

answer the research questions proposed in Section 1, I sequentially design four econometric specications. The baseline regression, namely specication I, is

ρij,t = µ + γρij,t−1 + λSizeij,t + Xij,t β +

PT

t=2 δt dt

+ uij,t , (1)

uij,t = ηij + ij,t , (i × j) ∈ (1, ...N )2 , i < j, t = 1, . . . , T, where

t.

ρij,t

is the daily returns correlation between stock market

The lagged dependent variable

ρi,t−1

i

and

j

in year

captures the potential dynamics of stock

5

market correlation. Sizeij,t is the joint mass of markets

i

and

j,

which is dened

as the sum of the logarithms of the market capitalizations in market year

t. X ij,t

is a

1×K

i

and

j

in

vector consisting of the variables that describe the market

linkages, such as information capacity, economic integration, nancial integration, and industrial similarity. Section 3.2.

dt

The measurement of these variables is described in

is the time dummy variable for year

xed eect and is independent of

ij,t . N

our study. The number of time periods,

t. ηij

is the cross-section

is the number of markets, which is 40 in

T,

is 16 (years).

Unobserved heterogeneity in the cross-section and time-series dimensions is controlled for by the xed eect

ηij

and the time dummy variable

dt .

The cross-

section xed eect makes the model specication parsimonious, because it is able to account for unincluded time invariant variables (e.g., geographical distance) that may be correlated with the included variables.

Likewise, the time dummy

variable is able to capture the impact of common shocks.

This is particularly

important for our study, because our sample contains several international crisis periods such as the Asian nancial crisis from 1997 to 1998 and the global nancial crisis beginning in 2007 when common shocks prevailed internationally. Although unobserved heterogeneity is controlled for by using xed eects, the baseline regression I assumes the inuences of the bilateral linkages to be homogeneous on all pairwise market correlations, which maybe implausible because of the dierent characteristics of these markets. Therefore, I extend the baseline specication I to specication II, in regression (2), in order to examine the interdependence mechanism for markets across dierent levels of economic development.

6

mh ρij,t = µ + γρij,t−1 + λSizeij,t + X ij,t β 0 + dmm ij X ij,t β 1 + dij X ij,t β 2 P + Tt=2 δt dt + ηij + ij,t , where

dmm ij

Likewise,

is 1 if both markets

dmh ij

is 1 if either

the impact of factor

i and j

i or j

Xij,t ,

(2)

are developing markets, and is 0 otherwise.

is a developing market, and is 0 otherwise. Hence,

respectively, on the correlations between developed

markets, on those between developing markets, and on those between developed and developing markets, is

β0 , β0 + β1 ,

and

β0 + β2 .

Furthermore, in order to examine the eect of the EMU on stock market integration, two more specications are designed. The following regression, namely specication III, extends specication I by introducing a dummy variable for pairs of countries that are simultaneously EMU members.

U + ρij,t = µ + γρij,t−1 + λSizeij,t + X ij,t β + φdEM ij where

U dEM ij

is one if both markets

i

and

j

PT

t=2 δt dt

+ ηij + ij,t ,

(3)

are members of the EMU, and zero

otherwise (Section 3.2 denes two other dummy variables for EMU countries in regard to dierent stages of the implementation of the EMU). If

φ

is positive,

we can say that the degree of interdependence among EMU stock markets is larger than can be explained by the linkages between markets when assuming the interdependence mechanism to be homogeneous across markets. The most extended specication, IV, examines the EMU eect while taking into account the heterogeneity across developing and developed countries:

mh ρij,t = µ + γρij,t−1 + λSizeij,t + X ij,t β 0 + dmm ij X ij,t β 1 + dij X ij,t β 2 P U +φdEM + Tt=2 δt dt + ηij + ij,t . ij Since EMU countries are a subset of the developed countries, a positive

7

(4)

φ

in

specication IV indicates that stock market correlations in EMU are more than can be explained by the interdependence mechanism of developed markets. If

φ loses its

scale and statistical signicance, integration among EMU stock markets is merely attributable to market linkages via the same mechanism as the interdependence of other developed markets. That is to say, there is no pure EMU eect. To estimate the dynamic panel regressions (1-4), I use the general method of moments (GMM). For the dataset includes 40 markets, there are 780 pairwise correlations in total. This means that the panel dataset has a very large number of cross-sections (780) and a relatively small number of time periods (16).

