You are on page 1of 18

Weak Form EIIiciency

In Indian Stock Markets



Hypothesis oI Market EIIiciency is an important concept Ior the investors
who wish to hold internationally diversiIied portIolios. With increased
movement oI investments across international boundaries owing to the
integration oI world economies, the understanding oI eIIiciency oI the
emerging markets is also gaining greater importance. In this paper we test the
weak Iorm eIIiciency in the Iramework oI random walk hypothesis Ior the two
major equity markets in India Ior the period 1991 to 2006. The evidence suggests
that the series do not Iollow random walk model and there is an evidence oI
autocorrelation in both markets rejecting the weak Iorm eIIiciency hypothesis.

INTRODUCTION

EIIiciency oI equity markets has important implications Ior the investment
policy oI the investors. II the equity market in question is eIIicient researching to
Iind mispriced assets will be a waste oI time. In an eIIicient market, prices oI the
assets will reIlect markets best estimate Ior the risk and expected return oI the
asset, taking into account what is known about the asset at the time. ThereIore,
there will be no undervalued assets oIIering higher than expected return or
overvalued assets oIIering lower than the expected return. All assets will be
appropriately priced in the market oIIering optimal reward to risk. Hence, in an
eIIicient market an optimal investment strategy will be to concentrate on risk and
return characteristics oI the asset and/or portIolio. However, iI the markets were
not eIIicient, an investor will be better oII trying to spot winners and losers in the
market and correct identiIication oI miss -priced assets will enhance the overall
perIormance oI the portIolio RutterIord(1993). EMH has a twoIold Iunction - as
a theoretical and predictive model oI the operations oI the Iinancial markets and as
a tool in an impression management campaign to persuade more people to invest
their savings in the stock market (Will 2006).
The understanding oI eIIiciency oI the emerging markets is becoming more
important as a consequence oI integration with more developed markets and Iree
movement oI investments across national boundaries. Traditionally more
developed Western equity markets are considered to be more eIIicient.
Contribution oI equity markets in the process oI development in developing
countries is less and that resulted in weak markets with restrictions and controls
(Gupta, 2006)
In the last three decades, a large number oI countries had initiated reIorm
process to open up their economies. These are broadly considered as emerging
economies. Emerging markets have received huge inIlows oI capital in the recent
past and became viable alternative Ior investors seeking international
diversiIication. Among the emerging markets India has received it`s more than Iair
share oI Ioreign investment inIlows since its reIorm process began. One reason
could be the Asian crisis which aIIected the Iast developing Asian economies oI
the time (also sometimes collectively called 'tiger economies). India was not
aIIected by the Asian crisis and has maintained its high economic growth during
the period (Gupta and Basu 2005).

Today India is one oI the Iastest growing emerging economies in the
world. The reIorm process in India oIIicially started in 1991. As a result, demand
Ior investment Iunds is growing signiIicantly and capital market growth is expected
to play an increasingly important role in the process. The capital market reIorms in
India present a case where a judicious combination oI competition, deregulation and
regulation has led to sustained reIorms and increased eIIiciency (Datar and Basu
2004).

At this transitional stage, it is necessary to assess the level oI eIIiciency oI the
Indian equity market in order to establish its longer term role in the process oI
economic development. However, studies on market eIIiciency oI Indian markets are
very Iew. They are also dated and mostly inconclusive. The objective oI this
study is to test whether the Indian equity markets are weak Iorm eIIicient or not.
EMH, similar to other theories that require Iuture expected prices or returns, use past
actual prices or returns Ior the tests. Sets oI share price changes are tested Ior seria l
independence. Random walk theory Ior equity prices show an equities market in
which new inIormation is quickly discounted into prices and abnormal or excess
returns cannot be made Irom observing past prices (Poshakwale 1996).

The next section oI this paper provides a brieI background oI the Indian equity
market and a brieI literature review oI studies testing market eIIiciency in emerging
markets. Section 3 explains the methodology used in this study and data
sources, Iollowed by the results oI the analysis in section 4. The last section
summarizes the conclusions and their implications.

