Modeling Asymmetric Volatility in Indian Stock Market

Pacific Business Review International
Volume 6, Issue 9, March 2014
Modeling Asymmetric Volatility in Indian Stock Market
CMA. Potharla Srikanth*
*Assistant Professor
Commerce Dept., Post Graduate College
Constituent College of Osmania University
Secunderabad. A.P.
The main feature of any financial instruments is stochastic nature of its
returns. The spread of outcome of return from the asset is called volatility
which influences numerous financial decisions. The main goal of estimating
the volatility is to assess the market risk. Volatility is one of the key parameters
for pricing financial derivatives. Estimation of volatility helps in risk
management and it also helps in efficient management of portfolios. In the
present study, asymmetric nature of volatility is modeled by applying two
popularly used asymmetric GARCH models i.e., GJR-GARH model and
PGARCH model. BSE-Sensex is used as a proxy for Indian stock market and
period of study is from 1st july, 1997 to 30th march,2013. Results of
Augmented Dickey Fuller test reveals that the natural logarithmic values of
Sensex returns and S&P 500 returns are stationary at their level form. Results
of analysis with selected asymmetric GARCH models reveals the presence of
leverage effect in Indian stock market and it also confirms the effect of
periodic cycles on the conditional volatility in the market. In order to test
whether the selected asymmetric GARCH models adequately captured the
persistence in volatility and to test whether residuals from the selected models
are free from ARCH effect, ARCH LM test is conducted. Results of ARCHLM test concludes that there is no ARCH effect left in residuals obtained from
both GJR-GARCH Model and PGARCH Model estimations.
Asymmetric volatility, Leverage effect, Indian stock market, GJR-GARCH
Model , PGARCH Model, Conditional volatility.
JEL Classification: C13, C52, C53, C58, G10, G17
In the recent past, there is a growing importance for estimating and analyzing
volatility. Volatility has its impact on many issues in economics and finance.
Volatility indicates the fluctuations detected in some phenomenon over a time.
In terms of modeling and forecasting literature, it means “the conditional
variance of the underlying asset return”(Tsay,2010). While estimating
regression equation under tradition OLS model, one fundamental assumption
is variance of all the squared residuals is homoscedastic, it means that all the
squared residuals from regression estimation have the same level of variance.
But, in the course of time, various empirical studies have showed that stock
returns exhibit heteroskedasticity which means variance of squared error
w w w. p b r. c o . i n
Pacific Business Review International
terms are not equal. For example, in the seminal work done by
Mandelbrot(1963), he found that there will be a tendency of large
changes in asset prices are succeeded by large changes and small
changes are succeeded by small changes and it can be either
positive or negative. After the seminal work of Mandelbrot(1963),
there was growing importance for the new innovative regression
models that can capture the heteroscedasticity effectively, so that
valid estimation can be done with the regression models.
investor base. Important reforms introduced in recent past
includes establishment of Securities Exchange Board of
India(SEBI) in 1992 as governing body; introduction of screen
based trading system, T+2 rolling settlement system,
demutualization of stock exchanges, Demat form of trading,
establishment of Clearing and Settlement houses, introduction of
Straight through processing system, Value at Risk(VaR) based
Margin system and many a more.
In 1982, Engle proposed Autoregressive Conditional
Heteroscedasticity(ARCH) model to analyse volatility by relating
conditional variance of error term to the linear combination of the
squared error term with one period lagged values. The major
limitation in the ARCH model was that the long lag lengths leads to
large parameters and hence to overcome this limitation in 1986
Bellerslev introduced Generalized Autoregressive Conditional
Heteroscedasticity(GARCH) model by modeling the conditional
variance to depend on its lagged values as well as squared lagged
values of the error term. Seminal works of Engle and Bellerslev led
to increased interest in the study of volatility. Various people like
academicians, investors, policy makers have started doing
research on modeling conditional volatility. Though, ARCH and
GARCH models are doing well in modeling volatility, major
limitation in these models was they could not capture asymmetric
effect. Asymmetric effect indicates that markets responds
differently to negative and positive news. This type of asymmetric
effect can be captured by applying various asymmetric GARCH
models like Exponential GARCH (EGARCH) model proposed by
Nelson in 1991 ; the GJR-GARCH model introduced by Glosten,
Jagannathan and Runkle in 1993 and Power GARCH(PGARCH)
model introduced by Higgins and Bera (1992) etc.,
Literature review
Origin and growth of Indian stock market
The origin of Indian stock market can be traced back to the end of
the eighteenth century in which long-tern negotiable instruments
were issued for the first time. However, after the enactment of
companies act in 1850 only the actual beginning of the of the stock
market in India had been started. The main feature of companies
act 1850 was the provisions relating to business organizations with
limited liability which motivated many investors in India to invest
in corporate securities. In 1875, 22 enterprising brokers under a
Banyan tree established the Native share and stock brokers
association at Bombay which was the predecessor of current
Bombay Stock Exchange. It was followed by the formation of
Associations/Exchanges in Ahmedabad(1894), Calcutta(1908)
and Madras(1937).
