Integration and Interdependence Among Equity Markets in South Asia: Measuring Through ARDL Bounds Approach

I. Introduction

The issue of international financial integration and interdependence occupies a vital place in the area of international finance. The recent freeing of capital controls across nations and the march towards greater economic integration around the world have invoked great interest among scholars and industry practitioners for the significant implications for international capital flows and portfolio diversification. However, freeing of capital controls and economic integration is a necessary, but not sufficient, condition for the integration of financial markets. Country-specific factors like the dissimilarities in the size of the market, financial market regulations, economic growth rate, transaction cost, etc., also determine the extent of integration of financial markets. It may be mentioned that economic integration has deepened in the world over the time which could be gauged by the rise in the number of Regional Trade Agreements (RTA) in force, which were 83 in 1999, leapfrogged to 194 in 2009 and 293 in 2019. It is the intensification of economic integration, coupled with liberal capital controls regime, that has created a ground for better financial integration.

The international financial integration literature has come a long way, ever since Kasa (1992) found a common stochastic trend among the major developed countries’ stock markets. Similar results of the rise in integration were reported by studies undertaken in other regions as well. For instance, Corhay et al. (1993) testified to almost similar results by taking a sample of select European stock markets. Chowdhury (1994) also reported nearly similar results by accounting for the data on newly industrialized economies of Asia. Berben and Jos Jansen (2009) find pieces of evidence of increased integration among a select group of countries in Europe and found little relevance of the European Union in promoting financial integration among member nations. Blackman et al. (1994) tested the existence of co-movement among the stock prices of major markets of the industrialized world in two sub-periods. One of them represented the period before the commencement of financial deregulation and technological progress, that is, the 1970s, and the other represented the period after these events, that is, the 1980s, and found that the latter period was more integrating than the former. It partly explains why Berben and Jos Jansen (2009) found a limiting role played by European Monetary Union. Another implication of Blackman et al. (1994) findings is that the gains accruing from portfolio diversification in international markets were observed to be lower in the later period, representing the period of financial deregulation.

Even though a vast body of the literature that deals with financial integration in developed countries is available, a growing body of literature has focused on the emerging nations of Asia, Africa and Latin America. Financial integration in these countries appears to be moving hand in hand with economic integration (Phylaktis & Ravazzolo, 2002). Interestingly, the markets of Southern African Customs Union demonstrated little evidence of integration (Piesse & Hearn, 2002). The Middle East and North African regions also did not exhibit much financial integration with the European Monetary Union and the USA (Marashdeh, 2005). One aspect that needs to be noted here is that although Latin American countries appear to be better integrated with the global financial market, there is little evidence of them being financially integrated with each other (Hunter, 2006). Majid et al. (2008) found an increased level of integration among Association of Southeast Asian Nations (ASEAN) countries’ stock markets as well as the integration of them with developed countries’ markets such as the USA and Japan. It has increased considerably, predominantly in the aftermath of the 1997 Asian financial crisis. Shamiri and Isa (2010) find pieces of evidence of volatility transmission from developed countries’ markets like that of the USA and Japan towards the Asia-Pacific markets. Palamalai and Devakumar (2013) find long-run relationships among the stock markets of emerging economies in the Asia-Pacific region. Wang (2014) reported that the linkages among the stock markets in East Asia had been strengthened in the aftermath of the global financial crisis of 2008–2009. Teng et al. (2016) reported that the economic activities of China influence the ASEAN economies’ stock markets. Ahmed and Singh (2016) found co-integrating relations among the stock markets of Regional Comprehensive Economic Partnership (RCEP) economies, which include ASEAN economies since all ASEAN countries are members of RCEP. Rijanto (2017) revealed that the linkages between ASEAN countries’ stock markets and the developed countries’ stock markets like the USA and the UK had been strengthened, following the financial crisis of 2008. Sehgal et al. (2018a) find an increased level of integration among the equity markets in the East Asia economic community.