It

is well-known that the xed eects estimator is biased and inconsistent for such dynamic panels (see Nickell, 1981).

The random eects GLS estimator is also

biased (see Baltagi, 2008). I therefore estimate the model with Arellano and Bover (1995)/Blundell and Bond (1998) GMM, which is designed for dynamic panels with a large number of cross sections and a small number of time periods. The Arellano and Bover (1995)/Blundell and Bond (1998) estimation is based on the study of Arellano and Bond (1991), which carries out a rst dierence transformation and uses GMM. Arellano and Bover (1995) and Blundell and Bond (1998) augment the Arellano and Bond (1991) estimator by assuming that the rst dierences of the instrumenting variables are uncorrelated with the xed eects.

This allows

the introduction of more instruments and dramatically improves the eciency (see Roodman, 2006). Another merit of the ArellanoBover/BlundellBond GMM estimation is the ability to estimate the coecients of the time invariant variables by introducing more moment equations. This is particularly useful when it comes to examine the EMU eect as a dummy variable for the EMU countries (dened in Section 3.2.6) varies over cross-sections but not over time.

8

3

Selected variables and data

This section presents the selected variables and data sources, focusing on the expected impacts of the bilateral linkages and the joint EMU participation.

3.1 National market indexes and correlations Our sample of interest contains the main market indexes of 40 economies.

The

indexes are extracted from the MSCI for the period from the beginning of 1995 to the end of 2010, denominated in U.S. dollars. Following Flavin et al. (2002), cross-country stock market interdependence is measured as the correlation of the daily logarithmic returns of the national market indexes in each year.

1

[Table 1 about here.]

As there is no convention for the classication of developed countries and developing countries, income level is used as a proxy for the level of economic development, and so high income countries are designated as developed economies and middle income countries, as developing economies. The sample consists of 28 high income countries and 12 middle income countries according to the World

2

Bank

(see Table 1).

Eleven countries, marked with

∗

, among the developed

countries are members of the EMU. The pairwise correlations of markets are categorized into three groups:

those among developed markets, those among

developing markets, and those between developed & developing markets.

1 Wälti

(2011) transforms the correlations by Fisher's z transformation. This transformed dependent variable yields the same statistically signicance for each independent variable as using the untransformed dependent variable in the present paper. For ease of interpretation, the untransformed correlation coecient is retained. 2 See http://data.worldbank.org/about/country-classications/country-and-lending-groups.

9

3.2 Selected determinants of stock market interdependence 3.2.1 Size The size of a stock market is measured as the logarithm of the total market capitalization of its listed stocks.

The correlations between markets with large

capitalization are expected to be large, as international stock markets are sensitive to shocks coming from large markets. The data for the market capitalization is collected from the World Bank's World Development Indicators Database. total size of two markets

Sizeij,t

i

and

j

The

is

= log (MarkCapi,t ) + log (MarkCapj,t ).

(5)

3.2.2 Information capacity To measure information capacity, wiring capacity is used since its measurement is less subjective than that of internal human capacity, following Mondria and Wu (2010). To measure a country's information capacity, I use three statistics from the World Bank's World Development Indicators' database: the number of telephone lines per 100 persons, the number of mobile cellular subscriptions per 100 persons, and the number of internet users per 100 persons.

The variable of information

capacity is calculated as the rst principal of the three statistics. The degree of cross-country information linkage between two markets is measured as the sum of their information capacity. Information capacity is expected to have a positive impact on stock market correlations. First, as illuminated by Portes and Rey (2005), market segmentation is mainly attributable to information asymmetries. A high information capacity provides investors with easier access to information and more advanced tools for

10

analyzing the information, and may therefore lead to less information asymmetries between domestic and foreign investors and hence less market segmentation. Second, a high information capacity may also foster information diusion between markets.

Markets with a high information capacity tend to react promptly

to external shocks and hence comove more with other markets.