BACKGROUND

Equity Market In India

The reIorm process in India began in early 1990s with stock exchanges and then
spread to banks, mutual Iunds, NBFCs and oI late, to insurance companies.
However, reIorms in equity market in particular commenced in mid-1980s (Datar &
Basu 2004). Mumbai (Iormerly known as Bombay) Stock Exchange (BSE) has
always played the dominant role in the equity market in India. Traditionally,
stock exchanges were governed by brokers leading to conIlict oI interest
situation between the interest oI common investors and those oI
brokers/owners oI stock exchanges. With the establishment oI National Stock
Exchange (NSE), a new institutional structure was introduced in India that could
ensure smooth Iunctioning oI market through a combination oI new technology and
eIIicient market design. The Securities Exchange Board oI India (SEBI) was set up
as a market regulator with statutory powers to control and supervise operations oI all
participants in the capital market viz. stock exchanges, stock brokers, mutual Iunds
and rating agencies. The development oI debt market is another signiIicant
development, which has been Iacilitated by deregulation oI administered interest
rates. Opening oI stock exchange trading to Foreign Institutional Investors (FIIs)
and permission oI raising Iunds Irom international market through equity
linked instruments have introduced a degree oI competition to domestic
exchanges and other market participants. Operations oI FIIs have Iacilitated
introduction oI best practices and research inputs in trading and risk management
systems.

Mumbai stock exchange (BSE), the premier stock exchange oI India is probably the
oldest stock exchanges in Asia, established in 1875. It was initially named as
'Native Share and Share Broker Association (Poshakwale 1996). Stability in
prices Ior the BSE was considered to be an important Ieature. During the period 1987
to 1994, average annual price Iluctuations oI ordinary shares on BSE were 25.1 as
compared with London Stock Exchange(22), and the New York Stock Exchange
(23.9) (Poshakwale 1996).

In this study, to determine market eIIiciency oI equity markets in India, we
considered two stock exchanges BSE and NSE. Market capitalization oI BSE in
July, 2006 was INR 19,871 billion and that oI NSE at the same time was INR 18,487
billion. BSE is the oldest stock exchange in India and has the longest data series
available. NSE is one oI the newer stock exchanges in India. The purpose oI
establishing NSE was to provide transparency and a better Iunctioning market Ior the
investors. Because oI governments support, NSE is Iast becoming more accessible
market to domestic and Ioreign investors. The perceived liquidity and accessibility oI
the NSE market is an important Iactor and may have diIIerent impact on the market
eIIiciency. High liquidity in the market is an important pre -condition Ior the market
eIIiciency, since a thinly traded market is not in a

position to adjust to the new inIormation quickly and accurately. Thus, analysis oI
two major equity markets in India together should provide a more comprehensive
andcomplete picture.

Studies On Market Efficiency

The eIIicient market hypothesis is related to the random walk theory. The idea that
asset prices may Iollow a random walk pattern was introduced by Bachelier in 1900
(Poshakwale 1996). The random walk hypothesis is used to explain the successive
price changes which are independent oI each other.

Fama (1991) classiIies market eIIiciency into three Iorms - weak, semi-strong and
strong. In its weak Iorm eIIiciency, equity returns are not serially correlated and
have a constant mean. II market is weak Iorm eIIicient, current prices Iully
reIlect all inIormation contained in the historical prices oI the asset and a trading rule
based on the past prices can not be developed to identiIy miss-priced assets. Market
is semi-strong eIIicient iI stock prices reIlect any new publicly available inIormation
instantaneously. There are no undervalued or overvalued securities and thus, trading
rules are incapable oI producing superior returns. When new inIormation is released,
it is Iully incorporated into the price rather speedily. The strong Iorm eIIiciency
suggests that security prices reIlect all available inIormation, even private
inIormation. Insiders proIit Irom trading on inIormation not already incorporated into
prices. Hence the strong Iorm does not hold in a world with an uneven playing Iield.