Though capital controls have been removed incrementally and partially during the 1980s and 1990s in South Asia, the studies undertaken, over the time, demonstrate some evidence of integration of financial markets (Narayan et al., 2004; Mohsin & Rivers, 2010; Perera & Wickramanayake, 2012; Tripathi & Seth, 2016). The research undertaken by Narayan et al. (2004) reported co-integrating relations among the region’s stock markets when Pakistan’s stock market was introduced as a dependent variable. Mohsin and Rivers (2010) used real interest rate parity approach to assess the level of financial integration in this region. They exhibited an improvement in the amount of financial integration in the post-liberalization era. Sharma and Bodla (2011) find a one-way causal relationship between the stock market of India and the stock market of Sri Lanka as well as Pakistan. Perera and Wickramanayake (2012) reported that integration of stock market is much stronger than the integration of bond market among selected countries in South Asia, and trade openness and political stability were found to be important determinants of regional financial integration (Arora & Ratnasiri, 2014). Abbas et al. (2013) find the existence of transmission of volatility at least one way, that is, from developed countries’ equity markets to the regional equity markets of Asia. Dasgupta (2014a) noted that the stock market of India is well integrated with the stock markets of the rest of the South Asian stock markets. Prakash and Kumar (2014) show that the equity markets in South Asian countries are well integrated with the global economy. Tripathi and Seth (2016), using Johansen’s co-integration test, find the existence of long-run relationship among all major stock markets in South Asia, that is, Bangladesh, India, Pakistan and Sri Lanka and the transmission of volatility among these stock markets under Autoregressive Conditional Heteroskedasticity-Generalized Autoregressive Conditional Heteroskedasticity (ARCH-GARCH) framework. On the other hand, Sehgal et al. (2018b) find a low level of integration among the equity markets in South Asia, using Copula GARCH and Diebold and Yilmaz’s approach.

It may be noted that the long-run asset price co-movements across different markets have significant implications for international portfolio diversification. Portfolio theory advocates the diversification of assets holdings across the world, provided that the returns from other markets are mostly unequal, as compared to the home country market, which may ultimately benefit the investors by apportioning their risks across markets (Grubel, 1968). The long-run asset price co-movements across the co-integrated markets indicate the incidence of a common stochastic trend that binds the markets together in the long run. As a result, in the long run, the potential for earning abnormal returns are arbitraged away in such markets. However, still, the possibility of capitalizing on diversification benefits in the short run cannot be ruled out. The existence of country-specific risk and the fact that returns are not perfectly correlated across countries indicate that diversification benefits are not eliminated in practice even in a more extended period. On the contrary, there is potential for reaping diversification benefits if the markets are not co-integrated.

In South Asia, economic liberalization and the freeing of capital controls for inflows during the 1980s and 1990s had created a ground for greater integration of financial markets in this region. Investors manage to minimize country-specific risks by following cross-country diversification of financial asset holdings but are still vulnerable to common shocks if the financial markets are integrated. Studies demonstrate that markets with considerable economic interrelations and geographic proximity exert substantial influence on each other (Janakiramanan & Lamba, 1998). Abbas et al. (2013) find volatility transmission within the financial markets in this region. Thus, it is imperative to examine the degree of financial integration at a regional level. This study examines the regional financial integration by employing equity prices of major equity markets in South Asian economies, namely Bangladesh, India, Pakistan and Sri Lanka. The biggest three stock exchanges in the world are located in the USA, Japan and the UK. The New York Stock Exchange with a market capitalization of US$13.4 trillion is the biggest stock exchange in the world, followed by Tokyo Stock Exchange with a market capitalization of US$3.8 trillion. The third biggest stock exchange is the London Stock Exchange with a market capitalization of US$3.6 trillion (The World’s Biggest Stock Exchanges, n.d.). Studies in this regard highlight that the co-movement of stock prices in the USA, Japan and the UK have increased after the stock market crash of 1987 (King & Wadhwani, 1990). Abbas et al. (2013) reported the transmission of shocks from developed countries’ markets towards four Asian markets, viz. China, India, Pakistan and Sri Lanka. Dunis and Shannon (2005) show that the Japanese market is more tightly integrated with the emerging markets of Southeast and Central Asia, including India, than the UK and the US Therefore, this region’s integration with developed countries’ equity markets is gauged by including three major equity markets of the USA, the UK and Japan.

