Professional Documents
Culture Documents
The financial sector liberalization in India was initiated on the basis of the recommendations of the
Narasimham Committee on Financial System. In particular, the reform of the Indian stock market was
guided by two broad themes—structural transformation and speedy access to information. A decade of
reform has brought about a radical transformation of the Indian stock market. This study employed the
GARCH-M model to test the efficiency of the stock market in India, along with various autocorrelation
tests for different sub-samples, using the daily data on S&P CNX Nifty from January 1, 1996 to March
31, 2005. The results suggest that there are periods of efficiency in its weak form, followed by
inefficiency. The observed inefficiency, particularly after the introduction of futures and option in returns
may be due to some dynamics in higher order moments of the residuals.
Introduction
The financial sector liberalization that started in India in the early 1990s was based on
the various recommendations provided in the Narasimham Committee Report on the
Financial System, submitted in September 1992. Financial market reform was an integral
part of the financial sector reform that brought gradual improvement in the functioning
of the Indian stock market. These reforms were aimed at enhancing competition,
transparency and efficiency in the Indian financial market.
The reform in the Indian stock market brought a paradigm shift in the functioning and
structure of the market that was guided by two broad themes—structural transformation
and speedy access to information. On one hand, massive institutional reforms were
undertaken to improve the functioning of the market. This resulted in greater
participation of both individual and institutional investors. On the other hand, advances
in information technology facilitated computerization on a large scale and Internet
trading facilitated trading from any part of the country.
Few notable developments in this direction were; the permission given to the private
sector to enter into the mutual fund industry (earlier dominated by Unit Trust of India
(UTI) since 1992-93). Since September 1992 foreign institutional investors were
permitted to invest in Indian capital market and Indian companies were allowed to raise
funds abroad through GDRs and ADRs. The Indian capital market integrated more with
the world market since 1999, when Indian companies started listing on foreign stock
exchanges, especially in the Nasdaq.
Besides this, certain institutional reforms such as, the setting up of the National Stock
Exchange (NSE) and the Securities and Exchange Board of India (SEBI), introduction of
† This paper was presented at the conference jointly organized by The Icfai-Philadelphia University
in December 2005 at Hyderabad.
* Former Research Associate, The Icfai University, Andhra Pradesh, India.
E-mail: ashman24@rediffmail.com
* * Reader, University of Hyderabad, Hyderabad, Andhra Pradesh, India.
E-mail: omkarss@uohyd.ernet.in
Random Walk
The random walk, in its simplest form represented as:
Pt Pt 1 t
where, Pt is the price at time ‘t’ and t is the error term which has zero mean and whose
values are independent of each other. The price changes, Pt Pt Pt 1 is thus simply t
and is hence independent of past price changes and current price is the summation of all
past errors:
t
Pt t
i 1
Thus, the random walk model implies that prices are generated by purely random
changes.
To test random walk hypothesis one of the most direct ways for an individual time
series is to check for autocorrelation.
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 73
Serial Correlation Analysis
The independence of the stock price changes is also determined by the Sample
Autocorrelation Function (SACF) which measures the amount of linear dependence
between observations in a time series that are separated by lag k . The SACF is defined
as:
nk
( xt x )( xt k x )
k t1
nk
t 1
( xt x )2
If the price changes of the stocks are independently distributed, the k will be zero for
all time lags.
The Q-Statistic
The Q-statistic is given as:
m
Q n ˆ k2
k 1
where, n=sample size and m=lag length. The Q statistics is use to test the joint
hypothesis that all the serial correlation for lags 1 through m are simultaneously zero
against the alternative hypothesis that at least some of them are not equal to zero. This
Q statistics in a large sample is approximately distributed as the chi-square distribution
with ‘m’ degree of freedom. If the calculated Q is greater than the critical Q value at the
chosen level of significance from the chi-square distribution, one can reject the null
hypothesis in favor of alternate hypothesis. The rejection of null hypothesis implies
dependency of the series, which violates the random walk process.
