VOLATILITY FORECASTING - A PERFORMANCE MEASURE OF GARCH TECHNIQUES WITH DIFFERENT DISTRIBUTION MODELS

International Journal of Soft Computing, Mathematics and Control (IJSCMC), Vol. 5, No. 2/3, August 2016
1
VOLATILITY FORECASTING - A
PERFORMANCE MEASURE OF GARCH
TECHNIQUES WITH DIFFERENT
DISTRIBUTION MODELS
Hemanth Kumar P.1
and Basavaraj Patil S.2
1
Computer Science Engineering, VTURRC, Belagavi
hemanth00kumar@gmail.com
2
Computer Science Engineering, VTURRC, Belagavi
dr.sharan9@gmal.com
Abstract
Volatility Forecasting is an interesting challenging topic in current financial instruments as it is directly
associated with profits. There are many risks and rewards directly associated with volatility. Hence
forecasting volatility becomes most dispensable topic in finance. The GARCH distributions play an important
role in the risk measurement and option pricing. The min motive of this paper is to measure the performance
of GARCH techniques for forecasting volatility by using different distribution model. We have used 9
variations in distribution models that are used to forecast the volatility of a stock entity. The different GARCH
distribution models observed in this paper are Std, Norm, SNorm, GED, SSTD, SGED, NIG, GHYP and JSU.
Volatility is forecasted for 10 days in advance and values are compared with the actual values to find out the
best distribution model for volatility forecast. From the results obtain it has been observed that GARCH with
GED distribution models has outperformed all models.
Keywords
Volatility, Forecasts, GARCH, Distribution models, Stock market
1. INTRODUCTION
Volatility plays a key role in finance it is responsible for option pricing and risk management.
Volatility is directly associated with risks and returns, higher the volatility the more financial
market is unstable. It may result in both High profits or huge loses if volatility is changing at higher
rate. Volatility directly or indirectly controls asset return series, equity prices and foreign exchange
rates. If the pattern of volatility clusters is studied for longer duration we observe that, once if
volatility reaches its highest point then it will continue for a longer duration. These are readily
recognized by Generalized Autoregressive Conditional Heteroscedasticity (GARCH) model
introduced by Bollerslev [1986]. The volatility models identify and track the volatility clusters that
are reaching either higher peaks or lower peaks by modeling the volatility clusters. In every period,
the arrival of cluster is demonstrated as another advancement term with fluctuation scaled up by the
data of profits and volatilities in the past periods. While considering the volatility dynamic with
standout lagged period, the GARCH (1, 1) model has turned into a workhorse in both scholarly and
practice because of its effortlessness and instinctive understanding.
DOI: 10.14810/ijscmc.2016.5301