The rest of the article is organized as follows: Section II analyses the experiences of the major countries in South Asia with the policies adopted for liberalization of the capital account. Section III discusses the data, measures and estimation techniques. Section IV analyses the results of the estimated models, while Section V sums up the findings.

II. Route Towards Capital Account Convertibility: Experiences of Major Countries in South Asia

A large number of studies world over have established that financial integration is contingent upon liberalization of capital account that facilitates cross-border capital flows, whether at the regional or the global level. This section endeavours to analyse the experiences of major countries in South Asia that have partially or wholly liberalized the capital account.

III. Data and Methodology

Methodology

The interaction between the equity markets of major economies in South Asia, viz. Bangladesh, India, Pakistan and Sri Lanka, and their interaction with major markets of the developed world, viz. Japan, the UK and the USA are examined with the help of autoregressive distributed lag (ARDL) bounds test of cointegration. For implementing the procedure suggested by Pesaran et al. (2001), the order of integration of the time series has to be determined, as the test results are not valid if any of the time series is integrated of the order 2 {I(2)}. The orders of integration of the time series are identified with the help of Augmented Dickey–Fuller (ADF) test and Phillips–Perron (PP) test, which are some of the existing unit root tests. The ADF, PP and ARDL bounds tests of cointegration are used with the help of EViews 9 software package, which is a statistical software used for data analysis. Nevertheless, Perron (1989) shows that it is challenging to reject the non-stationary null hypothesis if we apply the standard unit root test in a time series, which is stationary around a trend and contains a structural break. Since we are dealing with financial time series, which is extremely volatile to external shocks, the time series might likely be biased with one or more structural breaks. As the standard unit root tests (ADF and PP test) are not programmed to detect the presence of any structural break in the series, the study applies Zivot and Andrews unit root test, which could detect one structural break even if the break date is unknown. The Zivot and Andrews unit root test is estimated with the help of Stata 13 software package, which is a statistical software used for data analysis. A structural break can occur either in the intercept term or in the slope, or both, in a trend function. The study uses the procedure suggested by Zivot and Andrews (1992), allowing for a break in both intercept and slope, which is specified as follows:

Δ ln y t = α 0 + δ 1 ln y t − 1 + θ t + γ 1 D U t + μ 1 D T t + ∑ i = 1 p β 1 i Δ ln y t − i + ε t

(1)

where DU _t is a dummy variable for a mean shift occurring at a break-date (BD), taking the value of 1 if t is greater than BD otherwise 0, and DT _t is the corresponding trend shift variable.

This procedure chooses a point as a break-date among all possible available break-dates, where the computed t-statistic of δ₁ is the minimum. This procedure tests a null hypothesis of unit root without a structural break against an alternative hypothesis of stationary time series with a one-time break.

Bounds Testing Approach

ARDL bounds testing approach of cointegration is a multivariate approach, which overcomes the limitations suffered by earlier procedures suggested by Engle and Granger (1987) and Johansen and Juselius (1990) and also delivers robust results in case of small sizes of the sample. To implement the Johansen cointegration test, all the variables have to be non-stationary and have to be integrated of the same order. On the contrary, the bounds test approach need not require all the variables to be integrated of the same order, and it gives valid results even if one or more variables are stationary. However, the results are not valid in case any of the variables are integrated of order 2 {I(2)}. One advantage of this approach is that this approach could identify which variable is the dependent variable when cointegration is established in a particular system of the equations through the bounds testing procedure. This study estimates eight models through this approach for examining the relationship in the long run. The dynamics are probed by obtaining causal relations in the Granger sense under the vector error correction model (VECM) framework. The specifications of the models are as follows:

Bangladesh

ln ( E M B ) t = α 01 + α 11 ln ( E M I ) t + α 12 ln ( E M P ) t + α 13 ln ( E M S ) t + ε 1 t

(2)

ln ( E M B ) t = α 02 + α 21 ln ( E M I ) t + α 22 ln ( E M P ) t + α 23 ln ( E M S ) t − 1 + α 24 ln ( E M J ) t + α 25 ln ( E M U K ) t + α 26 ln ( E M U S A ) t + ε 2 t

(3)

India

ln ( E M I ) t = α 03 + α 31 ln ( E M B ) t + α 32 ln ( E M P ) t + α 33 ln ( E M S ) t + ε 3 t