ˆ k2 m
LB n(n 2)
k 1 n k
which has better small sample properties than the Q statistics. However, in a large sample
both follow the chi-square distribution with m degree of freedom.
ht 0 1 t21 2 ht 1
where, ht is the conditional variance and t is an error term.
0.15
0.10
0.05
Log Returns
0.00
–0.05
–0.10
–0.15
500 1000 1500 2000
No. of Observations
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 75
Besides this, the relatively large value of the Table 1: Descriptive Statistics
Kurtosis (7.79) suggests that data is leptokurtic or Mean –0.0003
heavily tailed and sharply peaked around the mean.
Median –0.0006
Jarque-Bera statistic also rejects the null of normality.
The above descriptive statistics corroborated the well Standard Deviation 0.0100
known fact that daily stock returns are not normal, Skewness 0.2700
but leptokurtic and skewed. Kurtosis 7.7900
Table 3B: SACFs, PACFs and Ljung-Box Statistic, Sample Period: Jan. 1997–Dec. 1997
Panel A: SACFs
1. 0.018 0.026 0.016 0.002 –0.074 –0.137 –0.012 –0.004 –0.039 0.060
11. –0.024 –0.124 0.084 0.082 –0.035 –0.064 0.088 –0.140 –0.013 –0.061
21. 0.060 –0.045 0.045 0.063 0.014
Panel B: PACFs
1. 0.018 0.026 0.015 0.001 –0.075 –0.135 –0.005 0.006 –0.034 0.058
11. –0.044 –0.150 0.089 0.089 –0.044 –0.062 0.067 –0.181 0.030 –0.030
21. 0.039 –0.037 0.041 –0.011 0.034
Panel C: Ljung-Box Statistic
Lags Ljung-Box Statistic P-value
Q(1) 0.078 0.78
Q(5) 1.66 0.89
Q(10) 7.69 0.65
Q(15) 15.73 0.40
Q(25) 28.33 0.29
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 77
Table 3C: SACFs, PACFs and Ljung-Box Statistic, Sample Period: Jan. 1998-Dec. 1998
Panel A: SACFs
1. 0.027 –0.047 0.023 –0.081 0.010 0.010 –0.024 0.039 0.082 0.085
11. 0.058 0.023 –0.061 0.026 –0.004 0.041 –0.034 –0.092 –0.048 0.007
21. 0.051 –0.064 –0.128 0.020 0.018
Panel B: PACFs
1. 0.027 –0.048 0.026 –0.085 0.018 0.000 –0.019 0.034 0.081 0.087
11. 0.058 0.033 –0.049 0.040 –0.007 0.053 –0.053 –0.092 –0.073 –0.014
21. 0.034 –0.083 –0.132 0.009 0.011
Panel C: Ljung–Box Statistic
Lags Ljung-Box Statistic P-value
Q(1) 0.18 0.66
Q(5) 2.59 0.76
Q(10) 6.83 0.74
Q(15) 9.03 0.87
Q(25) 19.32 0.78
Table 3D: SACFs, PACFs and Ljung-Box Statistic, Sample Period: Jan. 1999-Dec. 1999
Panel A: SACFs
1. –0.059 –0.027 –0.005 0.058 0.033 –0.101 –0.098 –0.034 0.031 0.108
11. –0.084 0.006 0.022 –0.045 –0.029 –0.107 –0.045 –0.065 0.034 0.036
21. –0.071 0.015 0.021 –0.109 0.031
Panel B: PACFs
1. –0.059 –0.030 –0.008 0.057 0.040 –0.094 –0.109 –0.056 0.016 0.124
11. –0.049 0.002 –0.004 –0.082 –0.041 –0.089 –0.051 –0.075 0.023 0.035
21. –0.061 –0.024 –0.019 –0.143 0.008
Panel C: Ljung-Box Statistic
Lags Ljung-Box Statistic P-value
Q(1) 0.18 0.66
Q(5) 2.59 0.76
Q(10) 6.83 0.74
Q(15) 9.03 0.87
Q(25) 19.32 0.78
11. –0.082 –0.070 –0.013 –0.049 0.056 0.057 0.142 –0.063 –0.143 –0.100
21. 0.022 –0.059 0.136 0.093 0.011
Panel B: PACFs
