“AN PHASEWISE EMPIRICAL

ANALYSIS OF INTEGRATION
BETWEEN NSE AND NYSE”

Harsh Purohit

Professor & Chair- ICICI Bank CBFSI, WISDOM

Faculty of Management Studies, Banasthali University, Banasthali Vidyapith

iic@banasthali.in

Hartika Chhatwal,
Sr. Lecturer, Department of Management,
Delhi Institute of Advanced Studies
haritika@rediffmail.com, (9810286030)

Himanshu Puri,

(Corresponding Author)
Assistant Professor, Department of Finance,
Institute of Marketing and Management (IMM)

purihiman@gmail.com, (9990103061)

1
 “AN PHASEWISE EMPIRICAL ANALYSIS OF
INTEGRATION BETWEEN NSE AND NYSE”

ABSTRACT
Integration of emerging stock exchanges with the developed stock exchanges has naturally
constituted a privileged field for Financial Research. Markets are said to be integrated if they
share a common trend and move together. This study also examines the extent of cross-
country integration between the stock market of India and US. This study captures the
movements of the National Stock Exchange of India (NSE) in comparison with the New
York Stock Exchange (NYSE) as well as examines the long term association and cause &
effect relationship between their indices. The Indices selected for the study are NSE 100 and
NYSE 100. The study also investigates the impact of volatility on NSE due to volatility in
NYSE. The objectives of the study are examined by employing ADF test to check the
stationarity, Johansen’s co-integration test for examining the long term relationship, Granger
test to investigate the cause and effect relationship and Garch (1, 1) model to check the
volatility in NSE due to volatility in NYSE. The daily closing data is taken for last eleven
years (April 2004 - March 2014) for the analysis. Since the period of financial crisis is
considered, the entire period (April 2004- March 2014) is divided into three sub-periods,
namely before crisis period (April 2004- August 2008), during crisis period (September
2008- December 2010) and after crisis period (January 2011- March 2014). The findings of
the study proved that there are structural break points in the specified data, tested through
Chow test. The series derived from NSE 100 and NYSE 100 were not stationary in the level
form, but there is evidence of stationarity in the first difference form. Long run relationship
between these series is also observed. There is bi-directional causality between NYSE and
NSE for before crisis period. For the period during crisis, there is no feedback from NYSE
return to NSE return and vice versa. For the period after crisis, there is a unidirectional
relationship from NYSE returns to NSE return. Empirical results also found that in all the
period before, during and after crisis there is persistence of high volatility. Thus, the
volatility not only persisted before the 2008 global financial crisis but also has increased.
Tests proves the existence of integration between the US and Indian markets.

2
Key Words: Volatility, NSE, NYSE, Integration, ADF Test, Johansen’s Co-Integration,
Granger Causality, GARCH
INTRODUCTION
Now, the entire world is recognised as a single market because of globalisation and
technological developments happening across the globe. Removal of structural bottlenecks,
introduction of new players/instruments, free pricing of financial assets, relaxation of
quantitative restrictions, improvement in trading, clearing and settlement practices and
greater transparency has transformed the financial system of economies. Dismantling of
various controls in the financial system has facilitated the integration of financial markets.
Even the emerging markets like India are no exception to this development. Development of
markets and institutions, effective price discovery which contributes to higher savings,
investment and economic progress are the outcome of financial integration between
economies. There has been huge hastening in financial globalization because of increasing
cross-country investment in foreign assets. Because of technological advancement and
global transactions, countries have become more economically interlinked, thereby
explaining this higher interdependency and integration between them. Stock market;
however, appear to be unique as they serve as economic barometers representing the growth
for the economy. The ever closer relations among international stock markets have
important implications for macroeconomic policies and outcomes. But, the integration
between financial markets also poses several risks like contagion and associated disruption
of economic activities that were evident during the crisis in Asia in the late 1990s and more
recently, in 2008 with national stock markets declining sharply because of US financial
crises. Analysts across the globe have recognised that they should continuously monitor
dependency and integration of their market on frequently basis because of such global melt
down affecting almost all of the economies together. Recognising the importance of such
integration, many researcher have came ahead to study this aspect in context of stock
markets of emerging economies being impacted by stock market of developing economies.
The present paper is also inclined towards analysing the integration between the stock
market in India, represented by National Stock Exchange (NSE) and stock market of US,
represented by New York Stock Exchange (NYSE). The study involves the testing of long-
term association and causality aspect between its indices. The impact on the emerging
economy’s stock exchange due to volatility in developed market exchange would also form
part of the study.
3
LITERATURE REVIEW
There have been numerous studies on market integration and interdependence. Few of them
have been studies for the paper. They are:

Eun and Shin (1989) checked the presence of substantial amount of interdependence
among national stock markets of USA, UK, Canada, Germany, Australia, France, Japan,
Switzerland and Hongkong. The study found a substantial amount of multi-lateral
interactions among the national stock markets. The paper used daily closing price data
during the period January 1980 through December 1985. The analysis indicated that
innovations in the U.S. were rapidly transmitted to other markets in a clearly recognizable
fashion, whereas no single foreign market can significantly explain the U.S. market
movements.

Masih, Abul and Rumi (1997) examined the dynamic linkage patterns among national
stock exchange prices of four Asian newly industrializing countries - Taiwan, South Korea,
Singapore and Hong Kong. The sample consisted on closing daily share price indices data of
the four stock markets from January 1982 to June 1994. They concluded that these markets
are not mutually exclusive of each other and significant short run linkages appear to run
among them.