The existing

literature, for example Ivkovich and Weisbenner (2007), has found the presence of an information diusion eect on investment behavior.

Furthermore, the

development of information capacity advances the form of stock trading services, noticeably reduces the transaction cost, and therefore encourages cross-border trading.

3.2.3 Financial integration In this study, nancial integration is measured by the Chinn-Ito nancial openness index introduced by Chinn and Ito (2006).

The index measures a country's

degree of capital account openness, based on the information of controls on crossborder nancial transactions reported in the IMF's Annual Report on Exchange Arrangements and Exchange Restrictions.

The evolution of capital account

openness facilitates cross-border equity holdings, and hence is expected to increase stock market integration. I measure the level of nancial integration between two countries as the sum of the values of their respective Chinn-Ito indexes.

3.2.4 Economic integration Economic integration facilitates the convergence of cash ows between countries, fosters business cycle synchronization, and hence is expected to increase stock market interdependence.

Since economic integration mainly refers to trade

11

unication, the economic integration of two countries is measured by the intensity of their bilateral trade. Following Beine and Candelon (2011), the relative trade intensity is dened as

Econij,t =

where

expij,t

from country

and

i

impij,t

expij,t + impij,t expij,t + impij,t + , expi,t + impi,t expj,t + impj,t

(6)

are, respectively, the values of the exports and imports

to country

and imports of country

i,

trade between countries

i

j . expi,t

and

impi,t

and similarly for and

j

are the values of the total exports

j.

Hence

Econij,t

is the intensity of

relative to their total value of trade.

The data for bilateral trade is taken from the STAN Bilateral Trade Database (source: OECD), which contains the values of annual imports and exports of goods for all countries in the sample. All values are in U.S. dollars at current prices.

3.2.5 Dissimilarity in industrial exposure Stock markets with similar industrial structure may be highly correlated, as they are more likely to be driven by the same global industry-specic factors. Following Asgharian and Bengtsson (2006), the risk exposures (betas) of national stock markets to the Datastream world industry indexes will be a proxy for industrial dissimilarity: the industrial dissimilarity of two markets is dened as the average absolute cross-market dierence in these betas.

3.2.6 Economic and Monetary Union The introduction of a common currency (i.e., euro) and a single monetary policy has standardized the pricing of nancial assets, reduced investors' transaction

12

costs, and thus eased cross-border investment among the EMU countries.

The

expected ease of cross-border investment might have increased the EMU stock market integration before the EMU was inaugurated in the start of 1999. Moreover, the EMU stock market integration might have further increased since the monetary

3

transition became fully accomplished in the start of 2002 .

Therefore,

the

eect of EMU on stock market integration is analyzed at dierent stages of the implementation of the EMU, by employing the following dummy variables in the econometric specications III and IV.

U dEM ij

1 = 0

EM U (1999−)

1 = 0

EM U (2002−)

1 = 0

dij,t

dij,t

As dened in equation (7),

if

i ∈ EMU

and

j ∈ EMU

for any

t, (7)

otherwise.

if

i ∈ EMU

and

j ∈ EMU

for

t ≥ year 1999, (8)

otherwise.

if

i ∈ EMU

and

j ∈ EMU

for

t ≥ year 2002, (9)

otherwise.

U dEM ij

is a time-invariant dummy variable for stock

market correlations between the countries that joined in the EMU at any time

t

4

in the sample period . Dierently, the dummy variables dened in equations (8) and (9) vary both over cross-sections and over time.

If all these three dummy

variables are included in specication III or IV, the coecient for

U dEM ij

describes

the eect of the EMU on stock market integration already before the inauguration of the EMU; a positive coecient for

EM U (1999−)

dij,t

3 On

indicates an increase in the EMU

1 January 2002, the euro notes and coins are introduced. joined in the EMU on January 1, 2001, whereas other EMU countries participated from January 1, 1999. 4 Greece

13

eect after the inauguration of the EMU; and a positive coecient for

EM U (2002−)

dij,t

indicates a further increase in the EMU eect after the completion of monetary transition.

