 Open Access
 Total Downloads : 488
 Authors : Abdalla Suliman Mhmoud, F. M. Dawalbait
 Paper ID : IJERTV4IS020150
 Volume & Issue : Volume 04, Issue 02 (February 2015)
 Published (First Online): 21022015
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
Estimating and Forecasting Stock Market Volatility using GARCH Models: Empirical Evidence from Saudi Arabia
A. S. Mhmoud
Assistant Professor, Department of Mathematics,
College of Arts and Sciences, Taif University, kingdom of Saudi Arabia
F. M. Dawalbait
Assistant Professor, Department of Mathematics, College of Arts and Sciences, Taif University, kingdom of Saudi Arabia
Abstract: In this paper we used symmetric and asymmetric GARCH models such as GARCH(1,1) ,EGARCH(1,1) and GRJGARCH(1,1) models to estimate and forecast volatility of Saudi stock market under various assumptions namely: Normal,Student t and GED distributions .The study carried out using daily closing prices index over the period of 1st January 2005 to 31st December 2012. The common measures of forecast evaluation of the models such as Root Mean Square Errors(RMSE), Mean Absolute Errors (MAE), Mean Absolute Percentage Errors(MAPE) and Theil Inequality Coefficient(TIC) were computed .The empirical results showed that the asymmetric GARCH models with a heavy tailed error distribution better than the symmetric GARCH model in the estimation the conditional variance equations. Moreover ,we found that the GRJ GARCH(1,1) model provide the best out of sample forecast for Saudi stock market . Finally, the empirical results reveal that conditional variance process is highly persistent and confirm the presence of leverage effect in returns of Saudi stock market.
1. INTRODUCTION
Over the last few decades volatility forecasting has been an important area of research in financial markets for academics, policy makers and market participants. Since 1982 when Engle proposed the Autoregressive Conditionally Heteroscedastic (ARCH) model, a lot of effort has been expended in improving volatility models since better forecasts translate into better pricing of options and better risk management [1]. Therefore, in this section we overview a number of papers that have investigated the performance of GARCH models with regard to nonnormal error distribution in mature stock .
There is a large number of research studies that examine stock market volatility carried out in the context of both developed and developing countries. However, such a study does not exist for Saudi Arabia [2] . We try to fit an adequate model to estimate and forecast stock volatility of Saudi Arabia
REVIEW OF SAUDI STOCK MARKET
The Saudi stock Market, the Tadawul, is the largest in the Gulf region, it was established in 1984. It lists 156 publicly traded companies (as of September 2, 2012) , divided into nine sectors, according to Bloombergs classification, each of which has its own subindex. In order of size, they are: Financials, Basic Materials (Petrochemicals), Industrials ,Telecoms, Consumer Goods, Consumer Services, Oil and Gas, Utilities, Healthcare . The Saudi Arabia Monetary Agency (SAMA) was responsible for supervising the market from 1984 until 2003. In July 2003, authority was handed over to the newlyformed Capital Market Authority (CMA). The CMA is now the sole regulator and supervisor of Saudi Arabias capital markets, and issues the necessary rules and regulations to protect investors and ensure fairness and efficiency in the market. The main index, the Tadawul All Share Index (TASI) reached its highest point at 20,634.86 on 25 February 2006 and a record low of 1140.57 Index points in May of 1995. During 20042007 period, interesting events emerge[3] . First, the index price started at 4,432 point during year 2004 and increased sharply to 20,600 point in 2006, which represent highest level during the last 18 years and only lasted for one day (25th February 2006) before the index decline. This sharp increase in the index price could be due to the good news effects, such as the European Union WTO and the sharp increase in the oil price ($70) in 2005. Secondly, the market index lost more than 10,554 points during a short period of time (from26th January 2006 to 1st May 2006), despite the continuous increase in oil price, which reached to $90 in 2008. By the end of 20082009, the stock market lost about 5,343 points where the price declined to the same level as in 2004.
LITERATURE REVIEW
There has been a large amount of literature on modeling and forecasting stock market volatility in both developed and developing countries around the world. Many econometric models have been used to investigate volatility characteristics. However, no single model is superior. K.Lakshmi estimated the conditional volatility of Saudi stock market by applying AR(1) GARCH(1,1) models to the daily stock returns data spending from August 1,2004 to October 31,2013. The results show that a linear GARCH(1,1) is adequate to estimate the volatility stock market of the country and Saudi stock market returns are characterized by volatility clustering and followed a non normal distribution .Moreover, the study showed that past returns play an important role in determining the current period return [2].