Bala and Mukund (2001) in their study examined the nature and extent of linkage between
the US and the Indian stock markets. They used the theory of co-integration to study the
interdependence between the Bombay stock exchange (BSE), the NYSE and NASDAQ. The
data consisted of daily closing prices for their three indices from January 1991 through
December 1999. The results investigated that the Indian stock market was not affected by
the movements in US markets for the entire sample period.

Mishra (2002) investigated the international integration of India’s domestic financial market
with the U.S. stock market. By applying the ordinary least squares (OLS) method and
cointegration technique, he found a positive correlation between NASDAQ and BSE. He
concluded that BSE was influenced by the movements of NASDAQ. But there is no
cointegrating vector between BSE and NASDAQ indexes, which shows that there is no
4
long-run relationship between these two stock exchanges.
Ahmad, Ashraf and Ahmed (2005) examined the inter linkages and causal relationship
between the Nasdaq composite index in the US, the Nikkei in Japan with that of NSE Nifty
and BSE Sensex in India using daily closing data from January 1999 to August 2004. The
study used Granger Causality and Johansen co-integration methods to examine short run and
long term relationship among the stock markets respectively. The results of Co-integration
test revealed that there was no long-term relationship of the Indian equity market with that
of the US and Japanese equity markets. Granger causality test suggested that there was a
unidirectional relationship from Nasdaq and Nikkei to Indian stock markets.

Wong, Agarwal and Du (2005) used the weekly data of BSE 200 (India), S&P 500 (US),
FTSE 100 (UK) and Nikkei 225 (Japan) from January 1991 to December 2003 to inspect the
long run equilibrium relationship and short run dynamic inter linkages between the Indian
stock market and world major developed stock market. The results showed that in the long
run, the Indian Market is integrated with the developed stock markets and is also influenced
by the variability in these markets. Both US and Japan Stock market causes the Indian stock
market but not the other way in the short run. In addition, they found that the Johansen
method is the best to reveal their co-integration relationship.

Lamba (2005) performs a comprehensive large sample analysis to investigate the presence
of long run relationships among South Asian equity markets and the developed equity
markets. The results reveal that Indian markets are influenced by developed equity market of
US, UK and Japan.

Mukherjee (2007) examined the integration of Indian stock markets with select few
international indices by examining the trends, similarities and patterns in activities and
movements. For the study, five stocks markets based on specific qualitative attributes were
classified namely Market capitalization, Number of listed securities, Listing agreements,
circuit filters and settlement. The stock exchanges considered were NSE, NYSE, Hang sang,
Russian and Korean. The study was conducted for the period 1995-2006. To test for
integration of the markets, statistical methods such as correlation analysis, exponential trend
analysis and the risk-return analysis was used. It was observed that when compared to
Russian stock markets, NSE was observed to be more volatile in nature. NSE mainly rose
5
because of tech boom till mid of 2000 which detrended back by 2001. Interest rate regimes
and other macroeconomic factors were found to be responsible for consistent uptrend of
NSE. From the study it was found that Hang Seng though very volatile in nature had less
correlation with the Indian markets uptil 2003. Hang seng rose during 1996 because of east
asian miracle. By year 2000, the index had crashed due to fear of recession. But, it was
found that hangsang index and NSE were reacting almost identically after 2003 which
means larger integration of the Indian economy in the foreign market and impact of
investments by FIIs and other foreign investors. NSE and NYSE markets were found to be
highly correlated after mid 2000 reflecting the beneficial effect of tech boom on both the
markets. NSE and Korean stock exchanges are found to be highly correlated though east
asian crisis had more impact on Korean stock exchange due to its integration with east asian
economies. The study find conclusive evidence of integration of the Indian markets with the
global markets much more in the recent years (maily after 2006) than the previous years as
India before 2000 was in its inception stage as a globalized economy, thus being directly
protected from foreign exposure.

Arshad Hasan et al (2008) examined the long term relationship between Karachi stock
exchange and equity markets of developed world for the period 2000 to 2006 by using
multivariate Co-integration analysis. Johansen and Juselius multivariate Co-integration
analysis indicates that markets are integrated and there exists a long term relationship
between these markets. However, pair wise Co-integration analysis shows that Karachi stock
market is not co-integrated with equity market of US, UK, Germany, Canada, Italy and
Australia. However, Karachi stock exchange is found to be integrated with France and
Japan.

Raj and Dhal (2009) investigated the nature of the financial integration of India’s stock
market with global and major regional markets. Co-integration relation suggests that the
Indian market’s dependence on global markets, such as the United States and the United
Kingdom, is substantially higher than on regional markets such as Singapore and Hong
Kong. VECM result shows that international market developments at regional and global
levels together could account for the bulk of the total variation in the Indian stock market.
Within Asia, the Singapore and Hong Kong markets have significant influence, while the
Japanese market has weak influence on the Indian market. The two influential global
markets, the United States and the United Kingdom, could have a differential impact on the
Indian market in the opposite direction, amid 6a structural shift in India’s integration with
these global markets.
Saif siddique (2009) examined the association between S & P CNX Nifty and selected
Asian and US stock markets. The author tries to examine the integration of stock markets in
longitudinal rather than cross sectional, thus adding to the literature. Interdependency among
global stock markets is studied primarily through correlation of returns, Co-integration and
the Granger Causality tests in this study. The study considers thirteen indices for the study
which are analyzed over two time frames mainly 1999-2004 and 2004-2009. No very clear
direction of relationships was found to exist in the sense of Granger Causality indicating the
fact that influence of few markets, especially that of the US, has eroded over a period of
time.