2
While applying GARCH models in monetary danger administration, the conveyance of GARCH
developments assumes a critical part. From the meaning of GARCH model, it is clear that the
restrictive circulation of future returns has the same shape as the appropriation of the advancements.
Subsequently, an unseemly model on the appropriation of advancements might prompt either
underestimation or overestimation of future dangers. Furthermore, diverse appropriations of
GARCH advancements might likewise prompt distinctive choice estimating results. This paper
looks at a current analysis structure on the dispersion of GARCH advancements, what's more,
exhibits its downside when applying to money related time arrangement. Further, we add to an
option technique, especially towards applications to money related time arrangement.
The recent work carried in the field of finance using Garch techniques are discussed in this section.
Francesco Audrino [2016][1] discuss about Volatility Forecasting on SP 500 data set considering
Downside Risk, Jumps and Leverage Effect. The paper forecast the leverage effect separated into
continuous and discontinuous effects, and past volatility are separated into good and bad leverages.
Momtchil Dojarliev [2014][2] researched on the volatility and value risk evaluation for MSCI North
American Index, the paper compares techniques such as Naïve, GARCH, AGARCH and BEKK
model in forecasting volatility. Out of all the techniques Naïve has the highest failure rate and
BEKK model has highest successful rate. Karunanithy Banumathy [2015][3] in their research work
modeling in Stock Market volatility using GARCH, Akaike Information Criterion (AIC) and
Schwarz Information Criterion (SIC), the study proves that GARCH and TGARCH estimations are
found to be most appropriate model to capture the symmetric and asymmetric volatility
respectively.
Amadeus Wennström [2014][4] research on volatility forecasting and their performance of 6
generally used forecasting models; the simple moving average, the exponentially weighted moving
average, the ARCH model, the GARCH model, the EGARCH model and the GJR-GARCH model.
The dataset used in this report are three different Nordic equity indices, OMXS30, OMXH25 and
OMXC20. The result of this research work suggests that EGARCH has better MSE (Mean Square
Error) rates compared to other techniques. Yiannis Dendramis [2012] [5] measure performance of
option parametric instability models, as EGARCH or GARCH models, can be extensively enhanced
in the event that they are joined with skewed conveyances of return innovations. The execution of
these models is observed to be like that of the EVT (compelling esteem hypothesis) methodology
and it is superior to that of their expansions taking into account Markov administration exchanging
effects with or without EGARCH effects. The paper …recommends that the execution of the last
approach can be additionally significantly enhanced on the off chance that it depends on …altered
residuals got through instability models which take into account skewed appropriations of return
developments.
BACK GROUND
Amid the most recent couple of decades have seen a huge number of various recommendations for
how to show the second momentum, are referred as Volatility. Among the models that have
demonstrated the best are the auto-regressive heteroskedasticity (Arch) group of models presented
by Engle (1982) and the models of stochastic change (SV) spearheaded by Taylor (1986). During
the last couple of years ARFIMA sort demonstrating of high-recurrence squared returns has
demonstrated exceptionally productive [6]. Forecasting the unpredictability of profits is key for
some ranges of finance, it is understood that financial return arrangement show numerous non-
ordinary qualities that cannot be caught by the standard GARCH model with a typical blunder
dissemination. In any case, which GARCH model and which error appropriation to utilize is still
open to address, particularly where the model that best fits the in-test information may not give the

3
best out-of-test instability gauging capacity which we use as the foundation for the determination of
the most effective model from among the choices. In this study, six mimicked examines in GARCH
(p,q) with six distinctive mistake circulations are completed[7].The Generalized Autoregressive
Conditional Heteroskedasticity (GARCH) model, intended to display instability bunching, displays
overwhelming tailedness paying little heed to the appropriation of on its development term. While
applying the model to money related time arrangement, the conveyance of advancements plays an
imperative part for risk estimation and option valuing [8].
Financial related returns arrangements are essentially described by having a zero mean, showing
high kurtosis and little, if any, connection. The squares of these profits frequently present high
connection and perseverance, which makes ARCH-sort models suitable for evaluating the
restrictive unpredictability of such procedures; see Engle (1982) for the original work, Bollerslev et
al (1994) for a review on instability models and Engle and Patton (2001) for a few expansions. The
ARCH parameters are typically assessed utilizing most extreme probability (ML) techniques that
are ideal when the information is drawn from a Gaussian circulation [9].
This paper looks at the anticipating execution of four GARCH (1, 1) models (GARCH, EGARCH,
GJR and APARCH) utilized with three dispersions (Normal, Student-t and Skewed Student-t). They
investigate and look at changed conceivable wellsprings of conjectures upgrades: asymmetry in the
contingent difference, fat-followed conveyances and skewed appropriations. Two noteworthy
European stock records (FTSE 100 and DAX 30) are considered utilizing day by day information
over a 15-years time span. Our outcomes propose that enhancements of the general estimation are
accomplished when topsy-turvy GARCH are utilized and when fat-followed densities are checked
in the contingent change. Also, it is found that GJR and APARCH give preferred figures over
symmetric GARCH. At long last expanded execution of the estimates is not obviously watched
when utilizing non-ordinary conveyances [10].
This paper breaks down the system, results and exactness of GARCH (1, 1) models used with three
appropriations (Normal, Student-t and Skewed Student-t). They examine and contrast different
conveyances with get high determining precision through rolling out improvements in asymmetry
restrictive change, skew and fat followed circulations. Two vital European stock records (FTSE 100
and DAX 30) are examined using each day data over a 15-years time span. Our results suggest that
improvements of the general estimation are expert when hilter kilter GARCH are used and when
fat-took after densities are considered in the prohibitive change. Likewise, it is found that GJR and
APARCH give favored guesses over symmetric GARCH. Finally extended execution of the gages
is not clearly watched while using non-common dispersals [11].
This paper thinks about 330 ARCH-sort models as far as their capacity to portray the restrictive
difference [12]. The models are looked at out-of-test utilizing DM–$ swapping scale information
and IBM return information, where the last depends on another information set of acknowledged
change. We discover no confirmation that a GARCH (1, 1) is beated by more refined models in our
investigation of trade rates, while the GARCH (1, 1) is obviously second rate compared to models
that can oblige an influence impact in our examination of IBM returns. The models are contrasted
and the test for unrivaled prescient capacity (SPA) and the rude awakening for information
snooping (RC). Our observational results demonstrate that the RC needs energy to a degree that
makes it not able to recognize "great" and "awful" models in their investigation [12].
2. VOLATILITY
In finance, Volatility is defines level of variety of an exchanging trade prices after some time as
measured by the standard deviation of profits. Historic volatility is gotten from time arrangement of
past business sector costs. A suggested unpredictability is gotten from the business sector cost of a
business sector exchanged subsidiary (specifically a choice). The image σ is utilized for

