Mathematical Theory and Modeling www.iiste.

org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

Characterization of Student’s T- Distribution with some Application

to Finance
Sandya Nilmini Kumari*, Abby Tan
Department of Mathematics, Universiti Brunei Darussalam, BE 1410, Negara Brunei Darussalam
* E-mail of the corresponding author: [email protected]
Abstract
The distributional form of the returns on the underlying assets plays a key role in finance under valuation theories for derivative
securities. Among them, Student t-distributions are generally applied in financial studies as heavy-tailed substitute to the normal
distribution. Therefore, distributions of logarithmic asset returns can often be fitted extremely well using Student t-distribution
with υ degree of freedom, such that 3 ≤ υ ≤ 5. The aim of this paper is to investigate the characterization behavior of Student
t-distributions and its related properties into finance which are based on computational aspects using Mathematica. Furthermore,
convolution, infinity divisibility and self-decomposability properties of Lévy-Student process are considered as background to the
option pricing. Finally, applications of modeling high frequency price returns are discussed.
Keywords: Characterization behavior, degree of freedom, heavy-tail, Lévy-Student process

1. Introduction
The distributional form of the returns on the underlying assets plays a key role in finance under valuation theories for derivative
securities. Among them, Stochastic processes with heavy-tailed marginal distributions, including Student’s t-distribution, are used
commonly for modeling in logarithmic stock returns and stochastic volatility in finance, econometrics, insurance, turbulence,
communication networks, etc (Aas & Haff 2006; Heyde & Leonenko 2005; Kumari & Tan 2013). Hence, it is frequently used to
model the asset returns for which the tails of the normal distribution are almost invariably found to be thin.
This distribution is increasing important in classical as well as in Bayesian modeling in Statistics. Both Normal and T-distribution
are members of the general family of symmetric distributions. Due to its central importance in statistical inference, the Student-t
offers a more viable alternative particularly when it comes to real market data. The applications of Student-t can be classified into
the fields of empirical modeling, cluster analysis, discriminant analysis, multiple regression, risk and dependence modeling, Lévy
processes & derivatives, portfolio selection and many more. Therefore in this paper, we concentrate on characterization of Student
t-distribution and its related properties into finance.
2. Characterization of Student T-Distribution
Considering the sample of independent observation X1 … … … Xn from the normal population with mean μ and variance σ2 for
testing the null hypothesis H0 ∶ μ = μ0 against the alternative H1 ∶ μ ≠ μ0 , Gosset (Student 1908) suggested the ‘t’ test
statistic
(x̅n −μ0 )
tn = sn , n ≥ 2, (01)
⁄
√n
1 1
where ̅
Xn = ∑ni=1 Xi and sn2 = ∑ni=1(Xi − ̅
X n )2 .
n n−1
He derived that the distribution law ℒ(t n ) = T1 (n − 1, 1, 0) , where T1 (υ, σ, μ) denotes the univariate Student's
t-distribution(ST) with with υ > 0 degree of freedom/ tail parameter, a scaling parameter σ2 > 0 and a location parameter
μ ϵ ℝ1 . The t-test and the associated theory became well-known through the work of Fisher (1927), who called the distribution
"Student's distribution".
A random variable T1 has a standard Student's t distribution (ST) with υ degrees of freedom and it can be written as a ratio:
Y
T1 = 2
, between a standard normal random variable and the square root of a Chi-square random variable with υ degrees
√χυ⁄
υ

of freedom (dividing by υ a Chi-square random variable with degrees of freedom). This is similar to obtaining a Gamma
random variable with parameters υ⁄2 and 2. Then its Probability Density function (PDF) can be defined as;

1
 Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

Γ(
υ+1
) 1 −(υ+1⁄2)
fυ (x) = 2
υ [1 + x 2 ] , t ϵ ℝ1 , (02)
√π υ Γ(2 ) υ

and Γ(z) is the Euler’s gamma.

