2506 Hooi Granger

8/11/2019 2506 Hooi Granger

1/23

EXCHANGE RATE AND STOCK PRICE INTERACTION IN MAJOR ASIAN MARKETS:EVIDENCE FOR INDIVIDUAL COUNTRIES AND PANELS ALLOWING FOR STRUCTURAL

BREAKS

Hooi Hooi Lean1, Paresh Narayan

2and Russell Smyth

3

Discussion Paper 25, 2006

ABSTRACT

This paper examines the relationship between exchange rates and stock prices in eight Asiancountries using cointegration and Granger causality tests over the period 1991 to 2005. We test forcointegration and Granger causality for both individual countries using the Gregory and Hansen(1996) cointegration test that accommodates a structural break in the cointegrating vector, and fora panel using the Westerlund (2006) panel Lagrange multiplier (LM) cointegration test that allowsfor multiple structural breaks in the level of the individual cointegrating equations. We find little

evidence of cointegration. Our results for individual countries suggest that the only country forwhich exchange rates and stock prices are cointegrated over the entire period is Korea wherethere is weak long-run uni-directional Granger causality running from exchange rates to stockprices. Employing the panel LM cointegration test with multiple structural breaks we find thatexchange rates and stock prices are not cointegrated. We conclude that for the eight countriesexchange rates and stock prices primarily have only a contemporaneous effect on each other thatis reflected in the short-run intertemporal co-movements between these financial variables.

JEL CODE: F31 G15

1Department of Economics, Monash University, Australia and Economics Program, School of Social

Sciences, Universiti Sains Malaysia, Malaysia.2School of Accounting, Finance and Economics, Griffith University, Australia.

3Department of Economics, Monash University, Australia.


2/23

EXCHANGE RATE AND STOCK PRICE INTERACTION IN MAJOR ASIAN MARKETS:EVIDENCE FOR INDIVIDUAL COUNTRIES AND PANELS ALLOWING FOR STRUCTURAL

BREAKS

1 INTRODUCTION

There are two competing perspectives on whether exchange rates Granger cause stock prices orvice-versa. The traditional approach states that exchange rates Granger cause stock prices. Thisapproach postulates that changes in the exchange rate will affect the competitiveness of a firm,which in turn will influence the firms earnings and net worth and stock prices in general (see eg.Dornbusch and Fischer, 1980). The portfolio approach states that stock prices Granger causeexchange rates. The portfolio perspective suggests that rising (falling) stock prices influence capitalflows from foreign investors who substitute local (foreign) currency for foreign (local) currency (seeeg. Frankel, 1993). Thus, a rise (fall) in stock prices will lead to an appreciation (depreciation) ofthe exchange rate due to an increase in the demand (supply) of foreign currency.

There is a growing empirical literature that examines the interaction between exchange rates andstock prices. There are studies for advanced market economies (see eg. Bahmani Oskooee andSohrabian, 1992; Ajayi and Mougoue, 1996; Nieh and Lee, 2001); countries in Central and EasternEurope (Bahmani-Oskooee and Domac, 1997; Grambovas, 2003); countries in South Asia (see eg.Smyth and Nandha, 2003; Mishra, 2004; Narayan and Smyth, 2005; Venkateshwarlu and Tiwari,

2005) and countries in North and East Asia (see eg. Yu, 1996; Abdalla and Murinde, 1997; Ajayi etal., 1998; Granger et al., 2000; Ibrahim, 2000; Wu, 2000; Hatemi-J and Roca, 2005; Ramasamyand Yeung, 2005).

Overall, the existing literature is inconclusive regarding the causal relationship between exchangerates and stock prices. This is true for the subset of studies on this topic for countries in Asia. Forexample, Abdalla and Murinde (1997) examined the relationship between exchange rates andstock prices in Korea, Philippines, India and Pakistan using monthly data over the period January1985 to July 1994. They found that there was a long-run equilibrium relationship betweenexchange rates and stock prices in India and the Philippines only and that exchange rates Grangercaused stock prices in India, while stock prices Granger caused exchange rates in the Philippines.Granger et al. (2000) examined the relationship between exchange rates and stock prices for nine

Asian countries during the Asian financial crisis and found that exchange rates and stock priceswere not cointegrated in any of the countries. Ibrahim (2000) considered the relationship betweenexchange rates and stock prices in Malaysia using monthly data from January 1979 to June 1996.He found that exchange rates and stock prices were not cointegrated when examined in a bivariatecontext, but, in a multivariate setting, exchange rates, foreign reserves, stock prices and the moneysupply were cointegrated and that the money supply and foreign reserves Granger caused theexchange rate. Hatemi-J and Roca (2005) used bootstrap causality tests with leveragedadjustments to examine the links between exchange rates and stock prices in Malaysia, Indonesia,Philippines and Thailand in the periods immediately before and during the Asian financial crisis.They found that prior to the crisis, exchange rates Granger caused stock prices in Indonesia andThailand, while the reverse was true in Malaysia, but during the crisis there was no significant linkbetween the variables.


3/23

The purpose of this paper is to provide further evidence on the relationship between exchangerates and stock prices for eight emerging and developed Asian markets; namely, Hong Kong,Indonesia, Japan, Korea, Malaysia, the Philippines, Singapore and Thailand, over the period 1991to 2005. Standard and Poors (2004) classify Hong Kong, Japan and Singapore as developedmarkets and the others as emerging markets. The main contribution of the paper is that in addition

to using the Gregory and Hansen (1996) test and Granger causality to examine the relationshipbetween exchange rates and stock prices for individual countries, for the first time in the literatureon exchange rate stock price interaction we examine the relationship between exchange rates andstock prices in a panel cointegration and Granger causality framework. Because of the possibilityof structural breaks, we use the panel Lagrange multiplier (LM) cointegration test proposed byWesterlund (2006) that can accommodate multiple structural breaks in the level of the individualcointegrating equations. Panel cointegration tests have increased power because they combineinformation from cross-sectional as well as time series data.