4
unpredictability, and compares to standard deviation, which ought not be mistaken for the
comparatively named difference, which is rather the square, σ2
.The three principle purposes of
estimating Volatility are for knowing risk, allocation of assets, make profit with financial trading. A
huge piece of volatility forecasting is measuring the potential future misfortunes of an arrangement
of advantages, and keeping in mind the end goal to gauge these potential misfortunes, gauges must
be made of future volatilities and relationships. In asset management, the Markowitz methodology
of minimizing risk for a given level of expected returns has turned into a standard methodology, and
obviously an assessment of the fluctuation covariance network is required to measure volatility.
Maybe the most difficult use of forecasting volatility is to utilize risk factors for building up a risk
oriented return model.
3. METHODOLOGY
The methodology can be split into 4 steps such as Data Acquisition, Data Preprocessing, Estimation
of Volatility, Forecasting using GARCH Techniques and Result Comparison. Data acquisition is the
first step, the stock market closing. The detailed explanation of steps is explained below.
Data Acquisition – this paper uses 10 years of stock market data for volatility forecasting. We have
selected the SP 500 index as the input dataset. SP 500 end of the day stock data is downloaded from
Yahoo Finance. This paper uses 10 years of historical data ranging from 5th
December 2005 to 4th
December 2015. 10 years of data set resulted in 2514 samples of data set. One row of data is
generated per day except on Saturday and Sunday as their will be no transactions on weekends.
Fig.1 Closing Prices of SP 500 Index for a period of 10 years
3.1 Data Preprocessing
The downloaded data consists of 6 columns such as Date, Open, High, Low, Close, Volume and
Adjusted Closing prices. The data when downloaded is in the recent date first order, the data is
arranged to contain recent date at the last to predict the volatility values for net 10 dates. The data is
checked for missing values or NA values, such data will be either replaced with mean or median or
deleted.
4.2 Estimating Volatility
The volatility is estimated from open, high, low and close values of stock data; generally volatility
is calculated as the standard deviation or returns of stock data. The volatility is calculated by
considering every 10 days as interval of stock data. Close Method is the most commonly used