Studies testing market eIIiciency in emerging markets are Iew. Poshakwale (1996)
showed that Indian stock market was weak Iorm ineIIicient; he used daily BSE index
data Ior the period 1987 to 1994. Barua (1987), Chan, Gup and Pan (1997) observed
that the major Asian markets were weak Iorm ineIIicient. Similar results were Iound
by Dickinson and Muragu (1994) Ior Nairobi stock market; Cheung et al
(1993) Ior Korea and Taiwan; and Ho and Cheung (1994) Ior Asian markets.
On the other hand, Barnes (1986) showed a high degree oI eIIiciency in Kuala
Lumpur market. Groenewold and Kang (1993) Iound Australian market semi-strong
Iorm eIIicient.

Some oI the recent studies, testing the random walk hypothesis (in eIIect testing Ior
weak Iorm eIIiciency in the markets) are; Korea (Ryoo and Smith, 2002; this study
uses a variance ratio test and Iind the market to Iollow a random walk process iI the
price limits are relaxed during the period March 1988 to Dec 1988), China, (lee et al
2001; Iind that volatility is highly persistent and is predictable, authors use GARCH
and EGARCH models in this study), Hong Kong (Cheung and Coutts 2001; authors
use a variance ratio test in this study and Iind that Hang Seng index on the Hong
Kong stock exchange Iollow a random walk), Slovenia (Dezlan, 2000), Spain
(Regulez and Zarraga, 2002), Czech Republic (Hajek, 2002), Turkey (Buguk and
Brorsen, 2003), AIrica (Smith et al. 2002; Appiah -kusi and Menyah, 2003) and
the Middle East (Abraham et al. 2002; this study uses variance ratio test and the runs
test to test Ior random walk Ior the period 1992 to 1998 and Iind that these markets
are not eIIicient).

METHODOLOGY & DATA

To test historical market eIIiciency one can look at the pattern oI short-term
movements oI the combined market returns and try to identiIy the principal process
generating those returns. II the market is eIIicient, the model would Iail to identiIy
any pattern and it can be inIerred that the returns have no pattern and Iollow a
random walk process. In essence the assumption oI random walk means that either
the returns Iollow a random walk process or that the model used to identiIy the
process is unable to identiIy the true return generating process. II a model is able to
identiIy a pattern, then historical market data can be used to Iorecast Iuture
market prices, and the market is considered not eIIicient.

There are a number oI techniques available to determine patterns in time series data.
Regression, exponential smoothing and decomposition approaches presume that the
values oI the time series being predicted are statistically independent Irom one
period to the next. Some oI these techniques are reviewed in the Iollowing
section and appropriate techniques identiIied Ior use in this study.

Runs test (Bradley 1968) and LOMAC variance ratio test(Lo and MacKinlay 1988)
are used to test the weak Iorm eIIiciency and random walk hypothesis. Runs test
determines iI successive price changes are independent. It is non-parametric and
does not require the returns to be normally distributed. The test observes the
sequence oI successive price changes with the same sign. The null hypothesis oI
randomness is determined by the same sign in price changes. The runs test only
looks at the number oI positive or negative changes and ignores the amount
oI change Irom mean. This is one oI the major weaknesses oI the test. LOMAC
variance ratio test is commonly criticized on many issues and mainly on the selection
oI maximum order oI serial correlation (Faust, 1992). Durbin-Watson test (Durbin
and Watson 1951), the augmented Dickey-Fuller test (Dickey and Fuller 1979) and
diIIerent variants oI these are the most commonly used tests Ior the random
walk hypothesis in recent years (Worthington and Higgs 2003; Kleiman, Payne
and Sahu 2002; Chan, Gup and Pan 1997).