Hassan B. Ghassan and Hassan R. Alhajhoj tested the effect of capital market liberalization on volatility of TASI. The results of return equation exhibit the existence of a positive relationship between return and risk, which indicates the high risk and explains the dynamics of shareholders behavior, especially on Saturday and Tuesday, where utmost important information is excreted[4].
Ajab Al Freedi, Ahmed Shamiri and Zaidi Isa in their study examined several stylized facts (heavytaileness, leverage effect and persistence) in volatility of stock price returns exploiting symmetric and asymmetric GARCH family models for Saudi Arabia. Their study was carried out using closing stock market prices over 15 years covering the period 1st January 1994 to 31th March 2009.The sample period was divided into three subperiods according to the local crisis in 2006. Their findings revealed that asymmetric models with heavy tailed densities improve overall estimation of the conditional variance equation. Moreover, they found that AR (1) GJR GARCH model with Studentt outperform the other models during and before the local crisis in 2006,while AR (1) GARCH model with GED exhibits a better performance after the crisis. Furthermore, their results revealed that the existence of leverage effect at 1 percent significance leve[3].
Elsheikh and Suliman use the Generalized Autoregressive Conditional Heteroscedastic models to estimate volatility (conditional variance) in the daily returns of the principal stock exchange of Sudan namely, Khartoum Stock Exchange (KSE) over the period from January 2006 to November 2010. The models include both symmetric and asymmetric models that capture the most common stylized facts about index returns such as volatility clustering and leverage effect. The empirical results show that the conditional variance process is highly persistent (explosive process), and provide evidence on the existence of risk premium for the KSE index return series which support the positive correlation hypothesis between volatility and the expected stock returns[5].
A paper by D.N. Vee and P.Gonpot aimed at evaluating volatility forecasts for the US Dollar/Mauritian Rupee exchange rate obtained via a GARCH (1,1) model under two distributional assumptions: the generalized Error
Distribution (GED) and the Studentst distribution. They make use of daily data to evaluate the parameters of each model and produce volatility estimates. The forecasting ability was subsequently assessed using the symmetric loss functions which are the Mean Absolute Error(MAE) and Root Mean Square Error (RMSE). They latter show that both disributions may forecast quite well with a slight advantage to the GARCH(1,1) GED for outofsample forecasts[6].
A paper by Dumitru and Cristiana compares several statistical models for daily stock return volatility in terms of sample fit and outofsample forecast ability. The focus is on U.S. and Romanian daily stock return data corresponding to the 20022010 time interval .They investigate the presence of leverage effects in empirical time series and estimate different asymmetric GARCH family models (EGACH, PGARCH and TGARCH) specifying successively a Normal, Student's t and GED error distribution. They find that GARCH family models with normal errors are not capable to capture fully the leptokurtosis in empirical time series, while GED and Students t errors provide a better description for the conditional volatility. presence in empirical time series. Finally, they report that volatility estimates given by the EGARCH model exhibit generally lower forecast errors and are therefore more accurate than the estimates given by the other asymmetric GARCH models[7].
Ahmed Shamiri and Zaidi Isa investigated the relative efficiency of several different types of GARCH models in terms of their volatility forecasting performance. They compared the performance of symmetric GARCH, asymmetric EGARCH and nonlinear asymmetric NAGARCH models with six error distributions (normal, skew normal, studentt, skew studentt, generalized error distribution and normal inverse Gaussian. Their results suggested that allowing for a heavytailed error distribution leads to significant improvements in variance forecasts compared to using normal distribution[8].
In an investigation by C. Kosapattarapim, YanXia and M. Michael employed six simulated studies in GARCH (p,q) with six different error distributions are carried out. The analysis was then carried out using the daily closing price data from Thailand (SET), Malaysia (KLCI) and Singapore (STI) stock exchanges. Their Results show that although the best fitting model does not always provide the best future volatility estimates the differences are so insignificant that the estimates of the best fitting model can be used with confidence. The empirical application to stock markets also indicated that a non normal error distribution tends to improve the volatility forecast of returns. They conclude that volatility forecast estimates of the best fitted model can be reliably used for volatility forecasting[9].