5
volatility calculation technique. This method works on closing prices of the stock data. The plot of
estimated volatility for SP 500 using Close price is as shown in Fig.1
VolatilityClose= cc = ∑ ……………………………………………….…….. (1)
Forecasting of Volatility - The volatility estimated from close technique is forecasted using Garch
technique. In this paper we apply Garch with different distribution models in order to forecast
accurately. The volatility is forecasted for 10 days in advance.
4.3 Result Comparison
The results of Garch technique with different models are compared with error measuring parameter
MSE. The distribution model with lowest MSE value is considered as the most accurate distribution
model compared to other models. The main aim of this research is to find out a distribution model
with lowest error.
Figure 1. Volatility of SP 500 Index for a period of 10 years
4. GARCH TECHNIQUES
Generally GARCH is referred as Generalized Autoregressive Conditional Heteroskedasticity model,
intended for volatility is clustering and displays heavy-tailedness depending upon the selection of
innovation term. While applying Garch model to forecast volatility, the distribution of innovation
terms plays a critical part for risk estimation and option pricing. GARCH models have been created
to clarify the unpredictability grouping. In the GARCH model, the development (or remaining)
conveyances are thought to be a standard typical dispersion, regardless of the way that this
presumption is frequently dismisses experimentally. Consequently, GARCH models with non-
ordinary advancement dissemination have been produced. In this research Garch techniques on

6
applying different innovation models to conclude about a better forecasting model. Financial
models with long tailed distributions and volatility clustering have been acquainted to overcome
issues with the authenticity of traditional Garch models. These traditional models of financial time
series lack the explanation of homoskedasticity, skewness, substantial tails, and instability grouping
of empirical asset returns.
In GARCH models, the probability density function is written in terms of the scale and location
parameters, standardized to have mean zero and variance equal to one.
αt = (µt, σt, ω) ……………………………………………………..………………………………….…….(2)
Where the conditional mean is given by
µt = µ(θ, xt) = E(yt|xt)……………………………………………………………………………....(3)
and the conditional variance is,
σ2
t = σ2
(θ, xt) = E((yt − µt)2
|xt) ……………………………………………………………………(4)
with ω = ω(θ, xt) denotes the others parameters of the distribution, perhaps a shape and skew
parameter. The conditional mean and variance are used to scale the innovations,
zt(θ) = yt − µ(θ, xt)σ/(θ, xt) ………………………………………………………..……………..(5)
Having conditional density which may be written as,
g(z|ω) = d/dzP(zt<z|ω) ……………………………………………………………………………..(6)
5.1 Student distribution
The GARCH-Student model was initially utilized portrayed as a part of Bollerslev (1987) as a
distinct option for the Normal appropriation for fitting the institutionalized developments. It is
depicted totally by a shape parameter ν, yet for institutionalization we continue by utilizing its 3
parameter representation as takes after
f(x) =
√
……………………………………………………………………………(7)
Where α, β, and ν are the area, scale and shape parameters separately and Γ is the Gamma capacity.
Like the GED dispersion depicted later, this is a unimodal and symmetric dispersion where the area
parameter α is the mean (and mode) of the dissemination while the change is
Var(x) = ………………………………………………………………………………………………..(8)
For the purposes of standardization we require that:
Var(x) = = 1 …………………………………………………………………………………..………...(9)

7
That implies
…………………………………………………………………………………………………….(10)
Substituting β into f(x) we obtain the standardized Student's distribution
f ) = f(z) =
√
(1+ ) …………………………………………………………….....(11)
5.2 Normal distribution
The Normal Distribution is a circular appropriation portrayed totally by it initial two minutes, the
mean and change. Formally, the arbitrary variable x is said to be typically disseminated with mean
µ and change σ2 with thickness given by
f(x) = ……………………………………………………………………………………………...(12)
Taking after a mean filtration or brightening process, the residuals ε, institutionalized by σ yield the
standard typical thickness given by
f( ) = f(z) = ( ) ………………………………………………………………………….……….(13)
To get the restrictive probability of the GARCH process at every point in time, the contingent
standard deviation σt from the GARCH movement progress, goes about as a scaling component on
the thickness, so that
LLt(zt; t) = f(zt) ……………………………………………………………………………………..…..(14)
Which outlines the significance of the scaling property. At last, the ordinary conveyance
has zero skewness and zero overabundance kurtosis.
5.3 Skew Normal distribution
In probability theory and statistics, the skew normal distribution is a continuous probability
distribution that generalizes the normal distribution to allow for non-zero skewness.
Let (x) denote the standard normal probability density function
(x) = ……………………………………………………………………………………………..(15)
With the cumulative distribution function given by
(x) = ∫ = ……………………………………………………………………….(16)
A stochastic procedure that supports the conveyance was depicted by Andel, Netuka and Zvara
(1984).[1] Both the dispersion and its stochastic procedure underpinnings were results of the
symmetry contention created in Chan and Tong (1986), which applies to multivariate cases past

