Download as doc, pdf, or txt
Download as doc, pdf, or txt
You are on page 1of 26

STOCK RETURNS AND VOLATILITY: EVIDENCE FROM SELECT EMERGING MARKETS Abstract This paper investigates the behavior

of stock returns and volatility in ten emerging markets and compares them with those of developed markets under different measures of frequency (daily, weekly, monthly and annual) over the period January 1, 2002 to December 31, 2006. The ratios of mean return to volatility for emerging markets are found to be higher than those of developed markets. Sample statistics for stock returns of all emerging and developed markets indicate that return distributions are not normal and return volatility shows clustering . In most cases, GARCH (1,1) specification is adequate to describe the stock return volatility. The significant lag terms in the mean equation of GARCH specification depend on the frequency of the return data. The presence of leverage effect in volatility behavior is examined using the TAR-GARCH model and the evidence indicates that is not present across all markets under all measures of frequency. Its presence in different markets depends on the measure of frequency of stock return data. 1. Introduction Emerging capital markets have come to play an increasingly important role in international investment in the wake of globalization of financial markets in recent years (Park and Agtmael, 1993). The gross domestic product (GDP) of emerging market countries as a group has been growing at roughly double the rate of advanced economies in recent years. Aggregate GDP of the developing countries grew 6.6 per cent in 2004, while the GDP of high-income nations grew at 3.1 per cent (IFC, 2006). For the five years ended December 2005, the emerging market stocks (represented by Morgan Stanley Capital International (MSCI) emerging markets index) provided returns of 140 per cent while developed international stocks (represented by MSCI Europe, Australasia and Far East (EAFE) index provided returns of 25 per cent and U.S. stocks (represented by S&P 500 index) only 3 per cent. (Source: MSCI and Standard & Poors). The market capitalization of emerging market countries has more than doubled over the past decade, growing from less than $ 2 trillion in 1995; it was set to exceed $5 trillion in 2006. The share of emerging markets is now more than 12 per cent of the world market capitalization. In fact, this share is steadily growing (IFC, 2006). Emerging Markets equity funds had net inflows of

$22.4bn in 2006, according to data from Emerging Portfolio Fund Research (EPFR), which compares with the previous record set in 2005 of $20.3bn. The interest of international investors in emerging markets is surging due to spectacular rates of stock returns in these markets vis--vis returns provided by developed markets and low correlation in stock prices and stock returns between emerging markets and developed markets. Divecha, Drach and Stefak (1992), Price (1994), Stanley (1995), Barry and Lockwood (1995) and Barry, Peavy and Rodriquez (1998) find that the emerging markets have historically experienced a high level of mean returns and volatility but low correlation with developed markets. The low degree of correlation implies the possible diversification benefits for investors of developed markets to invest in emerging markets. Harvey (1995) reports that emerging markets have high average returns, low overall volatility, low exposure to world risk factors and little integration. Professional investors seriously consider the appropriate proportion of equities from these markets to be included in an optimal internationally diversified portfolio (Fraser and Power, 1997). Emerging market stock returns and volatility are important to understand since investors seek increased diversification and opportunities for high returns (Michelfelder, 2005). Country and stock selection strategies are imperative if a global portfolio manager wants to achieve enhanced performance (Fernandez, 2005) This paper examines the behaviour of stock returns and volatility for ten select emerging markets over a five year period January 1, 2002 to December 31, 2006, and compares them to those of developed markets. Many cross-country empirical studies have been carried out in the past on stock market returns and volatility. These include Sharma and Kennedy(1977) for Bombay, London and New York; Huang (1995) for nine Asian countries; Huber (1997) for Vienna; Grieb and Reyes (1999) for Brazil and Mexico; Chang and Ting(2000) for Kuwait; Lee, Cheng and Rui (2001) for China; Apergis and Eleptheriou (2001) for Athens; Cheung and Coutts (2001) for Hongkong; Chiang and Doong (2001) for seven Asian stock markets; Abraham, Seyyed and Alsakran (2002) for Saudi Arabia, Kuwait and Bahrain; Karmakar (2003) for India; Lai, Balachandher and Nor (2003) for Malaysia; Abraham,;, Lima and Tabak (2004) for Hong Kong and Singapore.

This paper adds to the existing research by using both low frequency and high frequency data to assess the stock return behaviour. This enables discrimination of performance with respect to different data frequencies (Chiang and Doong, 2001). Unlike other studies, low frequency stock returns represent holding period returns and not returns between successive points representing end of the frequency period under consideration. The sample size for low frequency data is, therefore, quite large. Most of the existing research tests the applicability of first-lag Autoregressive [AR(1)] and Generalized Auto-regressive conditional heteroscadasticity [GARCH(1,1)] parameterization to the return data. This paper attempts to find the estimated ARMA (Autoregressive Moving Average) and GARCH structure that adequately fits the return data. The paper also uses more recent data starting from January 1, 2002 and focuses on highly capitalized emerging markets including the so-called BRIC (Brazil, Russia, India and China) countries. The results of the study can be used by global portfolio managers, particularly emerging market funds in achieving portfolio diversification and adopt specific trading strategies in emerging markets. Predictability of volatility is important in designing optimal asset allocation decisions as well as dynamic hedging strategies for options and futures (Baillie and Myers, 1991). The analysis can also be useful in making financial decisions concerning risk analysis, portfolio selection and derivative pricing. The paper is organized into following sections. Section 2 describes the data used in this study and presents some statistical properties of stock returns with different frequencies of the data. Section 3 provides the measures of observed volatility in the ten markets. Section 4 examines the presence of asymmetry in observed volatility in the ten markets. Section 5 contains concluding remarks. 2. The Data and basic statistics 2.1 Data sample: The universe of emerging markets considered is based on the Morgan Stanley Capital International (MSCI) Emerging Markets Index, which consists of indices of 25 emerging markets. The market capitalization of the premier stock exchanges of these 25 emerging markets as on December 31, 2005 is given in Table 2, in descending order of market capitalization.

Table 1: Market Capitalization of Emerging Stock Markets as on 3

December 31, 2005 ( billion US $) 1 S. Korea 718.01 . 2 India (Mumbai) 553.07 . 3 South Africa 549.31 . 4 Taiwan 476.02 . 5 Brazil (Sao Paulo) 474.65 . 6 Russia (MICEX) 399.83 . 7 China (Shanghai) 285.19 . 8 Mexico 239.13 . 9 Malaysia 180.15 . 10 Turkey 161.54 . 11 Chile 136.49 . 12 Thailand 123.89 . 13 Israel 122.58 .
(Pakistan), www.pse.cz (Czech www.Oanda.com.

