Interdependence and dynamic linkages ... - Wiley Online Library

Accounting and Finance 44 (2004) 419–439

Interdependence and dynamic linkages between the emerging stock markets of South Asia

Blackwell Oxford, Accounting ACFI AFAANZ, 30810-5391 Original unknown 44 P. 2004 Narayan UK Article Publishing, 2004. etand al. Finance /Published Accounting Ltd. byand Blackwell FinancePublishing 44 (2004) 00–00

Paresh Narayana, Russell Smythb, Mohan Nandhac a

Department of Accounting, Finance and Economics, Griffith University, Southport 4215, Department of Economics, Monash University, Caulfield East 3145, and cDepartment of Accounting and Finance, Monash University, Clayton 3800, Australia

b

Abstract The present article examines the dynamic linkages between the stock markets of Bangladesh, India, Pakistan and Sri Lanka using a temporal Granger causality approach by binding the relationship among the stock price indices within a multivariate cointegration framework. We also examine the impulse response functions. Our main finding is that in the long run, stock prices in Bangladesh, India and Sri Lanka Granger-cause stock prices in Pakistan. In the short run there is unidirectional Granger causality running from stock prices in Pakistan to India, stock prices in Sri Lanka to India and from stock prices in Pakistan to Sri Lanka. Bangladesh is the most exogenous of the four markets, reflecting its small size and modest market capitalization. Key words: Cointegration; Granger causality; South Asia; Stock market linkages JEL classification: G15

1. Introduction The integration of international financial markets has been a topic of growing interest in recent times. This interest has been sparked by several developments. One factor is the improved flow of capital across national borders associated with the freeing of control on asset market transactions. A second factor is that there has been a reduction in transaction costs and a commensurable increase in the flow of information. Third, there has been an increasing interest in which markets are leaders and which markets are followers including which market or An earlier version of the present article was presented at the Asian Business and Economics Research Unit South Asia workshop, held at Monash University in August 2003. We thank two anonymous referees of the Journal and participants in the workshop for their constructive comments. Received 6 July 2003; accepted 15 September 2003 by Robert Faff (Editor). © AFAANZ, 2004. Published by Blackwell Publishing.

420

P. Narayan et al. / Accounting and Finance 44 (2004) 419– 439

markets set the trend for specific geographical regions (Masih and Masih, 1999). Fourth, the level of integration of markets carries important implications for portfolio theory, which advocates that investors diversify their assets across national borders provided that returns to stocks in these markets are less than perfectly correlated with the domestic market. Beginning with Kasa (1992), who found that there is a single common trend driving the stock markets of Canada, Germany, Japan, the UK and the USA, a large published literature has emerged focusing on the level of integration between the world’s major stock markets. However, the published literature which tests for integration purely between emerging markets is relatively scant (see Christofi and Pericli, 1999; Chen et al., 2002; Ng, 2002). The lack of research on interdependencies among emerging markets is surprising given that, as Chen et al. (2002, p. 1114) note, emerging markets represent a useful separate data source to investigate the market integration hypothesis given their low correlation with developed markets. Therefore, the potential for data-snooping biases are reduced. The present article contributes to the meagre published literature on interdependencies between emerging markets through examining stock market linkages in South Asia, using daily stock price indices over the period 1995–2001. We assess the dynamic linkages between the stock price indices of Bangladesh, India, Pakistan and Sri Lanka using a temporal Granger causality approach where we bind the four stock price indices within a multivariate cointegration framework. To test for the existence of any long-run relationships we use the bounds test approach to cointegration within an autoregressive distributive lag (ARDL) framework, which was developed by Pesaran and Shin (1999) and Pesaran et al. (2001). We also extend the Granger causality analysis to examine the impulse response functions. We focus on South Asia as a region for four reasons. First, few studies have included South Asian countries when considering stock market linkages. Goldberg and Dalgado (2001) consider India as part of a broader study of stock market linkages in Latin America and South-East Asia.1 Elyasiani et al. (1998) include the Indian and Sri Lankan markets as part of a wider study of stock market linkages between Sri Lanka and her major trading partners, but do not test for long-run relationships. Second, South Asia as a region has undergone rapid market liberalization in recent times, which has opened up the South Asian countries to increased investment flows. Third, the South Asian countries are proximate in an economic and geographical sense as well as sharing a strong common historical heritage and given these close ties, ex ante it is reasonable to expect that there might be high levels of market linkages. 1

While the title of Goldberg and Delgado’s (2001) article suggests that they look at financial integration in South Asia, with the exception of India, the other South Asian countries (Indonesia, Malaysia and Thailand) which they consider are better described as being in South-East Asia.

© AFAANZ, 2004


421

Fourth, from an Australasian and, in particular, an Australian perspective, traditionally, South Asia was considered a secondary market relative to SouthEast Asia (Vicziany, 1993). In part this reflected the perception of Australasian companies that South Asia (and India especially) was a difficult market (Vicziany, 1993, 2000; Vicziany and Chaterjee, 1999). However, there are several similarities between Australia, New Zealand and India, which always underpinned the potential for improved investment and trade ties. These countries are the biggest democracies in the region, with a free press, well-established government institutions and an independent judicial system as well as sharing English as the main language of commerce. Australia, New Zealand and India also have several complementarities in traded goods. Australia and New Zealand can supply wool for India’s textile industry as well as food processing technologies, which can be of benefit to South Asia as the world’s largest producing region. Australia can also supply coal for steel-making and mining technologies for the development of India’s mineral resources (JSCFADT, 1998, p. 16). Because of these complementarities South Asia has become increasingly important to Australia and New Zealand over the last decade. Since the liberalization of the international sector of the Indian economy in 1991, and the establishment of the Australia-India Council in 1991, trade has increased and a growing number of Australian companies have entered the Indian market and become insiders (Vicziany, 2000). Trade between Australia and India alone increased from $A0.75bn in 1989–1990 to over $A2.5bn in 1998–1999 (Shand, 2000, p. 17) and by the end of 2003 India is expected to be among Australia’s 10 largest trading partners (Indo-Australian Business Committee, 2003). Trade between New Zealand and South Asia also increased through the 1990s. New Zealand’s exports to South Asia increased from $NZ231m in 1990 to $NZ413m in 2002, while New Zealand’s imports from South Asia rose from $NZ91.5m in 1990 to $NZ300m in 2002 (New Zealand Ministry of Foreign Affairs, 2003). The balance of the article is set out as follows. The next section discusses the economic intuition underpinning stock market linkages. Section 3 provides a brief overview of South Asian stock markets. The empirical specification is presented and the data discussed in Section 4. Section 5 examines the order of integration of the variables using the Zivot and Andrews (1992) sequential trend break model. Once the order of integration of the variables is established we proceed to present the results from cointegration, the short-run and long-run results from the ARDL model and Hansen’s (1992) suite of tests for long-run parameter stability. Section 5 concludes with the findings for Granger causality. Section 6 contains the impulse response functions, while Section 7 summarizes the main findings from the present study. 2. Economic intuition underpinning stock market linkages Portfolio theory advocates that investors diversify assets across national boundaries provided that returns to stocks in these other markets are not perfectly © AFAANZ, 2004