8
ordinariness, e.g. skew multivariate t dissemination and others. The dispersion is a specific instance
of a general class of circulations with likelihood thickness elements of the structure f(x)=2 φ(x)
Φ(x) where φ() is any PDF symmetric around zero and Φ() is any CDF whose PDF is symmetric
around zero
f(x)=2 φ(x) Φ(x) .............................................................................................................................................(17)
5.4 Generalized Error distribution
The Generalized Error Distribution (GED) is a 3 parameter distribution belonging to the exponential
family with conditional density given by,
f(x) = ……………………………………………………………………………………....(18)
With, and speaking to the area, scale and shape parameters. Since the conveyance is symmetric and
unimodal the area parameter is likewise the mode, middle and mean of the conveyance. By
symmetry, every odd minute past the mean are zero. The fluctuation what's more, kurtosis are given
by
Var(x) = …………………………………………………………………………………….(19)
Ku(x) = …………………………………………………………………………………….(20)
As abatements the thickness gets atter and atter while in the farthest point as! 1, the
dispersion tends towards the uniform. Uncommon cases are the Normal when = 2, the
Laplace when = 1. Institutionalization is straightforward and includes rescaling the
thickness to have unit standard deviation
Var(x) = = 1…………………………………………………………………………………. (20)
That implies √ …………………………………..…………………………………………. (21)
5.5 Skewed Distributions
Fernandez and Steel (1998) proposed introducing skewness into unimodal and symmetric
distributions by introducing inverse scale factors in the positive and negative real half lines. Given a
skew parameter the density of a random variable z can be represented as:
f (z|ξ) = [f(ξz) H(-z) + f( ………………………………………………………….….….(22)
Where ξ, H (.) is Heaviside function. The absolute moments, requirements for deriving central
moments are as follows
Mr= 2 ∫ …………………………………………………………………………...…….……. (23)
The mean and variance are defined as

9
E (z) = M1 (ξ - ) ……………………………………………………………………………...…………. (24)
Var(z) = (M2 – M1
2
)( 1
2
– M2 …………………………………………………………….(25)
5.6 Skew Student distribution
The Normal, Student and GED distributions have skew variants which have been standardized to
zero mean, unit variance by making use of the moment conditions given above.
5.7 Normal Inverse Gaussian distribution
The normal-inverse Gaussian distribution (NIG) is a continuous probability distribution that is
defined as the normal variance-mean mixture where the mixing density is the inverse Gaussian
distribution. The NIG distribution was noted by Blaesild in 1977 as a subclass of the generalized
hyperbolic distribution discovered by Ole Barndorff-Nielsen, in the next year Barndorff-Nielsen
published the NIG in another paper. It was introduced in the mathematical finance literature in
1997.
The Inverse Gaussian distribution are controlled by the location, and
, their relation is refereed as
………………………………………………………………………………...…………….(26)
The probability density function is given by
(
√
√
) …………………………………………….............……………………….(27)
Where Ki denotes Bessel function of third kind
5.8 Generalized Hyperbolic distribution
The General Hyperbolic distribution was popularized by Aas and Ha (2006) because of its
uniqueness in the GH family in having g one tail with polynomial and one with exponential
behavior. This distribution is a limiting case of the GH when | and where v is the
shape parameter of the Student distribution. The domain of variation of the parameters is R and
v> 0, but for the variance to be infinite v> 4, while for the existence of skewness and kurtosis, v > 6
and v> 8 respectively. The density of the random variable x is then given by:
f(x) =
( )
√ ( )
√
…………………………….…………….……………..(28)
5.9 Johnson’s reparametrized SU distribution
The reparametrized Johnson SU distribution, discussed in Rigby and Stasinopoulos (2005), is a four
parameter distribution denoted by JSU ( ), with mean and standard deviation for all values
of the skew and shape parameters and respectively.