G.R.Pasha,Tahira Qasim and Muhammad Aslam in their paper compare the performance of different GARCH models such as GARCH, EGARCH,GJR and APARCH models, to characterize and forecast financial time series volatility in Pakistan. The empirical results demonstrate that the use of asymmetry in the GARCH models and the assumption of fattail distributions for the innovations improve the volatility forecasts. Overall, EGARCH fits the
best while the GJR model, with both normal and non normal innovations, seems to provide superior forecasting ability over short and long horizons[10].
ht 0
q
i1
2 t 1
(3)
j
According to R. Engle, D. Lilien and R.Robins
Volatility is a key parameter used in many financial applications, from derivatives valuation to asset
This model is referred to as ARCH(q), where q refers to the order of the lagged squared returns included in the model. If we use ARCH(1) model it becomes:
management and risk management. Volatility measures the
size of the errors made in modeling returns and other
ht 0
u 2 t 1
(4)
1
financial variables. It was discovered that, for vast classes
Where
yt is observed time series. t is residual. 0
of models, the average size of volatility was not constant but changes with time and is predictable. Autoregressive
is constant.
j represents ARCH effect. q is length of
conditional Heteroscedasticity (ARCH)/generalized autoregressive conditional Heteroscedasticity (GARCH) models and stochastic volatility models are the main tools used to model and forecast volatility[1]
A paper by J. Frimpong and F. Oteng aimed to make models and forecasts volatility (conditional variance) on the Ghana Stock Exchange using a random walk (RW), GARCH(1,1), EGARCH(1,1), and TGARCH(1,1) models.
The unique three days a week Databank Stock Index (DSI) is used to study the dynamics of the Ghana stock market volatility over a 10year period. The competing volatility models were estimated and their specification and forecast performance compared with each other, using AIC
ARCH lags. Since ht is a conditional variance, its value
must always be strictly positive; a negative variance at any point in time would be meaningless. To have positive conditional variance estimates, all of the coefficients in the conditional variance are usually required to be non negative. Thus coefficients must be satisfy 1 0 .
GARCH Model: The model allows the conditional variance of variable to be dependent upon previous lags; first lag of the squared residual from the mean equation and present news about the volatility from the previous period which is as follows:
and LL information criteria and BDS nonlinearity q p
diagnostic checks. The DSI exhibits the
h 2 t i h
(5) Un
stylized characteristics such as volatility clustering,
t 0 1
i1
i1
i t i
leptokurtosis and asymmetry effects associated with stock der the hypothesis of covariance stationary, the
market returns on more advanced stock markets. The random walk hypothesis is rejected for the DSI. Overall, the GARCH (1,1) model outperformed the other models
unconditional variance ht can be found by taking the unconditional expectation of equation 5 . We find that
under the assumption that the innovations follow a normal
ht 0 1h 1h
(6)
distribution[11].
MATERIALS AND METHODS
Methodology: Autoregressive conditional
Where 1 and 1 are the ARCH effect and GARCH effect
Solving the equation 6 we have
heteroscedasticity (ARCH) and its generalization (GARCH) models represent the main methodologies that
h 0
1 1 1
(7)
have been applied in modeling and forecasting stock market volatility [5] .In this project different symmetric and asymmetric GARCH models are use .In the volatility modeling process using GARCH models, the mean and variance of the series are estimated simultaneously. This section briefly reviews this methodology
Volatility Models: Volatility model should be able to forecast volatility. Virtually all the financial uses of volatility models entail forecasting aspects of future returns. Typically a volatility model is used to forecast the absolute magnitude of returns. Volatility models can be
For this unconditional variance to exist, it must be the case that 1 1 1 and for it to be positive, we require that
0 0 .
GJR GARCH Model: The GJR model is a simple extension of GARCH with an additional term added to account for possible asymmetries [12] .Glosten, Jagananthan and Runkle (1993) develop the GARCH model which allows the conditional variance has a different response to past negative and positive innovations [13].
divided into symmetric and asymmetric models. In this q 2 2 p
paper we use one symmetric GARCH models which is
ht 0 i t i i t i dt 1 j ht j
(8)
GARCH(1,1) and two asymmetric GARCH models,
i1
i1
namely EGARCH(1,1) and GJRGARCH(1,1).
ARCH Model: ARCH models based on the variance of
Where is a dummy variable that is:
the error term at time t depends on the realized values of the squared error terms in previous time periods. The model is specified as:
1 if
d
t 1
0 if
t 1 0,
t 1 0,
bad news
good news
yt t
t ~ N (0, ht )
(1)
(2)
The coefficient is known as leverage term . In the model, effect of good news shows their impact by i ,
while bad news shows their impact by
. In
addition if 0 news impact is asymmetric and