422


correlated with the investor’s home market. Beginning with Grubel (1968), the benefits of international asset diversification, particularly in terms of risk reduction, have been much discussed in the published literature. The issue of whether stock markets are cointegrated carries important implications for portfolio diversification. Cointegrated stock markets imply that there is a common force, such as arbitrage activity, which brings the stock markets together in the long run. Therefore, testing for cointegration is a test of the level of arbitrage activity in the long run. Stock markets that are not cointegrated imply that arbitrage activity to bring the markets together in the long run is zero (Masih and Masih, 1997, 1999, 2002). The practical implication of finding that stock markets are cointegrated is that the potential for gaining abnormal profits through international diversification is limited in the long run. This is because if the markets are cointegrated, abnormal profits will be arbitraged away in the long run and, in the absence of barriers or potential barriers generating country risk and exchange rate premiums, we would expect similar yields for financial assets of similar risk and liquidity, irrespective of nationality or location (Von Furstenberg and Jeon, 1989). However, if markets are not cointegrated, there is no arbitrage activity to bring the markets together in the long run. Under these circumstances, there is potential for investors to obtain long-run gains through international portfolio diversification (Masih and Masih, 1997). Two provisos to the statement that cointegration implies limited opportunities for portfolio diversification are worth noting (Masih and Masih, 1999, p. 273). The first finding that stock markets are cointegrated does not preclude the possibility of investors making arbitrage profits through international diversification across these countries in the short term. Depending on the speed of adjustment, the short term can last a fair while. Second, because of the existence of varying degrees of financial risk of different securities and because various security cash flows covary less than perfectly across countries, while the diversification benefits in cointegrated markets may be very limited in the long run, they are unlikely to be eliminated in practice. The implications of finding cointegration for the Market Efficiency Hypothesis (MEH) are unclear. Granger (1986) argues that cointegration between two prices reflects an inefficient market. If two prices share a common trend in the long run, this implies predictability of each price’s movement, which indicates that one market may be caused by another and hence inefficiency exists. However, more recently, Dwyer and Wallace (1992), Caporale and Pittis (1998), Fernandez-Serrano and Sosvilla-Rivero (2001) and Masih and Masih (2002) argue that cointegration does not imply anything about efficiency. Masih and Masih (2002, p. 87) suggest that predictability, suggested by cointegration, implies nothing necessarily about inefficiency. A market is inefficient only if by using the predictability, investors can earn risk-adjusted excess returns, but if returns are generated it is unclear whether they are just compensation for risks incurred or are truly excess and risk-adjusted. © AFAANZ, 2004


423

3. Overview of South Asian stock markets India has more than 20 regional stock exchanges: of which, Mumbai (Bombay), Calcutta and Delhi account for more than 50 per cent of the listing. Mumbai is the major financial centre, accounting for two-thirds of the trading volume. The value traded in Mumbai was 3.5 per cent of the top 30 markets in the 1990s (Ariff and Khalid, 2000, p. 323). Ariff and Khalid (2000) suggest that the Mumbai Stock Exchange is among the top six emerging stock markets in the world following South Korea, Taiwan, Mexico, Thailand and Malaysia. Bekaert et al. (2001) date India’s integration into world equity markets as 1992. Around this time India introduced various liberalization measures including the first international equity offering, first American Depositary Receipts (ADR) and the relaxation of foreign investment laws related to Indian listed securities (Goldberg and Delgado, 2001, pp. 284–285). By the mid-1990s India had around 5,000 listed companies. However, the market remains dominated by the top 500 companies, which constitute more than 90 per cent of the market cap. These companies also account for almost 99 per cent of the traded value on the stock exchanges. Therefore, for all practical purposes, these top 500 companies represent the whole of Corporate India.2 Pakistan has three stock exchanges: the Islamabad and Lahore Stock Exchanges and, the leader, the Karachi Stock Exchange. Since the early 1990s, the government has taken a number of policy decisions for revamping the overall structure of the stock market to create a more conducive investment-friendly environment. These include introduction of various laws and rules for the protection of small investors, measures for improving efficiency and transparency in trade, curbing insider trading, strengthening the structure of the Securities and Exchange Commission of Pakistan, and bringing the market more in line with accepted international norms.3 The stock market initially reacted positively to these changes in 1991–1993, but political instability and an uncertain investment environment have hindered Pakistan’s attempts to develop its stock market (Ariff and Khalid, 2002). Sri Lanka has one of the oldest stock exchanges in the world with a history of share trading going back over 100 years. However, while formalized share trading commenced in Colombo in 1896, the share market remained small, elitist, and of marginal significance to the overall economy. In 1989 Sri Lanka liberalized foreign investment and, at the same time, the government’s focus turned to developing a modern capital market to raise funds for economic development. The Securities and Exchange Commission and the regulatory framework were strengthened through passage of important legislation such as 2

For more details on India see http://www.capitalmarket.com /Compendium 2001-2002.htm

3

For more details on Pakistan see http://www.pakistaneconomist.com /database1/private/ capital.htm © AFAANZ, 2004