Under the random walk hypothesis, a market is (weak Iorm) eIIicient iI most
recent price has all available inIormation and thus, the best Iorecaster oI Iuture price
is the most recent price. In the most stringent version oI the eIIicient market
hypothesis, ct is random and stationary and also exhibits no autocorrelation,
as disturbance term cannot possess any systematic Iorecast errors. In this study we
have used returns and not prices Ior test oI market eIIiciency as expected returns
are more commonly used in asset pricing literature (Fama (1998).


Returns in a market conIorming to random walk are serially uncorrelated,
corresponding to a random walk hypothesis with dependant but uncorrelated
increments. Parametric serial correlations tests oI independence and non- parametric
v
runs tests can be used to test Ior serial dependence. Serial correlation
coeIIicient test is a widely used procedure that tests the relationship between
returns in the current period with those in the previous period. II no signiIicant
autocorrelation are Iound then the series are expected to Iollow a random walk.

A simple Iormal statistical test was introduced was Durbin and Watson (1951).
Durbin-Watson (DW) is a test Ior Iirst order autocorrelation. It only tests Ior the
relationship between an error and its immediately preceding value. One way to
motivate this test is to regress the error oI time t with its previous value.

ut put-1 vt (1)
where vt ~ N(0,o2 ).

DW test cannot detect some Iorms oI residual autocorrelations, e.g. iI corr(ut, ut-1)
0 but corr(ut, ut-2) = 0, DW as deIined earlier will not Iind any autocorrelation. One
possible way is to do it Ior all possible combinations but this is tedious and
practically impossible to handle. The second-best alternative is to test Ior
autocorrelation that would allow examination oI the relationship between ut and
several oI its lagged values at the same time. The Breusch- GodIrey test is a
more general test Ior autocorrelation Ior the lags oI up to rth order.
Because oI the abovementioned weaknesses oI the DW test we do not use the
DW test in our study. An alternative model which is more commonly used is
Augmented Dickey Fuller test (ADF test). Three regression models (standard
model, with driIt and with driIt and trend) are used in this study to test Ior unit root in
the research, (Chan, Gup and Pan 1997; Brooks 2002). In this study we Iollowed the
test methodologies Irom Brooks (2002) with slight adjustments.

Augmented Dickey-Fuller (ADF) unit root test oI nonstationarity is conducted in the
Iorm oI the Iollowing
regression equation. The objective oI the test is to test the null hypothesis that 0 1
in:
991 :9
against the one-sided alternative 0 1. Thus the hypotheses to be tested are:

H0: Series contains a unit root against
H1: Series is stationary

In this study we calculate daily returns using daily index values Ior the
Mumbai Stock Exchange (BSE) and National Stock Exchange (NSE) oI India. The
data is collected Irom the Datastream data terminal Irom Macquarie University. The
time period Ior BSE is Irom 24th May 1991 to 26th May 2006 and Ior NSE 27th
May to 26th May 2006. Stock exchanges are closed Ior trading on weekends and this
may appear to be in contradiction with the basic time series requirement that
observations be taken at a regularly spaced intervals. The requirement however,
is that the Irequency be spaced in terms oI the processes underlying the series. The
underlying process oI the series in this case is trading oI stocks and generation oI
stock exchange index based on the stock trading, as such Ior this study the index
values at the end oI each business day is appropriate (French 1980).

Table 1 presents the characteristics oI two data sets used in this study. During
the period covered in this study, the mean return oI the NSE index is much lower
than that oI the BSE, similarly the variance oI NSE is lower as compared with BSE
index suggesting a lower risk and a lower average return at NSE as compared with
BSE. It is relevant to note that NSE was established by the government oI India to
improve the market eIIiciency in Indian stock markets and to break the monopolistic
position oI the BSE. NSE index is a more diversiIied one as compare d to the same oI
BSE. This can also be due to the unique nature oI Indias equity markets, the
settlement system on BSE was intermittent (,/, system up until 2nd July 2001)
and on NSE it was always cash.