1. 0.072 –0.010 –0.035 0.013 –0.081 0.018 0.016 0.043 0.034 0.017
11. –0.082 –0.056 0.002 –0.052 0.065 0.035 0.126 –0.079 –0.139 –0.062
Table 3F: SACFs, PACFs and Ljung-Box Statistic, Sample Period: Jan. 2001-Dec. 2001
Panel A: SACFs
1. 0.172 –0.134 –0.050 0.048 0.123 –0.068 –0.029 –0.009 0.076 0.059
11. 0.009 –0.094 0.014 0.143 0.001 –0.008 –0.115 0.028 0.000 –0.141
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 79
Table 3G: SACFs, PACFs and Ljung–Box Statistic, Sample Period: Jan. 2002-Dec. 2002
Panel A: SACFs
1. –0.026 –0.038 0.046 0.123 0.019 0.023 –0.035 0.080 0.035 –0.122
11. –0.009 –0.015 0.038 0.031 –0.106 –0.078 0.097 0.000 –0.036 –0.023
21. 0.022 0.034 0.024 0.016 0.075
Panel B: PACFs
1. –0.026 –0.039 0.044 0.124 0.030 0.032 –0.044 0.063 0.029 –0.121
11. –0.014 –0.045 0.040 0.057 –0.091 –0.076 0.066 0.017 0.003 –0.024
21. 0.004 0.020 0.046 0.057 0.048
Table 3H: SACFs, PACFs and Ljung-Box Statistic, Sample Period: Jan. 2003-Dec. 2003
Panel A: SACFs
1. 0.165 0.006 0.056 0.016 0.002 0.009 –0.007 –0.005 –0.019 0.002
11. 0.048 0.087 –0.004 –0.093 –0.001 –0.052 0.042 0.075 0.040 0.086
21. 0.036 0.046 0.034 0.047 –0.053
Panel B: PACFs
1. 0.165 –0.021 0.060 –0.003 0.001 0.006 –0.010 –0.002 –0.019 0.009
11. 0.047 0.076 –0.031 –0.094 0.021 –0.060 0.075 0.057 0.030 0.079
21. 0.001 0.036 –0.003 0.035 –0.066
Panel C: Ljung–Box Statistic
Lags Ljung-Box Statistic P-value
Q(1) 6.59 0.008
Q(5) 7.84 0.16
Q(10) 7.98 0.63
Q(15) 12.95 0.60
Q(25) 20.89 0.69
Table 3I: SACFs, PACFs and Ljung-Box Statistic, Sample Period: Jan. 2004-March 2005
Panel A: SACFs
1. 0.086 –0.225 0.059 0.142 -0.065 –0.116 –0.032 -0.061 0.011 0.148
11. -0.020 –0.112 0.065 0.123 -0.120 –0.098 –0.009 0.042 0.069 –0.012
21. 0.010 0.021 -0.001 0.071 -0.027
Panel B: PACFs
1. 0.086 –0.234 0.110 0.075 -0.058 –0.063 –0.060 –0.100 0.041 0.143
11. –0.036 –0.051 0.033 0.041 -0.097 –0.009 –0.068 0.037 0.096 –0.025
21. 0.046 –0.003 –0.057 0.076 –0.007
Panel C: Ljung-Box Statistic
Lags Ljung-Box Statistic P-value
Q(1) 2.34 0.12
Q(5) 27.35 0.00
Q(10) 40.48 0.00
Q(15) 55.93 0.00
Q(25) 63.49 0.00
Non-linearity in Returns
As mentioned earlier, if stock prices are non-linear, standard tests of efficiency like
autocorrelation and random walk which are incapable of capturing non-linearity may lead
to misleading inferences regarding efficiency. The EMH in the presence of non-linearity
is tested by the significance of the coefficients of 0 and 1 in the following equation:
Rt 0 1 Rt21 t
If the EMH holds, then, 0 1 0 and t is a white noise process. The Wald test
was conducted to test whether the coefficient of squared returns is statistically different
from zero. The results of Wald test is reported in Table 4. The Wald test results for the
entire sample period—January 1996 to March 2005—shows that it rejects the null
hypothesis: 0 = 1 =0 implying that the market was inefficient.