424


insider trading laws and a takeover and mergers code, which contributed to improving market integrity. Sri Lanka also introduced a computer based depository system, which made the Colombo Stock Exchange the first market in the region to operate in a paperless environment (USAID, 2003). After 1989, the rapid increase in foreign investment following liberalization contributed to a boom in the stock market. In 1990 and 1991 the Colombo Stock Exchange was considered to be one of the best performing in the world (Ariff and Khalid, 2000, p. 423). There was a 15-fold increase in annual turnover and an 8-fold increase in market capitalization between 1989 and 1994 (Elyasiani et al., 1998). Formal trading on what was to become the Dhaka Stock Exchange commenced in 1956 with 196 listed securities and a total paid up capital of about Taka4bn. Between 1971 and 1976 trading activities were suppressed because of the civil war and economic policies of the then government. When trading activities resumed in 1976, the Dhaka Stock Exchange had nine listed companies with a paid-up capital of Taka137.5m (Chowdhury, 1994). The number of listed companies in Bangladesh increased to 230 in 2001 (see Table 1); however, compared to the other South Asian markets, the Bangladesh stock market remains small. In Bangladesh, there are a large number of non-actively traded shares and only very limited roles for mutual funds and professionally managed broker and investment houses. Table 1 presents some summary statistics for the four South Asian markets and compares them with Australia and New Zealand. India has the largest number of listed companies by a big margin. In 2001 India had more than 4 times the number of listed companies in Australia and almost 40 times the number of listed companies in New Zealand. The market capitalization of the Indian market also dwarfs the other South Asian markets, both in levels and as a percentage of gross domestic product (GDP), although all the South Asian markets are less capitalized than Australia (in levels and as a percentage of GDP) and less capitalized than New Zealand (as a percentage of GDP). The average company size on South Asian markets is smaller than in Australasia. In 2001 Australia ranked 28th in the world with average company size of $US280.6m and New Zealand ranked 44th in the world with average company size of $US122.6. By contrast, India was 85th ($US19.1m), Pakistan was 92nd ($US6.6m), Sri Lanka was 93rd ($US5.6m) and Bangladesh was 94th ($US5m) (Standard and Poors, 2002, p. 27). Moreover, while India also has the highest turnover of the South Asian markets, turnover in India has been less than in Australia for most years, although the figures for 2001 are of similar magnitude. After India, Pakistan and Sri Lanka are the next largest markets in South Asia. Pakistan has more listed companies and higher turnover than Sri Lanka, but both countries have similar market capitalization as a percentage of GDP. Compared with India, Sri Lanka and Pakistan, Bangladesh is a small market with low levels of capitalization. In Bangladesh, since 1992 market capitalization as a percentage of GDP has been in the range 1–3.8 per cent. If the costs of information search are fixed across markets of different sizes, the incentive © AFAANZ, 2004

1992

1993

1994

Number of listed companies Bangladesh 145 153 170 India 2,781 3,263 4,413 Pakistan 628 653 724 Sri Lanka 190 200 215 Australia 1,030 1,070 1,186 New Zealand 123 136 173 Market capitalization ($USm) Bangladesh 314 453 1,050 India 65,119 97,976 127,515 Pakistan 8,028 11,602 12,263 Sri Lanka 1,439 2,498 2,884 Australia 144,634 204,866 218,865 New Zealand 15,348 25,597 27,217 Market capitalization as a percentage of GDP Bangladesh 1.0 1.4 3.1 India 24.7 35.1 38.6 Pakistan 16.4 22.4 23.5 Sri Lanka 14.8 24.2 24.6 Australia 46.1 66.9 63.1 New Zealand 38.3 58.6 53.0 Turnover ($USm) Bangladesh 11 15 107 India 20,705 21,779 27,376 Pakistan ,980 1,844 3,198 Sri Lanka 114 385 700 Australia 45,771 67,712 94,726 New Zealand 3,168 6,785 7,162

1995

1996

1997

1998

1999

2000

2001

183 5,398 764 226 1,178 169

186 5,999 782 235 1,190 158

202 5,843 781 239 1,159 132

208 5,860 773 233 1,162 131

211 5,863 765 239 1,217 124

221 5,937 762 239 1,330 144

230 5,795 747 238 1,334 145

1,338 127,199 9,286 1,998 245,218 31,950

1,551 122,605 10,639 1,848 311,981 38,641

1,537 128,466 10,966 2,096 295,785 30,511

1,034 105,188 5,418 1,705 328,949 25,028

865 184,605 6,955 1,584 427,683 28,106

1,186 148,064 6,581 1074 372 794 18,866

1,145 110,396 4,944 1332 374,269 17,779

3.6 34.9 15.17 15.3 65.1 53.2 158 21,962 3,210 221 98,654 8,407

3.8 30.9 16.6 13.3 74.7 59.2 772 96,153 6,054 134 145,395 8,882

3.8 30.5 17.4 13.9 70.4 47.7 385 158,301 11,369 309 171,531 10,793

2.4 24.5 8.5 10.9 78.3 47.27 789 148,239 9,038 281 161,080 10,314

1.9 41.5 11.9 10.1 106.0 50.8 789 278,828 21,057 209 194,336 11,251

2.5 32.4 10.7 6.6 98.9 41.5 768 509,812 32,974 144 226,325 10,784

741 249,298 12,455 153 240,667 8,428

425

Sources: Standard and Poors (2001). Figures for GDP for Australia and New Zealand are from United Nations (1999); ABS (2001, 2002, 2003); Statistics New Zealand (2002). GDP, gross domestic product.