Table 1:

Data characteristics - BSE and NSE - 1991-2006
Index Mean Variance Minimum Maximum ObservatioSkewness Kurtosis
BSE 0.068138 2.652610 - 18.1055858 3915 0.328482 9.051393
NSE 0.000591 0.000256 - 0.13291131 3915 -0.209051 8.192904



RESULTS

This study conducts a test oI random walk Ior the BSE and NSE markets in India,
using stock market indexes Ior the Indian markets. It employs unit root tests
(augmented Dickey-Fuller (ADF)). We perIorm ADF test with intercept and no
trend and with an intercept and trend. We Iurther test the series using the Phillips-
Perron tests and the KPSS tests Ior a conIirmatory data analysis.

In case oI BSE and NSE markets, the null hypothesis oI unit root is convincingly
rejected, as the test statisticis more negative than the critical value, suggesting that
these markets do not show characteristics oI random walk and
as such are not eIIicient in the weak Iorm. We also test using Phillip-Perron test and
KPSS test Ior conIirmatory data analysis and Iind the series to be stationary. Results
are presented in Table 2. For both BSE and NSE markets, the results are statistically
signiIicant and the results oI all the three tests are consistent suggesting these markets
are not weak Iorm eIIicient.

Table 2:

Results of ADF, PP
Index ADF Test Statistic
(5 lags with
ADF Test Statistic
(5 lags with
PP unit root test,
with intercept and
KPSS (Tau
statistic) for lag
BSE -24.80770 -24.80455 -56.22748 .13264, .13762,
NSE -24.16392 -24.16776 -54.82289 .11710, .12203,
Note: ADF critical values with an intercept and no trend are: -3.435, -2.863 and -
2.567 at 1, 5 and 10 levels; with and intercept and trend are: -3.966, -3.414 and
-3.129 at 1, 5 and 10 levels.
PP critical values are: -3.435, -2.863 and -2.567 at 1, 5 and 10 respectively.
KPSS critical values are: 0.216, 0.176, 0.146 and 0.119 at 1, 2.5, 5 and 10
levels. Null oI stationarity is accepted
iI the tests statistic is less than the critical value.
The null hypothesis in the case oI ADF and PP is that the series is non-stationary,
whereas in the case oI KPSS test it is series is stationary.


Results oI the study suggest that the markets are not weak Iorm eIIicient. DW test,
which is a test Ior serial correlations, has been used in the past but the explanatory
power oI the DW can be questioned on the basis that the DW only looks at the serial
correlations on one lags as such may not be appropriate test Ior the daily data.
Current literature in the area oI market eIIiciency uses unit root and test oI
stationarity. This notion oI market eIIiciency has an important bearing Ior the Iund
managers and investment bankers and more speciIically the investors who are s
eeking to diversiIy their portIolios internationally. One oI the criticisms oI the
supporters oI the international diversiIication into emerging markets is that the
emerging markets are not eIIicient and as such the investor may not be able
to achieve the Iull potential beneIits oI the international diversiIication.

CONCLUSIONS & IMPLICATIONS

This paper examines the weak Iorm eIIiciency in two oI the Indian stock
exchanges which represent the majority oI the equity market in India. We employ
three diIIerent tests ADF, PP and the KPSS tests and Iind similar results. The results
oI these tests Iind that these markets are not weak Iorm eIIicient. These
results support the common notion that the equity markets in the emerging
economies are not eIIicient and to some degree can also explain the less optimal
allocation oI portIolios into these markets. Since the results oI the two tests are
contradictory, it is diIIicult to draw conclusions Ior practical implications or Ior
policy Irom the study.