0 leverage effect exists. To satisfy nonnegativity condition coefficients would be 0 0 , i 0 , 0 and i i 0 . That is the model is still acceptable, even if i 0 , provided that i i 0. [12] EGARCH Model: Nelson (1991) proposed a GARCH class model named Exponential GARCH that allows for asymmetric effects and therefore solves one of the important shortcomings of the symmetric models. While the GARCH model imposes the nonnegative constraints on the parameters, EGARCH models the log of the conditional variance so that there are no restrictions on these parameters [14] .
DESCRIPTIVE STATISTICS
Table 1 presents descriptive statistics on TASI returns over the sample period. As expected for time series of returns, the mean is close to zero , while the daily volatility is represented by standard deviation is 0.01845, which is high value indicating high fluctuations of the TASI daily returns. The skweness coefficient is negative , reveal that the TASI returns has a heavy left tail ,while kurtosis is very high (39.70), suggesting that the distribution is highly leptokurtic confirming previous finding that stock returns are not normally distributed. The high value of Jarque Bbera corroborate that normality is rejected at a pvalue of almost 1 % . Moreover, Engle (1982) LM test statistics indicates the existence of ARCH effects in the residual series and therefore the variance of returns series is non constant for period under review .
log( h 2 t )) w log( p
t 1
) t 1
ht 1
t1
ht1
(9)
The Ljung Box Q statistics of order 20 on both returns and squared returns reflects a high serial correlation.
Where and are the ARCH and GARCH parameters .
is leverage parameter and w is constant
Minimum 
0.137730 
Maximum 
0.040780 
Mean 
0.0000391 
Median 
0.000545 
Standard deviation 
0.008450 
Skewness 
2.709126 
Kurtosis 
39.70133 
Q 20 
60.83 
Prob. of Q 20 
0.0000 
Q2 20 
1623 
Prob. of Q2 20 
0.0000 
Jarque Bera 
125706.2 
Prob. of Jarque Bera 
0.0000 
ARCH(2) 
55.43 
Prob. Of ARCH(2 
0.0000 
Note that the lefthand side is the logarithm of the conditional variance. This implies that the leverage effect is exponential, rather than quadratic, and that forecasts of the conditional variance are guaranteed to be nonnegative. The presence of leverage effects can be tested by the hypothesis
Table 1. Descriptive Statistics of TASI returns series
that < 0. If 0 , then the impact is asymmetric E
GARCH basically models the log of the variance (or standard deviation) as a function of the lagged logarithm of the variance/std dev and the lagged absolute error from the regression model. It also allows the response to the lagged error to be asymmetric, so that positive regression residuals can have a different effect on variance than an equivalent negative residua.
DATA FOR ANALYSIS
The data used in this paper consist of daily closing price of Saudi Arabia stock market index namely ,Tadawul All Share Index(TASI) over the period of 1st January 2005 to 31st December 2012. This yields of 2317 observations. The estimation process is run using data from 1st January 2005 to 30th June 2012, while the remaining data ( 1st July2012 to 31th December 2012) are used for the evaluation of the out sample forecast performance .The data series of stock market have been taken from the Saudi Stock Market Website (http://www.tadawul.com.sa) . Daily returns rt were calculated as a logarithm of TASI indices such as :
p
Fig 1 presents the patterns of daily logarithm returns of the TASI for the period under review . We observe that small returns changes tend to be followed by small changes and large returns changes tend to be followed by other large returns changes of same signs. This behavior of stock returns series indicates that there is a clear evidence of volatility clustering in TASI returns .The implication of volatility clustering is that volatility shocks to day will influence the expectation of volatility in the immediate future periods .Fig 2 shows the distribution of TASI log returns, which clearly indicate the departure from
r log( t )
(10)
normality with a high peaked distribution.
p
t
t 1
Where rt
is daily return of the index at time t, pt
and
pt 1 are the closing market index at the current t and previous day(t1) respectively .
.050
.025
.000
.025
.050
.075
.100
.125
.150
05 06 07 08 09 10 11 12
Fig 1 : Daily Returns series
900
800
700
600
Frequency
500
400
300
200
100
0
.15 .10 .05 .00 .05
Fig 2 : Daily Returns Distribution
Generally , the TASI returns series do not confirm that In the variance equation from table 3 , the first three
there is normal distribution, but display negative skewness and leptokurtic distribution. This predicated that a fatter tailed distribution such as as Studentt or GED distribution should generate better results than normal distribution
UNIT ROOT TEST FOR THE TASI DAILY RETURNS
To verify whether the daily returns are stationary series, the Augmented Dickey fuller (ADF) test was used. The results of test are present in table 2.The results reject the null hypothesis of a unit root test. Consequently, we Deduce that returns are stationary in level for the period under review.
Table 2: Stationary Test for daily returns series
TASI returns series 