© AFAANZ, 2004

Table 1 Summary statistics of Australasian and South Asian markets

426


to undertake research for mispricing will be greater in larger markets than smaller markets. In an information context, the higher the market capitalization, the greater the liquidity of the market and the more open and deregulated the economy, the more the market is expected to lead (Masih and Masih, 2002). Therefore, on the basis of the brief review in this section, it is reasonable to expect India and, to a lesser extent, Pakistan and Sri Lanka to be the leading markets in South Asia. 4. Empirical specification and data The long-run multivariate model estimated for the four countries is as follows: India: lnSPIt = α 0 + α 1lnSPPt + α 2lnSPSt + α 3lnSPBt + ε1t

(1)

Pakistan: lnSPPt = α 0 + α 1lnSPIt + α 2lnSPSt + α 3lnSPBt + ε2t

(2)

Sri Lanka: lnSPSt = α 0 + α 1lnSPIt + α 2lnSPPt + α 3lnSPBt + ε3t

(3)

Bangladesh: lnSPBt = α 0 + α 1lnSPIt + α 2lnSPPt + α 3lnSPSt + ε4t

(4)

Here lnSPI, lnSPP, lnSPS and lnSPB are the natural logs of the stock prices in India, Pakistan, Sri Lanka and Bangladesh, respectively, while the ε terms are serially independent random errors with mean zero and finite covariance matrix. Equations (1)–(4) are used to test whether the four South Asian markets are cointegrated. The null hypothesis for equations (1)–(4) is no cointegration, which implies that there is no arbitrage activity to bring the markets together in the long run. The stock price indexes included in the present study are SE All Share (Bangladesh), Bombay SE National 200 (India), Karachi SE 100 (Pakistan) and Colombo SE All Share (Sri Lanka). We use daily data (excluding weekends and holidays) for the period 2 January 1995 to 23 November 2001, which gives a total of 1,800 observations. Using daily data is preferable to using lower frequency data such as weekly or monthly data because longer time horizons can obscure transient processes to innovations which may last only a few days (Elyasiani et al., 1998). The start and end dates for the present study were determined by data availability. All data were obtained from Datastream. © AFAANZ, 2004


427

5. Cointegration, causality testing and elasticities 5.1. Unit root tests We begin with testing the order of integration of the variables. The reason for this is that while we use the ARDL approach to cointegration, which is applicable irrespective of the order of integration of the variables, for an unbiased estimation of Granger causality it is essential that all variables are integrated of the same order. To establish the order of integration of each of the variables we first applied the standard Augmented Dickey Fuller (ADF) and Phillips-Perron (PP) unit root tests. The ADF and PP tests, which are not reported here to conserve space, but are available from the authors on request, suggest that SPIt, SPPt, SPSt, and SPBt are each integrated of order one or I(1) However, Perron (1989) shows that the power to reject a unit root decreases when the stationary alternative is true and a structural break is ignored. As the timeframe incorporates several tumultuous events including the Asian financial crisis, continuing financial liberalization in the four countries, civil unrest in Sri Lanka and skirmishes in the Kashmiri border region between India and Pakistan the prospect that there is a structural break in the stock price data is increased. The effect of the Asian financial crisis was not uniform across the four countries. Similar to other Asian emerging economies, India was a recipient of a large amount of private capital in the 1990s, but it was able to avoid the Asian contagion to a large extent (Shetty, 2002). One important reason for this is that the short-term capital liabilities in India were lower than the worse affected countries in South-East Asia. To illustrate, in 1997, on the eve of the crisis, short-term debt as a percentage of total external debt was: 7.2 per cent in India; 19.9 per cent in the Philippines; 25 per cent in Indonesia; 27.9 per cent in Malaysia; and 41.5 per cent in Thailand (World Bank, 1999). With a relatively small amount of foreign debt, the banking sector in India was also much better placed than these crisis countries. In 1997, reserves as a proportion of total outstanding short-term debt was 3.7 in India, 2.5 in Malaysia, 1.5 in the Philippines, 1 in Thailand and 0.6 in Indonesia (World Bank, 1999). Bangladesh’s economy was also relatively unscathed by the Asian financial crisis. The reason for this is that Bangladesh only has a small external economy, which was not linked by large trade with the countries which were affected by the currency problems (Ariff and Khalid, 2002). The stock price index, however, shows a sharp drop at the height of the crisis in South-East Asia in midto-late 1997. The Asian financial crisis had more widespread repercussions in Pakistan than Bangladesh or India. The financial crisis manifested itself in near default on external loans in 1996–1997 and a technical default on many external debt payments after 1998, which was compounded by the imposition of sanctions following India and Pakistan’s nuclear explosions (Hasan, 1998). While Sri Lanka was the first of the South Asian countries to undertake financial liberalization in the late 1970s, the benefits of reform have been tempered © AFAANZ, 2004

428


by ongoing civil war including an assassination attempt on the Prime Minister prior to the national election in 2000, which undermined investor confidence. To allow for the effects of such structural breaks Perron (1989) proposed three alternative models. These are: (i) model A (the crash model), which allows for a one-time structural break in the intercept of the trend function; (ii) model B (the changing growth model), which allows for a structural break in the slope; and (iii) model C (the crash-cum-growth model), which allows for a structural break in intercept and slope. Perron (1989) treated the structural break as exogenous. However, we follow the more recent published literature and treat the break point as being endogenous. We use two forms of the Zivot and Andrews (1992) sequential trend break model. In Perron’s (1989) original terminology, these are models A and model C. Model A has the following form: k

∆yt = κ + φyt −1 + βt + θ1DUt +

∑ d j ∆yt− j + εt

(5)

j =1

Model C takes the following form: k

∆yt = κ + φyt −1 + βt + θ1DUt + γ 1DTt +

∑ d j ∆yt− j + εt

(6)

j =1

Here ∆ is the first difference operator, εt is a white noise disturbance term with variance σ 2, and t = 1, . . . , T is an index of time. The ∆yt−j terms on the righthand side of equations (5) and (6) allow for serial correlation and ensure that the disturbance term is white noise. DUt is an indicator dummy variable for a mean shift occurring at time TB and DTt is the corresponding trend shift variable, where 1 if t > TB DUt =  0 otherwise and t − TB if t > TB DTt =  otherwise 0 The breakpoint is searched for over the range of the sample (0.15T, 0.85T). The null hypothesis is that φ = 0 in equations (5) and (6), which implies that the series {yt} is an integrated process without a structural break. The alternative hypothesis is that φ < 0, which implies that {yt} is break point stationary. The break point is selected by choosing the value of TB for which the t-statistic for φ is minimized. The results for the Zivot and Andrews (1992) tests are reported in Table 2. They find no additional evidence against the unit root hypothesis relative to the unit root tests without a structural break. In each case we are unable to reject the unit root null hypothesis at the 5 per cent level or better, confirming the series are I(1). © AFAANZ, 2004


429

Table 2 Zivot and Andrews test for unit roots with one structural break Stock price Model A INDIA (TB) φ θ γ k PAKISTAN (TB) φ θ γ k SRI LANKA (TB) φ θ γ k BANGLADESH (TB) φ θ γ k