The remainder of the paper is set out as follows. In the next section we provide an overview of theexchange rate regimes and stock markets in each of the eight Asian countries and describe thedata used in the study. Following this, we begin by considering the interaction between exchangerates and stock prices in a cointegration and Granger causality framework for individual countries.We then proceed to examine the interaction between stock prices and exchange rates for the eightAsian countries in a panel cointegration and Granger causality framework. Foreshadowing ourmain findings, we find that exchange rates and stock prices primarily have only acontemporaneous effect on each other, reflected in the short-run interaction effects between thesefinancial variables.

2 OVERVIEW OF THE MARKETS

We use weekly stock market indices and nominal exchange rates in terms of local currency relativeto the US dollar for Hong Kong, Indonesia, Japan, Korea, Malaysia, the Philippines, Singapore andThailand. Weekly data are used to avoid representation bias from some thinly-traded stocks. Usinglocal currency per US dollar is important to avoid rounding-up errors, which is particularly pertinentfor countries with large denominations such as Indonesia and Korea (Ramasamy and Yeung,2005). The sample is from January 1, 1991 to June 30, 2005, which means there are 757observations in total. All data are extracted from DataStream and transformed into logarithmicscale prior to analysis.

Table 1 provides some key indicators of the stock markets in these countries. Japan, Hong Kongand Korea have the three largest stock markets of the countries based on market capitalization,

value traded and the number of listed companies. Malaysia, Singapore and Thailand sit behindthese three in terms of size with the stock markets in Indonesia and the Philippines being muchsmaller according to all indicators. Table 2 shows the exchange rate classification for each of thecountries over the timeframe of the study. Singapore, Korea, Indonesia, Japan and the Philippineshad a floating or managed floating exchange rate over the entire period. The Hong Kong dollar waspegged to the US dollar. The Thai Baht was fixed to a basket of currencies until the Asian financialcrisis and has been subject to a managed float since July 1997. The Malaysian Ringgit was subjectto a managed float until the Asian financial crisis and has been pegged to the US dollar sinceSeptember 1998.


4/23

Table 1: Key Indicators of Stock Markets of Countries in the Sample, 2003

Indonesia Korea HongKong

Japan Malaysia Philippines Sin

Market Capitalization(a)

54,659 (36th

) 329,616 (15th

) 714,597 (9th

) 3,040,665(2

nd)

168,376(23

rd)

23,565 (44th

) 145

Total Value Traded

(a)

14,774 (36

th

) 682,706 (7

th

) 331,615(15th

) 2,272,989(2nd

) 50,135 (29

th

) 2,635 (47

th

) 87,8

Number of Listed

Domestic Companies

333 (29th

) 1,563 (8th

) 1,029 (11th

) 3,116 (6th

) 897 (13th

) 234 (39th

) 475

Average Company Size(a)

164.1 (49th

) 210.9 (43rd

) 694.5 (18th

) 975.8 (15th

) 187.7 (46th

) 100.7 (57th

) 305

Notes: (a) Figures are in $US million.(b) Figures in parenthesis are world rankings

Source: Standard and Poors (2004)

Table 2: Exchange Rate Classification of Countries in the Sample

Country Currency Exchange Regime Classification

Thailand Baht FB until 07/97 then MF

Singapore Dollar MF

Philippines Peso F

Malaysia Ringgit MF until 09/98 then FU

Korea Won MF until 12/97 then F

Indonesia Rupiah MF until 08/97, F until 2000 then MF

Hong Kong Dollar FU

Japan Yen MF

Notes: F is freely floating; MF is a managed float; FB is fixed to a basket, FU is pegged to the $US.Sources: IMF http://www.imf.org/external/np/mfd/er/index.aspand the Chinese University of Hong Kong,

http://intl.econ.cuhk.edu.hk/exchange_rate_regime/index.php?cid=8
http://www.imf.org/external/np/mfd/er/index.asphttp://www.imf.org/external/np/mfd/er/index.asp


5/23

3 COINTEGRATION AND GRANGER CAUSALITY FOR INDIVIDUAL COUNTRIES

3.1 Methodology

3.1.1 Univariate LM Unit Root Test

We begin through examining the stationary properties of the exchange rates and stock priceseries. Most existing studies of the relationship between exchange rates and stock prices use theAugmented Dickey-Fuller (ADF) or Phillips-Perron unit root tests to ascertain the order ofintegration of the series. A problem with these tests is that neither allows for the possibility of astructural break. Perron (1989) showed that the power to reject a unit root decreases when thestationary alternative is true and a structural break is ignored. Perron (1989) developed an ADF-type unit root test with one exogenous structural break and Zivot and Andrews (1992) developed

an ADF-type unit root test with one endogenous structural break. Granger et al. (2000) andHatemi-J and Roca (2005), which are two studies which examine the relationship betweenexchange rates and stock prices that use a unit root test with a structural break, employ the Zivotand Andrews (1992) and Perron (1989) unit root tests respectively. However, both of these testshave the limitation that the critical values are derived while assuming there is no break under thenull. Nunes et al. (1997) showed that this assumption leads to size distortions in the presence of aunit root with a break. As a result, utilizing ADF-type tests one might conclude that a time series istrend stationary, when in fact it is non-stationary with a break, meaning that spurious rejectionsmight occur. To examine the stationarity properties of the data for individual countries we employthe LM unit root test with one structural break proposed by Lee and Strazicich (2004). In contrast tothe Perron (1989) and Zivot and Andrews (1992) ADF-type unit root tests, the LM unit root test hasthe major advantage that its properties are unaffected by the existence of a structural break under

the null (see Lee and Strazicich, 2001, 2004).