14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25.

Poland Indonesia Egypt Colombia Argentina Pakistan (Karachi) Philippines Jordan Czech Republic Peru Hungary Morocco

93.61 81.43 79.51 50.50 47.59 45.93 39.82 37.91 36.82 34.35 32.58 27.08

Source: World Federation of Exchanges, www.fin.rus.com (Russia), Federation of Euro-Asian Stock Exchanges Republic), Casablanca-bourse.com (Morocco), www.bvl.com.pe (Peru), www.ammanstockex.com (Jordan).Local currency figures have been converted using exchange rates provided by

The sample for the study consists of ten top emerging markets based on market capitalization as on December 31,2005. In the case of a country having more than one stock exchange, the exchange with the highest market capitalization is chosen. The choice of a particular index from a stock exchange is based on its prominence in the financial markets. The stock exchanges and the indices for the 10 economies in our sample are given in Table 2.The developed markets are represented by the MSCI developed markets index, which is based on the stock indices of 23 countries.

Table 2: Stock Exchanges and Indices of Select Emerging Markets covered in the study S. No. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Name of Country South Korea India South Africa Taiwan Brazil Russia China Mexico Malaysia Turkey Name of Stock Index Seoul Composite Bse Sensex Jse Index Taiwan Weighted Bovespa Micex 10 Shanghai Composite IPC KLSE ISE National-100

The data for each of these 10 markets corresponds to the daily closing stock market index values of these marketss for the period from January 1, 2002 to December 31, 2006 The closing values of the MSCI developed market index will also be taken for the same period. Daily stock returns are calculated as
P Rt = ln t *100 t P 1

:
(1)

where Rt is the daily stock return on a given day, Pt is the closing index value on a given day, Pt-1 is the closing index value on the previous day and ln is the natural logarithm

.
5

The weekly returns are logarithmic returns for a holding period of one week (5 trading days). The monthly and annual returns are calculated as simple returns over a holding period of one month (21 trading days) and one year (252 trading days) respectively using the following formula:
Pt Pt 1 Pt 1

Rt =

(2)

where Rt is the monthly/annual stock return, Pt is the closing index value on a given day, Pt-1 is the closing index value on a day that precedes the given day by a month/year.

2.2 Basic statistics The basic statistics for daily, weekly, monthly and annual returns are reported in Table 3A, 3B, 3C and 3D respectively. These include the mean return, standard deviation (S.D.) of return, ratio of mean return to standard deviation, skewness (SK), kurtosis (KUR), Ljung-Box Q(24) value for the returns, Kolmogrov-Smirnov D-Statistic (D-Stat) for the returns and Augmented DickeyFuller (ADF) unit root statistic for the returns.
Table 3A: Sample statistics of daily stock returns: 1/1/2002-31/12/2006

Country

Mean return

S.D. Of return 1.51 1.35 1.10 1.33 1.72 2.09 1.50 1.17 0.67 2.46 0.83

Mean/ S.D. 0.036 0.086 0.062 0.020 0.055 0.077 0.029 0.097 0.057 0.033 0.024

SK.

KUR.

Q(24)

D-Stat.

ADF Unit Root -33.60 -34.18 -26.58 -33.86 -33.92 -35.52 -37.35 -32.67 -30.66 -38.93 -32.49

S.Korea India S.Africa Taiwan Brazil Russia China Mexico Malaysia Turkey Developed

0.055 0.116 0.068 0.027 0.094 0.160 0.044 0.113 0.038 0.082 0.020

-0.32 -0.74 -0.18 -0.17 -0.23 -0.48 0.67 -0.04 0.11 -0.34 0.11

4.69 10.41 5.35 5.51 3.62 5.82 11.66 5.27 4.76 45.11 6.90

24.16 64.67 24.82 29.49 21.89 15.73 37.39 30.77 44.86 27.53 46.35

0.070 0.050 0.030 0.070 0.006 0.061 0.088 0.040 0.050 0.070 0.060

World
Notes: Q(24) is the Ljung-Box Q-statistic that tests the joint significance of the autocorrelations of the daily return series up to the 24th order. The critical value are 36.42 and 42.98 for the significance levels of 0.05 and 0.01 respectively. D-Stat. is the Kolmogorov-Smirnov D-statistic for the null hypothesis of normality. The critical values are 0.023 and 0.026 for significance levels of 0.05 and 0.01 respectively. ADF unit root is the Augmented DickeyFuller Test statistic for the null hypothesis of unit-root non-stationarity. The critical values are 1.95 and 2.58 for significnce levels of 0.05 and .01 respectively. Table 3B: Sample statistics of weekly stock returns: 1/1/2002-31/12/2006

Country

Mean return

S.D. of return 2.95 2.67 2.22 2.72 3.36 4.11 2.79 2.37 1.49 4.21 1.79

Mean/ S.D 0.077 0.157 0.117 0.038 0.095 0.156 0.056 0.192 0.101 0.064 0.044

SK.

KUR.

Q(24)

D-Stat

ADF Unit Root - 9.63 - 8.54 - 7.97 - 8.76 - 8.90 -10.34 - 7.64 - 7.37 -11.37 - 7.13 - 6.35

S.Korea India S.Africa Taiwan Brazil Russia China Mexico Malaysia Turkey Developed

0.227 0.419 0.260 0.102 0.320 0.642 0.155 0.456 0.150 0.270 0.078

-0.36 -0.79 -0.51 -0.31 -0.45 -0.50 0.54 -0.57 0.18 0.17 -0.07

3.72 5.71 4.55 4.45 3.49 5.64 4.62 5.07 4.52 5.72 7.08

976.1 1175.7 1062.2 1077.7 1007.1 1104.1 1127.0 1004.0 1232.1 1094.5 1241.7

0.055 0.052 0.053 0.051 0.045 0.081 0.038 0.059 0.042 0.041 0.071

World Notes: (Refer Table 3A)

Table 3C: Sample statistics of monthly stock returns: 1/1/2002-31/12/2006

Country

Mean return

S.D. of return 6.05 6.34 4.69 5.85 7.38 8.99 6.36 5.21 3.56 9.63 3.79

Mean/ S.D. 0.212 0.381 0.297 0.110 0.289 0.383 0.135 0.450 0.215 0.216 0.127

SK.

KUR.