23 April 1999 −0.0084 (−3.1782) 0.0044*** (2.4819) 5 22 April 1998 −0.0106 (−3.8579) −0.0064*** (−2.9654) 6 17 February 1997 −0.0090 (−4.4437) 0.0024*** (2.9766) 2 19 September 1997 −0.0060 (−3.0048) −0.0070*** (−2.8541) 7

Model C 26 April 1999 −0.013* (−4.6039) 0.0092 (0.0000) 4.2609*** (−3.9123) 5 22 April 1998 −0.0123 (−4.1679) −0.0069 (0.0000) −3. 1864* (1.5757) 6 12 May 1998 −0.0073 (−4.2915) −0.0024 (0.0000) −3.3684* (−1.8712) 6 14 February 1997 −0.0086 (−3.8477) −0.0134 (0.0000) −4.6190*** (−2.9717) 7

Note: The figure in parentheses are t-statistics The critical values for the structural break dummy variables follow the asymptotic standard normal distribution. Critical values for tφ are Model A: −5.34 (1%), − 4.8 (5%), − 4.58 (10%); Model C −5.57 (1%), −5.08 (5%), − 4.52 (10%) from Zivot and Andrews (1992). *** (*) denotes statistical significance at the 1 (10)% level. TB, time of break.

The break dates are statistically significant for model A and statistically insignificant for model C in each case. The break dates in India occur in April 1999 and have a positive coefficient. This was at a time when world stock markets were improving and was just prior to George Soros’ announcement in May 1999 that the Asian financial crisis was over (Ariff and Khalid, 2000, p. 473). In contrast, the break dates in Pakistan have a negative coefficient and occur in late April 1998. India and Pakistan conducted nuclear tests in May 1998, which had an adverse effect on the stock markets of both countries. The statistically significant break date for Bangladesh, with a negative coefficient, is on 19 September 1997, where a plunge on the Stock Exchange All Share followed substantial drops in South-East Asian stock markets in the same month. On September 8 1997, the Manilla stock market dropped 9.3 per cent and the Jakarta stock market dropped 4.5 per cent. On October 8 1997 Malaysia spent $US20bn to prop up its share market (Ariff and Khalid, 2000, p. 471). The statistically significant break date in Sri Lanka, with a positive coefficient, is in © AFAANZ, 2004

430


February 1997 prior to the financial crisis. This is associated with various financial sector and regulatory reforms in Sri Lanka during 1996–1997 including the introduction of an over-the-counter board system and partial capital account liberalization. 5.2. Multivariate cointegration We use the ARDL approach to cointegration in preference to the Engle and Granger (1987) or Johansen (1988) methods because the latter are not robust for small sample sizes. While we have 1800 observations, increasing the number of observations through using daily data does not add robustness to the cointegration results because what matters is the length of the period, rather than the number of observations. However, Pesaran and Shin (1999) show that with the ARDL framework, the ordinary least squares estimators of the shortrun parameters are T -consistent and the ARDL based estimators of the longrun coefficients are super-consistent in small sample sizes. The ARDL procedure involves two stages. The first stage is to establish that a long-run relationship exists among the variables in equations (1)–(4). The second stage involves estimating the long-run and short-run coefficients of equations (1)–(4) conditional on whether the variables are cointegrated. The mathematical derivation of the long-run and short-run parameters can be found in Pesaran et al. (2001). To implement the bounds test consider a vector of two variables: zt where zt = ( yt , xt′ )′ , yt is the dependent variable and xt is a vector of regressors. The data generating process of zt is a p-order vector autoregression. For cointegration analysis it is essential that ∆yt be modelled as a conditional Error-Correction Model: ∆yt = β0 + π yy yt −1 + π yx ⋅x xt −1 +

p

q

i =1

j=0

∑ ϑi ∆yt−i + ∑ φ′j ∆ xt− j + θwt + µt

(7)

Here, π yy and π yx are long-run multipliers, β0 is the drift and wt is a vector of exogenous components. Lagged values of ∆yt and current and lagged values of ∆x t are used to model the short-run dynamic structure. The ARDL procedure tests for the absence of any level relationship between yt and xt through exclusion of the lagged levels variables in equation (7). The null and alternative hypotheses for the absence of a conditional level relationship between yt and x t are: H0: π yy = 0, π yx·x = 0′,

(8)

H1: π yy ≠ 0, π yx·x ≠ 0′ or π yy ≠ 0, π yx·x = 0′ or π yy = 0, π yx·x ≠ 0′.

(9)

These hypotheses can be examined using the standard F-statistic. While the Ftest has a non-standard distribution, critical values are reported in Pesaran et al. (2001). Critical value bounds exist for all classifications of the regressors into purely I(1), purely I(0) or mutually cointegrated. If the computed F-statistic falls outside the critical bounds, a conclusive decision can be made regarding © AFAANZ, 2004


431

cointegration without knowing the order of integration of the regressors. If the estimated F-statistic is higher than the upper bound of the critical values then the null hypothesis of no cointegration is rejected. If the estimated F-statistic is less than the lower bound of the critical values then the null hypothesis of no cointegration cannot be rejected. Turning to the results from the unrestricted error correction model, for equation (1) FSPI(.) = 1.6771; for equation (2) FSPP(.) = 4.2371; for equation (3) FSPS(.) = 2.1490; and for equation (4) FSPB(.) = 0.6963. From Pesaran et al. (2001, p. 300), the relevant upper bound critical value is 3.7 and the relevant lower bound critical value is 2.8 at the 5 per cent level of significance. An advantage of the ARDL approach is that we can tell which series is the dependent variable from the F-test when cointegration exists. The F-test here shows that the null hypothesis of no cointegration among the variables in equation 2 cannot be accepted because FSPP(.) exceeds the upper bound critical value at the 5 per cent level. Therefore, there is a long-run relationship between the variables when stock prices in Pakistan is treated as the dependent variable. However, for equations (1), (3) and (4) the F-statistic is less than the lower bound critical value and the null hypothesis of no cointegration is accepted. Therefore, we conclude that there is no long-run relationship between the variables when stock prices in Bangladesh, India or Pakistan are taken as the dependent variable. The implication of the finding of cointegration when stock prices in Pakistan is the dependent variable is that the gains from international diversification for investors with long holding periods in South Asia are limited. However, as discussed in Section II, different securities will have varying degrees of financial risk and security cash flows will covary less than perfectly across national borders. This means that although the benefits of portfolio diversification in cointegrated markets might be restricted in the long run, in practice they are unlikely to be eliminated. In addition, investors can still reap arbitrage profits through international diversification of stocks in these countries in the short term. As far as the MEH, our finding that the four markets are cointegrated suggests that since each stock price series contains information on the common stochastic trends, which bind all the stock prices together, the predictability of one country’s stock prices can be enhanced considerably through utilizing information on the other countries’ stock prices. But because predictability does not imply anything about the risk-adjusted excess rate of return, this does not necessarily say anything about market efficiency (Masih and Masih, 2002, p. 87). 5.3. Elasticities and parameter stability The long- and short-run elasticities together with the relevant diagnostic tests for the short-run model when stock prices in Pakistan are the dependent variable are reported in Table 3. In both the long run and short run, an increase in stock prices in India and Bangladesh exert a positive and statistically significant impact on stock prices in Pakistan. An increase in stock prices in Sri Lanka has © AFAANZ, 2004