The LM unit root test can be explained using the following data generating process

(DGP):t t ty Z e= + 1t t te e, = + . Here, consists of exogenous variables andtZ t is an error

term with classical properties. Lee and Strazicich (2004) developed two versions of the LM unitroot test with one structural break. Using the nomenclature of Perron (1989), Model A is known asthe crash model, and allows for a one-time change in the intercept under the alternative

hypothesis. Model A can be described by [ ]'

1, ,t tZ t D= , where 1tD = for and zero

otherwise; T

1,B

t T +

B is the date of the structural break, and ' = (B 1, 2, 3). Model C, the crash-cum-growth model, allows for a shift in the intercept and a change in the trend slope under the

alternative hypothesis and can be described by [ ]'

1, , ,t t tZ t D DT= , where forand zero otherwise.

t BDT t T= 1,Bt T +

The LM unit root test statistic is obtained from the following regression:

tttt SZy ++= 1

where ttxttZyS = , ;T,...,t 2= are coefficients in the regression of ty on tZ ; x is

given by tt Zy ; and and represent the first observations of and respectively. The

LM test statistic,

1y 1Z ty tZ

% , is given by t-statistic for testing the unit root null hypothesis that 0= . The

location of the structural break is determined by selecting all possible break points for the

minimum t-statistic as follows:

( )BT


6/23

( ) ( )iinf inf

=% % , where TTB= .

The search is carried out over the trimming region (0.15T, 0.85T), where T is sample size. Toselect the lag length, we used the general to specific procedure proposed by Hall (1994). We setthe maximum number of lags equal to eight and used the 10 per cent asymptotic normal value of1.645 to ascertain the statistical significance of the last first-differenced lagged term. After decidingthe optimal lag length for each breakpoint, we determined the break where the endogenous LM t-test statistic is at a minimum. Critical values for the LM unit root test with one structural break aretabulated in Lee and Strazicich (2004).

3.1.2 Cointegration

Once the order of integration of each variable is ascertained, we test for cointegration.

St= + Et+ ut (1)where Stand Etdenote the natural log of stock index and exchange rate and utis the error term.Gregory and Hansen (1996) propose three models for testing cointegration where they allow forthe existence of a structural break in the cointegrating vector.

The first contains a level shift (Model C):

1 2t t t t uS D E = + + + 1,..., n=

S D t E

= + + + + 1,...,t n

, t (2)

The second model contains a level shift and trend (Model C/T):

1 2 0 1t t t t u , = (3)

Here for0tD = t < and for1tD

= t . The intercept before the level shift is denoted as 1 ,

while 2 is the change in intercept due to the level shift.

The third model allows for a regime shift (Model C/S):

1 2 0 1 2t t t t t t uS D t E E D = + + + + + 1,...,t n, = (4)

Here, 1 and 2 are as in Equations 2 and 3. 1 denotes the cointegrating slope coefficient

before the regime shift and 2 denotes the change in the slope coefficient. In order to test for

cointegration between Stand Etwith structural change, i.e. the stationarity of in Equations 24,

Gregory and Hansen (1996) propose a suite of tests. These statistics are the commonly used ADF

statistics and extensions of the

tu

Z and tZ test statistics proposed by Phillips (1987). These

statistics are defined as:

( )* infT

ADF ADF

=

( )* infT

Z Z

=

( )* inft tT

Z Z

=


7/23

As the break point,, is unknown a priori, the model is estimated recursively allowing the breakpoint to vary between (0.15T, 0.85T), where T is the sample size. The null hypothesis of nocointegration is examined using the three statistics with interest in the smallest values for the threestatistics across all break points required to reject the null.

3.1.3 Granger Causality

Once it is established whether or not there is a long-run relationship between the series, we testwhether there is Granger causality between exchange rates and stock prices. If exchange rate andstock price are cointegrated, an error correction term should be included in the bivariateautoregression model as follows (Granger, 1988):

= =

++++=n

i

m

i

ttitiitit ECTESS1 1

111210 (5)

= =

++++=m

i

n

i

ttitiitit ECTESE1 1

212210 (6)

Here is changes in the exchange rate andtE tS is changes in the stock price. ECTt-1, which isSt-1 Et-1, is an error correction term derived from the long run cointegrating relationship inEquation (1). The error correction term can be estimated by using the residual from a cointegratingregression. The estimated 1 and 2 denote the speed of adjustment. If cointegration does notexist, the error correction term is dropped from the bivariate autoregression model. The decisionrule is reject (accept) H0:21= 22 = ..=2m= 0, meaning exchange rates do (do not) Grangercause stock prices, and reject (accept) H0:11= 12 = ..=1m= 0, meaning stock prices do (do not)Granger cause exchange rates. The lag structure is determined with the minimum final predictionerror criterion.

4 RESULTS AND DISCUSSION

The results for the LM unit root test with one structural break are presented in Tables 3 and 4. Inboth Model A and Model C the exchange rate and stock price index in each of the countries isintegrated of order one (I(1)) at the 5 per cent level or better. We briefly discuss the location of thebreakpoints. In Model A, the break in the intercept is statistically significant at the 5 per cent levelor better for both the exchange rate and stock index for each of the eight countries. In Model C,except for stock prices in Indonesia and Korea, the break in the intercept and/or slope isstatistically significant at the 5 per cent level or better in each case. With the exception of stock

prices in Hong Kong and Japan in Model A and stock prices in Indonesia and exchange rates inJapan and Hong Kong in Model C, the structural break is associated with the Asian financial crisis.The structural break in stock prices in Hong Kong and Japan in Model A occurs at the time of the11 September 2001 terrorist attacks in New York and Washington. The structural break inexchange rates for Hong Kong in Model C occurs a few months prior to the collapse of the internetbubble. The structural break in Indonesian stock prices is nestled between the Enron andWorldCom collapses as well as the 11 September 2001 and Bali bombing terrorist attacks.