Geert Bekaert et .,al(2000) investigated leverage effect and
time-varying risk premium explanations of the asymmetric
volatility. They conclude that mechanism behind the asymmetry
for the high and the medium leverage portfolio is covariance
asymmetry. They also conclude that negative shocks increase
conditional covariance substantially, whereas positive shocks
have a mixed impact on conditional covariance. Christos
Floros(2008) examined the application of asymmetric GARHC
models for modeling volatility and explaining financial market
risk. They used two major indices i.e., CMA General index from
Egypt and TASE-100 index from Israel. Results for the study
provide strong evidence that asymmetric GARCH models can
better explain volatility in two countries stock markets. Results of
the study also concludes that increased risk will not necessarily
lead to increased return in the market. E. Abounoori et (2011)
studied the nature of stock market volatility in Tehran Stock
exchange and in their study they estimated GJR-GARCH model
with Gaussian innovations and fat-tailed distributions. Results of
their study concludes that effect of bad news on the volatility is
stronger than the effect of good news. Suliman Zakaria (2012)
made an attempt to model volatility in Saudi stock market TASI
index. He applied various asymmetric GARCH models like
EGARCH, TGARCH and PGARCH. He observed persistence of
conditional volatility and the results of his studies were in favour
of 'positive correlation hypothesis' which established positive
relationship between volatility and expected stock return. His
studies also confirms the presence of leverage effect in market
returns. Emenike Kalu O et (2012) analysed the response of
volatility to negative and positive news in Nigerian stock exchange
(NSE) by using daily closing prices from January 2nd 1996 to
December 30th 2011. Results of their study supported the presence
of asymmetric effect in the NSE stock returns but the study did not
confirm the presence of leverage effect. The study provides
evidence in support of positive news producing higher volatility in
the immediate future than negative news with the same magnitude.
Data and methodology
After the initiation of liberation measures in 1991, it was found
necessary to make the Indian stock market trading system on par
with international standards. On the basis of recommendations of
Pherwani committee, the National Stock Exchange was
incorporated in 1992. The key promoters of NSE are Industrial
Development Bank of India, Industrial Credit and Investment
Corporation of India, Industrial Finance Corporation of India, all
Insurance Corporations, selected commercial banks and others.
The period of study is from 1st july, 1997 to 30th september,2012
and BSE-Sensex is used as surrogate index for Indian stock market
and S&P 500 index is used as indicator of US stock market
conditions. Natural logarithmic values are computed for dailywise returns of Sensex and S&P500 and expressed in percentages.
Descriptive statistics used in the study are mean, median,
maximum, minimum, standard deviation, skewness and kurtosis.
Over the last 125 years, securities market in India is striving hard to
make its trading system operationally and informationally
efficient and to conform itself to international standards. Many
reforms have been initiated in Indian securities market which
helped to augment its growth in terms of trading volume and
Unit root test has been conducted to test the stationarity of dailywise log returns and after analyzing the basic feature of time series
data i.e., stationarity, the next step is to model the asymmetric
volatility present in the market returns by employing two
asymmetric GARCH models i.e., GJR-GARCH model and
w w w. p b r. c o . i n
Volume 6, Issue 9, March 2014
PGARCH model.
GJR-GARCH model is proposed by Glosten, Jagannathan and
Runkle in 1993. It allows the conditional variance to respond
differently to the negative and positive innovations. The model
specification is as follows,
Mean Equation
In equation(1), Rt is the Sensex return at time 't' and )log(t2s is the
natural logarithmic value of GARCH; )1(AR is first order
Sum of
indicates persistence of shocks in the
volatility. If its value is less than one, it signifies that shock is not
expected to last for a long period and if its value is closer to one,
then shock is expected for a long period. If it is equal to one, then
shock is going to effect volatility for indefinite future.