432


Table 3 Long-run and short-run results from the ARDL model Long-run coefficients (lnSPPt is the dependent variable) Regressor lnSPIt lnSPBt lnSPSt

Coefficient 0.7738*** 0.2564** 0.1498

t-statistic 5.0630 2.2569 0.7581

Short-run coefficients Regressor lnSPIt lnSPBt lnSPSt ECTt−1

Coefficient

t-statistic

0.1105*** 0.0026** 0.2062*** −0.0103***

4.2561 2.0014 3.9339 −3.6667

Diagnostic tests R2 -2 σ Durbin–Watson χ2Auto (1) χ2Norm (2) χ2Hetero (3) χ2RESET (4)

0.0281 0.0254 0.0180 1.9076 0.0329 2.3827 2.6289 2.3887

Notes: lnSPP, lnSPI, lnSPB and InSPS are the natural logs of the stock prices in Pakistan, India, Bangladesh and Sri Lanka, respectively The critical values for χ2(1) = 3.84 and χ2(2) = 5.99 at the 5% significance level. **(***) denotes statistical significance at the 5 (1) per cent level. ARDL, autoregressive distributive lag. Variables are defined in the text.

an insignificant impact on stock prices in Pakistan in the long run but a statistically significant impact in the short run. In the short-run model the one period lag on the error correction term is statistically significant with a negative sign. This confirms the results from the cointegration test that there is a long-run relationship between the stock prices of the four countries when stock prices in Pakistan is the dependent variable. The small magnitude of the coefficient on the lagged error-correction term suggests that once shocked, convergence to equilibrium is very slow. While the adjusted coefficient of determination is small, which is an expected outcome from this exercise, the diagnostic tests perform well, supporting the overall validity of the short-run model. To test the stability of the long-run parameters we used the suite of tests suggested by Hansen (1992) which can be applied when all variables are I(1). Hansen (1992) recommends three tests for parameter stability: SupF, MeanF, and L c. These tests vary in terms of the alternative hypothesis, but all have © AFAANZ, 2004


433

the same null hypothesis that the parameters are stable. When the calculated probability values are greater than 0.05 the null hypothesis is accepted. The results of the Hansen (1992) tests for the long-run equation where stock prices in Pakistan is the dependent variable suggested the parameters were stable with all test statistics greater than 0.05.4 5.4. Granger causality Following Granger (1969) SPIt (stock prices in India) are said to be Grangercaused by SPPt (stock prices in Pakistan) if the information in the past and present values of SPPt helps to improve the forecast of the SPIt variable; that is, if MSE(Yt | Ωt ) < MSE(Yt | Ω′t ) where MSE is the conditional mean square root of the forecast of SPt. Here, Ωt denotes the set of all relevant information up to time t, while Ω′t excludes the information in the past and present values of SPIt. When the Bangladesh, Indian and Sri Lankan markets are taken as the dependent variable it is inappropriate to include a lagged error-correction term because there is no long-run relationship when these variables are the dependent variable. However, where the series are integrated of order one, in the presence of cointegration, Vector Autoregression estimation in first differences will be misleading (Engle and Granger, 1987). Therefore, when the Pakistani market is the dependent variable the Granger causality test is augmented with a lagged error correction term and estimated within a Vector Error-Correction framework. In each case the dependent variable is regressed against past values of itself and other variables and the optimal lag length is chosen on the basis of the Schwarz Bayesian Criterion. Table 4 presents the results for short-run and long-run Granger causality. The Wald F-test of the explanatory variables indicates the significance of the shortrun causal effects, while the t-statistic on the coefficient of the lagged errorcorrection term when Pakistani stock prices is the dependent variable indicates the significance of the long-run causal effects. In the Pakistani stock price equation the coefficient on the lagged error correction term is significant at the 1 per cent level with a negative sign. Therefore, in the long run, stock prices in Bangladesh, India and Sri Lanka Granger-cause stock prices in Pakistan. This means that the Pakistani market bears the burden of any disturbance in the long-run equilibrium relationship. The error correction coefficient is −0.107, which implies that once shocked, convergence to equilibrium is slow. The finding that there is a single long-run relationship when the Pakistani stock price is the dependent variable and that Pakistan is the most endogenous (dependent) market in the long run suggests that stock prices in Pakistan are dependent on movements in stock prices in the other three countries. The likely reason for this lies in the haphazard implementation of government policies which have retarded the attractiveness of the Pakistani market to domestic and

4

The results of the Hansen (1992) test are available from the authors on request.