(4)

ln ( E M I ) t = α 04 + α 41 ln ( E M B ) t + α 42 ln ( E M P ) t + α 43 ln ( E M S ) t + α 44 ln ( E M J ) t + α 45 ln ( E M U K ) t + α 46 ln ( E M U S A ) t + ε 4 t

(5)

Pakistan

ln ( E M P ) t = α 05 + α 51 ln ( E M B ) t + α 52 ln ( E M I ) t + α 53 ​ ln ( E M S ) t + ε 5 t

(6)

ln ( E M P ) t = α 06 + α 61 ln ( E M B ) t + α 62 ln ( E M I ) t + α 63 ln ( E M S ) t + α 64 ln ( E M J ) t + α 65 ln ( E M U K ) t + α 66 ln ( E M U S A ) t + ε 6 t

(7)

Sri Lanka

ln ( E M S ) t = α 07 + α 71 ln ( E M B ) t + α 72 ln ( E M I ) t + α 73 ln ( E M P ) t + ε 7 t

(8)

ln ( E M S ) t = α 08 + α 81 ln ( E M B ) t + α 82 ln ( E M I ) t + α 83 ln ( E M P ) t + α 84 ln ( E M J ) t + α 85 ln ( E M U K ) t + α 86 ln ( E M U S A ) t + ε 8 t

(9)

where ln is the natural log, EMB _t is the returns of the equity market of Bangladesh at time t, EMI _t is the returns of the equity market of India at time t, EMP _t is the returns of the equity market of Pakistan at time t, EMS _t is the returns of the equity market of Sri Lanka at time t, EMJ _t is the returns of the equity market of Japan at time t, EMUK _t is the returns of the equity market of the UK at time t and EMUSA _t is the return of the equity market of the USA at time t.

The existence of a long-run relationship between the dependent and independent variables under this approach can be ascertained by first estimating Equations (2)–(9) by ordinary least squares and testing the joint significance of the coefficients of the lagged levels of variables with the help of F-statistics. That is, the null hypothesis of H₀: δ₁ = δ₂ = 0 against the alternative one H₁: δ₁ ≠ δ₂ ≠ 0 is tested. Cointegration or long-run relations are established when the computed F-statistics exceed the value of the upper bound (computed by Pesaran et al. [2001]) at the appropriate level of significance; here, we will reject the null hypothesis. On the contrary, if the computed F-statistics is lower than the lower bounds value (calculated by Pesaran et al. [2001]) at the appropriate level of significance, then the null hypothesis is accepted, that is, cointegration is not established. The result will be inconclusive if the computed F-statistics falls in between the upper and lower bounds value at the appropriate level of significance. The generic form of the ARDL model is as follows:

Δ ln y t = α 0 + δ 1 ln y t − 1 + δ 2 ln x t − 1 + ∑ i = 1 p β 1 i Δ ln y t − i + ∑ j = 0 q β 2 j Δ ln x t − j + ε t

(10)

where Δ is the difference operator, ln is the logarithmic form, δs are the long-run coefficients, βs are the short-run coefficient and ε_t is the white noise error term. The structural lags p and q are determined by using minimum Akaike information criterion (AIC).

Once cointegration is established in the long run through bounds test, short-run coefficients associated with the long-run estimates are estimated through VECM. It is a vector autoregression estimated in first-difference form along with one additional term known as error correction term (ECT). Causality in the Granger sense is obtained within this framework with the help of the Wald test. The specification of VECM is as follows:

Δ ln y t = β 0 + ∑ i = 1 p β 1 i Δ ln y t − i + ∑ i = 0 q β 2 i Δ ln x t − i + φ E C T t − 1 + ε t

(11)

where β s are the short-run coefficients of the model, φ is the speed of adjustment parameter and ECT is the error correction term.

Causality in the Granger sense is obtained by implementing the Wald test on the current and lagged terms of the short-run coefficients of the independent variables estimated under the VECM framework. Where the null hypothesis is that the set of coefficients of the variables are not statistically significant, that is, (β₁₁ = β₁₂ = 0) is tested against an alternative hypothesis, that is, the set of coefficients of the variables are statistically significant (β₁₁ ≠ β₁₂ ≠ 0). If the null hypothesis is accepted, then it can be concluded that the independent variable does not cause variation in the dependent variable in the short run.