10
The probability density function is given by
√ ( )
……………………………………………………….……………..(29)
6. Results
This research mainly focused on exploring Garch techniques with different distribution models.
Garch techniques were tested with 9 different distribution models on the same data set to forecast
10 days in advance. The future 10 forecasted values are tabulated in table and compared with rest of
the other models. The Actual volatility values are calculated from the original stock data by
considering 10 days more of SP 500 closing prices than used for forecasting with Garch techniques.
The actual values of volatility and forecasted volatility with different Garch distribution models are
tabulated in the table. The accuracy of the forecasting techniques is measured using Mean Square
Error (MSE) parameter. The MSE of the forecasting models are tabulated in the table. The Garch
distribution model with lowest MSE value for all the 10 forecasted value is considered as accurate
forecast model. From the table it is evident that Garch technique with GED distribution is having
lowest MSE values. Hence Garch technique with GED model can be considered as accurate
technique compared to other models.
Table 1. Actual and Forecasted volatility values for 10 days
Day
Forecast
Actual
Volatility
Std Norm SNorm GED SSTD SGED NIG GHYP JSU
T+1 0.1146 0.1140 0.1163 0.1165 0.1129 0.1142 0.1131 0.1136 0.1137 0.1137
T+2 0.1143 0.1149 0.1181 0.1185 0.1131 0.1152 0.1135 0.1143 0.1145 0.1144
T+3 0.1023 0.1157 0.1198 0.1204 0.1132 0.1162 0.1138 0.1149 0.1152 0.1151
T+4 0.1061 0.1165 0.1214 0.1221 0.1134 0.1172 0.1142 0.1155 0.1159 0.1158
T+5 0.1063 0.1173 0.1229 0.1239 0.1136 0.1182 0.1146 0.1161 0.1166 0.1165
T+6 0.0897 0.1181 0.1244 0.1255 0.1137 0.1191 0.1149 0.1167 0.1173 0.1172
T+7 0.0841 0.1189 0.1257 0.1270 0.1139 0.1201 0.1153 0.1173 0.1179 0.1178
T+8 0.0793 0.1196 0.1270 0.1285 0.1141 0.1210 0.1157 0.1179 0.1186 0.1185
T+9 0.0611 0.1204 0.1283 0.1299 0.1142 0.1218 0.1160 0.1185 0.1193 0.1191
T+10 0.0649 0.1211 0.1295 0.1312 0.1144 0.1227 0.1164 0.1190 0.1199 0.1197