4

Empirical Analysis

This section analyzes the interdependence among the selected markets over the sample period from 1996 to 2010. I begin with analyzing which linkages matter for stock market comovement and how the mechanisms of interdependence dier in dierent groups of markets. Then I examine whether the EMU increases stock market integration at dierent stages of the implementation of the EMU when controlling for the selected linkages. Finally, I check the sensitivity of the results to the large economies.

4.1 The eects of market linkages and the heterogeneity This section implements an empirical analysis based on various econometric model specications (described in Section 2) for the sample period from 1996 to 2010. In the GMM estimation, three lags of the dependent variable are adopted as instruments and robust standard errors are used.

In addition, I will take into

account potential endogenous independent variables. Bilateral trade and capital account openness are likely to be endogenous, as they may be inuenced by shocks (e.g., political risk and currency crises) that may aect at the same time stock market returns. Industrial dissimilarity is also potentially endogenous, as its proxy is estimated from the returns of stock market indexes. Two lags of these regressors are used as instruments.

The issue of potential endogeneity in a panel gravity

14

model is also tackled in Wälti (2011).

He regresses each endogenous variable

on dierent sets of exogenous variables and then uses the predicted values of endogenous variables in the main estimation.

Wälti (2011)'s method, however,

is subject to estimation errors. I commence by showing the importance of controlling for the common trend in the sample, which includes several international nancial crisis periods.

The

rst and the second columns in Table 2 present the estimates for the baseline specication I without and with time xed eects respectively. Comparing these two columns, we can see that the sign of the coecient diers for economic integration, nancial integration, and industrial dissimilarity.

This is because

ignoring the common trend in the data biases the estimates of the causal eects of the market linkages.

As shown in Table 3, the yearly-specic xed eect in

specication I is highly signicant for most years in the sample period. Specically, stock market correlations increase signicantly in the 1997-1998 Asian nancial crisis period and in the recent global nancial crisis period, while bilateral linkages are being controlled for.

[Table 2 about here.]

[Table 3 about here.]

When time eects are controlled for by using a dummy variable for each year, the coecients for all the selected linkages become statistically signicant. Consistent with the expectation, information capacity, economic integration, and capital account openness have positive impact on stock market interdependence, and less industrial dissimilarity is associated with higher stock market interdependence.

15

Next, I examine the existence of heterogeneity in the linkage eect by using specication II (displayed in the last column of Table 2), which regards the eect on the interdependence among developed markets as the benchmark. The t statistics of

β0

in specication II show that interdependence among developed markets

can by explained by information capacity, nancial integration, and industrial dissimilarity but not by economic integration.

The coecient associated with

the interdependence between developing markets,

β1 ,

is statistically signicant

for information capacity, economic integration, and industrial dissimilarity; the coecient for the interdependence between developing and developed markets,

β2 ,

is

signicant

for

nancial

integration

and

industrial

similarity.

indicate that the importance of linkage diers between markets.

These

Specically,

the eects of information capacity and industrial structure are more substantial for developed markets.

The impact of economic integration is much larger on

the interdependence between developing markets.

Financial integration appears

to be more important for the interdependence between developed markets and that between developing markets than for that between developed and developing markets. To show the magnitude and statistical signicance of the linkage eects across markets, I show the marginal eects of the linkages on the three groups of market interdependence, i.e.,

β0 , β0 +β1 , and β0 +β2 , in Table 4.

The statistical signicance

of the marginal eects being dierent between the groups indicates that the market interdependence for developed markets and that for developing markets are not driven by the same linkages.

Specically, for every unit increase in information

capacity, the correlation between two developed markets and the correlation between a developed market and a developing market will increase signicantly,

16

by 0.83 percentage points and 0.73 percentage points respectively, contrast to an insignicant increase in the correlation between developing markets.

Similarly,

the impact of industrial dissimilarity is insignicant or small on the correlations between developing markets and other markets, whereas the correlation among two developed markets will increase by 11.3 percentage points for every unit decrease in industrial dissimilarity. This implies that industry factors do not account for the comovement of developing markets as they do for the comovement of developed markets.

The benet of international diversication across developing markets

thus does not come from the degree of dierence in their industrial structures. In contrast, developing markets are strongly linked by real economic activities though not by information capacity or industrial similarity.