Table 1 reports the descriptive statistics of the variables. It shows that the variables used in the study are not volatile, as the standard deviations of the variables are below their respective means. Table 1 further shows that the distribution of the sample is quite symmetrical since the coefficients of skewness for all variables lie between ‘+’ or ‘−’1. Kurtosis indicates that the distribution of two variables, namely EMP and EMUSA are too flat since the coefficients are less than −1.

The results of ADF and PP tests in both at levels and in first differences are reported in Table 2. Here, the non-stationarity null hypothesis is tested against the stationary alternative hypothesis. The results presented in Table 2 reveal that at levels, all the variables are non-stationary at the appropriate level of significance.

The computed value of test statistics (t-statistics for ADF test and adjusted t-statistics for PP test) is less than the critical value at an appropriate significance level as signified by respective probability values. However, when the same tests are administered on the variables at first differences, the null hypothesis of non-stationary time series is rejected, and the stationary alternative hypothesis is accepted for all the variables at an appropriate level of significance. Since the computed value of test statistics exceeds the critical value at appropriate levels of significance signified by its respective probability values, it could, therefore, be inferred that at levels, all the variables are non-stationary. When the same tests are administered on the first differenced variables again, these variables become stationary, that is, all the variables are integrated of the order 1 I(1). Since financial time series are volatile due to exposure to external shocks, Zivot and Andrews unit root test is used to check whether the series suffers from any break, which is reported in Table 3.

OPEN IN VIEWER

Table 2.

Results of ADF and PP Tests

Variables	Augmented Dickey–Fuller Test		Phillips–Perron Test
	Level	First Difference	Level	First Difference
LnEMB	−1.71558 (00.7359)	−8.818311 (0.0000)	−1.874989 (0.3427)	−9.487521 (0.0000)
LnEMI	−2.659546 (0.256)	−10.18815 (0.0000)	−1.652953 (0.4518)	−10.7756 (0.0000)
LnEMP	−2.523683 (0.3163)	−10.06228 (0.0000)	−1.906985 (0.3279)	−10.39465 (0.0000)
LnEMS	−2.734541 (0.2258)	−9.975215 (0.0000)	−2.474735 (0.1248)	−10.36819 (0.0000)
LnEMJ	−2.828443 (0.1915)	−9.255192 (0.0000)	−0.311527 (0.9181)	−9.667693 (0.0000)
LnEMUK	−1.363037 (0.8647)	−11.51006 (0.0000)	−2.169361 (0.2188)	−12.15367 0.0001
LnEMUSA	−2.828348 (0.1915)	−10.29896 (0.0000)	0.552475 (0.9877)	−10.85689 (0.0000)

Source: The authors.

Note: Probability values are given within the parentheses ‘( )’.

OPEN IN VIEWER

Table 3.

Results of Zivot and Andrews Test

Variables	Level	Break Date
LnEMB	−3.851	2011m10
LnEMI	−3.709	2011m8
LnEMP	−3.447	2013m4
LnEMS	−3.26	2012m12
LnEMJ	−3.926	2012m12
LnEMUK	−3.285	2015m8
LnEMUSA	−4.18	2015m8

Source: The authors.

Note: The critical values are −5.57 at 1%, −5.08 at 5%, and −4.82 at 10% level of significance, respectively.

OPEN IN VIEWER

Table 4.

Outcome of the Bounds Test at a Regional Level

Dependent Variable	Independent Variable	Computed F-statistic
LnEMB	LnEMI	2.410141
	LnEMP
	LnEMS
LnEMI	LnEMB	3.359417*
	LnEMP
	LnEMS
LnEMP	LnEMB	0.798546
	LnEMI
	LnEMS
LnEMS	LnEMB	3.173923
	LnEMI
	LnEMP
Lower Bound Value	Upper Bound Value	Critical Value
3.65	4.66	***
2.79	3.67	**
2.37	3.2	*

Source: The authors.

Note: The asterisks indicate the level of significance: ***1%, **5% and *10%.