Normal distribution is recovered as υ → ∞. This reflects the fact that the estimated variance converges in probability to the true
variance when υ grows without limit. The tails of the t distribution become more pronounced for smaller values of υ. Whereas
for finite υ the tails of the density function decay as an inverse power of order υ + 1, and is therefore fat-tailed relative to the
Normal. Therefore, extreme outcomes are more likely than Normal distribution. It is implied by the Fig.1. When υ = 1, the t
distribution is identical to another classical distribution, the Cauchy distribution. Its tails are so fat that they prevent it from having
a mean (as well as any higher order moment).
A random variable (X) has a general Student's t-distribution and it can be written as a linear transformation of a standard Student's
t random variable (T1 ) as, X = μ + σ T1 . The PDF of X can be obtained using the formula for the density of a function of an
absolutely continuous variable (X = g(T) = μ + σ T1 is a strictly increasing function of T1 , since σ is strictly positive) as;

dg −1 (x) x− μ 1
fX (x) = fT1 (g −1 (x)) = fT1 ( )
dx σ σ
υ+1 −(υ+1⁄2)
Γ( ) 1 x− μ 2
fX (x) = 2
1 [1 + ( ) ] , x ϵ ℝ1 . (03)
√π σ υ Γ(2υ) υ σ

This distribution results from compounding a Gaussian distribution with mean μ and unknown precision (the reciprocal of the
variance), with a gamma distribution with parameters υ⁄2 and υ⁄2. In other words, the random variable x is assumed to have
a normal distribution with an unknown precision distributed as gamma. This is marginalized over the gamma distribution. The
reason for the usefulness of this characterization is that the gamma distribution is the conjugate prior distribution of the precision
of a Gaussian distribution. As a result, the three-parameter Student's t distribution arises naturally in many Bayesian inference
problems.) Therefore, the standard ST-distribution can be formulated using the following integration as;
υ
Γ( )2
υ 1 x− μ 2 υ υ
∞ − ( ) −
fμ,σ2,υ (x) = ∫0 υ 2 2 e 2y σ 2y y −2−1 dy . (04)
Γ( )√2π yσ
2

Fig. 2 illustrates the structural behavior of the ST(υ, σ, μ) distribution when parameters are changing. (refer the ‘Appendix A’ for
Mathematica coding).
2.1 Simulation of General ST-distribution
We use four methods to simulate, 4 different samples from ST distribution. Fig. 3 shows the results of sampling from ST
distribution with 4 degrees of freedom, using following approaches;
I. 10000 independent samples are taken from five standard normal variables 1, 2, , and then calculate the ( ) as;

1
( ) = , which has the ST distribution with 4 degrees of freedom. (Fig. 3 (a)).
( 2+ + + )⁄√

2
II. Take 10000 samples from independent standard normal variable 1 and Chi-squared variable , with 4 degrees of freedom

1
and calculate the ( ) as; ( ) = 2⁄
, which is the sample from the ST distribution with 4 degrees of freedom, (Fig. 3(b)).
√

III. Take 10000 samples from independent standard normal variable 1 and Inverse-gamma variable Γ( , ), with = , and
22

calculate sample from the ST distribution with 4 degree of freedom as; √ Γ(2,2) , (Fig. 3 (c)). ( ) = 1
IV. Bailey (1994) discovered that the T distribution could be sampled by a very elegant modification to the well known Box-Muller
method, and its polar variant, for the Normal distribution. With some modification of polar algorithm for generating
(υ) -variates can be summarized as follows;
a) Generate two uniform variates 1 and 2 from [0, 1] and replace 1 and 2 as 1 = 2 1 − 1 and 2 = 2 2 − 1.
b) Let = 12 + 22 , If > 1 return to step (a.) for resample.
( −2⁄ −1)
c) Then calculate T as; = 1√ (Fig. 3(d)).

2
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

(see ‘Appendix B’ for Mathematica cording)

2.2 Moments
The general ST distribution is symmetrical around zero whereby all odd moments vanish. The k th moment exists if and only if
k < υ;
0 k odd; 0 < 𝑘 < 𝑣
k+1 υ−k k⁄
Γ( )Γ( )υ 2
2 2
k even; 0 < 𝑘 < 𝑣
E(X k ) = 1
√π Γ(2υ) (05)
Not Define k odd; 0 < 𝑣 < 𝑘
{ Infinite k even; 0 < 𝑣 < 𝑘

In this case, the variance of the distribution is Var(T) = E(X 2 ) = υ⁄υ − 2 and it is finite for υ > 2. Due to symmetry, the
E(X ) (υ−2)
skewness of the distribution is zero and kurtosis is given by K(T) = = and it is finite for υ > .
[E(X2 )]2 υ−