11
Table 2.Mean square error of forecasted values
Error
rates
Std Norm SNorm GED SSTD SGED NIG GHYP JSU
T+1 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000
T+2 0.00000 0.00001 0.00002 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000
T+3 0.00018 0.00031 0.00033 0.00012 0.00020 0.00013 0.00016 0.00017 0.00017
T+4 0.00011 0.00023 0.00026 0.00005 0.00012 0.00007 0.00009 0.00010 0.00009
T+5 0.00012 0.00028 0.00031 0.00005 0.00014 0.00007 0.00010 0.00011 0.00010
T+6 0.00081 0.00120 0.00128 0.00058 0.00087 0.00064 0.00073 0.00076 0.00076
T+7 0.00121 0.00173 0.00184 0.00089 0.00129 0.00097 0.00110 0.00114 0.00114
T+8 0.00163 0.00228 0.00242 0.00121 0.00173 0.00132 0.00149 0.00154 0.00153
T+9 0.00351 0.00452 0.00474 0.00282 0.00369 0.00302 0.00329 0.00338 0.00337
T+10 0.00316 0.00418 0.00440 0.00245 0.00334 0.00265 0.00293 0.00303 0.00301
6.1 Results Comparison
There has been a tremendous research in GARCH models in volatility forecasting, all these
forecasting researches are based on usage of different GARCH techniques such as EGARCH,
TGARCH and GARCH with maximum likelihood estimation. According to Ghulam Ali [2013]
[16] results on Garch model suggests that with wider tail distribution, the TGARCH model is
reasonable for explaining the data. GARCH with GED distributions have comparative advantage
over GARCH with normal distribution. The research paper of Yan Goa [2012] [17] compares
various GARCH techniques and the observation according to that paper are as follows: GED-
GARCH model is better than t-GARCH, and t-GARCH is better than N-GARCH. Yiannis
Dendramis [5] the GARCH model performance in stocks suggests that the skewed t-student and
GED distributions constitute excellent tools in modeling distribution features of asset returns.
According to Abu Hassan [18] [2009] suggest among GARCH with norm and s-norm and t-norm
distributions t norms outperforms the other distribution models.
The results in this paper also suggests that the GARCH model with GED distributions have minimal
mean square errors and it outperforms other GARCH distribution models.
7. Summary
The aim of this research work is to forecast volatility with high accuracy using different
distributions of Garch techniques. This paper uses SP 500 indices stock market end of the day data
for a period of 10 years for volatility forecasting. The volatility was calculated using standard
deviation of returns over period of time. The volatility was given as input for Garch techniques with
different distribution parameter. The research work uses 9 Garch distribution models that are used
to forecast the volatility. The different GARCH distribution models used in this paper are Std,
Norm, SNorm, GED, SSTD, SGED, NIG, GHYP and JSU. Future values of Volatility are
forecasted for 10 days in advance and values are compared with the actual values to find out the
best distribution model. Based on the results obtained it has been observed that GARCH with GED
distribution models predicts volatility with least error compared to other models. The future work of
this research can be the application of Hybrid distribution models to forecast volatility. The Hybrid
distribution models may involve combining two or more techniques to improve the results.

12
Table. 1 contains the actual volatility values for the next 10 days recoded after the values are
obtained. The Table.2 contains the MSE rate of the forecasting techniques and the actual values, the
result suggest GARCH with GED distribution models predicts volatility more accuracy compared to
other techniques. This technique is highlighted in bold as shown in the tables.
REFERENCES
[1] Francesco Audrino and Yujia Hu, “Volatility Forecasting: Downside Risk, Jumps and Leverage
Effect” , Econometrics Journal, Vol 4, Issue 8, 23 February, 2016.
[2] Momtchil Dojarliev and Wolfgang Polasek, “ Volatility Forecasts and Value at Risk Evaluation for
the MSCI North American Index”, CMUP, 2014.
[3] Karunanithy Banumathy and Ramachandran Azhagaiah, “Modelling Stock Market Volatility:
Evidence from India”, Managing Global Transitions, Vol 13, Issue 1, pp. 27–42, 2015.
[4] Amadeus Wennström, “Volatility Forecasting Performance: Evaluation of GARCH type volatility
models on Nordic equity indices”, June 11 2014.
[5] Yiannis Dendramis, Giles E Spunginy and Elias Tzavalis, “Forecasting VaR models under di¤erent
volatility processes and distributions of return innovations”, November 2012.
[6] ChaiwatKosapattarapim, Yan-Xia LinandMichael McCrae, “Evaluating the volatility forecasting
performance of best fitting GARCH models in emerging Asian stock markets” Centre for Statistical
& Survey Methodology,Faculty of Engineering and Information Sciences, 2011.
[7] Anders Wilhelmsson, “Garch Forecasting Performance under Different Distribution Assumptions”,
Journal of Forecasting, Vol 25, pp. 561–578, 2006.
[8] PengfeiSun and Chen Zhou, “Diagnosing the Distribution of GARCH Innovations”, Online Journal,
07 May 2013.
[9] Fernando Perez-Cruz, Julio A Afonso-Rodriguez and Javier Giner, “Estimating GARCH models
using support vector machines”,Institute of Physics Publishing, Quantitative Finance Volume 3
(2003) 1–10.
[10] Jean-Philippe Peters, “Estimating and forecasting volatility of stock indices using asymmetric
GARCH models and (Skewed) Student-t densities”, Université de Liège, Boulevard du Rectorat,7,
B-4000 Liège, Belgium.
[11] TimoTeräsvirta,“ An Introduction to Univariate GARCH Models”, Economics and Finance, No. 646,
December 7, 2006
[12] Peter R. Hansen and AsgerLunde, “A Forecast Comparison Of Volatility Models: Does Anything
Beat A Garch(1,1)?”, Journal Of Applied Econometrics, Vol 20, pp 873-889, 2005.
[13] Robert Engle, “GARCH 101: The Use of ARCH/GARCH Models in Applied Econometrics”, journal
of Economic Perspectives, Vol 15, No 4, pp, 157, 2001.
[14] Luc Bauwens,ASebastien Laurent and Jeroen V. K. Romboutsa, “Multivariate Garch Models: A
Survey” Journal Of Applied Econometrics, Vol 21, pp. 79–109, 2006.