8/23

Table 3: LM Unit Root Test, Model A

TB k St-1 BBt

Hong KongStock Index

09/05/01 4 -0.0071(-1.6883)

-0.1394

***

(-4.1187)

Hong Kong

Exchange Rate

10/15/97 5 -0.0093

(-1.5906)

0.0014**

(2.1133)

Indonesia

Stock Index

11/05/97 4 -0.0131

(-2.7313)

-0.0815**

(-2.3903)

Indonesia

Exchange Rate

12/31/97 6 -0.0151

(-2.4151)

0.6095***

(15.2351)

Japan

Stock Index

09/05/01 0 -0.0201

(-2.7734)

-0.1025***

(-3.4530)

Japan

Exchange Rate

03/25/98 2 -0.0101

(-2.0706)

0.0387***

(2.6384)

Korea

Stock Index

12/17/97 7 -0.0183

(-2.9282)

-0.1776***

(-4.2098)

Korea

Exchange Rate

12/03/97 8 -0.0148

(-2.7565)

0.2735***

(18.2783)

Malaysia

Stock Index

12/10/97 8 -0.0115

(-2.4497)

-0.1143***

(-3.0960)

Malaysia

Exchange Rate

12/31/97 8 -0.0087

(-2.0188)

0.1254***

(11.2545)

Philippines

Stock Index

12/02/98 8 -0.0032

(-1.2439)

-0.0882**

(-2.3819)

Philippines

Exchange Rate

12/10/97 8 -0.0068

(-1.6185)

0.1388***

(10.5677)

Singapore

Stock Index

08/26/98 8 -0.0094

(-2.1430)

-0.0941***

(-3.2258)

Singapore

Exchange Rate

06/03/98 2 -0.0049

(-1.4670)

0.0388***

(5.5894)

Thailand

Stock Index

07/22/98 3 -0.0038

(-1.3735)

-0.0929**

(-2.2774)

Thailand

Exchange Rate

02/11/98 4 -0.0087

(-2.2671)

0.0717***

(4.7404)

Notes: Critical values for the LM test at 10%, 5% and 1% significant levels = -3.211, -3.566, -4.239.Critical values for other coefficients based on standard t distribution = 1.645, 1.96, 2.576.*(

**)

***denote statistical significance at the 10%, 5% and 1% levels respectively.


9/23

Table 4: LM Unit Root Test, Model C

TB K St-1 BBt Dt

Hong Kong

Stock Index

11/26/97 7 -0.0220(-3.0240)

0.0697**

(2.0020)

-0.0113***

(-2.9604)

Hong KongExchange Rate

01/05/00 5 -0.0515(-3.8332)

0.0002(0.2619)

0.0004***

(3.5324)

Indonesia

Stock Index

06/19/02 4 -0.0181(-3.2664)

-0.0655*

(-1.9255)

0.0028

(0.8926)

Indonesia

Exchange Rate

12/10/97 7 -0.0278(-3.3658)

0.2496***

(5.9187)

0.0120**

(2.4183)

Japan

Stock Index

12/13/00 0 -0.0252(-3.1060)

-0.0812***

(-2.7338)

-0.0037

(-1.4012)

Japan

Exchange Rate

09/06/95 2 -0.0194(-2.8549)

0.0363**

(2.4817)

0.0034**

(2.5662)

KoreaStock Index

08/27/97 5 -0.0236(-3.3290) -0.0573(-1.3763) -0.0062*

(-1.6849)

Korea

Exchange Rate

11/05/97 6 -0.0311*

(-4.2912)

-0.0164

(-0.9969)

0.0063***

(3.1000)

Malaysia

Stock Index

10/08/97 8 -0.0270(-3.7040)

-0.0426

(-1.1801)

-0.0131***

(-3.1281)

Malaysia

Exchange Rate

10/15/97 8 -0.0255(-3.2360)

0.0744***

(5.7949)

0.0044***

(2.6993)

Philippines

Stock Index

06/25/97 8 -0.0214(-3.1550)

-0.0165

(-0.4507)

-0.0215***

(-3.5362)

Philippines

Exchange Rate

09/10/97 7 -0.0301(-3.4372)

0.0235*

(1.7046)

0.0077***

(3.4616)

Singapore

Stock Index

08/06/97 8 -0.0212(-3.1696)

-0.0347

(-1.1908)

-0.0086***

(-2.7546)

Singapore

Exchange Rate

08/06/97 2 -0.0174(-2.7215)

0.0215***

(3.0600)

0.0028***

(3.1349)

Thailand

Stock Index

07/16/97 3 -0.0136(-2.6187)

-0.0681*

(-1.6586)

-0.0126**

(-2.2501)

Thailand

Exchange Rate

08/13/97 6 -0.0222(-3.4091)

0.0247*

(1.7614)

0.0047***

(2.7111)

Critical values

Location of break, 0.1 0.2 0.3 0.4 0.5

1% significance level -5.11 -5.07 -5.15 -5.05 -5.11



Notes: The critical values are symmetric around and (1-).*(

**)

***denote statistical significance at the 10%,

5% and 1% levels respectively.

Given that exchange rates and stock prices are I(1) for each of the countries we proceed to test forcointegration with a structural break in the cointegration vector using the Gregory and Hansen

(1996) test. The results are presented in Table 5. There are a range of break points across the teststatistics and models. We begin by discussing the location of the structural break. In general the


10/23

break occurs in either 1993/94 when there was a bout of investor profit taking from these marketsdespite generally positive economic conditions and strong corporate earnings growth throughoutthe region; during the Asian financial crisis or in the period between 2000 and 2002, which was aperiod of global economic downturn precipitated by a slowdown in the US economy. This periodcontained a number of events that drove stock prices lower including revelation of fraudulentpractices at Enron and WorldCom, the end of the internet bubble and terrorism and wars.