Kurtosis measures the "fatness" of the tails of a distribution. Positive excess kurtosis means that distribution has fatter tails than a
normal distribution. Fat tails means the probability of big positive and negative returns realizations is higher than that suggested
by normal distribution. When calculating kurtosis, a result of +3.00 indicates the absence of kurtosis (distribution is mesokurtic).
For simplicity in its interpretation, some statisticians adjust this result to zero (i.e. kurtosis minus 3 equals zero), and then any
reading other than zero is referred to as excess kurtosis. Negative numbers indicate a platykurtic distribution while positive
numbers indicate a leptokurtic distribution. The likelihood of large gains or large losses on an investment is high when fatness of
the tails is higher. Excess kurtosis indicates that the volatility of the investment is itself highly volatile.
The k th moment of the standard ST distribution is given in equation (06) and it can be illustrated by using the following Fig. 4.
1 1 μ2
(μ2 + νσ2 ) 2F1 [−n, (−2n + ν), − , − 2 ] +
2 2 νσ
[ 2 ]
( nμ 2 − ν(μ2 + σ2 )) F [−n, 1 (−2n + ν), 1 , − μ ])
−n νn σ2n (1 + 2n)! Γ (
−2n + ν 2 1 2 2 νσ2
) ν k odd; 0 < k < v
2 (1 + 2n)μ(1 + 2n − ν)Γ(1 + n)Γ( )
k
E(X ) = 2
2
−n νn σ2n (2n)! Γ[1 (−2n + ν)] F [−n, 1 (−2n + ν), 1 , − μ ]
2 2 1 2 2 νσ2
ν k even; 0 < k < v
Γ[1 + n]Γ[ ]
2
Not Define k odd; 0 < v < k
{ Infinite k even; 0 < v < k

(06)
(where, 2F1 is a Hypergeometric function)
2.3 Characteristic Function
The characteristic function (CF) ψ(u), of the random variable X ~ ST(μ, σ2 , υ) is given by;
v Kv(√vσ|u|) v

ψ(u) = eiuμ 21−2 2

v (√vσ|u|)2 ; u ∈ℝ (07)
Γ( )
2

(where K v (… ) is the modified Bessel function of the second kind) and it can be derived as follows with Lemma 01and 02.
2

ψ(u) = E[eiux ] = E[eiu( μ+ δϵ) ] = eiuμ E[eiu(δϵ) ]1

1 D
Lemma 01: The random variable X~ ST(μ, σ, ν) has the representation: X ⇒ μ + δϵ.

1 1
Where; the independent random variables ϵ and δ 2 are standard normal N(0,1) and inverse gamma distribution RΓ( υ , σ2 ) respectively and
2 2

μ is a constant. The probability density function of fδ2 is given by:

3
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

Let, X = δϵ
∞ ∞ ∞
1 x
ψ(u) = eiuμ ∫ eiux fx (x) = eiuμ ∫ eiux ∫ fδ (t)fϵ ( ) dt dx2
−∞ −∞ −∞ |t| t

∞ ∞ 1 1 −(x⁄t)2 ⁄2 ∞ ∞ 1 x
= eiuμ ∫−∞ [∫−∞ eiux |t| e dx]] fδ (t)dt = eiuμ ∫−∞ [∫−∞ eiux |t| fϵ ( ) dx]] fδ (t)dt
√2π t

∞ ∞ 1 1−(x⁄t)2⁄ ∞
2
−(u ⁄2)t2
= eiuμ ∫−∞ [2 ∫0 Cos(ux) |t| e 2 dx]] fδ (t)dt = eiuμ ∫−∞ e fδ (t)dt
√2π

∞ 2
−(u ⁄2)t2
= eiuμ ∫ e 2tfδ2 (t 2 )dt
0
υ⁄
2 2 1
2 2 2
∞ −(u ⁄2)t2 ∞ −(u ⁄2)t̅ (2σ ) −υ⁄ −1 σ ⁄(2t̅)
Let t 2 = t̅ then = eiuμ ∫0 e fδ2 (t̅)dt̅ = eiuμ ∫0 e 1 t̅ 2 e dt̅
Γ( υ)
2