It is important to note that the BSE moved to a system oI rolling settlement with
eIIect Irom 2nd July 2006 Irom the previously used ,Badla system. The ,Badla
system was a complex system oI Iorward settlement which was not transparent and
was not accessible to many market participants. The results oI the NSE are similar
(NSE had a cash settlement system Irom the beginning) to BSE suggesting that
the changes in settlement system may not signiIicantly impact the results. On the
contrary a conIlicting viewpoint is that the results oI these markets may have been
inIluenced by volatility spillovers, as such the results may be signiIicantly
diIIerent iI the changes in the settlement system are incorporated in the analysis.
The research in the area oI volatility spillover has argued that the volatility is
transIerred across markets (BrailsIord, 1996), as such the results oI these
markets may be interpreted cautiously. For Iuture research, using a
computationally more eIIicient model like generalized autoregressive
conditional heteroskesdasticity (GARCH) could help to clear this.








One-SampIe KoImogorov-Smirnov Test

nifty sensex
N 1490 1490
Normal Parameters
a,b
Mean .07147 .07529
Std. Deviation 1.827344 1.831062
Most Extreme Differences Absolute .082 .084
Positive .071 .072
Negative -.082 -.084
Kolmogorov-Smirnov Z 3.153 3.239
Asymp. Sig. (2-tailed) .000 .000
a. Test distribution is Normal.
b. Calculated from data.






#:ns Test

nifty sensex
Test Value
a
.07147 .07529
Cases < Test Value 701 702
Cases >= Test Value 789 788
Total Cases 1490 1490
Number of Runs 728 712
Z -.801 -1.639
Asymp. Sig. (2-tailed) .423 .101
a. Mean


:tocorreIations
Series:nifty
Lag Autocorrelation Std. Error
a

Box-Ljung Statistic
Value df Sig.
b

1 .056 .026 4.603 1 .032
2 -.022 .026 5.316 2 .070
3 -.010 .026 5.457 3 .141
4 -.018 .026 5.925 4 .205
5 -.026 .026 6.942 5 .225
6 -.058 .026 11.925 6 .064
7 .033 .026 13.592 7 .059
8 .069 .026 20.757 8 .008
9 .017 .026 21.167 9 .012
10 .008 .026 21.272 10 .019
11 -.007 .026 21.336 11 .030
12 .004 .026 21.364 12 .045
13 .034 .026 23.101 13 .040
14 .049 .026 26.719 14 .021
15 .012 .026 26.946 15 .029
16 .019 .026 27.466 16 .037
:tocorreIations
Series:nifty
Lag Autocorrelation Std. Error
a

Box-Ljung Statistic
Value df Sig.
b

1 .056 .026 4.603 1 .032
2 -.022 .026 5.316 2 .070
3 -.010 .026 5.457 3 .141
4 -.018 .026 5.925 4 .205
5 -.026 .026 6.942 5 .225
6 -.058 .026 11.925 6 .064
7 .033 .026 13.592 7 .059
8 .069 .026 20.757 8 .008
9 .017 .026 21.167 9 .012
10 .008 .026 21.272 10 .019
11 -.007 .026 21.336 11 .030
12 .004 .026 21.364 12 .045
13 .034 .026 23.101 13 .040
14 .049 .026 26.719 14 .021
15 .012 .026 26.946 15 .029
16 .019 .026 27.466 16 .037
a. The underlying process assumed is independence (white noise).
b. Based on the asymptotic chi-square approximation.



:tocorreIations
Series:sensex
Lag Autocorrelation Std. Error
a

Box-Ljung Statistic
Value df Sig.
b

1 .072 .026 7.795 1 .005
2 -.037 .026 9.804 2 .007
3 -.011 .026 10.001 3 .019
4 -.028 .026 11.158 4 .025
5 -.027 .026 12.224 5 .032
6 -.051 .026 16.163 6 .013
7 .022 .026 16.902 7 .018
8 .073 .026 24.820 8 .002
9 .021 .026 25.453 9 .003
10 .016 .026 25.855 10 .004
11 .003 .026 25.868 11 .007
12 .004 .026 25.887 12 .011
13 .032 .026 27.415 13 .011
14 .044 .026 30.308 14 .007
15 .004 .026 30.337 15 .011

You might also like