For a unit increase

in bilateral trade intensity, the correlation between two developing markets will increase by 69.9 percentage points, which is more than twofold of the increase in the other two groups of market correlations.

Moreover, nancial integration

matters for the correlations among developed markets and those between developed and developing markets.

A unit increase in nancial integration will raise the

correlations of these two groups by 2.24 percentage points and 1.61 percentage points respectively, while having insignicant impact on the correlations between developed and developing markets.

[Table 4 about here.]

In addition, specications I and II (with time eects) correctly capture the dynamics of the stock market correlations as the ArellanoBond test does not reject the hypothesis of no second-order serial correlation in the rst-dierenced errors. The eect of the one-year lagged dependent variable remains positive and

17

statistically signicant across all the specications, which conforms to the nding of Kim et al. (2005) that stock market integration is persistent and that one of its main determinants is the existing level of integration. In addition, the eect of market size is positive, implying that markets with a large capitalization are more inuential on other markets.

4.2 The EMU eect This section examines whether joint EMU membership matters for stock market interdependence.

By using specications III and IV that control for bilateral

linkages (and their heterogeneous impacts), I investigate whether stock market interdependence in the EMU is higher than can be explained by the predened linkages across markets. Particularly, I examine whether the EMU inauguration and monetary transition have impact on the EMU stock market interdependence, by using dierent subsets of dummy variables for the joint EMU membership (dened in equations (7-9) in Section 3.2.6) in specications III and IV. First,

I investigate the impact of the EMU before and after the EMU

inauguration in 1999. As argued previously in the paper, the expected ease of crossborder investment may have increased the EMU stock market interdependence already before the EMU inauguration. Therefore, I examine the impact of joint EMU membership both before and after the EMU inauguration by using the dummy variables coecient for

U dEM ij

U dEM ij

and

EM U (1999−)

dij

dened in equations (7) and (8).

The

in the rst column of Table 5 shows that the correlations

among the EMU markets on average are 28 percentage points higher than the correlations of other kinds before the EMU inauguration, while the bilateral

18

linkages are being controlled for. This impact does not seem to dier a lot when the heterogeneous eects of the bilateral linkages are also controlled. This suggests the existence of a pure EMU eect that is not explained by the predened market linkages. Moreover, the signicantly positive coecient for

EM U (1999−)

dij,t

indicates

that the EMU joint membership further raises pairwise correlations among the EMU markets by about 9 percentage points after the EMU inauguration. Furthermore, I examine whether the EMU stock markets become even more integrated after the monetary transition being fully accomplished in the start of 2002, by adopting another dummy variable for the years from 2002.

EM U (2002−)

dij,t

for joint EMU membership

The third column of Table 5 shows that, in addition

to the increase induced by the expectation and by the EMU inauguration, the correlations between EMU markets on average increase by another 3.6 percentage points after monetary transition. However, according to the last column of Table 5, the increase in correlation after the monetary transition becomes much smaller and statistically insignicant when the heterogeneity between developed markets and developing markets is taken into consideration. This suggests that the increase of the EMU market interdependence after the monetary transition can be attributable the strengthening of the bilateral linkages.

[Table 5 about here.]

4.3 Sensitivity analysis To examine the sensitivity of the ndings to the selection of markets, I exclude the U.S market and the Chinese market, which are the giants respectively from developed markets and from developing markets, from our sample, and check

19

the robustness of the estimates of specications IIV. The signs and scales of the linkage coecients and the cross-market heterogeneous coecients(available upon request) are similar to those in the estimation results shown in Table 2. Furthermore, similar to the results shown in Table 5, there is pure positive EMU eect on stock market interdependence.

The EMU eect increases after

the inauguration of the EMU, but not after the monetary transition.

5

Conclusion

This paper uses a dynamic panel gravity model to explain the mechanism of interdependence between 40 national stock markets using four cross-market linkages:

information capacity, nancial integration, economic integration, and

similarity in industrial structure.

The overall magnitudes of the linkage eects

were analyzed, as well as their heterogeneity across markets. contribute to the overall stock market correlations.