Turning to the findings for cointegration, for Hong Kong, Indonesia, Japan, Singapore andThailand, the null hypothesis of no cointegration between exchange rates and stock prices is notrejected with any of the test statistics for any of the three models (C, C/T, C/S). For Korea andMalaysia the null hypothesis of no cointegration between exchange rates and stock prices isrejected with the ADF* statistic using Model C/T at the 5 per cent and 10 per cent levelrespectively. For the Philippines the null hypothesis of no cointegration between exchange ratesand stock prices is rejected with theADF* statistic for all three models at the 5 per cent level and

with the *t

Z statistic with Models C and C/T at the 10 per cent level. Thus a clear finding is that

exchange rates and stock prices do not hold a long run equilibrium relationship, meaning they donot move together and may drift apart in the long run, in Hong Kong, Indonesia, Japan, Singapore

and Thailand. The results for Korea, Malaysia and the Philippines are inconclusive, but we proceedto conducting the Granger causality testing on the basis that there is a long run equilibriumrelationship between exchange rates and stock prices in Korea, Malaysia and the Philippines.

i

Table 5: Gregory and Hansen (1996) Test for Cointegration with a Structural Break

Country Model ADF*

k TB Z*t TB Z

* TB

Hong Kong C -3.9649 3 06/16/93 -3.6605 06/09/93 -23.50 06/09/93

C/T -4.0918 2 11/14/01 -3.8197 09/26/01 -28.24 09/26/01

C/S -4.2964 2 07/07/93 -4.4660 07/19/95 -39.02 07/19/95

Indonesia C -2.7780 4 01/29/03 -2.1248 03/05/03 -10.17 03/05/03C/T -4.6871 6 12/20/00 -3.7377 06/21/00 -26.77 06/21/00

C/S -3.4392 3 02/19/03 -2.9652 03/05/03 -17.54 03/05/03

Japan C -4.0416 3 04/04/01 -3.7856 03/14/01 -26.88 03/14/01

C/T -3.9460 3 04/04/01 -3.6968 03/14/01 -28.30 02/14/96

C/S -4.1375 7 05/02/01 -3.8666 03/14/01 -28.09 03/14/01

Korea C -4.2127 7 12/30/98 -3.2467 01/20/99 -20.67 01/20/99

C/T -5.0100**

7 09/25/02 -3.4587 04/28/93 -23.51 04/28/93

C/S -4.5935 7 10/01/97 -3.3086 01/20/99 -21.22 01/20/99

Malaysia C -3.9179 6 02/16/94 -3.4324 10/08/97 -23.35 09/08/93

C/T -4.7449* 7 07/21/93 -3.9223 09/08/93 -30.08 09/08/93

C/S -3.9417 6 02/16/94 -3.4719 11/12/97 -23.06 09/08/93

Philippines C -4.8352**

3 05/26/93 -4.4506*

08/11/93 -34.82 08/11/93

C/T -5.2978**

3 05/26/93 -4.7301*

08/11/93 -39.78 08/11/93

C/S -4.9671**

3 05/26/93 -4.5205 08/11/93 -37.47 06/09/93

Singapore C -4.0738 0 02/03/99 -4.1580 01/27/99 -33.49 01/27/99

C/T -4.2186 0 02/03/99 -4.3115 01/27/99 -35.96 01/27/99

C/S -4.1959 8 01/27/99 -4.2424 01/27/99 -34.69 01/27/99

Thailand C -4.0441 2 04/23/03 -3.7952 04/23/03 -25.75 04/23/03

C/T -4.5345 2 10/23/96 -4.2376 08/07/96 -34.53 08/07/96

C/S -4.1395 2 06/26/02 -3.8229 06/26/02 -28.25 07/02/97

Note:*(

**) (

***) denotes statistical significance at the 10(5)(1)% level.


11/23

Table 6 show the results for Granger causality between stock prices and exchange rates over thewhole time period. The F-test indicates the significance of the short-run causal effects. ForIndonesia, Korea and Thailand there is short-run bi-directional Granger causality betweenexchange rates and stock prices. For Hong Kong, Malaysia and Singapore there is short-run uni-directional Granger causality running from exchange rates to stock prices. In the Philippines thereis short-run uni-directional Granger causality running from stock prices to exchange rates and inJapan there is neutrality between the exchange rate and stock prices. For Korea, Malaysia and the

Philippines the t-statistic for coefficient on the lagged error correction term indicates thesignificance of the long-run causal effects. If exchange rates and stock prices are cointegrated,there must be Granger causality in at least one direction, but it does not indicate the direction oftemporal causality between the variables (Granger, 1988). For Malaysia and the Philippines, the t-statisticon the long-run disequilibrium terms are statistically insignificant, suggesting there is nolong-run co-movement between exchange rates and stock prices in these countries. This findingreflects the inconclusive results from the cointegration tests. In Korea, in the long run there is aweak causal effect with uni-directional Granger causality running from exchange rates to stockprices at the 10 per cent level, consistent with the traditional view.

Table 6: Granger Causality of Stock Indices and Exchange Rates (Full Sample Period)

Country Granger Cause n M F-statistica

t-statisticb

Hong Kong ES 4 3 2.48*

n.a.

SE 5 1 0.05 n.a.

Indonesia ES 3 2 8.77***

n.a.

SE 6 5 4.04***

n.a.

Japan ES 1 1 0.60 n.a.

SE 3 1 0.79 n.a.

Korea ES 3 6 3.13***

-1.71*

SE 5 1 20.64*** 0.54

Malaysia ES 5 2 4.15**

-0.52

SE 3 1 0.66 0.99

Philippines ES 5 1 0.05 -0.77

SE 5 6 2.43**

0.36

Singapore ES 5 5 5.60***

n.a.

SE 3 1 0.00 n.a.

Thailand ES 4 5 2.66**

n.a.

SE 4 3 5.24***

n.a.

Notes: n and m are the optimal lag lengths.Implies Granger cause, e.g. E S implies exchange rate Granger causes stock index.a) F statistic for testing H0:21= 22 = ..=2m= 0orH0:11= 12 = ..=1m= 0b) t statistic for testing H0: 1= 0 or H0:2= 0 in ECM model.Note:

*(

**) (

***) denotes statistical significance at the 10(5)(1)% level.


12/23

Table 7: Granger Causality of Stock Indices and Exchange Rates (Sub-sample Period 1)

Country Granger Cause n m F-statistica

t-statisticb

Hong Kong ES 1 1 0.45 n.a.