Q(24)

KS DStat

ADF Unit Root -6.17 -4.06 -4.57 -4.38 -6.09 -4.89 -5.26 -4.69 -5.61 -5.25 -4.36

S.Korea India S.Africa Taiwan Brazil Russia China Mexico Malaysia Turkey Developed

1.282 2.413 1.391 0.643 2.135 3.441 0.861 2.345 0.764 2.082 0.480

-0.52 -0.64 -0.29 -0.08 -0.43 -0.34 0.63 -0.88 -0.22 0.28 -0.80

3.16 4.25 3.41 3.42 2.79 3.05 3.08 3.99 3.26 4.24 5.26

6901.8 7830.4 6791.9 6828.3 7137.6 6934.8 5701.6 7458.7 7515.4 7187.9 7137.7

0.054 0.061 0.033 0.027 0.070 0.056 0.080 0.088 0.038 0.033 0.090

World Notes: (Refer Table 3A)

Table 3D: Sample statistics of annual stock returns: 1/1/2002-31/12/2006

Country

Mean return

S.D. of return 23.51 26.90 23.16 18.48 30.48 38.70 26.99 20.04 12.80 27.82 13.90

Mean/ S.D. 0.750 1.402 0.921 0.450 1.212 1.236 0.094 1.681 0.627 1.402 0.655

SK.

KUR.

Q(24)

KS DStat

ADF Unit Root -1.68 -2.17 -0.93 -1.50 -2.73 -1.78 -2.07 -1.15 -1.44 -2.20 -0.53

S.Korea India S.Africa Taiwan Brazil Russia China Mexico Malaysia Turkey Developed

17.63 37.72 21.32 8.31 36.94 47.82 2.53 33.69 8.02 39.01 9.11

-0.28 0.08 -0.71 0.39 0.60 0.14 1.59 -0.98 0.30 -0.21 -1.27

2.77 2.47 2.70 3.38 3.61 1.72 5.29 4.04 3.55 2.50 4.74

20443 20313 20977 18561 20138 20132 13055 18201 19250 15226 18620

0.079 0.062 0.102 0.070 0.119 0.128 0.191 0.184 0.065 0.069 0.218

World Notes: (Refer Table 3A) Mexico has the highest ratio of mean returns to standard deviation under different measures of frequency. Except for annual returns, Taiwan has the lowest ratio of mean returns to standard deviation. For annual returns, China has the lowest ratio. Barring a few exceptions, the ratio of mean return to standard deviation for emerging markets is higher compared to the developed markets under all measures of frequency of return. Values of skewness of return distributions indicate heavy tails in these distributions under all measures of frequency. Values of kurtosis are high especially for high frequency return data. This suggests that, big shocks of either sign are more likely to be present and the stock-return series may not be normally distributed (Chiang and Doong, 2001) Values of Q(24) statistic show the presence of first-order autocorrelation for all markets in case of weekly, monthly and annual returns. This indicates the predictability of returns in these markets. However, in case of daily returns, first-order autocorrelation is present only in case of India, China , Malaysia and the developed world. Whether the daily return data for other markets represents a random walk remains to be tested. Return distributions are not normal as indicated by the D-statistic for all markets under all measures of frequency except in case of daily stock return distribution of Brazil. The ADF unit root test statistic values signify rejection of null hypothesis of non-stationarity in daily, weekly 9

and monthly return distributions. The annual return distributions are, however, found to be nonstationary requiring differencing of the return distributions to make them stationary. 3.Stock return volatility Traditional estimates of volatility are based on the standard deviation of the daily stock price changes and assume that variance is constant (homoscedastic) over time. It has been observed that stock volatility exhibits clustering phenomenon i.e. large changes tend to be followed by large changes and small changes tend to be followed by small changes. This is commonly referred to as the ARCH (Auto Regressive Conditional Heteroscedasticity) effect. The ARCH effect or volatility clustering in the time series of stock prices is present when the variance of the error term at time t is dependant on the lagged squared disturbance. The presence of the ARCH effect in the return series is ascertained by running the following regression ut2 = 0 + 1ut21 + where ut = rt r (3)

rt = daily / weekly / monthly / annual return


The coefficient is found to be statistically significant in all cases except daily return data of

Korea and Taiwan, indicating the presence of the ARCH effect. Performance of Engles ARCHTEST on the return series also confirms the presence of heteroscadasticity in the return series. Table 4 shows the value of Engles ARCHTEST statistic at a lag of 24.

Table 4: Values of Engles ARCHTEST on stock returns

Country Korea India S.Africa `Taiwan Brazil

Daily Returns 158.08 321.57 196.05 121.52 113.69

Weekly Returns 339.23 408.52 376.00 468.17 398.98

Monthly Returns 836.70 931.05 876.75 953.24 851.91

Annual Returns 945.40 968.70 966.72 948.74 957.56

10

Russia China Mexico Malaysia Turkey Developed World

132.36 195.35 83.05 83.05 436.68 285.27

438.18 604.36 464.67 603.77 393.82 558.81

883.48 936.64 924.27 1001.20 971.39 1005.6

951.28 908.05 970.77 961.10 928.46 1005.2

Engles ARCHTEST statistic tests the null hypothesis of no heteroscadasticity and is distributed as 2 with degrees of freedom equal to the lag. The critical value at 24 degrees of freedom at 5% significance level is 36.42. The statistical evidence shows that the null hypothesis cannot be accepted at 5% level of significance. Volatility under heteroscadasticity can be modeled using the ARCH (Auto Regressive Conditional Heteroscedasticity) model and different types of GARCH (Generalized Auto Regressive Conditional Heteroscedasticity) models. The ARCH model was first introduced by Engle (1982). A generalization of the above model (GARCH) was introduced by Bollerslev(1986). ARCH model is more suited for economic data. In financial markets, GARCH models are preferred over ARCH models. The GARCH process can model an infinite ARCH process with very few parameters by putting some constraints on the coefficients (Karmakar, 2006) The GARCH model requires specification of a conditional mean equation and a conditional variance equation. The mean equation can be modeled as a simple conditional mean equation of the form: Rt = c + t (4) (5)

t = Rt t

2 where Rt is the market return in period t and t It-1 GED (0, t , ). The conditional density

function of error term is modeled as a Generalized Error Distribution (GED). The choice of a GED density is dictated by the inability of gaussian GARCH processes to account for the leptokurtosis of most return series, an issue that is likely to be even more relevant when using
2 emerging market data (De Santis, 1997). t and t are the conditional mean and conditional

variance of Rt based on the information set It-1 . It-1 includes past available information relevant to

11

2 the formation of both t and t . is a measure of tail thickness which is equal to two for the

normal density and smaller than two for leptokurtic distributions and t is the error term . The conditional mean equation can also be modeled as pth order autoregressive (AR) process or qth order moving average (MA) process or (p,d,q) order autoregressive integrated moving average (ARIMA) process where d is the order of differencing required to achieve stationarity of the process. The order of the auto-regressive (AR) and the moving average (MA) terms in the mean equation are determined by the autocorrelation function (ACF) and the partial autocorrelation function (PACF) of the return series. The conditional variance equation in the GARCH (p,q) model takes the form:

t2 = 0 + 1 t21 + +q t2q + 1t21 + + p t2 p


0 >0, 1 , 2, .. q 0, 1, 2 , . p 0

(6)

All parameters are restricted to non-negative values to ensure positive conditional variances i.e.