All the linkages

However, the mechanism of

the interdependence between developed markets diers from that of developing markets.

Specically, information capacity and industrial structure similarity

have signicant impact on the correlations of a developed market with other markets, whereas the correlations of a developing markets with others is driven by economic integration. Financial integration is important for interdependence among developed markets and that among developing markets, but not for that between developed and developing markets. While allowing for heterogeneous mechanisms of interdependence, the eect of joining the European Economic and Monetary Union (EMU) on stock market integration is examined.

EMU stock market integration cannot be entirely

20

explained by the same interdependence mechanisms as with other developed markets, as the impact of EMU exists before the inauguration of EMU and further increases after that while bilateral linkages being controlled for.

This

implies that there is a pure positive eect of common EMU membership on stock market interdependence. The increase in correlations between stock markets of the EMU member countries are not only attributable to factors such as faster information transmission, larger industrial similarity and greater nancial integration.

However, the impact of the EMU does not seem to increase after

monetary transition.

References Arellano, M. and Bond, S. (1991).

Some tests of specication for panel data:

Monte Carlo evidence and an application to employment equations.

The Review

of Economic Studies, 58(2):277297. Arellano, M. and Bover, O. (1995).

Another look at the instrumental variable

estimation of error-components models.

Journal of Econometrics, 68(1):29 51.

Asgharian, H. and Bengtsson, C. (2006). Jump spillover in international equity markets.

Journal of Financial Econometrics, 4(2):167203.

Baltagi, B. H. (2008).

Econometric Analysis of Panel Data.

John Wiley and Sons

Ltd, 4th edition.

Beine, M. and Candelon, B. (2011). Liberalisation and stock market co-movement between emerging economies.

Quantitative Finance, 11(2):299312.

21

Beine, M., Cosma, A., and Vermeulen, R. (2010). integration:

The dark side of global

Journal of Banking and Finance,

Increasing tail dependence.

34(1):184 192.

Blundell, R. and Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models.

Journal of Econometrics, 87(1):115 143.

Chinn, M. and Ito, H. (2006). What matters for nancial development? capital controls, institutions, and interactions.

Journal of Development Economics,

81(1):163192.

Flavin, T. J., Hurley, M. J., and Rousseau, F. (2002). Explaining stock market correlation: a gravity model approach.

The Manchester School, 70(S1):87106.

Forbes, K. J. and Chinn, M. D. (2004).

A decomposition of global linkages in

nancial markets over time.

The Review of Economics and Statistics, 86(3):705

722.

Ivkovich, Z. and Weisbenner, S. (2007). Information diusion eects in individual investors' common stock purchases: Covet thy neighbors' investment choices.

Review of Financial Studies, 20(4):13271357. Kim, S. J., Moshirian, F., and Wu, E. (2005). Dynamic stock market integration driven by the European Monetary Union: An empirical analysis.

Journal of

Banking and Finance, 29(10):2475 2502. Mondria, J. and Wu, T. (2010).

The puzzling evolution of the home bias,

information processing and nancial openness.

and Control, 34(5):875 896. 22

Journal of Economic Dynamics

Nickell, S. (1981).

Biases in dynamic models with xed eects.

Econometrica,

49(6):14171426.

Portes, R. and Rey, H. (2005).

The determinants of cross-border equity ows.

Journal of International Economics, 65(2):269 296. Pretorius,

E.

(2002).

interdependence.

Economic

determinants

of

emerging

stock

market

Emerging Markets Review, 3(1):84 105.

Roll, R. (1992). Industrial structure and the comparative behavior of international stock market indices.

Roodman, D. (2006).

Journal of Finance, 47(1):341. How to do Xtabond2: an introduction to dierence and

system GMM in Stata.

Sims, C. A. (2006).

SSRN eLibrary.

Rational inattention:

beyond the linear-quadratic case.

American Economic Review, 96(2):158163. Wälti, S. (2011). Stock market synchronization and monetary integration.

Journal

of International Money and Finance, 30(1):96110. Yang, J., Min, I., and Li, Q. (2003). EMU matter?

European stock market integration: does

Journal of business nance & accounting, 30(9-10):12531276.