Ashwin G. Modi et al (2010) used the daily data from July 1, 1997 to June 30, 2008, to
examine the stock market indices of India (SENSEX), Hong Kong (HANGSENG), Mexico
(MXX), Russia (RTS), Brazil (BVSP), UK (FTSE-100) and US (DJIA and NASDAQ).
Correlation and Co-integration technique has been employed to study the short term and
long-term relationships between the markets. The results indicate that there is considerable
volatility in the correlations between the eight stock markets. The result reveals that MXX,
DOWJONES and NASDAQ are least dependent on other markets, whereas DOWJONES is
the most influential market.

Iqbal, Khalid and Rafiq (2011) attempted to find out dynamic relationship using Johansen
co-integration procedure for long run relationship and Granger Causality test. No integration
was found among US, Pakistan and India. However, the Granger Causality test showed the
evidence of unidirectional causality running from NYSE to Bombay and Karachi stock
exchange.

Srivastava et al (2012) studied the short and long-term relationships in three Asian markets,
namely Hongkong, Singapore and Japan along with four other global markets of the USA,
UK, Germany and France, perceived to be driving Indian Stock prices. By using co-
integration techniques they found that Indian stock markets are very much integrated with
other global markets in short run but less integrated in long run.

Mandaviya (2014) contemplated with the objective

7 of studying the degree of stock market
integration and degree of influence of Indian stock market by global stock market. In this
study, top 10 major stock exchanges is selected as a market capitalization and month-wise
average prices of BSE-Sensex, Nifty, NYSE, NASDAQ, S&P500, TOPIX, FTSE100,
HangSeng, SSE Composite index, DAX, S&P 500and Dow Jones. Multiple Correlations and
Regression has been computed for the select stock market indices. Statistical Significance of
the correlation has been tested by applying correlation t-test and regression t- test. The
results of these studies support the view that there is a substantial integration and
dependence between domestic and international financial markets. BSE – Sensex has
witnessed greater fluctuation between domestic and international very high co-efficient of
variance compared to other selected indices. BSE highest correlation was recorded between
Sensex and Nasdaq, Topix, Hang Seng, Dax. NSE highest correlation was recorded between
Nifty and Sensex, Topix, Hang Seng,

OBJECTIVE OF THE STUDY

The specific objectives of this present study are as follows:
1. To determine the presence of a structural break in NSE 100 and NYSE 100 data to
divide in three sub groups
2. To examine the long term relationship between return series of NSE 100 and NYSE
100
3. To examine the causality between NSE and NYSE
4. To examine the spill over affect on NSE due to volatility in NYSE

HYPOTHESES
The following hypotheses are formulated:
1. H0: There are no structural breakpoints in the data
2. H0: There is no long term relationship between return series of NSE 100 and NYSE
100
3. H0: There is no significant causal relationship between NSE and NYSE
4. H0: There is no significant volatility in NSE due to volatility in NYSE

DATA COLLECTION
The secondary data has been used for the present paper. The data for the study had been
collected for the eleven years, i.e. from April 2004-March 2014, from www.nseindia.com
8
and Yahoo Finance. Since the period of financial crisis is considered, the entire period (April
2004- March 2014) is divided into three sub-periods, namely before crisis period (April
2004- August 2008), during crisis period (September 2008- December 2010) and after crisis
period (January 2011- March 2014). Collected data is based on the daily movements of
NSE’s and NYSE’s indices such as NSE 100 and NYSE 100. The data has been analyzed
using E-Views 7 software.

RESEARCH METHODOLOGY
As the data has been sub divided, it was important to determine the presence of a structural
break in NSE and NYSE data. The Chow test has been used for the same purpose to begin
with. Then next step for any time-series analysis is to test for stationarity, i.e., the return
series of NSE 100 and NYSE 100 should be stationary. For this, Augmented Dickey Fuller
Test and Phillips-Perron Tests are available. In this study, we would be using ADF Test. The
next step is to check whether there exists a long-term relationship between the two variables
or not which can be done using the Johansen Co- integration Test followed by Granger
Causality test to determine the causal relationship. Finally, Garch (1, 1) Model has been
used to test for the volatility in the NSE due to NYSE. These entire tests have been carried
on the E-views software. For empirical analysis, raw series are converted into series of
returns. As due to a number of statistical reasons, it is preferable not to work with the prices
series directly. Returns have an additive advantage that they are unit-free. Price returns are
taken as continuously compounded returns as for log-returns frequency of compounding of
the return does not matter, which makes the returns series more comparable. Also,
compounded returns are time-additive. The returns series is calculated as follows:
pt
rt = 100% ∗ ln⁡( )
pt−1

Where, rt is the continuously compounded return in time period t, pt denotes the asset price
at time t and ln represents the natural logarithm.

Testing For Stationarity of Price Series

Most of the financial series are non-stationary in the level form. The prior step to carry out
any analysis in time-series is to test for stationarity. An examination of whether a series is
stationary or not is essential for two reasons:

9
 1. A stationary or otherwise of a series strongly influences its behavior and properties.
For stationary series, shock or the error term dies away gradually while for a non-
stationary series, the persistence of a shock will always be infinite.
2. The use of non-stationary data can lead to spurious regressions. It means that when
the standard regression techniques are applied to non-stationary data, the end results
could show a high R2 and significant coefficient estimates but is valueless.