The conditional variance of returns is expressed as a linear function of past volatility shocks (errors, t i for i=1,q) and past conditional variances ( t i for i=1,p). This implies that past
2 2

volatility shocks have positive but decreasing effects on the future market volatility. This is consistent with volatility clustering phenomenon. The GARCH (p, q) process is stationary when: (1+ 2+..+ q ) + (1+ 2 +..+ p) < 1 (7) The sum of the coefficients of equation (7) measures the speed of decay of shocks to volatility. If the sum is close to one, shocks to the current volatility remain important for long periods into the future. If the sum is equal to one, then shocks to volatility persist forever and the unconditional variance is not determined by the model. Engle and Bollerslev (1986) call this type of process Integrated GARCH (Chou, 1988). If the sum is greater than one it results into negative variance. The final GARCH (p ,q) specification is decided by looking at the properties of the standardized residuals (residuals divided by their conditional standard deviation). If the model is correctly specified, the standardized residuals should be independent and identically distributed with mean

12

zero and variance one. If the mean equation is correctly specified, all Q-statistics should not be significant. If the variance equation is correctly specified, all Q2- statistics should not be significant. We have calculated Q and Q2 statistics at a lag of 24. Maximum of nine lags has been used for the GARCH terms to test whether the GARCH process can be fitted to the data. The results of fitting the GARCH (p, q) specification to the daily, weekly, monthly and annual market return data are reported in Table 5A, 5B, 5C and 5D respectively.

Table 5A: GARCH (p,q) Model Estimates for daily stock returns : 1/1/2002-31/12/2006 Country Structure of mean equation Coefficient s of mean equation Significant lags SKOR ARMA (0, 0) IND ARMA (1, 0) SAFR ARMA (0, 0) TW ARMA (0, 0) BZL ARMA (1, 0) RUS ARMA (0, 0) CH ARMA (0, 0) MEX ARMA (1, 0) MAL ARMA (1, 0) TUR ARMA (1, 0) D.W. ARMA (1, 0)

AR(1)

AR(1)

AR(1)

AR(1)

13

a1 Order of variance equation Coefficient s of variance equation 0 1 2 1 2 Q(24) Q2(24) Log likelihood

GARCH (1,1)

0.111 GARCH (1,1)

GARCH (1,1)

GARCH (1,1)

GARCH (2,1)

GARCH (1,1)

GARCH (1,1)

0.085 GARCH (1,1)

0.114 GARCH (2, 2)

GARCH (1,1)

0.116 GARCH (1,1)

0.222* 0.064* 0.926 19.83 24.33 -2155.2

0.120 0.145 0.781 26.27 31.24 -1982.9

0.026 0.077 0.902 24.46 16.73 -1806.3

0.011* 0.044 0.952 16.80 29.18 -1978.6

0.070 -0.019* 0.076 0.917 17.43 24.33 -2384.7

0.300 0.164 0.772 13.64 21.34 -2579.5

0.191 0.119 0.793 25.58 8.40 -1968.9

0.065 0.092 0.859 24.95 29.48 -1896.3

0.014 0.057 0.094 -0.043* 0.869 20.99 33.85 -1181.9

0.011 0.104 0.886 15.99 3.35 -2632.4

0.004* 0.060 0.934 10.03 21.07 -1339.5

Notes : SKOR =S.Korea , IND= India, SAFR= S.Africa, TW=Taiwan, BZL= Brazil, RUS = Russia, CH= China,
MEX= Mexico, MAL = Malaysia, TUR = Turkey, D.W. = Developed World Q(24) is the Ljung-Box Q-statistic that tests the joint significance of the autocorrelations of the daily return series up to the 24th order. The critical value are 36.42 and 42.98 for the significance levels of 0.05 and 0.01 respectively. * not significant at 5% level of significance

Table 5B: GARCH (p,q) Model Estimates for weekly stock returns : 1/1/2002-31/12/2006 Country Structure of mean equation Coefficient s of mean equation Significant lags SKOR ARMA (0, 3) IND ARMA (0, 3) SAFR ARMA (4, 4) TW ARMA (4, 4) BZL ARMA (0, 3) RUS ARMA (0, 3) CH ARIMA (3, 3) MEX ARMA (3, 3) MAL ARMA (3, 3) TUR ARMA (3, 3) D.W. ARMA (3, 3)

MA (1, 2,3)

MA (1,2,3)

AR(1) MA

AR(1) MA(4)

MA (1,2,3)

MA (1,2,3)

AR(2) MA (1,2,3)

AR(1) MA (1,2,3)

MA (1,2,3)

MA (1,2,3)

AR(1) MA (1,2,3)

14

a1 a2 b1 b2 b3 b4 Order of variance equation Coefficient s of variance equation 0 1 2 1 Q(24) Q2(24) Log likelihood

(1,3,4) 0.999 0.907 0.819 0.645 GARCH (1,1) 0.911 0.824 0.652 GARCH (1,1) 0.088 -0.147 -0.754 GARCH (1,1)

0.913 0.948 0.828 0.682 -0.698 GARCH (1,1) GARCH (1,1) 0.894 0.807 0.690 GARCH (1,1) 0.113 0.886 0.727 0.634 GARCH (1,1)

0.105 0.874 0.832 0.735 GARCH (2,1) 0.931 0.766 0.626 GARCH (1, 1) 0.905 0.872 0.748 GARCH (1,1)

0.115 1.005 0.969 0.959 GARCH (1,1)

0.055 0.084 0.900 19.16 21.09 -2355.4

0.119 0.108 0.838 29.60 33.87 -2170.5

0.043 0.100 0.873 20.11 26.01 -1939.4

0.015* 0.047 0.948 17.81 25.17 -2126.5

0.075* 0.043 0.937 9.74 23.74 -2548.7

0.309 0.143 0.809 18.14 23.45 -2744.1

0.120 0.091 0.865 34.43 25.71 -2124.4

0.092 0.057* 0.060 0.829 18.50 33.23

0.044 0.136 0.810 19.49 11.32 -1434.2

0.181 0.101 0.870 17.47 21.03 -2737.3

0.007 0.069 0.920 13.40 15.80 -1349.2

Notes: Refer Table 5A

Table 5C: GARCH (p,q) Model Estimates for monthly stock returns : 1/1/2002-31/12/2006 Country Structure of mean equation Coefficients of mean equation SKOR ARMA (21, 0) IND ARMA (20,20) SAFR ARMA (20, 20) TW ARMA (20, 20) BZL ARMA (21, 21) RUS ARMA (21, 0) CH ARMA (21, 21) MEX ARMA (21, 21) MAL ARMA (21, 0) TUR ARMA (21, 0) D.W. ARMA (1,20)