The result accepts the unit root null hypothesis without a structural break. Since Zivot and Andrews unit root test does not provide any additional evidence, it could be concluded that the underlying series is integrated of the order 1 I(1).

Table 4 reports the outcome of bounds test administered on Equations (2), (4), (6) and (8), where the natural log of equity market indices of four major countries of South Asia, viz. Bangladesh, India, Pakistan and Sri Lanka are introduced as a dependent variable in Equations (2), (4), (6) and (8), respectively, for estimating the degree of financial integration in this region at a regional level. The computed F-statistics of the Equation (2) is 2.410141. The computed F-statistics falls in between lower critical bound (LCB) value and upper critical bound (UCB) value at 10 per cent level of significance. It implies that the result is inconclusive for the existence of a long-run relationship between the equity markets when equity market of Bangladesh (EMB) is taken as the dependent variable. Since the bound test is inconclusive; we will not proceed any further with regard to Equation (2). Equation (4) confirms the existence of a long-run relationship between the equity markets as the computed F-statistics (3.359417) exceeds UCB at 10 per cent level of significance, where equity market of India (EMI) is introduced as a dependent variable. Since cointegration or long-run relationship is established here at a regional level, we can estimate the Wald Statistics for examining the causality in the short run under VECM framework.

In Equation (6) where equity market of Pakistan (EMP) is introduced as a dependent variable, it exhibits no long-run relationship. Since the computed F-statistics (0.798546) is below the LCB, the null hypothesis of no cointegration is accepted, and we will not proceed any further with regard to Equation (6). The computed F-statistics (3.173923) of Equation (8) in which the equity market of Sri Lanka (EMS) is the dependent variable also falls in the inconclusive zone at 10 per cent level of significance. Here too, we will not proceed any further with regard to Equation (8). It could be concluded that the long-run relationship exists in South Asia only when India is taken as a dependent variable. The finding is in contrast with the results of Narayan et al. (2004), which established a cointegrating relationship between the four stock markets when Pakistan is taken as a dependent variable. Since the study periods are different, the finding of Narayan et al. (2004) cannot be taken as conclusive and, therefore, cannot be generalized.

The practical implication of this finding of a long-run relationship when EMI is taken as a dependent variable is that the potential for earning diversification benefits by international investors are somewhat limited in these markets in the long run. However, as the asset price co-move, co-vary, less than ideally, across countries and the existence of country-specific risk suggests that diversification benefits are not eliminated in the cointegrated markets in the long run in practice. It does not rule out the possibility of investors getting diversification benefits in the short term. Table 5 reports the outcome of bounds test administered on Equations (3), (5), (7) and (9), where the natural log of equity market indices of four major countries of South Asia, viz. Bangladesh, India, Pakistan and Sri Lanka are introduced as the dependent variable in Equations (3), (5), (7) and (9), respectively, for estimating the degree of financial integration of this region at an international level by including the developed countries’ equity markets such as that of Japan, the UK and USA.

OPEN IN VIEWER

Table 5.

Outcome of the Bounds Test at a Global Level

Dependent Variable	Independent Variable	Computed F-statistic
LnEMB	LnEMI	4.573457***
	LnEMJ
	LnEMP
	LnEMS
	LnEMUK
	LnEMUSA
LnEMI	LnEMB	5.188735***
	LnEMJ
	LnEMP
	LnEMS
	LnEMUK
	LnEMUSA
LnEMP	LnEMB	2.676171
	LnEMI
	LnEMJ
	LnEMS
	LnEMUK
	LnEMUSA
LnEMS	LnEMB	8.691081***
	LnEMI
	LnEMJ
	LnEMP
	LnEMUK
	LnEMUSA
Lower Bound Value	Upper Bound Value	Critical Value
2.88	3.99	***
2.27	3.28	**
1.99	2.94	*

Source: The authors.

Note: The asterisks indicate the level of significance: ***1%, **5% and *10%.