ADF statistic 
Critical values 

1% 
5% 
10% 

43.17877 
3.43313 
2.86265 
2.56741 
EMPIRICAL RESULTS AND DISCUSSION
In this section, we present the estimates of different GARCH models for TASI returns series . The basic estimation models consist of two equations; one for the mean, which is a simple
autoregressive model and the other for the variance represented in the GARCH(1,1), EGARCH(1,1) and
GJRGARCH(1,1) models . A maximum likelihood approach is used to estimate the models, with three underlying error distributions namely: normal, Studentt and Generalize Error Distribution (GED). Table3, 4 and 5 present the estimation results for the parameters of GARCH, EGARCH and GJRGARCH models while tables 6 – 8 report diagnostic tests.
coefficients 0 (constant) , 1 (ARCH term) and
1 (GARCH term) are highly significant , at standard level
. Moreover, the tail coefficients are significant justifying the use of non normal densities, the sum of
1 and 1 is close to unity , suggesting that shocks to the conditional variance are highly persistence, indicating that large changes in returns tend to followed by large changes and small changes tend to be followed by small changes , this mean that volatility clustering is observed in TASI index returns series.
The results presentd in table 4and 5 show that all asymmetric coefficients are statistically significant at 1% significance level, implies that the use of asymmetric GARCH models seem to be justify .The leverage effect
term in the asymmetric models EGARCH(1,1) and GJR
GARCH(1,1)are statistically significant with negative sign
, denoting that negative shocks tend to produce a higher volatility in the immediate future than positive shocks of same sign ,indicating that the existence of leverage effect is observed in returns of TASI market index.
Table 3: parameter estimation of AR(1)GARCH model
Norm al 
Studentt 
GED 