23

Table 1:

Selected markets High-income economies

Middle-income economies

(Developed markets)

(Developing markets)

1

Australia

1

Argentina

2

Austria

2

Brazil

3

Belgium

3

Chile

4

Canada

4

China

5

Czech Republic

5

India

6

Denmark

6

Indonesia

7

Finland

7

Malaysia

8

France

8

Mexico

9

Germany

9

The Philippines

10

Greece

10

Russia

11

Hong Kong

11

Thailand

12

Hungary

12

Turkey

13

Ireland

14

Israel

15

Italy

16

Japan

17

Korea

18

The Netherlands

19

New Zealand

20

Norway

21

Poland

22

Portugal

23

Singapore

24

Spain

25

Sweden

26

Switzerland

27

UK

28

USA

*

*

*

*

*

*

*

*

*

*

*

Note :

The sample contains the main market indexes of 40 economies, among which 28 are high income countries and 12 are middle income countries according to the classication of the World Bank. * indicates that the country is a member of the EMU.

24

Table 2:

Main results over the sample period 1996-2010

(II) 0.370∗∗∗ (23.28) µ -0.269∗∗ (-2.26) β0 Info 0.00829∗ (1.90) Econ 0.218 (1.40) Fin 0.0224∗∗∗ (3.77) Ind -0.113∗∗∗ (-6.47) Size 0.00561∗∗ (2.24) β1 Info×dmm -0.00532∗ (-1.85) Econ×dmm 0.478∗∗∗ (2.82) mm Fin×d -0.00632 (-0.80) Ind×dmm 0.120∗∗∗ (5.26) β2 Info×dmh -0.000943 (-0.33) mh Econ×d 0.0455 (0.21) Fin×dmh -0.0202∗∗∗ (-2.89) mh Ind×d 0.0963∗∗∗ (5.25) AR(1) 0.0000 0.0000 0.0000 AR(2) 0.0000 0.2588 0.6571 Pseudo R2 0.7548 0.8521 0.8600 Note : Info, Econ, Fin, and Ind denote information capacity, economic integration, nancial integration, and industrial dissimilarity respectively. The rst column presents the results of specication I (1) without a time dummy. Specication I in the second column includes time xed eects. The third column is specication II (2), where bilateral linkages interact with the dummy dmm for correlation between two developing countries and with the dummy dmh for the correlation between a developing country and a developed country. The AR tests yield the p-values of the ArellanoBond test with the null hypotheses of no rst-order serial correlation and no second-order serial correlation in the rst-dierenced errors. The pseudo R2 is the square of the correlation between the original dependent variable and its tted value. t statistics in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

ρt−1

(I) 0.402∗∗∗ (31.10) -1.351∗∗∗ (-13.89) 0.0476∗∗∗ (19.98) -0.0615 (-0.33) -0.00135 (-0.42) 0.0750∗∗∗ (6.65) 0.0211∗∗∗ (10.42)

(I) 0.343∗∗∗ (19.64) -0.230∗ (-1.74) 0.0204∗∗∗ (4.38) 0.352∗∗ (2.15) 0.0107∗∗∗ (3.68) -0.0474∗∗∗ (-4.15) 0.00323 (1.17)

25

Table 3:

Time eects over the sample period 1996-2010

(I) year1997

(II)

0.0852

∗∗∗

(10.57) year1998

0.161

∗∗∗

(19.61) year1999

-0.0504

∗∗∗

(-6.35) year2000

0.0282

∗∗∗

(2.81) year2001

0.0303

∗∗∗

(2.78) year2002

∗∗ 0.0241 (2.07)

year2003

-0.0382

∗∗∗

(-3.35) year2004

0.123

∗∗∗

(9.58) year2005

0.0177 (1.32)

year2006

0.159

∗∗∗

(10.94) year2007

0.178

∗∗∗

(12.38) year2008

0.193

∗∗∗

(12.66) year2009

0.143

∗∗∗

(9.24) year2010

0.183

∗∗∗

(11.81)

Note :