15

Significant lags c a1 a2 a3 a10 a11 a18 a19 a20 a21 b19 b20 b21 Order of variance equation Coefficients of variance equation 0 1 1 Q(24) Q2(24) Log likelihood

AR (1,19, 21) 2.072 0.950

AR(1,2, 3,18,19, 20) MA(20) 3.017 1.001 -0.100 0.085

AR(1,19 ,20) MA(20) 2.002 0.969

AR(1) MA(20) 0.499 0.983

AR(1,18, 21) MA(19, 20) 2.776 0.961

AR(1,19,2 0,21) 4.070 0.963

AR(19, 20,21) MA(21) -0.784* 0.982

AR(1) MA(20, 21) 2.633 0.994

AR(1, 18,20, 21) 0.492* 0.977

AR(1,10 ,11,18, 20,21) 2.526 0.953

AR(1) MA (20) 0.073 0.997

-0.165 0.147

-0.081 -0.181

-0.084 -0.189 0.188 0.070 -0.119 -0.128 GARCH (1,1) -0.120 -0.219 0.318 -0.068 -0.174 0.234 -0.859 -0.036 GARCH (1,1)

-0.082 -0.116 0.172

0.101 -0.073 -0.107 -0.224 0.291 -0.951

-0.218 GARCH (1,1) GARCH (1,1)

-0.350 GARCH (1,1)

-0.944 GARCH (1,1)

GARCH (1,1)

-0.094 GARCH (1,1)

GARCH (1, 1)

GARCH (1,1)

GARCH (1,1)

0.020* 0.045 0.949 19.83 24.33 -2155.2

0.109 0.100 0.862 25.80 46.25 -2281.6

0.030 0.068 0.918 26.88 18.90 -2090.3

0.014 0.046 0.948 16.06 29.40 -1998.2

0.088* 0.036 0.947 30.60 31.95 -2699.9

1.181* 0.145 0.709 19.19 48.58 -2913.2

0.177 0.096 0.860 18.20 32.97 -2223.8

0.122 0.114 0.807 28.04 21.22 -1966.3

0.224 0.207 0.569 28.66 7.95 -1587.9

0.145 0.073 0.910 21.75 48.38 -2860.7

0.004* 0.056 0.939 23.64 23.73 -1382.8

Notes: Refer Table 5A

Table 5D: GARCH (p, q) Model Estimates for annual stock returns Country Structure of mean equation Coefficients of mean equation SKOR ARIMA (0, 1, 0) IND ARIMA (2, 1, 2) SAFR ARIMA (0, 1, 0) TW ARIMA (0 ,1, 0) BZL ARIMA (0, 1, 0) RUS ARIMA (0, 1, 0) CH ARIMA (0,1, 0) MEX ARIMA (0, 1, 0) MAL ARIMA (1, 1, 0) TUR ARIMA (0, 1, 0) D.W. ARIMA (0,1,0)

16

Significant lags a1 a2 b1 b2 Order of variance equation Coefficients of variance equation 0 1 1 Q(24) Q2(24) Log likelihood

GARCH (1,1)

AR(1,2) MA(1,2) -0.932 -0.694 1.007 0.724 GARCH (1,1)

AR(1) 0.164

GARCH (1,1)

GARCH (1,1)

GARCH (1,1)

GARCH (1,1)

GARCH (1,1)

GARCH (1,1)

GARCH (1, 1)

GARCH (1, 1)

GARCH (1,1)

0.068* 0.059 0.930 23.94 30.48 -2240.2

0.264 0.123 0.842 30.45 12.77 -2301.8

0.089 0.062 0.911 27.79 14.10 -2000.1

0.047 0.048 0.941 32.43 15.32 -2048.2

0.184 0.064 0.921 26.92 26.90 2567.1

0.542 0.133 0.844 22.41 36.31 -2773.4

0.196 0.129 0.841 35.22 9.09 -1955.3

0.128 0.083 0.889 15.33 8.26 -2149.9

0.013 0.058 0.930 21.76 14.06 -1371.5

0.258 0.073 0.914 20.73 17.31 -2734.4

0.014* 0.069 0.921 27.00 13.06 -1542.5

Notes: Refer Table 5A

Structure of the mean equation differs among return data for different frequencies. For daily return data, only AR(1) term is significant in three markets and no ARMA structure is evident in other markets. In case of weekly return data, both AR and MA terms are significant up to a maximum lag of four. No constant term is used in the mean equation of daily and weekly stock return data as the values appear to fluctuate around zero. The lag of significant terms increases to a maximum of twenty one for monthly return data The mean equation for monthly stock returns is modeled using a constant term. In case of annual return data , except in two markets, the mean equation does not show any ARMA structure but the return series is differenced once to make it stationary in all cases. Structure of the variance equation, however, does not show much variation across return data for varying frequency. GARCH (1,1) process is adequate to describe the return data in most of the emerging markets and also the developed world. In some cases, the GARCH process is either not stable (daily return data of Malaysia) or does not adequately fit the return data (weekly return data of Mexico and monthly return data of India ,Russia and Turkey). In these cases volatility can be modeled using the integrated GARCH model or an Exponentially Weighted Moving Average (EWMA) model using the smoothing constant value equal to the GARCH beta.