The null hypothesis for the sub-sample period January 1996 to December 1996 was also
rejected at 1% level. However, the Wald test results for sub-sample January 1997 to
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 81
Table 4: Wald Test
Coefficients F-statistic Chi-square
Full Sample January 1996 to March 2005
0=1=0 9.91 19.82
(0.000052) (0.000050)
January 1996 to December 1996
0=1=0 22.55 45.10
(0.00) (0.00)
January 1997 to December 1997
0=1=0 0.23 0.47
(0.78) (0.78)
January 1998 to December 1998
0=1=0 1.42 2.84
(0.24) (0.24)
January 1999 to December 1999
0=1=0 3.66 7.32
(0.027) (0.025)
January 2000 to December 2000
0=1=0 6.14 12.28
(0.002) (0.002)
January 2001 to December 2001
0=1=0 9.47 18.94
(0.00) (0.00)
January 2002 to December 2002
0=1=0 0.38 0.77
(0.67) (0.67)
January 2003 to December 2003
0=1=0 3.71 7.43
(0.02) (0.02)
January 2004 to March 2005
0=1=0 32.32 64.65
(0.00) (0.00)
December 1997 and January 1998 to December 1998, show that the null hypothesis
cannot be rejected at 10% level in each case. Thus, the market was efficient
i.e., non-linearity was absent in stock returns over the period 1997 to 1998. However, the
Wald test indicates that the market was inefficient over the period 1999 to 2001 but was
efficient during January 2002 to December 2002. Again, the Wald test for sub-samples
January 2003 to December 2003 and January 2004 to March 2005 shows that the market
was inefficient as the null hypothesis was rejected at 1% level. In reject years, it seems that
the market overreacted to good and bad news.
GARCH-M Model
This study also employs GARCH-M model to test market efficiency. When the market
risk premium is too high, the volatility will also follow, leading to a false rejection of
inefficiency. To test this, GARCH-M model is estimated which incorporates time-varying
risk premium.
0 –0.000 –0.000 0.00 0.00 0.00 –0.00 –0.00 – 0.00 –0.00 0.00
(–0.98) (–0.80) (0.42) (0.20) (0.12) (–0.37) (–0.10) (–.033) (–2.17) (0.88)
1 0.106 0.081 0.18 –0.001 0.23 0.10 0.002 0.07 0.15 0.12
(4.48) (1.12) (2.60) (–0.01) (3.12) (1.40) (0.03) (0.93) (1.62) (1.43)
2 –0.024 –0.06 –0.01 –0.04 –0.10 –0.02 –0.08 0.04 –0.01 –0.06
(–1.06) (–1.00) (–0.19) (–0.81) (–1.60) (–0.31) (–1.01) (0.56) (–0.26) (–0.76)
3 0.051 0.13 0.03 0.02 –0.002 0.01 –0.04 –0.002 0.01 0.05
(2.44) (2.36) (0.42) (0.32) (–0.04) (0.21) (–0.52) (–0.02) (0.17) (0.79)
–1.16 –0.89 –21.80 –9.38 –3.43 3.48 –7.14 4.31 21.40 –31.01
(–0.45) (–0.14) (–0.96) (–0.23) (–0.50) (0.37) (–0.60) (0.37) (1.94) (–0.84)
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 83
It is a known fact that financial data often possess time varying volatilities
characterized by GARCH and its variants. BDS would be an appropriate test for testing
the null of no autocorrelation in returns of higher order. Under BDS test, in principle no
distributional assumptions are made about the data under the null hypothesis other than
that it is independent identically distributed (i.i.d) and can be interpreted as a test for
nonlinearity.
In the first step, the ARCH (1) is fitted to the data, thus eliminating non-linearity
of order two from the data. In the second step, BDS test is applied by running it on the
residuals of the ARCH (1) model so that any dependence found in the residuals must be
non-linear in nature of higher order.