Mean Equation 

0 
0.000232 (0.0345) 
0.000532 (0.0000) 
0.000569 (0.0000) 
1 
0.248065 (0.0000) 
0.066963 (0.0021) 
0.030028 (0.0927) 
Variance Equation 

0 
0.00000188 (0.0000) 
0.00000120 (0.0000) 
0.00000131 (0.0000) 
1 
0.262093 (0.0000) 
0.210373 (0.0000) 
0.208464 (0.0000) 
1 
0.750247 (0.0000) 
0.812193 (0.0000) 
0.801742 (0.0000) 
3.56198 (0.0000) 
0.946829 (0.0000) 
Normal 
Studentt 
GED 

Mean Equation 

0 
0.000408 (0.0014) 
0.000505 (0.0000) 
0.000542 (0.0000) 
1 
0.0.082199 (0.0002) 
0.7427 (0.0013) 
0.031784 (0.0693) 
Variance Equation 

0 
0.986622 (0.0000) 
0.637683 (0.0000) 
0.729641 (0.0000) 
1 
0.357857 (0.0000) 
0.272684 (0.0000) 
0.273951 (0.0000) 
1 
0.925928 (0.0000) 
0.955900 (0.0000) 
0.947474 (0.0000) 
0.230582 (0.0000) 
0.073230 (0.0000) 
0.096531 (0.0000) 

3.706843 (0.0000) 
0.958149 (0.0000) 
Figures in parentheses are pvalue Table 4: parameter estimation of AR(1) EGARCH model
Figures in parentheses are pvalue
Table 5: parameter estimation of AR(1) GJR GARCH model
Normal 
Studentt 
GED 

Mean Equation 

0 
0.000408 (0.0014) 
0.000505 (0.0000) 
0.000542 (0.0000) 
1 
0.0.082199 (0.0002) 
0.7427 (0.0013) 
0.031784 (0.0693) 
Variance Equation 

0 
0.986622 (0.0000) 
0.637683 (0.0000) 
0.729641 (0.0000) 
1 
0.357857 (0.0000) 
0.272684 (0.0000) 
0.273951 (0.0000) 
1 
0.925928 (0.0000) 
0.955900 (0.0000) 
0.947474 (0.0000) 
0.230582 (0.0000) 
0.073230 (0.0000) 
0.096531 (0.0000) 

3.706843 (0.0000) 
0.958149 (0.0000) 
Figures in parentheses are pvalue
Table 6 to 8 show models diagnostics of different GARCH models for the returns of TASI index. Ljung BoxPierce statistics Q2 (20) on the squared standardized residuals is
nonsignificant at 5% level, indicating that all models seem to be a adequately in describing the dynamics of the series. Moreover, LM test for presence of ARCH effect at lag 2 exhibit the absence of ARCH effect in standardize residual.
These results reveal that all variance equations are specified correctly. As expected asymmetric EGARCH and GJRGARCH models have both smaller values of AIC and bigger log likelihood function than the traditional GARCH model , therefore we conclude that the EGARCH and GJRGARCH models better the estimation the TASI returns series than the traditional GARCH model. Overall, using AIC and log likelihood values as model selection criteria, for the GARCH models , the results show that GJR GARCH model with all three distributions provide the best estimation for the TASI returns series. When we analyze the densities we find that ,the GED distribution clearly outperform the Student t and Gaussian distributions for all models estimated in this study.
Table 6 : Model diagnostics of AR(1) GARCH Model
Normal 
Studentt 
GED 