0.0858

∗∗∗

(12.11) 0.157

∗∗∗

(20.97) -0.0459

∗∗∗

(-6.11) 0.0431

∗∗∗

(5.03) 0.0475

∗∗∗

(5.28) 0.0447

∗∗∗

(4.47) -0.0156 (-1.60) 0.147

∗∗∗

(13.32) 0.0383

∗∗∗

(3.15) 0.183

∗∗∗

(15.27) 0.195

∗∗∗

(14.78) 0.215

∗∗∗

(14.72) 0.163

∗∗∗

(10.92) 0.208

∗∗∗

(14.29)

Info, Econ, Fin, and Ind denote information capacity, economic integration, nancial integration, and industrial dissimilarity respectively. The rst column presents the coecients for the year-specic dummy variable for specication I (1). The second column presents the coecients for the year-specic dummy variable for specication II (2). t statistics in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. 26

Table 4:

The impacts of market linkages on the three groups of market

interdependence

Info Econ Fin Ind

(Developed)

(Developing)

(Develped vs Developing)

β0

β0 + β1

β0 + β2

0.0030 ∗∗ 0.696 ∗∗∗ 0.0161

∗∗ 0.0073 ∗ 0.264

0.0083

∗

0.218 ∗∗∗ 0.0224 ∗∗∗ -0.113

0.007

Note :

0.0022 ∗∗ -0.017

This table shows the marginal eects of the linkages based on the estimates for specication II (2). Info, Econ, Fin, and Ind denote information capacity, economic integration, nancial integration, and industrial dissimilarity respectively. t statistics in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

27

Table 5:

The EMU eect

ρt−1 U dEM ij EM U (1999−)

dij,t

(III) 0.316∗∗∗ (16.58) 0.286∗∗∗ (4.00)

(III) 0.332∗∗∗ (21.00) 0.277∗∗∗ (3.99)

(IV) 0.316∗∗∗ (16.59) 0.270∗∗∗ (3.78)

(IV) 0.332∗∗∗ (21.00) 0.274∗∗∗ (3.95)

0.0889∗∗∗ (3.75)

0.0955∗∗∗ (4.17)

0.0762∗∗∗ (3.38)

0.0923∗∗∗

EM U (2002−)

0.0361∗ (1.84) 0.0183∗∗∗ (3.61) 0.0180 (0.09) 0.00692∗∗ (2.32) -0.0389∗∗∗ (-3.34) 0.00463 (1.45)

0.00667 (0.38) ∗∗∗ ∗ Info 0.0164 0.00854 0.00884∗ (3.29) (1.77) (1.90) Econ 0.0132 -0.392 -0.387 (0.06) (-1.37) (-1.36) Fin 0.00726∗∗ 0.0205∗∗∗ 0.0204∗∗∗ (2.42) (3.31) (3.30) ∗∗∗ ∗∗∗ Ind -0.0389 -0.0717 -0.0715∗∗∗ (-3.33) (-3.59) (-3.61) Size 0.00486 0.00910∗∗∗ 0.00906∗∗∗ (1.52) (3.32) (3.30) mm Info×d -0.00478 -0.00473 (-1.53) (-1.51) Econ×dmm 1.098∗∗∗ 1.092∗∗∗ (3.67) (3.71) mm Fin×d -0.00411 -0.00408 (-0.38) (-0.49) Ind×dmm 0.0802∗∗∗ 0.0804∗∗∗ (3.24) (3.34) mh Info×d -0.000497 -0.000461 (-0.16) (-0.14) Econ×dmh 0.521 0.515 (1.62) (1.62) Fin×dmh -0.0163∗∗ -0.0163∗∗ (-2.17) (-2.23) Ind×dmh 0.0561∗∗∗ 0.0558∗∗∗ (2.76) (2.74) Note : This table presents the results for specications III (3) and IV (4). Info, Econ, Fin, and Ind denote information capacity, economic integration, nancial EM U (1999−) U integration, and industrial dissimilarity respectively. dEM , dij,t , and ij

dij,t

EM U (2002−)

dij,t are dummy variables (dened in equations (7-9) in Section 3.2.6) for the correlations between two EMU countries at dierent stages of the EMU implementation. t statistics in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

28