This test for independence is based on estimation of correlation integrals at various
dimensions. BDS statistic estimation is non-parametric, which asymptotically follows a
normal distribution.
Tm ( )
BDS n m 1
Vm ( )
where n is the total number of observations, Tm(Cm(C1(, Cm( and C1( are
the correlation integrals, Vm( is the standard error of Tm( and and m are the distance
and dimension respectively, whose values are chosen by the investigator. In most of the
cases the values of used are 0.5 and the value of m is fixed in line with the number of
observations; for example using only 5 m if 500 n .
S&P Nifty returns are filtered for linear dependence using ARCH (1). To study
whether the higher order dependence structure can be sufficiently accounted by ARCH
(1), ARCH (1) standardized residuals are then tested for independently identically
distributed (i.i.d) using BDS test. If the null hypothesis of i.i.d. is rejected by the BDS test,
we then suspect that observed inefficiency in returns may be due to some dynamics in
higher order (greater than second) moments of the residuals. Towards this end, this study
tested for autocorrelation of ˆt3 and ˆt4 by Ljung-Box statistic to detect 3rd and 4th order
dependences in the residual series.
The results of BDS test is reported in Table 6. The first column of the table provides
the values of the distance, , measured in terms of half (0.5). The second column gives
present number of embedding dimensions, m and the value of test statistics are given in
the 3rd column of Table 6. The BDS test results show that for = s.d of linear filtered
data with m = 3, 4 and 5 the null of i.i.d was rejected for the period after the introduction
of futures and option trading that indicates that the inefficiency observed in returns may
be due to some dynamics in higher
Table 6: BDS Test
order moments of the residuals.
After Futures Trading After Option Trading
Autocorrelation analysis of the m BDS Statistic m BDS Statistic
residual of 3rd and 4th degree by the 0.5 2 0.076 2 0.206
help of the Ljung-Box statistic is 0.5 3 2.56* 3 1.91*
reported in Table 7. The results 0.5 4 3.24* 4 2.28*
show that the null hypothesis of no 0.5 5 4.10* 5 3.25*
autocorrelation can be rejected at Note: * Indicates significance at 1% level.
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 85
rejected at lag 1, 10, 15, 20 and 25 at 1% level for cubic residuals. The Ljung-Box statistic
for 4th degree residuals shows that the null hypothesis of no autocorrelation cannot be
rejected at lags 1, 5, 10, 15 and 25 at 10% level of significance implying no fourth degree
autocorrelation in the residuals of S&P returns for the period after the introduction of
option trading.
Conclusion
The study concludes that there is clear tendency towards market efficiency in the Indian
context. Though the market is not efficient over the entire sample period, the results of
sub-samples show that the market was efficient over certain sub-samples. Second,
inefficiency is due to non-linearity in returns and, higher order of non-linearity cannot
be ruled out. Third, The GARCH-M coefficient shows insignificant negative relationship
between stock returns and risk premium implying predictability, and this predictability is
due to inefficient pricing rather than changes in the perception of risk premia. The CAPM
does not hold in the Indian context. In short, there are periods of efficiency in its weak
form followed by inefficiency. Thus, there are traces of tendency towards market efficiency
in the Indian stock markets due to a radical reform undertaken in the last decade. The
observed inefficiency after the introduction of futures and option in returns may be due
to some dynamics in higher order moments of the residuals.
Reference # 01J-2007-02-05-01
References
1. Campbell J Y and Shiller R J (1988), “Stock Prices, Earnings and Expected
Dividends”, Journal of Finance, Vol. 43, pp. 661-676.
3. Cowles A (1944), “Stock Market Forecasting”, Econometrica, Vol. 12, pp. 206-214.
5. Cutler D M, Poterba M and Summers L H (1989), “What Moves Stock Prices”, Journal
of Portfolio Management, Vol. 15, pp. 4-12.
Are There Trends Towards Market Efficiency? A Study of the Indian Stock Market 87