Higgins and Bera proposed Power GARCH (PGARCH) model in
1992. Unlike other GARCH models, PGARCH model used
standard deviation instead of the variance. In this model, power
parameter is estimated unlike other models in which it is generally
imposed on the model itself. The specification of the model is as
Mean Equation
Variance Equation
In equation (2)
is squared standard deviation at time t, w is
constant and it implies unconditional volatility( or long run
is squared value of one day lagged error term and
is its coefficient;
bad news i.e.,
In equation(3), Rt is the Sensex return at time 't' and
is the natural logarithmic value of GARCH; )1(AR is first order
Variance Equation
is a dummy variable which gives value '1' for
<0 and value '0' for good news i.e.,
g is the coefficient for the product of
and it tests the
null hypothesis that g = 0 which indicates that the news effect is
symmetric. If g<0 (or negative) it indicates that positive shocks
In equation(4),
standard deviation raised to the power of
d;w is constant and it implies unconditional volatility;
modulus value of one day lagged error term and
is its
is one day lagged error term raised to the power
of d and 1gis its coefficient; 1-tds is one day lagged standard
tend to produce more volatility in the near future than negative
shocks. If g>0 ( or positive), it points out that negative shocks tend
to produce more volatility in near future than positive shocks.
Results and Discussion
is one day lagged value of squared standard deviation and 1b is its
coefficient which indicates volatility clustering.
Descriptive statistics of Sensex returns and S&P 500 returns
from 1st July, 1997 to 30th September, 2012.
w w w. p b r. c o . i n
deviation raised to the power of dand 1b is its coefficient. fis the
coefficient of one day lagged log returns from S&P500.
Pacific Business Review International
Table 1 presents descriptive statistics of Sensex returns and S&P
500 returns during the study period. As shown in the table, mean
return of Sensex is 0.033% with a standard deviation of 1.666%,
whereas the mean return of S&P 500 is 0.0065% with a standard
deviation of 1.3115%. Skewness of the distributions of Sensex
returns and S&P 500 returns are negative which indicates longer
left tail of the distribution showing large number of high values in
the distribution and comparatively higher value of negative
skewness of S&P500 returns indicates that US stock market is
more volatile than Indian stock market during the period. Kurtosis
of the distributions of Sensex returns and S&P 500 returns are
leptokurtic which implies that the values in the distribution are
closer to their mean value and lesser erratic swings present in the
Presence of non-stationarity is a common problem in time-series
data. Hence, stationarity of the data series is tested by using
augmented version of Dickey Fuller test. Results of the test
concludes that Sensex returns and S&P 500 returns are stationary
at their level form (p<0.01).
w w w. p b r. c o . i n
Volume 6, Issue 9, March 2014
Table 3 presents results of analysis under GJR-GARCH model. In
the mean equation, constant value is positive and significant
(p<0.05). Coefficient of log value of GARCH is negative and
significant which highlights that the exponential nature of
conditional volatility has its impact on stock market
returns(p<0.05). Statistically significant positive value of AR(1)
indicates the presence of first order positive autoregression effect
on mean returns(p<0.05). In the variance equation, constant which
represents unconditional volatility is positive and significant
(p<0.01) which points out the presence of seasonality effect on
Indian stock market. Squared value of one day lagged residual,
which surrogates the impact of recent news effect, is positive and
significant indicating the quick reflection of recent news in the
stock prices in the market(p<0.05). Asymmetry parameter is the
product of one day lagged squared value of error term and dummy
variable which gives value '1' for bad news and value '0' for good
news. When 'good news' is present, then g will become zero and the
model collapses to the standard GARCH form. In the present
analysis, asymmetry parameter is positive and
significant(p<0.05). According Carter(2007), this type of situation
indicates that negative shocks (i.e, bad news) have a large effect on
conditional volatility than the positive shocks(i.e., good news) of
same magnitude. One period lagged GARCH which proxies the
impact of old news is positive and significant indicating the
significant impact of old news on the market volatility and it also
indicates the presence of stylized facts such as volatility clustering
in conditional variance(p<0.05). Sum of /2+1+1gbaindicates
the persistence of shocks on the volatility. In the present case, its
value is 0.9807 which is closer to one signifying that shock is
expected to last for a long period. ARCH-LM test for
Heteroskedasticity is conducted on residuals estimated from the
GJR-GARCH model. Results of the test indicates that residuals
derived from the regression estimation are free from
heteroscedasticity (p>0.05) and variance equation is well specified
for the GJR-GARCH model.
Table 4 shows the analysis results of PGARCH model. In the mean
equation, constant is positive and significant. Effect of log value of
GARCH and first order autoregression are same as in the case of
GJR-GARCH model(p<0.05). In the variance equation,
unconditional volatility represented by constant is positive and
significant(p<0.05). Modulus value of one period lagged residual
is positive and significant indicating significant impact of absolute
value of error term on the volatility ignoring positive and negative
signs of the error term(p<0.05). The estimated value of power term
coefficient is 1.6094 which is positive and significant (p<0.05).