17

The log likelihood values are large which implies that the GARCH specification represents the return behaviour by capturing the temporal dependence of volatility. The small values of estimated coefficients imply that large market surprises induce relatively small revisions in the future volatility. The estimated values of coefficient in the conditional variance equation are considerably larger than that of implying that the prediction of volatility is dominated by the autoregressive component. The persistence of the conditional variance process measured by

i + i is high and close to 1. This implies that shocks to conditional variance take a long
i =1 i =1

time to die out and that current information is relevant for predicting future volatility over a long horizon. The Ljung-Box Q statistics for 24th order serial correlation in the level and squared standardized residuals are also reported, which indicate that the estimated models fit the data well. 4. Leverage Effect In the GARCH model, the variance dynamics are treated as symmetric, wherein a positive shock and a negative shock of the same magnitude will have the same effect on the present volatility. Christie (1982) and Schewart (1989) however pointed out that downward movements in the market are often followed by higher volatility than upward movements of the same magnitude.This is known as the leverage effect. Glosten, Jagannathan and Runkle (1993) and Nelson (1991) explicitly treat this asymmetry in variance in their extension of the GARCH model using the Threshold Autoregressive GARCH model (TAR-GARCH model) and Exponential GARCH (EGARCH model) respectively . The TAR-GARCH specification is attractive since fewer parameters need to be estimated. Moreover, in their study of daily stock returns in the Japanese market, Engle and Ng (1993) find that the parameterization of TARGARCH is the most promising one (Chiang and Doong, 2001) Glosten, Jagannathan and Runkle (1993) and Zakoian (1994) proposed the Threshold Autoregressive GARCH model (TAR-GARCH model) to specify the asymmetry of volatility. The TAR-GARCH (p,q) model is specified as follows:
p q

t2 = w + ( i + i Dt i ) t21 + i t2i
i =1 i =1

(8)

18

w 0, i 0 i 0 , i 0

(9)

where the dummy variable Dt 1 is equal to zero for a positive shock or movement ( t 1 > 0 ) and 1 for a negative shock ( t 1 < 0 )
2 Provided that > 0, the TGARCH model generates higher values for t given t 1 < 0 than for

a positive shock of equal magnitude. As with the ARCH and GARCH models, the parameters of the conditional variance are subject to non-negativity constraints. Leverage effect presence can be tested by the hypothesis =0 where the impact is asymmetric when 0. The results of fitting TAR-GARCH model to daily, weekly, monthly and annual market return series are reported in table 6A, 6B, 6C and 6D respectively.

Table 6A: TAR-GARCH Model Estimates for daily stock returns: 1/1/2002-31/12/2006 Country Structure of mean equation Coefficients of mean equation SKOR ARMA (0, 0) IND ARMA (1, 0) SAFR ARMA (0, 0) TW ARMA (0, 0) BZL ARMA (0, 0) RUS ARMA (0, 0) CH ARMA (0, 0) MEX ARMA (1, 0) MAL ARMA (1, 0) TUR ARMA (0, 0) D.W. ARMA ( 1,0)

19

Significant lags a1 Order of variance equation Order of asymmetry for TARGARCH Coefficients of variance equation 0 1 2 1 2 1 Q(24) Q2(24) Log likelihood

AR(1) GARCH (1,1) 1 0.135 GARCH (1,1) 1 GARCH (1,1) 1 GARCH (1,1) 1 GARCH (2,1) 1 GARCH (1,1) 1 GARCH (1,1) 1

AR(1) 0.087 GARCH (1,1) 1

AR(1) 0.110 GARCH (2, 2) 1 GARCH (1,1) 1

AR(1) 0.113 GARCH (1,1) 1

0.061 -0.006* 0.902 0.152 25.30 33.07 -2140.9

0.212 0.005* 0.715 0.322 22.81 38.54 -1963.0

0.010* 0.086 0.939 -0.072 25.88 19.72 -1801.2

0.003* 0.068 0.966 -0.070 20.10 35.03 -1971.6

0.149 -0.076 0.072 0.900 0.108 19.72 37.04 2368.0

0.290 0.178 0.775 -0.031* 13.40 20.54 -2579.3

0.190 0.096 0.784 0.075* 26.25 9.33 -1967.6

0.033 0.109 0.908 -0.096 24.27 27.88 -1886.5

0.000 0.120 -0.115 1.647 -0.653 0.000 19.99 35.32 -1156.2

0.122 0.088 0.881 0.040* 16.14 5.10 -2632.8

0.000 0.090 0.953 -0.090 10.00 19.14 -1329.6

Notes: Refer Table 5A

Table 6B: TAR-GARCH Model Estimates for weekly stock returns: 1/1/2002-31/12/2006 Country Structure of mean equation Coefficients of mean SKOR ARMA (0, 3) IND ARMA (0, 3) SAFR ARMA (4, 4) TW ARMA (4, 4) BZL ARMA (0, 3) RUS ARMA (0, 3) CH ARMA (3, 3) MEX ARMA (3, 3) MAL ARMA (3, 3) TUR ARMA (3, 3) D.W. ARMA (3,3)

20

equation Significant lags a1 a2 b1 b2 b3 b4 Order of variance equation Order of asymmetry for TARGARCH Coefficients of variance equation 0 1 2 1 1 Q(24) Q2(24) Log likelihood

MA (1,2,3)

MA (1,2,3)

AR(1) MA(1,3, 4) 0.998 -0.089 -0.150 -0.752 GARCH (1,1) 1

AR(1) MA(4) 0.912

MA (1,2,3)

MA (1,2,3)

AR(2) MA(1,2,3) 0.113 0.886 0.727 0.634 GARCH (1,1) 1

AR(1) MA(1,2,3) 0.105

MA (1,2,3)

MA (1,2,3)

AR(1) MA (1,2,3) 0.113 1.003 0.992 0.984 GARCH (1,1) 1

0.899 0.814 0.640 GARCH (1,1) 1

0.907 0.824 0.651 GARCH (1,1) 1

0.945 0.830 0.680 -0.694 GARCH (1,1) 1 GARCH (1,1) 1

0.895 0.807 0.690 GARCH (1,1) 1

0.874 0.832 0.735 GARCH (2,1) 1

0.930 0.765 0.625 GARCH (1, 1) 1

0.904 0.872 0.748 GARCH (1,1) 1

0.085 0.021* 0.887 0.135 20.17 22.14 -2436.0

0.152 0.040* 0.829 0.134 29.02 32.53 -2162.7

0.027 0.115 0.903 -0.071 20.31 27.16 -1935.5

0.010* 0.061 0.954 -0.036 17.93 26.77 -2125.2

0.134 -0.016* 0.935 0.093 10.57 23.02 -2538.3

0.317 0.129 0.808 0.031* 18.41 24.33 -2743.9

0.123 0.097 0.865 -0.015* 33.94 24.32 -2124.3

0.087 0.063* 0.057* 0.836 -0.013* 18.50 32.80 -2037.3

0.043 0.107 0.809 0.069* 20.78 12.15 -1432.7

0.182 0.115 0.868 -0.024* 17.20 21.02 -2737.0

0.002* 0.095 0.943 -0.085 9.59 15.89 -1333.6

Notes: Refer Table 5A

Table 6C: TAR-GARCH Model Estimates for monthly stock returns : 1/1/2002-31/12/2006 Country Structure of mean equation SKOR ARMA (21, 0) IND TARCH process does not fit SAFR ARMA (20, 20) TW ARMA (20, 20) BZL ARMA (21, 21) RUS ARMA (21, 0) CH ARMA (21, 21) MEX ARMA (21, 21) MAL ARMA (21, 0) TUR ARMA (21, 0) D.W. ARMA (1,20)