A stationary series can be defined as one with a constant mean, constant variance and
constant covariance for each given lag (Brooks, 2008) and thus conventional statistical
results are appropriate while a non-stationary series has time varying mean and variance.
Testing for stationarity implies testing for the presence of unit roots. A non-stationary series,
yt must be differenced d times before it becomes stationary, then the series is said to be
integrated of order d, i.e., if yt ~ I(d), then Δd yt ~ I(0) implies that applying the difference
operator , Δ, d times, leads to an I(0) process, i.e., presence of no unit roots. An I (0) is a
stationary process while an I (1) is a non-stationary process with one unit root. Consider an
AR (1) process-

Yt = ⁡∅⁡Yt−1 + ⁡ μt

Where, µt is the error term that follows a white-noise process.

H0: ∅ = 1 indicates the presence of unit root and thus the non-stationarity of the series.

H1: ⁡∅ < 1 indicates absence of unit roots and thus stationarity of the series.

Taking first difference of the above series i.e.,

Yt − Yt−1 = (∅ − 1)⁡Yt−1 + ⁡ μt

Or, ∆⁡Yt = ⁡θ⁡Yt−1 + ⁡ μt [θ = φ-1]

Here, ∅ = 1 is equivalent to a test of ⁡θ= 0 so that not rejecting ⁡θ= 0 implies presence of unit
root and of ⁡θ< 1 implies stationarity of the series.

Augmented Dickey Fuller Test

Dickey and Fuller were the first to test for a unit root in a time-series analysis. The
Augmented Dickey Fuller test uses the following10regression-

Δyt = a + αyt-1 + ∑ki=1 bi Δyt-i + εt …………………………………………..(1)

Δyt = a + βt +αyt-1 + ∑ki=1 bi Δyt-i + εt ……………………………......……..(2)

The regression test for unit root in yt, where Δyt-i is the lagged difference to accommodate
serial correlation in the errors, εt. k is the appropriate lag length.

The null and alternate hypotheses of the regression are

H0 : α = 0

H1 : α < 0

Not rejecting the null hypothesis indicates the presence of a unit root i.e., the series is non-
stationary while rejecting the null implies a mean stationary process in case of equation (1)
and a trend stationary process in case of equation (2). If ΔYt is stationary, it is called a
differenced stationary process. If ΔYt is stationary while Yt is not, then Yt is called
integrated of first order I(1). Presence of a unit root implies a permanent effect of random
shocks and variance is time-dependent i.e., increases with time.⁡The test statistics for the
original Dickey Fuller test are defined as

̂
θ
Test statistics = s.e⁡θ̂

The test statistics do not follow the usual t-distribution under the null hypothesis, since the
null hypothesis is that of non-stationarity, but rather it follows a non- standard distribution.
This test statistic is then compared to the value of Dickey Fuller test. If this test statistic is
less than the critical value, then the null hypothesis is rejected confirming the absence of unit
root and thus stationarity of the series. However, if this test statistic is greater than the
critical value, presence of unit root and thus non-stationarity is confirmed.

Johansen’s Co-integration Test

A set of variables is said to be co-integrated, if a linear combination of them is stationary.

Spot and futures prices are expected to be co-integrated as they are the prices of the same
asset in different points in time and hence will be affected by given any information. A co-
integrating relationship may also be seen as a long-term or equilibrium phenomenon, since it
is possible that co-integrating variables may deviate from their relationship in short run, but
it will return in the long run. The aim of this test is to determine whether a long-term
relationship exists between the variables or not.11Johansen’s methodology takes its starting
point in the vector auto regression (VAR) of order p given by
 yt = µ + A1 yt−1 + ⋯ … … + Ap yp−t + εt
Where yt is an n×1 vector of variables that are integrated of order one – commonly denoted
I(1) and εt is an n 1 vector of errors.

Maximum Eigen Value Test

The Eigen values are the squared canonical correlation between a linear combination of
stationary Δx t and linear combination between non-stationary x t 1 . This interpretation is
intuitively appealing because this correlation will be high only if the linear combination of
x t 1 , is itself stationary. Otherwise, a non-stationary variable cannot have a high correlation
with a stationary variable. Therefore, higher the Eigen value, higher will be the stationarity
of the particular linear combination of the non-stationary variable. Only those Eigen values
indicate the co-integrating relationship among the variables which are significantly different
from zero. The corresponding (normalized to a variable) Eigen vector of an Eigen value is
the potential co-integrating vector  . However, this vector represents a co-integrating
relationship only if its Eigen value is different from zero. Once the co-integrating vector (in
the form of eigenvector) is known, the error-correction vector can easily be estimated using
its OLS estimator.

Trace Test
Maximum likelihood estimator gives us k number of Eigen-values, but all of them will not
be significantly different from zero. Let we assume only r Eigen values are different from
zero. Now there are following possibilities:

1. r = 0, it means there is no co-integrating relationship among the variables. Therefore

the VAR should be estimated without error correction term.
2. r = k, this can happen only when xt is stationary rather than non-stationary.
3. r < k, then there are only r co-integrating relationship among the variables. This is the
most obvious situation and in this case only r Eigen values are different from zero and
remaining (k-r) Eigen values are non-distinguishable from zero.

Johansen suggests trace test (ML based test) to determine the number of non-zero Eigen
values. Trace test examines the null hypothesis that the co-integration rank is equal to r
12
against the alternative hypothesis that co-integration rank is k . The test is conducted in
inverse sequence, x i.e., r  k , k  1, k  2....0 . The test statistic is computed as follows:

k
Trace  T  ln( 1   )
i  r 1
i

Although both of these statistics are based on likelihood ratio approach, these do not follow
the standard  2 -distribution. Rather they have non-standard distribution. Before
implementing Johansen’s test we have to take two important decisions: (i) what should be
the order of the VAR i.e. ‘p’, and (ii) should we include deterministic parameters with or
without imposing co- integration restrictions.