SE 5 1 0.02 n.a.

Indonesia ES 3 2 9.39*** n.a.

SE 6 5 3.59***

n.a.


SE 3 1 0.02 n.a.

Korea ES 3 6 3.57***

-0.98

SE 5 5 4.47***

0.50

Malaysia ES 1 5 7.32***

-0.49

SE 6 6 4.85***

0.89


S

E 1 1 1.56 1.54Singapore ES 5 5 6.96

***n.a.

SE 4 6 2.35**

n.a.


n.a.

SE 2 3 4.11***

n.a.

Notes: See notes to Table 6

Table 8: Granger Causality of Stock Indices and Exchange Rates (Sub-sample Period 2)

Country Granger Cause n m F-statistica

t-statisticb

Hong Kong ES 4 3 2.66**

n.a.

SE 1 2 3.52**

n.a.

Indonesia ES 1 2 2.09 n.a.

SE 1 1 5.84**

n.a.


SE 1 6 2.22**

n.a.

Korea ES 1 1 0.25 -1.93*

SE 1 1 5.06**

-1.89*

Malaysia ES 5 6 8.16***

-1.99**

SE 6 4 5.23*** 3.31***


SE 5 6 2.85***

0.35

Singapore ES 1 6 1.93*

n.a.

SE 3 1 4.21**

n.a.


n.a.

SE 1 1 3.23*

n.a.

Notes: See notes to Table 6


13/23


14/23

Indonesia

0

2

4

6

8

10

12

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,e

xchange

rate

Japan

0

2

4

6

8

10

12

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,exchange

rate

Korea

0

1

2

3

4

5

6

7

8

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,exchange

rate


15/23

Malaysia

0

1

2

3

4

5

6

7

8

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,

exchange

rate

Philippines

0

1

2

3

4

5

6

7

8

9

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,exchange

rate

Singapore

0

1

2

3

4

5

6

7

8

9

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,exchange

rate


16/23

Thailand

0

1

2

3

4

5

6

7

8

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

stock

index,e

xchange

rate

Over the entire period, except for Korea, exchange rates and stock prices were not significantly

related in the long run. This result could mean that the transmission of information between thesemarkets was efficient or that the markets were segmented except for short-run contemporaneousco-movements in the markets other than Japan. Figure 1 plots the behaviour of exchange ratesand stock prices in each country. Apart from Korea, the two variables appear to be diverging,rather than converging, giving credence to the notion that the markets were segmented. Thisfinding indicates that it would only be possible to use the foreign exchange market as a hedge forinvestments in the stock market and vice-versa in the short-run except for Korea. The result alsosuggests that except for the short-run, for countries other than Korea, the stock market could notbe used as a policy base for intervention to stabilize the foreign exchange market and vice-versa.The only country for which there is a long-run equilibrium relationship is Korea, where therelationship between exchange rates and stock prices is consistent with the traditional view. Thisfinding suggests that in Korea, stock prices could be used to hedge foreign exchange investment

and that intervention in the exchange rate could be used to stabilize stock prices.

5 PANEL COINTEGRATION AND GRANGER CAUSALITY

5.1 Methodology

5.1.1 Panel LM Stationarity Test

We first implemented the panel stationarity test suggested by Hadri (2000), which is an extension

of the Kwiatowski et al. (1992) test. The Hadri (2000) test is based on the residuals from theindividual ordinary least squares (OLS) regressions:

itiiit ty ++= (7)

Given the residuals from the individual regressions, the LM statistic is:

( )

= =

N

it

i fTtSRN

LM1 0

22

1

1 (8)

where are the cumulative sum of the residuals,itSR


17/23

( ) =

=t

s

iti tSR

1

(9)

0f is the average of the individual estimators of the residual spectrum at frequency zero:

Nff

N

ii

== 100

(10)

Hadri (2000) shows that under mild assumptions,

( )( 1,0N

LMNZ

=

) (11)

5.1.2 Panel LM Cointegration Test

We employ the panel LM cointegration test with multiple structural breaks proposed by Westerlund(2006). Consider the following stock price-exchange rate long-run model:

,ititiijit eES ++= (12)

,re ititit +=

,rr iti1itit +=

The index denotes structural breaks. One can allow for at most breaks or

regimes, that are located at dates , where

1M,...,1j i+= iM1M i+ iiM1i T,...,T 1T 0i = and . The location

of the structural breaks are specified as a fixed fraction

TT 1iiM =+

( )1,0ij of such thatT TT ijij = andfor . Westerlund (2006) determines the structural breaks endogenously from

the data using the Bai and Perron (1998) technique, which globally minimises the sum of squaredresiduals to obtain the location of breaks:

ij1ij 1ijij

0:H i0 = for all ,N,...,1i=versus

0:H i1 for and1N,...,1i= 0i = for N,...,1Ni 1+=

The alternative hypothesis allows to differ across the cross-sectional units.i


18/23

The panel LM test statistic is defined as follows:

( ) ( ) =

+

= +=

N

1i

1iM

1j

ijT

11ijTt

2it

22.1i

21ijij STTMZ

where and , where is any efficient estimate of

. We use the fully modified ordinary least squares estimator (FMOLS) suggested by Phillips and

Hansen (1990) to estimate . The test statistic is written as a function of breaks to denote that it

is constructed for a certain number of breaks.

21i122i' 21i' 1.1i22.1i = +== t

11ijTk*ikit eS *ite

ite

ite

5.1.3 Panel Granger Causali ty

Once it is established whether exchange rates and stock prices are cointegrated, we examine thedirection of causality between exchange rates and stock prices within a panel data framework. Thepanel Granger causality test regression models are as follows:

= =

++++=n

1j

m

1j

it1itijitijjitijiit ECTESS (14)

= =

++++=m

1j

n

1j

it1itiiitijiitijiit ECTESE (15)

All variables are as defined above. If exchange rates and stock price are found not to becointegrated using the panel cointegration test, the error correction terms will be omitted.