Equations (3), (5) and (9) confirm the existence of a long-run relationship between the equity markets as the computed F-statistics (4.573457, 5.188735 and 8.691081) are above UCB at 1 per cent level of significance. Equation (7), where the EMP is the dependent variable, does not exhibit any cointegrating relations, as the computed F-statistics falls in the inconclusive zone. Since cointegration or long-run relationship has been established, we proceed further by estimating the Wald statistic for examining the causality under VECM framework for Equations (3), (5) and (9). In case of Equation (7) where computed F-statistics fall in the inconclusive zone at 5 per cent level of significance, we will not proceed any further.

The finding of cointegrating relations, when equity market indices of three markets from the developed world are introduced, is interesting. The results change drastically with three equations, that is, Equations (3), (5) and (9), exhibiting long-run relationship compared to only one equation, that is, Equation (4) for accessing regional integration. It implies that the South Asian region is well integrated with the developed world. The aforementioned pieces of evidence suggest that integration at the global level is much higher than the integration at the regional level for this region. The practical implication of this finding is that the potential for earning diversification benefits by international investors are somewhat limited in these markets in the long run. It, however, does not rule out the possibility of investors getting diversification benefits in the short term. Table 6 reports the results of VECM, and the associated Wald statistics estimated to examine the possibility of causality among the variables in the short run.

The coefficient of ECT is negative and significant at 1 per cent level of significance, which again confirms the existence of long-run relations among the four variables when EMI is taken as a the dependent variable. Therefore it could be concluded that, independent variables introduced in Equation (4) that is, EMBEMP EMS affects the dependent variable that is, EMI in the longrun. It means that India bears the burden of any disturbances in the long-run equilibrium. The magnitude of the coefficient of ECT shows the speed of adjustment towards the long-run equilibrium. Since the scale is not very large, we can conclude that the rate of convergence towards equilibrium is moderate. This finding implies that when there is disequilibrium in the long-run relations, EMI adjusts itself to restore equilibrium in the long run. Further, Table 6 shows that the Wald test statistics are significant for EMB, EMP and EMS, and it can be concluded that the EMB, EMP and EMS granger cause EMI in the short run. Table 7 reports the results of VECMs, and the associated Wald Statistic estimated to apprehend causality among the variables.

OPEN IN VIEWER

Table 6.

Granger Causality Results Based on Vector Error Correction Model

Dependent Variable	Independent Variable	Lag	Wald Statistics	ECT
ΔLnEMI	ΔLnEMB	3	6.72905*	–0.318894***
	ΔLnEMP	7	13.68377*
	ΔLnEMS	7	16.00235**

Source: The authors.

Note: The asterisks indicate the level of significance: ***1%, **5% and *10%. The Wald test follows a ϰ² distribution.

OPEN IN VIEWER

Table 7.

Granger Causality Results Based on Vector Error Correction Model

Dependent Variable	Independent Variable	Lag	Wald Statistic	t-Statistic	ECT
ΔLnEMB	ΔLnEMI	8	38.07363***		−0.56335***
	ΔLnEMJ	8	24.54868***
	ΔLnEMP	0		0.417219
	ΔLnEMS	0		−0.637758
	ΔLnEMUK	8	15.07215*
	ΔLnEMUSA	2	0.742276
ΔLnEMI	ΔLnEMB	0		1.501884	−0.07849***
	ΔLnEMJ	8	52.84204***
	ΔLnEMP	7	14.66787**
	ΔLnEMS	8	40.05011***
	ΔLnEMUK	8	38.38833***
	ΔLnEMUSA	6	17.66378***
ΔLnEMS	ΔLnEMB	0		−1.362779	−0.73462***
	ΔLnEMI	0		2.487483**
	ΔLnEMJ	6	36.39847***
	ΔLnEMP	2	9.446981***
	ΔLnEMUK	5	27.06998***
	ΔLnEMUSA	3	5.795838

Source: The authors.

Note: The asterisks indicate the level of significance: ***1%, **5% and *10%. The Wald test follows a ϰ² distribution.