Vector Error Correction Model

Once it is known that there exists a long-run relationship between the variables or not, the
next step is to check the significance of the coefficients through Ordinary Least Square
Method which will further determine the long-run causality between the variables. But if
there is no co-integrating relationship between the variables, then standard Vector
Autoregressive model is used to determine the causality.

Ordinary Least Square Method

Ordinary least squares (OLS) is a method for estimating the unknown parameters in a linear
regression. The method minimizes the sum of squared vertical distances between the
observed responses in the dataset and the responses predicted by the linear approximation.
The OLS estimator is consistent when the regressors are exogenous and there is no multi-co-
linearity, and optimal in the class of linear unbiased estimators when the errors are
homosckedastic and serially uncorrelated. Under these conditions, the method of OLS
provides minimum variance mean unbiased estimation when the errors have finite variances.
Under the additional assumption that the errors be normally distributed, OLS is
the maximum likelihood estimator. Suppose the data consists of n observation { yi, xi }ni=1.
Each observation includes a scalar response yi and a vector of predictors (or regressors) xi.
In a linear regression model, the response variable is a linear function of the regressors:

13
where β is a p×1 vector of unknown parameters; εi's are unobserved scalar random variables
(errors) which accounts for the discrepancy between the actually observed responses yi and
the "predicted outcomes" x′iβ and so x′ β is the dot product between the vectors x and β.
This model can also be written in matrix notation as

y = x⁡β + ⁡ε

Where y and ε are n×1 vectors, and X is an n×p matrix of regressors, which is also
sometimes called the design matrix.

As a rule, the constant term is always included in the set of regressors X, say, by
taking xi1 = 1 for all i = 1… n. The coefficient β1corresponding to this regressor is called
the intercept.

The paper employs OLS Method to determine significance of coefficients. If the coefficient
of the independent variable is significant then, it is concluded that long-run causality exists
between the variables.

Granger Causality Test

A statistical approach proposed by Clive Granger (1969) to assess whether there is any
potential predictability power of one indicator for the other (Foresti, 2007). A time series is
said to Granger cause other if the past values of the former improve the forecast of the latter
(Enders, 2008). A Granger causality test is testing for the causal relationship between two
stationary series X t and Yt in the following two equations:

Granger causality test seeks to answer whether changes in causes changes in, If, lags of the
former should be significant in the equation for the latter i.e., ≠ 0. If this is the case and not
vice-versa (i.e., = 0) it would be said that Y t Granger causes X t causes X or that there
exists unidirectional causality from Y t to X t . On the other hand, if X t causes Yt , lags of
X t should be significant in the equation for Y14
t . If both sets of lags were significant, it
would be said that there exists „bi-directional causality‟ or „bi-directional feedback‟.Also, if
there exists uni-directional Granger causality from Y t to X t , then Y t is said to strongly
exogenous in the equation of X t If neither set of lags are statistically significant in the
equation for the other variable, then it is said to be independent of each other. Granger
causality really means only a correlation between the current value of one variable and the
past values of other. It does not mean that movements of one variable cause movements of
another

GARCH (1, 1) Model

Volatility, as measured by variance of returns, is a crude measure of total risk of financial
assets. Since a financial time series does not have constant variance of errors, so a model
that does not assume constant variance and which describes how variance of errors evolves
overtime is required. An Auto Regressive Conditionally Heteroscedastic (ARCH) model
developed by Engel in 1982 is one of those models. Under ARCH model, the conditional
variance of the error term depends on the lagged values of the squared errors. (Brooks,
2008) Later the Generalized ARCH model (GARCH) was developed independently by
Bollerslev(1986) and Taylor(1986) where conditional variance not only depends on lagged
squared residuals but also on its own lagged values. The GARCH model is described by two
equations:

Mean Equation: Y = C + ϵt
Variance Equation: σ2t = ⁡ α0 + ⁡ α1 ε2t−1 + ⁡ βσ2t−1

Since it is a variance equation, its value must always be strictly positive. For an ARCH
model, there has to be non-negativity constraints i.e., α0 > 0,⁡ α1 > 0.
σ2t is known as the conditional variance. It is a one-period ahead estimate for the variance
based on any relevant past information.
βσ2t−1 is the last period’s variance which represents the GARCH term.
α1 ε2t−1 is the information about volatility during the previous period – an ARCH term.

This is the GARCH (1, 1) model as only one lag of the error term and conditional variance is
taken. Only one lag is taken because the more the parameters to be estimated, the large the
conditional variance. Also, everything else equal, the more the parameters, the more likely it
15
is that one or more of them will have negative estimated values. In general, a GARCH (1, 1)
model is sufficient to capture volatility clustering in the data and any higher order model is
rarely used in estimating volatility for financial series. A GARCH model is better than an
ARCH model as the former is more parsimonious than the latter and avoids over-fitting.

Hence, the model explains the current period’s conditional variance is the weighted function
of a long-term average value (constant in equation), last period’s forecasted conditional
volatility and last period’s squared residuals. The sum of the coefficients of ARCH and
GARCH term is usually close to unity signifies that the volatility of asset returns is highly
persistent. The effect of any shock in volatility dies out a rate of (1-α1 – β). If (α1 + β) ≥ 1,
then the effect of shock is permanent and it is termed as ‘non-stationarity in variance’ and
(α1 + β) = 1 is termed as ‘unit root in variance’. The volatility will be defined only if (α1 +
β) < 1. Therefore, this condition is imposed while estimating the GARCH model.