Q2 (20) 
0.972 (1.000) 
0.6475 (1.000) 
0.688 (1.000) 
ARCH(2) 
0.160 (0.923) 
0.0596 (0.9706) 
0.068239 (0.9665) 
AIC 
7.115337 
7.459543 
7.446527 
BIC 
7.102347 
7.443956 
7.430939 
LogLike 
7799.852 
8177.930 
8163.670 
Table 7 : Model diagnostics of AR(1) EGARCH Model
Normal 
Studentt 
GED 

Q2 (20) 
0.7627 (1.0000) 
0.47200 (1.00) 
0.5205 (1.000) 
ARCH(2) 
0.1511 (0.9227) 
0.043468 (0.9785) 
0.057271 (0.9718) 
AIC 
7.105337 
7.451015 
7.441348 
BIC 
7.102347 
7.432829 
7.423162 
LogLike 
7843.501 
8169.587 
8158.997 
Table 8: Model diagnostics of AR(1) GJRGARCH
Normal 
Studentt 
GED 

Q2 (20) 
0.88 (1.00) 
0.7102 (1.000) 
0.7723 (1.000) 
ARCH(2) 
0.162 (0.922) 
0.046820 (0.9542) 
0.112581 (0.9453) 
AIC 
7.183363 
7.465408 
7.455380 
BIC 
7.16775 
7.447223 
7.437194 
LogLike 
7875.374 
8185.355 
8174.369 
FORECAST PERFORMANCE
To evaluate and compare forecast performance of different GARCH models , several evaluation criteria were computed .In this study a variety of statistics have been used to evaluate and compare forecast performance. They include Root Mean Suare Errors(RMSE), Mean Absolute Errors (MAE), Mean Absolute Percentage Errors(MAPE) and Theil Inequality Coefficient(TIC).The model exhibits the lowest value of the errors is considered to be the best one. In this paper the length of the outsample period is chosen to be 124 days . The results reported in Tables 9,10and 11 show that the model exhibit the better
forecasting performance is GJR GARCH for all three distributions .From these results, we can conclude that volatility forecast of the TASI index may be improved by using asymmetric GARCH models with non normal distributions.
Table 9: Forecast performance out of sample – Normal
GARCH 
EGARCH 
GJRGARCH 

MSE 
0.008509 
0.008435 
0.008435 
MAE 
0.005064 
0.005054 
0.5050 
MAPE 
161.8994 
129.6981 
121.4941 
TIC 
0.903665 
0.912480 
0.933341 
Table 10: Forecast performance out of sample Student tt
GARCH 
EGARCH 
GJRGARCH 

MSE 
0.008443 
0.008442 
0.008439 
MAE 
0.005050 
0.005049 
0.005051 
MAPE 
135.8133 
133.9952 
131.7062 
TIC 
0.915369 
0.916719 
0.91653 
Table 11: Forecast performance out of sample – GED
GARCH 
EGARCH 
GJRGARCH 

MSE 
0.008475 
0.008453 
0.008451 
MAE 
0.005043 
0.005043 
0.005043 
MAPE 
133.4086 
131.0476 
130. 0740 
TIC 
0.931388 
0.933566 
0.934030 
CONCLUSION
We find that most of the financial markets, characterized by feature of volatility, this property means a large and abnormal fluctuations. In, for instance, prices of shares and bonds .Naturally enough, these fluctuations are undesirable by investors and decisionsmakers as they result in a state of uncertainty in financial transactions, and, thus, negatively affecting the economy. To deal with such problems, there is a need to use statistical models that take into account these fluctuations, one of these is the General Auto Regressive conditional heteroscedasticity models(GARCH models.
The main goal of this study is to compare the performance of different GARCH models including symmetric and asymmetric models in estimating and forecasting the volatility of Saudi stock market index . We found that the daily return of TASI exhibit the stylized fact such as volatility clustering, leptokurtosis , departure from normality and existence of heteroscdasticity in residuals series. The models were estimated assuming various assumptions namely: Normal, Student t and GED distributions. The sum of the parameters estimates the GARCH (1,1) is close to unity , revealing that a high degree of persistent in the conditional volatility of stock
returns on Saudi stock market. According to AIC and Log likelihood function among the competing models , asymmetric GARCH models(EGARCH and GRJGARCH) fit the conditional variance equations better than symmetric GARCH model. We have examined the empirical performance of the models for forecasting volatility in Saudi stock market and we found that the GRJ GARCH(1,1) model outperformed the other models. We recommend that GRJ GARCH can be use for estimating and forecasting the daily returns volatility of Saudi stock market.
ACKNOWLEDGMENT
The authors would like to express their sincere thanks to Taif university, Saudi Arabia for supported of this work
REFERENCES