The coefficient of ECT is negative and significant at 1 per cent level of significance for all the three models, which again confirms the existence of long-run relationship when equity markets of Bangladesh, India and Sri Lanka are taken as dependent variables. Therefore, in the long run, the equity markets of India, Pakistan, Sri Lanka, the UK (EMUK), the USA (EMUSA) and Japan (EMJ) Granger-cause equity markets of Bangladesh, equity markets of India, Japan, Pakistan, Sri Lanka, the UK and the USA Granger-cause equity market of India and equity markets of Bangladesh, India, Japan, Pakistan, the UK and the USA Granger-cause equity market of Sri Lanka. This finding implies that when there is disequilibrium in the long-run relationship, EMB, EMI and EMS adjust themselves to restore equilibrium in the long run. The magnitude of the coefficient of ECT shows the speed of adjustment towards the long-run equilibrium. Since the scale is not very large when EMI is taken as the dependent variable, we can conclude that the speed of convergence towards equilibrium is slow. However, when EMB and EMS are taken as dependent variables, the magnitude becomes pretty large, and the speed of convergence towards equilibrium becomes faster. It can be seen from the results that EMP is the most exogenous market among the four markets since it does not exhibit a long-run relationship with any of the other markets considered for this study. This finding can be attributed to the political instability, which Pakistan had experienced in the recent past that might have retarded the attractiveness of this market to overseas investors, and also the strained relations of Pakistan with India might have hampered further capital inflows.

Further, Table 7 demonstrates that when EMB is taken as a dependent variable, the Wald test statistics emerge as significant for equity markets of India, Japan and the UK, and it can be concluded that the EMI, Equity Market of Japan (EMJ) and Equity Market of United Kingdom (EMUK) Granger cause EMI in the short run. When EMI is taken as a dependent variable, the Wald test statistics are statistically significant for equity markets of Japan, Pakistan, Sri Lanka, the UK and the USA, suggesting that EMJ, EMP, EMS, EMUK and Equity Market of United States of America (EMUSA) granger cause EMI in the short run. Finally, when EMS is taken as a dependent variable, the Wald test statistics are statistically significant for equity markets of India, Japan, Pakistan, and the UK, which means EMI, EMJ, EMP and EMUK granger cause EMS in the short run. Here, for testing the stability of the parameters, the cumulative sum of recursive residual (CUSUM) and the square of CUSUM, CUSUMSQ tests have been applied. Figures 1–8 show the plot of CUSUM and CUSUMSQ tests. The results indicate the absence of any instability of the coefficients because the plots of the CUSUM and CUSUMSQ statistics fall inside the critical bands of the 5 per cent confidence interval of parameter stability.

OPEN IN VIEWER

Figure 1.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 2.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 3.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 4.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 5.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 6.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 7.

Source: Authors’ calculations.

OPEN IN VIEWER

Figure 8.

Source: Authors’ calculations.

This article examines the degree of integration among the major financial markets in South Asia at the regional level, and, at the same time, it estimates the degree of integration of this region with major developed countries’ equity markets. The bounds testing procedure finds the existence of a long-run relationship among the four markets when the equity market of India is taken as the dependent variable. The practical implication of this finding is that the potential for earning diversification benefits by international investors will be somewhat limited in these markets in the long run. However, as the asset price co-move, co-vary, less than perfectly, across countries, and with the existence of country-specific risks, it implies that diversification benefits are not eliminated in the cointegrated markets in the long run in practice. It does not rule out the possibility of investors getting diversification benefits in the short term. Although the bounds testing procedure finds some evidence of integration at a regional level, shreds of evidence suggest that integration at the global level is much higher than the integration at the regional level for this region. All the major countries in South Asia had embarked upon the path of liberalization in the 1980s and 1990s, and all the countries included in the analysis had undergone partial, if not full, capital account liberalization, particularly concerning inflows of foreign capital. Trade agreements like the South Asian Free Trade Area had failed to stimulate trade in large scale due to the existence of non-tariff barriers, which had prevented the growth of trade despite the fall in tariff inhibiting greater economic integration within the region (Raihan, 2016). Thus, freeing of capital controls and economic integration are necessary but not sufficient condition for greater financial integration. Limited economic integration within the region might have contributed to the lesser degree of financial integration at the regional level. Another interesting finding is that Pakistan does not exhibit cointegrating relations with the rest of the equity markets in South Asia, and results are inconclusive when developed countries’ equity markets are introduced to the estimated models. This finding might be attributed to the political instability, which Pakistan had experienced in the recent past, and also the strained relations with India, which might have further hampered capital inflows.