6 RESULTS AND DISCUSSION

The results of the Hadri (2000) panel LM stationarity test are reported in Table 9. The panel LMstationarity test statistic for stock prices in the panel of eight countries is 22.4011, while the panelLM stationarity test statistic for the panel of eight countries for exchange rates is 50.7661, whichare both significant at 1 per cent. Thus, the joint null hypothesis of stationarity for both series isrejected, implying that both series are panel non-stationary.

Table 9 Hadri's (2000) Panel LM Stationarity Test

Panel LM Stationarity Test

Hadri z-statistic Stock Prices 22.4011 (p=0.000)

Hadri z-statistic Exchange Rates 50.7661 (p=0.000)


19/23

Table 10: Panel LM Cointegration Test with Multiple Structural Breaks

Panel LM Statistics 9.888 (full panel) 13.993 (excluding Korea)

Location of Structural Breaks

TB1 TB2 TB3 TB4 TB5

Hong Kong 12/1/1993 1/31/1996 4/1/1998 5/31/2000 7/31/2002

Indonesia 12/1/1993 1/31/1996 4/1/1998 5/31/2000 7/31/2002

Japan 12/1/1993 1/31/1996 4/1/1998 5/31/2000 7/31/2002

Korea 12/1/1993 1/31/1996 4/1/1998 5/31/2000 7/31/2002

Malaysia 11/24/1993 1/24/1996 3/25/1998 5/24/2000 7/24/2002

Philippines 11/24/1993 1/24/1996 3/25/1998 5/24/2000 7/24/2002

Singapore 11/24/1993 1/24/1996 3/25/1998 5/24/2000 7/24/2002

Thailand 11/24/1993 1/24/1996 3/25/1998 5/24/2000 7/24/2002Notes: The bootstrapped critical value at the 1% level is 2.218.

The results of the panel LM cointegration test are reported in Table 10. The panel LM cointegrationtest indicates that there are five structural breaks. The first structural break occurs in either Januaryor November 1993, the second structural break occurs in January 1996, the third structural breakoccurs in March or April 1998, the fourth structural break occurs in May 2000 and the fifth structuralbreak occurs in July 2002. These dates correspond broadly with the breaks identified in theGregory and Hansen (1996) test for individual countries and are associated with investor profittaking in 1993/94, the Asian financial crisis in 1998 and global economic downturn in 2000-2002.We examine whether exchange rates and stock prices in the full panel as well as a smaller panelexcluding Korea are cointegrated. In the smaller panel we exclude Korea given the earlier findingsuggesting there is a long-run relationship between exchange rates and stock prices in thatcountry. The panel test statistic is 9.888 for the full panel of eight countries and 13.993 for a panelof seven countries. The bootstrapped critical value at the 1 per cent level is 2.218. These resultssuggest that we are able to reject the null hypothesis that all the countries of the panel in the panelof eight or panel of seven (excluding Korea) are cointegrated. Thus, even allowing for multiplestructural breaks in the panel cointegration model we find there is no long-run equilibriumrelationship between exchange rates and stock prices.

Given the panel cointegration test did not reveal any evidence for a long-run relationship betweenstock prices and real exchange rates, in specifying the panel Granger causality model, we use aVAR framework. We obtain a panel F-test statistic of 7.2 for stock prices Granger causingexchange rates and a panel F-test statistic of 3.8 for causality running from exchange rates tostock prices. Both test statistics are statistically significant at the 1 per cent level. Taken together,

the panel Granger causality results suggest bi-directional short-run causality among stock pricesand exchange rates in the eight Asian countries.

7 CONCLUSION

The traditional and portfolio approaches represent competing hypotheses concerning therelationship between exchange rates and stock prices. We find little support for either hypothesisbased on the long-run results. There is little evidence that a long-run equilibrium relationshipbetween exchange rates and stock prices exists in the Asian markets studied for individual

countries and no evidence of cointegration for the countries as a panel, even allowing for structuralbreaks in the cointegrating equation. Most investors believe that exchange rates and stock prices


20/23

represent avenues to predict the future path of each other. Our findings, though, indicate that thepredictive power of the two financial assets is restricted to the short-run and even then it does nothold for all countries and sub-periods.

In this respect our results are similar to findings by studies such as Granger et al. (2000) for Asiancountries and Bahmani-Oskooee and Sobrabian (1992) and Nieh and Lee (2001) for advancedmarket economies. We go further than these studies in that we seek to exploit the extra power in

the cross-sectional dimension of the data in testing for panel cointegration and still fail to findevidence of a long-run equilibrium relationship between exchange rates and stock prices. Givenour failure to find cointegration, one direction for future research could be to examine whetherexchange rates and stock prices are cointegrated with other variables such as foreign exchangereserves and the money supply and consider causality between these variables, similar to theapproach adopted by Ibrahim (2000) for Malaysia.


21/23

REFERENCES

Abdalla, I. and Murinde, V. (1997) Exchange Rates and Stock Price Interactions in EmergingFinancial Markets: Evidence on India, Korea, Pakistan and the Philippines, AppliedFinancial Economics, 7:25-35.

Ajayi, RA., Friedman J. and Mehdian, SM. (1998) On the Relationship between Stock Returns andExchange Rates: Tests of Granger Causality, Global Finance Journal, 9: 241-251.

Ajayi, RA. and Mougoue, M. (1996) On the Dynamic Relation Between Stock Prices andExchange Rates, Journal of Financial Research, 19: 193-207.

Bahmani-Oskooee, M. and Domac, I. (1997) Turkish Stock Prices and the Value of the TurkishLira, Canadian Journal of Development Studies, 18:139-150.