21

Coefficients of mean equation Significant lags c a1 a10 a11 a18 a19 a20 a21 b19 b20 b21 Order of variance equation Order of asymmetry for TARGARCH Coefficients of variance equation 0 1 1 1 Q(24) Q2(24) Log likelihood

AR (1,19,21) 1.761 0.949

AR (1,19, 20) MA(20) 2.535 0.970

AR(1) MA(20) 0.502 0.986

AR (1,18,21) MA(19, 20) 2.734 0.096

AR (1,19,20, 21) 4.563 0.971

AR (1,19,20, 21) MA(21) -0.689* 0.981

AR(1) MA (20,21) 3.285 0.100

AR (1,18,20, 21) 0.520* 0.977

-0.085 -0.165 0.152 -0.348 GARCH (1,1) 1 GARCH (1,1) 1 -0.947 GARCH (1,1) 1 -0.185 0.184 0.072 -0.117 -0.127 GARCH (1,1) 1 -0.113 -0.223 0.324 -0.068 -0.172 0.231 -0.858 -0.036 GARCH (1,1) 1

-0.083 -0.117 0.173

AR (1,10,11, 18,20, 21) 2.898 0.956 0.099 -0.072 -0.103 -0.231 0.296

AR(1) MA(20) 0.670 0.995

-0.951 GARCH (1,1) 1 GARCH (1, 1) 1 1 GARCH (1,1)

GARCH (1,1) 1

-0.093 GARCH (1,1) 1

0.027* 0.019* 0.951 0.043 30.57 31.62 -2444.9

0.017* 0.104 0.938 -0.092 30.11 21.49 -2081.3

0.007* 0.072 0.960 -0.070 18.07 33.13 -1989.5

0.090* 0.031 0.948 -0.008* 30.69 31.67 -2699.8

1.160* 0.190 0.712 -0.084* 20.29 51.22 -2906.1

0.0171 0.104 0.866 0.025* 17.67 30.38 -2233.6

0.047 0.134 0.908 -0.133 27.86 27.00 -1957.9

0.242 0.230 0.558 -0.041* 28.49 7.83 -1587.8

0.131 0.094 0.913 -0.043* 22.27 51.02 -2859.4

0.006* 0.082 0.963 -0.089 26.06 20.90 -1368.8

Notes: Refer Table 5A

Table 6D: TAR-GARCH Model Estimates for annual stock returns : 1/1/2002-31/12/2006 Country Structure of mean equation Coefficients SKOR ARIMA (0, 1, 0) IND ARIMA (2, 1, 2) SAFR ARIMA (0, 1, 0) TW ARIMA (0,1, 0) BZL ARIMA (0, 1, 0) RUS ARIMA (0, 1, 0) CH ARIMA (0, 1, 0) MEX ARIMA (0, 1, 0) MAL ARIMA (1, 1, 0) TUR ARIMA (0, 1, 0) D.W. ARIMA (0,1,0)

22

of mean equation Significant lags a1 a2 b1 b2 Order of variance equation Order of asymmetry for TARGARCH Coefficients of variance equation 0 1 1 1 Q(24) Q2(24) Log likelihood

GARCH (1,1) 1

AR(1,2) MA(1,2) -0.931 -0.694 1.006 0.724 GARCH (1,1) 1

AR(1) 0.159

GARCH (1,1) 1

GARCH (1,1) 1

GARCH (1,1) 1

GARCH (1,1) 1

GARCH (1,1) 1

GARCH (1,1) 1

GARCH (1, 1) 1

GARCH (1, 1) 1

GARCH (1,1) 1

0.028* 0.071 0.949 -0.050 24.97 28.30 -2237.0

0.260 0.125 0.843 -0.004* 30.40 12.85 -2301.8

0.089 0.060 0.910 0.007* 27.82 14.27 -2000.1

0.038* 0.057 0.951 0.035* 32.01 15.22 -2046.7

0.110 0.073 0.945 -0.058 29.40 25.31 2563.3

0.441 0.160 0.865 -0.088 22.73 36.40 --2771.1

0.126 0.145 0.885 -0.107 33.90 13.75 -1951.7

0.120 0.106 0.897 -0.057* 15.50 9.17 -2148.5

0.010* 0.081 0.945 -0.069 24.83 17.66 -1365.5

0.257 0.074 0.914 -0.005* 20.73 17.05 -2734.4

0.010 0.090 0.938 -0.070 27.59 9.99 -1538.0

Notes: Refer Table 5A In some markets, the TAR-GARCH effect (indicated by non-zero values of 1) is present under high frequency return data and in other markets it is present in low frequency return data. The appropriateness of the TAR-GARCH(1,1) specification to describe the conditional variance seems to depend on the frequency of the data. The results are in agreement with the study made by Chiang and Doong , 2001. For the developed world, TAR-GARCH process does not fit the return data under all measures of frequency as the sum of 1 and 1 is greater than one in all cases.

5. Conclusion In this paper, we examine the behaviour of stock returns and volatility of ten emerging markets over a five year period from January 1, 2002 to December 31,2006 at daily, weekly, monthly 23

and annual intervals and compare them with those of developed markets. We find that the distributions of stock returns show a high degree of leptokurtosis and are not normal. The stock return series are stationary under measures of daily, weekly and monthly frequencies but are non-stationary at annual frequency. First-order autocorrelation is observed in weekly, monthly and annual stock returns for all markets but such autocorrelation exists only in a few markets in case of daily stock returns. This is indicative of stock return predictability at low frequency intervals. There is no difference in the distributional characteristics of stock returns between emerging markets and developed markets. However, the ratio of stock returns to volatility (measured by standard deviation of return series) for emerging markets is higher than that of developed markets. This points to the benefit of including stocks from emerging markets in global investment portfolios. Application of Engles ARCH test indicates presence of volatility clustering in all markets under all measures of frequency. GARCH(1,1) appears to explain the volatility pattern in most of the cases. The coefficients of the mean equation in GARCH specification are significant at different lags depending on the interval at which the stock return is measured. The coefficients of the variance equation in GARCH specification indicate that volatility in stock returns evolves in a persistent fashion. Further analyzing the leverage effect in volatility by using a TAR-GARCH model indicates that the appropriateness of the TAR-GARCH specification to describe the conditional variance depends on the frequency of the stock return data. The TAR-GARCH process does not, however, fit the stock return data of developed markets.