When one period lagged residual and one period lagged standard
deviation are raised to the power of d, both are positive and
significant(p<0.05) which indicates the influence of power term in
ARCH and GARCH. The results of residual diagnostic tests
confirms that the residuals obtained from the PGARCH model are
not heteroskedastic (p>0.05).
w w w. p b r. c o . i n
Pacific Business Review International
In financial markets research, forecasting of volatility is crucial
task and it is essential to evaluate the performance of the selected
forecasting models. In the present study, Root Mean Squared
Error(RMSE), Mean Absolute Error(MAE), Mean Absolute
Percentage Error(MAPE) and Theil Inequality Coefficient(TI) are
used to evaluate the forecasting performance of the selected
asymmetric GARCH models and they are presented in table 5.
Based on RMSE, MAE and TI, PGARCH model outperforms
GJR-GARCH models. However, the values of forecast evaluation
measures for the two models have very low difference, so it can be
concluded that both the models are well suited for Indian stock
Final findings
In the present study, modest attempt has been made to model the
asymmetric volatility in the Indian stock market. BSE-Sensex is
used a proxy for Indian stock market. Findings of the analysis
brings out the fact that the returns from Indian stock market have
the impact of asymmetric volatility. A common phenomenon in
stock markets is that the market is more volatile to price declines
than to price rises and this type of stylized behaviour of stock
returns indicates 'leverage' effect. Leverage effect in Indian stock
market analyzed by applying two popularly used asymmetric
models i.e., GJR-GARCH and PGARCH. GJR-GARCH model
results concludes that shocks due to bad news have strong effect on
conditional volatility compared to shocks due to good news and
findings under GJR-GARCH model are consistence with the
studies by made E. Abounoori at el (2011). PGARCH model
results highlights significant influence of power term on the
conditional volatility. In the present era of global integration of
emerging stock markets like India with other world major stock
markets, presence of leverage effect in the Indian stock market
indicates that any negative shocks in the international markets can
easily spillover to Indian markets and which adversely affect the
Indian market and it was clearly evident from global financial
crisis in 2008. Hence, there is an urgent need for strengthening
control mechanism like setting circuit breakers in the market more
effectively, proper control on FII transaction and establishing good
surveillance system in the stock markets. The Government of India
should frame its monetary policy, fiscal policy and foreign trade
policy in such a way that it should give due care for promoting the
development of financial markets in the country and at the same
time the market should be well-regulated.
Carter, R. Hill, E William, and C. Lim, (2007), “ Principles of
Econometrics”, 3rd edition, New York, John wiley and
Sons, Inc.
Chirstos Floros(2008), “ Modelling Volatility using GARCH
models; Evidence from Egypt and Isreal”, Middle
Eastern Finance and Economics Journal, issue 2, pp.3141.
E n g l e , R . F. ( 1 9 8 2 ) , “ A u t o r e g r e s s i v e C o n d i t i o n a l
Heteroskedasticity with Estimates of the Variance of
U.K. Inflation,” Econometrica, 50, 987-1008.
Emenike Kalu O. and Aleke Stephen Friday(2012), “ Modeling
Asymmetric Volatility in the Nigerian Stock Exchange”,
European Journal of Business and Management, Vol.4,
No.12, pp.52-59
E. Abounoori and Y. Nademi (2011), “ The Asymmetric Effect of
News on Tehran Stock Exchange Volatility”,
International Journal of Trade, Economics, and Finance ,
Vol.2, No.4, August, 2011, pp 323-325.
Geert Bekaert and Guojun Wu (2000), “ Asymmetric Volatility and
Risk in Equity markets”, The Review of Financial
Studies, Vol.13, No.1,(spring 2000) pp.1-42.
Glosten, L.R., R. Jagannathan and D. Runkle (1993), “On the
Relation Between the Expected Value and the Volatility
of the Nominal Excess Return on Stocks,” Journal of
Finance, 48, 1779-1801.
Higgins, M.L. and A.K. Bera (1992), “A Class of Nonlinear ARCH
Models,” International Economic Review, 33, 137-158.
Mandelbrot,B., 1963, “The variation of Certain speculative
prices”, Journal of Business 36, pp394-414.
Suliman Zakaria Suliman Abdalla(2012), “Modelling Stock
Returns Volatility: Empirical Evidence from Saudi Stock
Exchanges”, International Research Journal of Finance
and Economics, Issue 85, pp.166-179.
Tsay, R.S., 2010” Analysis of Financial Time Series”, 3rd Edition,
New York, United States of America, John Wiley & Sons,
w w w. p b r. c o . i n