Bahmani-Oskooee, M. and Sohrabian, A. (1992) Stock Prices and the Effective Exchange Rate ofthe Dollar,Applied Economics, 24:459-464.

Bai, J., and Perron, P. (1998) Estimating and Testing Linear Models with Multiple StructuralChanges,

Econometrica,66: 47-78.

Dornbursh, R. and Fischer, S. (1980) Exchange Rates and Current Account,American EconomicReview, 70: 960-971.

Frankel, J. (1993) Does Foreign Exchange Intervention Matter? The Portfolio Effect, AmericanEconomic Review, 83:1356-1369.

Grambovas, CA. (2003) Exchange Rate Volatility and Equity Markets, Eastern EuropeanEconomics, 41:24-48.

Granger, CWJ. (1988) Causality, Cointegration and Control, Journal of Economic Dynamics andControl, 12:551-559.

Granger, CWJ., Huang, BN. and Yang, CW. (2000) A Bivariate Causality Between Stock Prices

and Exchange Rates: Evidence from Recent Asian Flu, Quarterly Review of Economicsand Finance, 40:337-354.

Gregory, AW. and Hansen, B. (1996) Residual Based Tests for Cointegration in Models withRegime Shifts, Journal of Econometrics, 70:199-226.

Hadri, K. (2000) Testing for Stationarity in Heterogeneous Panel Data, Econometrics Journal,3:148-161.

Hall, A. (1994) Testing for a Unit Root in Time Series with Pretest Data Based Model Selection,Journal of Business and Economic Statistics, 12:461-470.

Hatemi-J, A. and Roca, E. (2005) Exchange Rates and Stock Prices: Interaction During Good andBad Times: Evidence from the ASEAN4 Countries,Applied Financial Economics, 15:539

546.

Ibrahim, M. (2000) Cointegration and Granger Causality Tests of Stock Price and Exchange RateInteractions in Malaysia,ASEAN Economic Bulletin, 17:36-47.

Kwiatowski, D., Phillips, PCB., Schmidt, P., Shin, Y. (1992) Testing the Null Hypothesis ofStationarity Against the Alternative of a Unit Root, Journal of Econometrics, 54:91-115.

Lee, J. and Strazicich, MC. (2001) Break Point Estimation and Spurious Rejections withEndogenous Unit Root Tests, Oxford Bulletin of Economics and Statistics, 63:535-58.

Lee, J. and Strazicich, MC. (2004) Minimum LM Unit Root Test with One Structural Break,Manuscript, Department of Economics, Appalachian State University.

Mishra, AK. (2004) Stock Market and Foreign Exchange Market in India: Are they Related? SouthAsia Economic Journal, 5:209-232.


22/23

Narayan, PK and Smyth, R. (2005) Exchange Rates and Stock Prices in South Asia: EvidenceFrom Granger Causality Tests, ICFAI Journal of Applied Finance, 11:31-37.

Nieh, CC. and Lee, CF. (2001) Dynamic Relationship Between Stock Prices and Exchange Ratesfor G7 Countries, Quarterly Review of Economics and Finance, 41:477-490.

Nunes, L., Newbold, P. and Kaun, C. (1997) Testing for Unit Roots with Structural Breaks:Evidence on the Great Crash and the Unit Root Hypothesis Reconsidered, Oxford Bulletinof Economics and Statistics, 59:435-448.

Perron, P. (1989) The Great Crash, the Oil Price Shock and the Unit Root Hypothesis,Econometrica, 57:1361-1401.

Phillips, P. and Hansen, B. (1990) Statistical Inference in Instrumental Variable Regression withI(1) Processes, Review of Economic Studies, 57:99-125.

Ramasamy, B. and Yeung, MCH. (2005) The Causality Between Stock Returns and ExchangeRates Revisited,Australian Economic Papers, 44:162-169.

Smyth, R. and Nandha, M. (2003) Bivariate Causality Between Exchange Rates and Stock Prices

in South Asia,Applied Economics Letters, 10:699704.Standard and Poors (2004) Global Stock Market Factbook (New York: Standard and Poors).

Venkateshwarlu, M. and Tiwari, R. (2005) Causality Between Stock Prices and Exchange Rates:Some Evidence for India, ICFAI Journal of Applied Finance,11:5-15.

Westerlund, J. (2006) Testing for Panel Cointegration with Multiple Structural Breaks, OxfordBulletin of Economics and Statistics, 68:101-132.

Wu, Y. (2000) Stock Prices and Exchange Rates in a VEC Model The Case of Singapore in the1990s, Journal of Economics and Finance, 24:260-274.

Yu, Q. (1996) Stock Prices and Exchange Rates: Experiences in Leading East Asian FinancialCentres Tokyo, Hong Kong and Singapore, Singapore Economic Review, 41:47-56.

Zivot, E. and Andrews, D. (1992) Further Evidence of the Great Crash, the Oil-price Shock andthe Unit-root Hypothesis, Journal of Business and Economic Statistics, 10:251-270.


23/23

Notes

iGiven that the results of the Gregory and Hansen (1996) test were not conclusive across all three teststatistics and models we also conducted the Granger causality tests for Korea, Malaysia and thePhilippines without the ECT. The contemporaneous results for exchange rates and stock prices for thesethree countries, which are available on request, are quantitatively the same as those reported.

iiAfter dividing the sample into two sub- sample periods we first examined the order of integration for each

series using the LM unit root test with one structural break for both sub-periods and found all variables to beI(1). We then tested for cointegration for each country in each sub-period using the Johansen (1988) andGregory and Hansen (1996) cointegration tests and found ambiguous evidence that exchange rates andstock prices were cointegrated in both sub-sample periods for Korea, Malaysia and the Philippines. Thus,we included an error correction term in Equations (5) and (6) for both sub-sample periods for these threecountries.

2506 Hooi Granger

Documents

Transcript of 2506 Hooi Granger