References:
Apergis, Nicholas and Sophia Eleptheriou (2001), Stock Returns and Volatility: Evidence from Athens Stock Market Index, Journal of Economics and Finance, Spring 2001, 25,1, 50-61.

24

Abraham, Abraham, Fazal J. Seyyed and Sulaiman A. Alsakran (2002), Testing the Random Walk Behaviour and Efficiency of the Gulf Stock Markets, The Financial Review, 37(2002), 469-80. Baillie, Richard T., Robert J. Myers (1991), Bivariate Garch Estimation of the Optimal Commodities Futures Hedge: Summary, Journal of Applied Econometrics, April-June, 6, 2,109-24. Barry, B., J.W. Peavy III, and M. Rodriquez (1998), Performance Characteristics of Emerging Capital Markets, Financial Analysts Journal January/February, 54(1), 72-80. Barry, B. and L.J.Lockwood, New Directions in Research on Emerging Capital Markets (1995), Financial Markets, Institutions and Instruments, December 1995, 4(5), 15-36. Bollerslev, T., R.Y.Chou and K.F.Kroner (1992), Arch Modeling in Finance, Journal of Econometrics, 52, 5-59. Bollerslev, T. (1986), Generalised Autoregressive Conditional Heteroscedasticity, Journal of Econometrics, 31, 30727. Box, G., and G. Jenkins (1976), Time Series Analysis, Forecasting and Control, San Francisco, CA; Holden-Day. Chang, Kuo-Ping and Kuo-Shiuan Ting (2000), A Variance Ratio Test of the Random Walk Hypothesis for Taiwans Stock Market, Applied Financial Economics, 2000, 10, 525-32. Chiang, Thomas C. and Shuh Chyi Doong (2001), Empirical Analysis of Stock Returns and Volatility: Evidence from Seven Asian Stock Markets Based on TAR-GARCH Model, Review of Quantitative Finance and Accounting, Nov. 2001; 17,3, 301-18. Cheung, Kwong-C. and J. Andrew Coutts (2001), A Note on Weak Form Market Efficiency in Security Prices: Evidence from Hong Kong Stock Exchange, Applied Economic Letters, 8, 407-10. Chou, Yeutien Ray (1988), Volatility Persistence and Stock Valuations: Some Empirical evidence using GARCH, Journal of Applied Econometrics Oct-Dec. 3(4), 279-94. Christie, A.A. (1982), The Stochastic Behaviour of Common Stock Variances: Value, Leverage and Interest Rate Effects, Journal of Financial Economics, 10, 407-432. De Santis, G. and Selahattin Imrohoroglu (1997), Stock returns and Volatility in Emerging Financial Markets, Journal of International Money and Finance, 16(4) , 561-79 Divecha, A.B., J. Drach and D.Stefek (1992), Emerging Markets: A Quantitative Perspective Journal of Portfolio Management Fall ,19(1), 41-56. Engle, R.F. (1982), Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of UK Inflation, Econometrica ,50, 987-1008. Engle, R.F. and T.P.Bollerslev (1986), Modelling the Persistence of Conditional Variances, Econometric Review, 5, 1-50. Fact Sheet: IFC and Emerging Markets at a Glance retrieved August 16, 2006 from < http:// www.ifc.org/ifcext/50th anniversary.nsf/Content/Fact_sheet_English > Fernandez, Nuno (2005), Portfolio Disaggregation in Emerging Market Investment, Journal of Portfolio Management, 31 (2), 41-49. Fraser, P. and D. Power (1997), Stock Return Volatility and Information: An Empirical Analysis of Pacific Rim, U.K. and U.S. equity markets, Applied Financial Economics ,7, 241-53. Glosten, L.R., R. Jagannathan and D.E. 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.

25

Grieb, Terrance and Mario G. Reyes (1999), Random Walk Tests For Latin American Equity Indexes and Individual Firms, The Journal of Financial Research, Vol. XXII, No.4, 371-83. Harvey, C.R. (1995), Predictable Risk and Returns in Emerging Markets, Review of Financial Studies, 773-816. Huang , Bwo-Nung (1995), Do Asian Stock Market Prices Follow Random Walks? Evidence from the Variance Ratio Test, Applied Financial Economics, 5, 251-56. Huber Peter (1997), Stock Market Returns in Thin Markets: Evidence from the Vienna Stock Exchange, Applied Financial Economics, 7, 493-98. Hull, John C. (2006), Options, Futures and Other Derivatives, Prentice Hall of India. Karmakar, M. (2003), Heteroscedastic Behaviour of the Indian Stock Market: Evidence and Explanation, Journal of Academy of Business and Economics, 1, 27-36. Karmakar, M. (2006), Time-varying Volatility and Leverage Effect in Financial Markets of Asia-Pacific Countries, The ICFAI Journal of Applied Finance, 12(6), 19-29. Lai, Ming-Ming, K. Guru Balachandher and Fauzias, Mat Nor(2003), An Examination of the Random Walk Model and Technical Trading Rules in the Malaysian Stock Market, Quarterly Journal of Business & Economics, Vol 41, Nos. 1 & 2, 81-104. Lee, Cheng F., Gong-meng Chen and Oliver M. Rui (2001), Stock Returns and Volatility on Chinas Stock Markets, The Journal of Financial Research, Vol. XXIV, No. 4, 523-43. Lima, Eduardo Jose Araujo and Benjamin Miranda Tabak (2004), Tests of the Random Walk Hypothesis for Equity Markets: Evidence from China, Hong Kong and Singapore, Applied Economic Letters, 2004, 11, 255-58. Michelfelder, Richard A.(2005), Volatility of Stock Returns: Emerging and Mature Markets, Managerial Finance, 31, 2, 66-86. Nelson, D.B. (1991), Conditional Heteroskedasticity in Asset Returns: A New Approach, Econometrica, 59, 347370. Park, K.H. and Agtmael A.W.(1993), The Worlds Emerging Stock Markets: Structure, Developments, Regulations and Opportunities. Chicago: Probus Publishing Company. Price M.P.(1994), Emerging Stock Markets, New York: McGraw- Hill. Schewart, G.W. (1989), Why Does Stock Market Volatility Change Over Time?, Journal of Finance, 44, 115-1153. Sharma, J.L. and Robert E. Kennedy (1977), A Comparative Analysis of Stock Price Behaviour of the Bombay, London and New York Stock Exchanges, Journal of Financial and Quantitative Analysis, September, 391-413. Stanley, M.T.(1995), Guide to Investing in Emerging Markets. Chicago: Richard D Irwin Zakoian, J.M.(1994) Threshold Heteroskedastic Models, Journal of Economic Dynamics and Control, 18, 931-955.

26

You might also like