© AFAANZ, 2004

434


Table 4 Granger causality tests: F-statistics (probability values) Dependent variable

lnSPIt

lnSPIt lnSPPt lnSPSt lnSPBt

0.4262 (0.7342) 1.1462 (0.3291) 0.2294 (0.8759)

lnSPPt

lnSPSt

lnSPBt

6.0938*** (0.0004)

2.3399* (0.0716) 1.7242 (0.1600)

0.2719 (0.8457) 0.2747 (0.8437) 0.4378 (0.7259)

2.9311** [0.0324] 0.9581 (0.4117)

ECTt−1 (t-statistics)

−0.0107*** (−3.8162)

0.1304 (0.9421)

Note: lnSPI, lnSPP, lnSPS and lnSPB are the natural logs of the stock prices in India, Pakistan, Sri Lanka and Bangladesh, respectively. *** denotes statistical significance at the 1% level. not applicable. Variables are defined in the text.

overseas investors. The lead-lag relationship between stock markets is related to the liquidity of the market, market capitalization and the level of transaction costs which depend, at least partly, on the level of openness and deregulation of the economy. One expects that those markets which are leaders would have relatively higher levels of liquidity and market capitalization and lower levels of transaction costs (Masih and Masih, 2002). While the size of Pakistan’s stock market is similar to that in Sri Lanka, it lags behind India. In 1991 Pakistan introduced regulatory measures to improve the effectiveness of its stock markets and attract more capital. Pakistan first established an auction market for short-term treasury bills and long-term federal investment bonds. In addition, a secondary market for government securities was also established in the early 1990s. These policies were initially successful in increasing trading in 1991–1993. Market capitalization on the Karachi Stock Exchange alone increased from PRs38bn in 1987–1988 to PRs200bn in 1991–1992. Similar trends occurred on the Islamabad and Lahore Stock Exchanges (Ariff and Khalid, 2000, p. 370). However, political instability, which resulted in some reversals of government policies, in the following years resulted in a downward trend in trading. In 1992 there were 178 new companies listed in Pakistan and in 1993 there were 110 new companies listed, but by 1997 the number of new listings fell to 36 (Ariff and Khalid, 2000, p. 370). In 1997 a newly elected government introduced a new reform agenda and there were some signs of recovery up to May 1998, when Pakistan exploded a nuclear bomb. This resulted in economic sanctions and generated severe foreign exchange rate shortages. In the short run there is unidirectional Granger causality from stock prices in Pakistan to India, stock prices in Sri Lanka to India and from stock prices in Pakistan to Sri Lanka. Therefore, while Pakistan is the most dependent market in the long run it has a transitory effect on the Indian and Sri Lankan markets, © AFAANZ, 2004


435

reflected in the short-run results. Overall the results from the Granger causality analysis suggest that Bangladesh is the most exogenous of the four South Asian markets. For the other three markets at least one channel of Granger causality is active, either in the short run through the joint test of lagged differences (India and Sri Lanka) or in the long run through a statistically significant errorcorrection term (Pakistan). 6. Impulse response functions To this point our analysis has been restricted to within-sample tests. These can be used to examine the Granger exogeneity or endogeneity of the dependent variable in the sample period, but cannot ascertain the degree of exogeneity of the variables beyond the sample period (Masih and Masih, 2002). To determine the relative strength of the Granger-causal chain we consider generalized impulse response functions. Impulse response functions display graphically the expected response of each market to shocks in that market and shocks in neighbouring markets. This function enables characterization of the dynamic interactions among variables and allows us to observe the speed of adjustment of variables in the system. The impulse response of stock prices in Bangladesh, India, Pakistan and Sri Lanka to one-standard deviation shocks in stock prices in Bangladesh, India Pakistan and Sri Lanka over a 1-year period are presented in Figure 1. Beginning with the impulse response function for stock prices in Pakistan, a shock to stock prices in India initially leads to a rise in stock prices in Pakistan, which flattens out and gradually starts to decline after 2–3 months. Shocks to stock prices in Bangladesh exert a small, but constant, positive influence on stock prices in Pakistan, while shocks to stock prices in Bangladesh begin with a small positive effect, which declines over time. In the impulse response function for India, shocks to stock prices in Sri Lanka have a small positive effect, while shocks to stock prices in Bangladesh have a slightly larger negative effect on the Indian stock market. Shocks to stock prices in Pakistan initially have a small positive effect on stock prices in India for the first 1–2 months and then have a negative effect thereafter. From the impulse response function for Sri Lanka, it can be observed that shocks to stock prices in Bangladesh and Pakistan have a positive effect on stock prices in Sri Lanka, while shocks to stock prices in India have a positive effect for the first 50 days and then have a negative effect. Own shocks explain most of the variation in Bangladesh stock prices, with shocks to stock prices in Pakistan having a small positive impact and shocks to stock prices in Sri Lanka having a small negative effect. 7. Conclusion Our first major finding is that there is a long-run relationship between the stock prices of the four countries when stock prices in Pakistan is the dependent © AFAANZ, 2004

436


Figure 1 Impulse response functions. (a) Impulse response of SPP (stock prices in Pakistan) to one standard deviation innovation in SPP, SPI (stock prices in India), SPS (stock prices in Sri Lanka) and SPB (stock prices in Bangladesh) (b) Impulse response of SPI to one standard deviation innovation in SPI, SPP, SPS and SPB (c) Impulse response of SPS to one standard deviation innovation in SPS, SPI, SPP and SPB

© AFAANZ, 2004


437

Figure 1 (cont’d). (d) Impulse response of SPB to one standard deviation innovation in SPB, SPI, SPS and SPP

variable. The second main finding is that in the long run, stock prices in Bangladesh, India and Sri Lanka Granger-cause stock prices in Pakistan, meaning that Pakistan bears the burden of adjustment in the long-run equilibrium relationship. In the short run there is unidirectional Granger causality from stock prices in Pakistan to India, stock prices in Sri Lanka to India and from stock prices in Pakistan to Sri Lanka. Our third finding, from the Granger causality analysis is that Bangladesh is the most exogenous of the four South Asian markets. The impulse response functions support the finding that Bangladesh is the most exogenous market and further suggest that the Indian market is more exogenous than either the Pakistani or Sri Lankan markets. Since, at least the beginning of the 1990s South Asian markets have undergone rapid market liberalization, which has opened up the region to higher investment and trade flows. This should make our findings interesting to investors. The fact that investment and trade flows from Australasia to South Asia have increased so rapidly over this period means the findings are of particular interest to investors in these countries. References Ariff, M. and A. Khalid, 2000, Liberalization, growth and the Asian financial crisis (Edward Elgar, Cheltenham). Australian Bureau of Statistics (ABS), 2001, 2002, 2003, Yearbook Australia (Australian Bureau of Statistics, Canberra). Bekaert, G., C. R. Harvey and R. Lumsdaine, 2001, Dating the integration of world equity markets, Working Paper (Stanford University, Stanford, CA). Caporale, G. and N. Pittis, 1998, Cointegration and predictability of asset prices, Journal of International Money and Finance 17, 441– 453. Chen, G. M., M. Firth and O. M. Rui, 2002, Stock market linkages: evidence from Latin America, Journal of Banking and Finance 26, 1113 –1141. Chowdhury, A. R., 1994, Statistical properties of daily returns from Dhaka stock exchange, Bangladesh Development Studies 26, 61–76. © AFAANZ, 2004