ANALYSIS AND FINDINGS

Descriptive Statistics
Table 1 shows statistical moments of daily stock returns. For the whole sample period (April
2004-March 2014), the study found that Indian stock market provided higher stock returns.
In terms of risk adjusted return (average stock adjusted to standard deviation), the securities
listed in NSE provided higher returns with higher standard deviation. Skewness and kurtosis
measure provide insights about the underlying statistical distribution of stock returns. It is
evident that skewness is negative and kurtosis is positive for both the markets during the
sample period. Skewness and kurtosis exhibits more or less a similar pattern of statistical
distribution. The Jarque-Bera statistic, defined over skewness and kurtosis measures, is very
high for both the markets, implying that stock returns differ significantly from the normal
distribution. This implies that in each stock market there are opportunities for investors to

Table 1: Descriptive Statistics

N Minim Maxi Mean Std. Skewness Kurtosis
um mum Deviation
Statist Statisti Statist Statistic Statistic Statisti Std. Statisti Std.
ic c ic c Error c Error
retnse 2493 -.1305 .1633 .000523 .0162524 -.252 .049 9.646 .098
retnyse 2493 -.1023 .1153 .000180 .0136393 -.418 .049 10.435 .098
Valid N 16
2493
(listwise)
benefit from abnormal returns.

Chow Test Results

In order to endogenously determine the presence of a structural break in NSE and NYSE data,
we conduct a Chow test.

H0: There are no structural breakpoints in the data

H1: There are structural breakpoints in the data

The test is done for the year 2008 and 2011 as these are the year’s corresponding to the
advent and end of crisis in U.S. The results are reported in the following Table 2.

Table 2: Chow Test Results

PERIOD F-statistic Prob. F
2008 6.171888 0.0021
2011 4.756617 0.0087

The null hypothesis is rejected in both the years as the p-value is less than 0.5 indicating
that there are structural breakpoints at the specified time. Chow test does not mention
specific point in time. Since, there exist structural breaks in the returns from NSE 100 for the
same year as in returns from NYSE 100.

Results for the Stationarity of Series

We test for stationarity of the NSE 100 and NYSE 100 series for all the three periods (pre,
post and during financial crisis). While testing for the stationarity of series using ADF test
the hypotheses are:
H0: Presence of unit root i.e., non-stationary series.

H1: No unit roots i.e., stationary series.

The critical values of the ADF at levels are tabulated in Table 3.

Table 3: Critical Values

1% 5% 10%
Critical Values -1.6158 -2.5685
17 -3.1303
The logarithms of series are tested for stationarity at levels. The results for the same are
reported in table 4.

Table 4: Results of Stationarity at Levels

Periods Before crisis During Crisis After Crisis
NSE -1.069614 -0.290463 -1.398342
NYSE -1.479606 -2.086963 -0.684744

As can be seen by comparing tables 3 and 4, the calculated ADF test statistic is greater than
the critical value, so the null hypothesis that the series has one unit root is not rejected. This
implies that both the series are non-stationary at levels. Therefore, to carry out further
analysis, it is necessary to make the series stationary. The results for stationarity at first
difference are tabulated in table 5.

Table 5: Results of Stationarity at First Difference

Periods Before crisis During Crisis After Crisis
NSE -31.22066 -22.83140 -26.11852
NYSE -37.02131 -19.94584 -29.55015

Comparing table 3 and 5 shows that the calculated ADF test statistics is less than the critical
value, so the null hypothesis of presence of one unit root is rejected. This implies that both the
series becomes stationary at first difference. In other words, it can be said that the ADF test
confirms the stationarity of returns from NSE 100 and NYSE 100.

Johansen’s Co-integration Test Results

Before testing for co-integration, it is necessary to determine the optimal lag length. As
mentioned before the starting point of the test is the vector auto regression (VAR) of order
“p”. This “p” is the optimal lag length. The optimal lag length of VAR for all the three
periods are tabulated in table 6.

Table 6: Results for optimal lag length

Before Crisis During Crisis After Crises
Lags SBC 18 SBC SBC
0 32.04771 32.16930 30.39162
1 22.71366 24.17303* 22.83594*
 2 22.68932* 24.20595 22.85635
3 22.70552 24.19878 22.88172
4 22.71400 24.23638 22.90622
5 22.72162 24.27967 22.93540
6 22.73490 24.31717 22.95522
7 22.74331 24.33881 22.98049
8 22.75840 24.37385 23.01013
Note: * indicates the optimal lag length chosen by Schwartz Bayesian criteria.

As can be seen from table 6, the optimal lag length, “p” for before crisis period is 2 and for
during crisis period and after crisis period is 1. The test requires the maximization of eigen
value and trace test which will determine the number of co-integrating equations. The
hypothesis used for the test is:

H0: There is no long run relationship between return series of NSE 100 and NYSE 100.

H1: There exists a long run relationship between return series of NSE 100 and NYSE 100.

The analysis is done for all the three periods and results are reported in table 7.

Table 7: Unrestricted Co-integration Rank Test

Number of
PERIODS Hypothesised Maximum Eigen Trace Statistics p-value
equations Statistics
Before Crisis None 14.99737 17.75331 0.0445
At most 1 2.755942 2.755942 0.0969
During Crisis None 34.46842 35.91168 0.0000
At most 1 1.443260 1.443260 0.2296
After Crisis None 25.18349 25.19011 0.0013
At most 1 0.006622 0.006622 0.9346
NOTE: MacKinnon-Haug-Michelis (1999) p-values

As can be seen from the above table 7, for all the three periods both maximum eigen
statistics and trace statistics have p-value less than 0.05, so the null hypothesis that there is
no long run relationship between NSE and NYSE indices in logarithm is rejected, i.e., there
exists a long-run relationship between the two data series. However, for the null hypothesis
19between the two series, both the maximum
that there exits at most one co-integration equation
eigen statistics and trace statistics have p-value greater than 0.05. So, finally we conclude that
there is only one co-integrating equation and hence there exists a long-run relationship
between NSE and NYSE returns.