13
[15] AricLaBarr, “Volatility Estimation through ARCH/GARCH Modeling” Institute for Advanced
Analytics North Carolina State University, 2014.
[11] David Ardiaa, LennartHoogerheideb, “GARCH Models for Daily Stock Returns: Impact of
Estimation Frequency on Value-at-Risk and Expected Shortfall Forecasts”, Tinbergen Institute
Discussion Paper, Vol 3, pp. 47, 2013.
[12] Jin-ChuanDuan, “The Garch Option Pricing Model”, Mathematical Finance, Vol 5, No 1, pp. 13-32,
1992.
[13] Da Huang, Hansheng Wang And Qiwei Yao, “Estimating GARCH models: when to use
what?”,Econometrics Journal, Vol 11, pp. 27–38, 2008.
[14] Ivo Jánský and Milan Rippel, “Value at Risk forecasting with the ARMA-GARCH family of models
in times of increased volatility”, Institute of Economic Studies, Faculty of Social Sciences Charles
University in Prague, 2007.
[15] Christian Schittenkopf, Georg Dorffner and Engelbert J. Dockner, “Forecasting Time-dependent
Conditional Densities: A Semi-nonparametric Neural Network Approach, Journal of Forecasting,
Vol 8, pp. 244-263, 1996.
[16] Ghulam Ali, “EGARCH, GJR-GARCH, TGARCH, AVGARCH, NGARCH, IGARCH and
APARCH Models for Pathogens at Marine Recreational Sites”, Journal of Statistical and
Econometric Methods, vol. 2, no.3, 2013, 57-73.
[17] Yan Gao, Chengjun Zhang and Liyan Zhang, “Comparison of GARCH Models based on Different
Distributions, Journal Of Computers, Vol. 7, NO. 8, August 2012, pp 1967-73.
[18] Abu Hassan Shaari Mohd Nor, Ahmad Shamiri & Zaidi Isa, “Comparing the Accuracy of Density
Forecasts from Competing GARCH Models”, Sains Malaysiana 38(1)(2009): 109–118.

VOLATILITY FORECASTING - A PERFORMANCE MEASURE OF GARCH TECHNIQUES WITH DIFFERENT DISTRIBUTION MODELS

More Related Content

Similar to VOLATILITY FORECASTING - A PERFORMANCE MEASURE OF GARCH TECHNIQUES WITH DIFFERENT DISTRIBUTION MODELS (20)

Recently uploaded (20)

VOLATILITY FORECASTING - A PERFORMANCE MEASURE OF GARCH TECHNIQUES WITH DIFFERENT DISTRIBUTION MODELS