R. Engle, D. Lilien and R. Robins, "Estimating Time Varying Risk Premia in the Term Structure: The ARCHM Model", Econometrica,vo.55, 1987, pp. 391407.

K. Lakshmi, "Stock Market Volatility in Saudi Arabia :An Application of Univariate GARCH Model", Asian Social Science journal , vol 10, no.10, 2014, pp.142152

Ajab Al Freedi, Ahmed Shamiri and Zaidi Isa , " A Study on the Behavior of Volatility in Saudi Arabia Stock Market Using Symmetric and Asymmetric GARCH Models" , Journal of Mathematics and Statistics,vol.8,no.1, 2012 ,pp.98106 .

Hassan B. Ghassan and Hassan R. Alhajhoj ," Effect of Capital Market liberalization on Volatility of TASI ", Journal of Development and Economic Policies , vol.14, no.2, 2012 , PP 739.

Ahmed Elsheikh and Suliman Zakaria ," Modeling Stock Market Volatility Using GRACH Models Evidence From Sudan" , International journal of Business and Social Science , vol. 23, no.2, pp.114122

D.N. Vee , and P.Gonpot, "Forecasting Volatility of USD/MUR Exchange Rate using A GRACH(1,1) Model with GED and Student'st errors'', Univariate of Maurittus Research Journal, Vol. 17, 2011, pp .114.

M.Dumitru,and T.Cristiana ,"Asymmetric Conditional Volatility Models: Empirical Estimation and Comparison of Forecasting Accuracy", Romanian Journal of Economic Forecasting ,vol 3, 2010
, pp.7491 .

Ahmed Shamiri and Zaidi Isa," Modeling and Forecasting Volatility of Malaysian Stock Markets", Journal of Mathematics and Statistics ,vol.5,no.3, 2009,
pp. 234240.

C. Kosapattarapim, YanXia and M. Michael'' Modelling Conditional Volatility Forecasting Performance of Best Fitting GARCH Model in Emerging Asian Stock Markets '', Unpublished .

G.R.Pasha,Tahira Qasim ,and Muhammad Aslam, "Estimating and Forecasting of Financial Time Series in Pakistan with GARCH type Models" , The Labore Journal of Economic
,vol 12,no. 2, 2007, pp115149.

G.R.Pasha,Tahira Qasim ,and Muhammad Aslam, "Estimating and Forecasting of Financial Time Series in Pakistan with ARCH type Models" , The Lahore Journal of Economic ,vol 12,no. 2, 2007, pp115149

C. Brooks and S. Burke," Information Criteria for GARCH Model Selection: An application toHigh Frequency Data", European Journal of Finance ,vol. 9,no.6,pp, 2003,PP .557580 .

L. Glosten, R.Jagannathanand D. Runkle ," On the Relation between the Expected Value and the Volatility of the Nominal Excess Returns on Stocks" Journal of Finance,vol. 48, no. 5 , 1993, PP
.17791791.

M.Dumitru, and T.Cristiana, T.(2010), Asymmetric Conditional Volatility Models: Empirical Estimation and Comparison of Forecasting Accuracy, Romanian Journal of Economic Forecasting
,vol 3, 2010, pp.7491.