Granger Causality Results

Table 8: Granger Causality Results
Before Crisis During Crisis After Crisis
Chi-square Probabi Chi-square Probabi Chi-square Probabi
Null hypothesis lity lity lity
NYSE does not 24.72963 0.0000 0.271407 0.6024 16.32345 0.0003
Granger cause
NSE
NSE does not 18.66132 0.0001 0.042747 0.8362 2.367658 0.1239
Granger cause
NYSE

.
As can be seen from above table 8, for before crises period, NYSE returns Granger causes NSE
returns. And also, the null hypothesis that NSE returns does not Granger cause NYSE return is
rejected because the p-value is 0.0001 which is less than 0.05. Thus, there is bi-directional
causality between NYSE and NSE for before crisis period. For the period during crisis, both
the null hypothesis is not rejected (p-value > 0.05). This implies that there is no feedback from
NYSE return to NSE return and from NSE return to NYSE return. For the period after crisis,
the null hypothesis that NYSE returns does not Granger cause NSE return is rejected (p-
value < 0.05) implying a unidirectional relationship from NYSE returns to NSE return.

Results for the presence of volatility GARCH (1, 1)

Before testing for the presence of volatility, it necessary to check for volatility clustering which
describes the tendency of large changes in asset prices to be followed by large changes in the
same asset price and small changes by small changes in the same asset price.

FIGURE 1: VOLATILITY CLUSTERING

20
 .2

.1

.0
.2

-.1
.1

-.2
.0

-.1

-.2
04 06 08 10 12 14

Residual Actual Fitted

As can be seen from the graph, from the beginning of 2004 to the beginning of 2007, small
changes in index returns are followed by small changes index returns. However, during
crisis, large changes in prices are followed by large changes. In other words, the current
level of volatility tends to be positively correlated with its immediately preceding period’s
level. This suggests the presence of volatility clustering and hence, suggests the use of
ARCH and GARCH to check for the persistence of volatility shocks.

As mentioned before, the test of presence of volatility using GARCH (1, 1) consists of a mean
equation and a variance equation.
Mean equation: R t = C + ⁡ εt

Variance equation: ⁡σ2t = ⁡ α0 + ⁡ α1 ε2t−1 + ⁡ βσ2t−1

Where, Rt⁡ is the return series

In the first step, using mean equation, price returns are regressed on the constant and then
using the residuals from the mean equation, the forecast variance of the current time period
is predicted using the variance equation. If (α1 + β) is close to unity, then it implies
persistence of high volatility shocks. The results of GARCH (1, 1) model are tabulated in
table 9.

Table 9: GARCH (1, 1) Results 21

Equation Variable Before Crisis During Crisis After Crisis
(Apr 2004- Aug 2008) (Sep 2008- Dec 2010) (Jan 2011- March 2014)
Coefficient Probability Coefficient Probability Coefficient Probability

C(1) 0.001772 0.0000 0.001117 0.0301 0.000370 0.2987

MEAN
C(2) 0.091327 0.0119 0.022818 0.4790 0.024320 0.5163
C(3) 1.14E-05 0.0000 2.57E-06 0.1413 9.13E-07 0.1487
VARIANCE C(4) 0.183630 0.0000 0.097275 0.0004 0.047088 0.0002
C(5) 0.783718 0.0000 0.895511 0.0000 0.945180 0.0000

In the table, C (1) and C (2) are the coefficients of the mean equation. C (1) represents the
constant term and C(2) is the coefficient of the error term in mean equation.

Further note that

Ht = C(3) + C(4)*RESID(-1)^2 + C(5)*GARCH(1)
Where, Ht is the variance of the residual term i.e., error term derived from the mean
equation C (3) is the constant term i.e.,

RESID(-1)^2 = previous period‟s squared residual (error derived from the mean equation). It
is known as previous day’s dependent variable information about volatility. This is the
ARCH term. GARCH (-1) is the last period‟s variance given by with C(5) as the coefficient.

Our interest lies in the significance of the ARCH and GARCH coefficients. In all the three
periods, the coefficients, C (4) and C (5) i.e., ARCH and GARCH coefficients respectively
are significant. This implies that both the last period’s squared residuals and the conditional
variance are internal shocks to the volatility of the dependent variable. In the period before the
crisis, during crisis and after crisis, the sum of coefficients is 096, 0.99 and 0.99(approximately)
respectively; this is close to unity, implying persistence of high volatility during these
periods. Thus, the volatility not only persisted before the 2008 global financial crisis but
also has increased. The NSE in India is a mirror of what happened with US economy.

CONCLUSION
The present study investigated the long run relationship and short term linkages between
22
NSE and NYSE by using the daily data of NSE 100 and NYSE 100 Composite from April
2004 – March 2014. Our main findings are as follows: Indian Stock market is statistically
significantly integrated with US Stock Market. There exist long run relationship between
NSE and NYSE. Granger causality test proved that there is bi-directional causality between
NYSE and NSE for before crisis period. For the period during crisis, there is no feedback
from NYSE return to NSE return and vice versa. For the period after crisis, there is a
unidirectional relationship from NYSE returns to NSE return. Both the markets are highly
volatile as indicated by the results of GARCH (1, 1) model. Both long run equilibrium and
short run dynamics detected in this study indicates and confirms that Indian Financial
liberalization has opened the cross financial movements across borders and hence Indian
stock markets exhibits similar and tighter co-movements with other Markets especially USA
markets and that they are more integration than ever due to closer financial and economical
linkages. Co-integrated markets imply financial stability among stock markets.