438


Christofi, A. and A. Pericli, 1999, Correlation in price changes and volatility of major Latin American stock markets, Journal of Multinational Financial Management 9, 79 – 93. Dwyer, G. and M. Wallace, 1992, Cointegration and market efficiency, Journal of International Money and Finance 11, 318 –327. Elyasiani, E., P. Perera and T. N. Puri, 1998, Interdependence and dynamic linkages between stock markets of Sri Lanka and its trading partners, Journal of Multinational Financial Management 8, 89 –101. Engle, R. F. and C. W. J. Granger, 1987, Cointegration and error correction representation: estimation and testing, Econometrica 55, 251–276. Fernandez-Serrano, J. L. and S. Sosvilla-Rivero, 2001, Modelling evolving long-run relationships: the linkages between stock markets in Asia, Japan and the World Economy 13, 145–160. Goldberg, C. and F. Delgado, 2001, Financial integration of emerging markets: an analysis of Latin America versus South Asia using individual stocks, Multinational Financial Journal 5, 259 –301. Granger, C. W. J., 1969, Investigating causal relations by econometric models and crossspectral methods, Econometrica 37, 424 – 438. Granger, C. W. J., 1986, Developments in the study of cointegrated economic variables, Oxford Bulletin of Economics and Statistics 48, 213 –228. Grubel, H., 1968, Internationally diversified portfolio: welfare gains and capital flows, American Economic Review 58, 89–94. Hansen, B. E., 1992, Tests for parameter stability in regressions with I(1) processes, Journal of Business and Economic Statistics 10, 321–335. Hasan, P., 1998, Pakistan at the threshold of the 21st century: how to shape a better economic future, Pakistan Development Review 37, 85 –122. Indo-Australian Business Committee, 2003, Australia (Indo-Australian Business Committee, New Delhi). Johansen, S., 1988, Statistical analysis of cointegrating vectors, Journal of Economic Dynamics and Contrology 12, 231–254. Joint Standing Committee on Foreign Affairs, Defence and Trade (JSCFADT), 1998, Report on Australia’s trade relationship with india (AGPS, Canberra). Kasa, K., 1992, Common stochastic trends in international stock markets, Journal of Monetary Economics 29, 95 –124. Masih, A. M. M. and R. Masih, 1997, Dynamic linkages and the propagation mechanism driving major international stock markets: an analysis of the pre-and-post-crash eras, Quarterly Review of Economics and Finance 37, 859– 888. Masih, A. M. M. and R. Masih, 1999, Are Asian stock market fluctuations due mainly to intra-regional contagion effects? Evidence based on Asian emerging stock markets, Pacific Basin Finance Journal 7, 251–282. Masih, A. M. M. and R. Masih, 2002, Propagative causal price transmission among international stock markets: evidence from the pre-and-post globalization period, Global Finance Journal 13, 63–91. New Zealand Ministry of Foreign Affairs and Trade, 2003, Statistics New Zealand, Overseas Trade (New Zealand Ministry of Foreign Affairs and Trade, Wellington, New Zealand). Ng, T. H., 2002, Stock market linkages in South-East Asia, Asian Economic Journal 16, 353–377. Perron, P., 1989, The great crash, the oil price shock and the unit root hypothesis, Econometrica 57, 1361–1401. Pesaran, M. H. and Y. Shin, 1999, An autoregressive distributed lag modelling approach to cointegration analysis, in: S. Strom, ed., Econometrics and Economic Theory in the 20th Century: The Ragnar Frisch Centennial Symposium (Cambridge University Press, Cambridge). © AFAANZ, 2004


439

Pesaran, M. H., Y. Shin and R. Smith, 2001, Bounds testing approaches to the analysis of level relationships, Journal of Applied Econometrics 16, 289 –326. Shand, R., 2000, India: A new beginning, Basham lecture, Faculty of Asian Studies (Australian National University, Canberra). Available: http://www.anu.edu.au /asianstudies/ bash2.html. Accessed 10 September 2003. Shetty, A. G., 2002, The Asian financial crisis and India, Finance India 16, 541–555. Standard and Poors, 2002, Emerging Stock Markets Factbook 2002 (Standard and Poors, New York). Statistics New Zealand, 2002, New Zealand Official Yearbook (Statistics New Zealand, Wellington). United Nations, 1999, Statistical Yearbook 46th Issue 1999 (United Nations, New York). USAID [Online]. Available: http://www.usaid.gov/regions/ane/newpages/perspectives/ srilanka /srilmkts.htm. Accessed 10 September 2003. Vicziany, M., 1993, Australian companies in India: The ingredients for successful entry into the Indian market, in: M. Vicziany, ed., Australia-India: Economic Links Past, Present and Future (Indian Ocean Centre for Peace Studies, Perth). Vicziany, M., 2000, Coal, beer, blood and condoms, Australian Contributions to India’s Economic Growth, South Asia, Special Issue 23, 239 –253. Vicziany, M. and S. Chaterjee, 1999, Australian business attitudes to India in the late 1990s, In: S. Neelmagham, ed., Enterprise Management: New Horizons in Indo-Australian Collaboration (Tata McGraw-Hill, New Delhi). Von Furstenberg, G. M. and B. N. Jeon, 1989, International stock price movements: links and messages, Brookings Papers on Economic Activity 1, 125–179. World Bank, 1999, Global Development Finance (World Bank, Washington DC). Zivot, E. and D. Andrews, 1992, Further evidence of the great crash, the oil-price shock and the unit-root hypothesis, Journal of Business and Economic Statistics 10, 251–270.

© AFAANZ, 2004