REFERENCES
 Ahmad, Ashraf and Ahmed(2005), “Is the Indian Stock Market Integrated with the
US and Japanese Markets?: An Empirical Analysis”, South Asia Economic Journal,
Vol.6(2), pp.193-206
 Arshad H, Saleem H M N & Shoaib A (2008), “Long-Run Relationships between an
Emerging Equity Market and Equity Markets of the Developed World an Empirical
Analysis of Karachi Stock Exchange”, International Research Journal of Finance and
Economics, Vol 1, No. 16, pp. 52-62.
 Bala A, & Mukund KS (2001), “Interrelationship between Indian and US Stock
Markets”, Journal of Management Research, Vol.1, No.3, pp.141-148.
 Bracker K, Docking D S & Koch P D (1999), “Economic determinants of evolution
in international stock market integration”, Journal of Empirical Finance, Vol. 6, pp.
1-27.
 Mukherjee D, “Comparative Analysis of Indian Stock Market with International
Markets”, Great Lakes Herald – April 2007 Volume 1, Issue 1, pp. 39-71
 Eun and Shim,(1989), “International Transmission of Stock Market Movements”,
Journal of Financial and Quantitative Analysis, Vol. 24(2), pp.241-255

23
 Hadi M A H, (2003), “Financial Integration of Stock Markets in the Gulf: A
Multivariate Cointegration Analysis”, International Journal of Business, Vol.8, No.3,
pp. 335-346.
 Ilhan M, Joe K, Linguo G & Meric Gulser (2012), “Co-movements of and Linkages
between Asian Stock Markets”, Business and Economics Research Journal, Vol. 3
No. 1, pp. 1-15.
 Iqbal A, Khalid. N, and Rafiq. S., (2011), “Dynamic Interrelationship among the
Stock Markets of India, Pakistan and United States”, International Journal of Human
and Social Sciences, Vol. 6, Issue No. 1
 Janak R, & Sarat D, (2005), “Integration of India’s stock market with global and
major regional markets”, BIS papers No. 42, pp. 203-236
 Johansen S (1988), “Statistical analysis of cointegration vectors”, Journal of
Economic Dynamics and Control, No. 12, pp. 231-254
 Lamba, AS, (2005), “An Analysis of the Short- and Long-Run Relationships
between South Asian and Developed Equity Markets”, International Journal of
Business, Vol. 10, Issue 4
 Lim K, Lee H & Liew K, (2003), “International diversification benefits in ASEAN
stock markets: a revisit”. Mimeo, Lebuan School of International Business and
Finance, Universiti Malasya Sabbah, Faculty of Economics and Management,
University Putera Malaysia
 Liu Y A, Pan M S & Shieh J C P (1998), “International Transmission of Stock Price
Movements: Evidence from the U.S. and Five Asian-Pacific Markets”, Journal of
Economics and Finance, Vol.22, No.1, pp. 59-69
 Mandaviya J, (2014), “Indian Stock Market Influenced By Global Stock Market- A
Study Of Select World Major Stock Markets”, Samzodhana – “Journal of
Management Research , Vol. 2, Issue 1
 Masih A, & R Masih, (2001), “Long and short term dynamic causal transmission
amongst international stock markets”, Journal of International Money and Finance,
No. 20, pp. 563-587
 Masih, A.M.M., & Masih, R., (1997), “A Comparative Analysis of the Propagation
of Stock Market Fluctuations in Alternative Models of Dynamic Casual Linkages”,
Applied Financial economics, Vol. 7, Issue 1, pp 59-74
24
 Mishra A K, (2002), “International financial integration of domestic financial
markets: A study of India”, The ICFAI Journal of Applied Finance, Vol. 8, pp. 5-15
 Modi G A, Patel K B, Patel R N (2010), “The Study on Co-Movement of Selected
Stock Markets”, International Research Journal of Finance and Economics, No. 47,
pp. 164-179
 Mukherjee T K & Naka A (1995), “Dynamic relations between macroeconomic
variables and the Japanese stock market: An application of a vector error correction
model”, Journal of Financial Research, No. 18, pp. 223-237
 Mukherjee D, (2007), “Comparative Analysis of Indian Stock Market with
International Markets”, Great Lakes Herald, Vol. 1, pp. 39-71
 Padhi P, (2011), “On the Linkages among Selected Asian, European and the US
Implied Volatility Indices”, NSE Working Paper, pp. 1-35
 Sharkasi, Ruskin A, Heather J & Martin C (2005), “Interrelationships among
International Stock Market Indices: Europe, Asia and the Americas”, International
Journal of Theoretical and Applied Finance, Vol.8, No.5, pp. 1-18
 Dr Saif Siddiqui, “Examining Associations between S&P CNX Nifty and selected
Asian & US Stock Markets”, NSE Research papers, 2009
 Srivastava A, Qamar F, & Khan M, (2012), “An analysis of Cointegration on Indian
Stock Market with Global Stock Markets”, Finance India, Vol. XXVI, No. 2, pp.
513-530.
 Wong WK, Agarwal A, and Du J, (2005), “Financial Integration for Indian Stock
Market: A Fractional Co-integration Approach”, Working Paper No. 0501,
Department of Economics, National University of Singapore

25