⁄2υ
1
( σ2 ) ∞ 2 2
−(u ⁄2)t̅ σ ⁄(2t̅ ) (−υ⁄ −1 )
=e iuμ 2 ∫ e e t̅ 2 dt̅
1
Γ ( υ) 0
2
Kυ⁄ (√vσ|u|) υ⁄
2
ψ(u) = eiuμ 2(1−υ⁄2) 2
υ (√vσ|u|)
Γ( )
2

3. Empirical Evidence
Some empirical results as given in below is an indication of the suitability of the Lévy student process for financial data. Daily
closing prices of Gold Future index and S&P 500 index are considered from Janaury-2002 to December-2012, which are traded in
the NYMEX, with a total of 2870 and 2769 observations respectively. In the analysis, logarithmic returns were used.
The maximum likelihood estimates of the parameters are given in table 01. For the assessment of goodness of fit, the
Anderson-Darling (AD) and Person χ2 tests are utilized with the log-likelihood estimator and results are recorded in Table 01.
The smaller the value of AD and χ2 means that closer to empirical distribution and fitted one. Obviously, the statistics for Lévy
process are smaller than the value for Brownian motion. Higher log-likelihood value gives better fit. Further, the corresponding
empirical densities, Gaussian and Student T density, for Gold Future Index and S&P 500 index are shown in Fig. 5 in separately.
Therefore, Student-t distribution is more realistic distribution in describing the financial data than Gaussian, for the historical data
investigated. As a result, we can be used to student process for asset price modeling.
4. Financial Applications
4.1 Convolution
Convolutions and related operations are found in many applications of mathematics and engineering. The concept of convolution
is needed to many derivatives in Lévy processes applications.
Corollary 1: The PDF for n-fold self-convolution can be found by; either convolution integrals:
∞ ∞
∫ ∫ ⋯ f(x1 , x2 , … )g(y1 − x1 , y2 − x2 , … ) dx1 dx2 ⋯ (08)
−∞ −∞
th
or the inverse Fourier transform of the n power of the characteristic function of the original function as,

1 ⁄2 2
( 𝜎2 ) −𝜐⁄ −1 𝜎 ⁄(2𝑥)
2
𝑓𝛿2 (𝑥) = { 1 𝑥 2 𝑒 𝑓𝑥>0
Γ( 𝜐)
2

For > 0 , f0 𝑓𝑥 ≤0
δ (x) = 2x fδ2 (x) , the density function of δ.

2
Lemma 02: Let X1 and X2 be independent stochastic random variable (absolutely continuous) with density function fX1 and fX2 . Then the
random variable X1 . X2 is absolutely continuous with density function (consider product of r. v.) as:

∞ 1 x
fX1 .X2 (x) = ∫−∞ |t| fX1 (t)fX2 ( ) dt ; x ∈ ℝ
t

4
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

1 ∞
∫ e−ixt ψ(x)n dx.
√2π −∞
(09)

In general, the Student t-distributions are not closed under convolution (Nadarajah & Dey 2005), because the sums of independent
identically distributed (iid) t variables are not ‘t-distributed’. Therefore, analytical solutions can be derived in some special cases
only. Nadarajah & Dey (2005) provided analytical solutions for the density function fυ1,υ2 of the 2 fold-convolution for any odd
integer values of υ1 and υ2 . They showed that there are no similar analytical solutions for even υ. However, numerical solution
can be found for any integer υ using the characteristic function technique.
The n-fold self-convolution of a Student’s t-distribution for odd υ (=3, 5, 7, ...) can be formulated simply and their expression are
given in below. However, n-fold convolution do not hold for even υ (=4, 6, ...) except υ = 2 case.
The PDF for a 2-fold self-convolution:

(2) 1 1
−ν 1 + 2v 1+ν 1 1 + ν 2 + ν u2
fu (u) = 22
2 √πΓ [ ]Γ[ ] 2F1 [ + ν, , ,− ]
ν 2 2 2 2 2 ν
√νΓ [2]

(2) ⁄2 u2
ν = 2; fu (u) = π 2F1 [2 , 2 , 2, − 8 ]
2

(2) 00√ (8 00+120u2 +u )

ν = 5; fu (u) = (10)
π(20+u2 )

The PDF for a 4-fold self-convolution with 𝜈 = 5:

2 + 12 56960000u + 73728000u6 +
(118 50858700000 + 152166 000000u
( ) 800√
fu (u) = ) (11)
π(80+u2 )9 308 00u + 810u10 + u12
8

The PDF for a 6-fold self-convolution with 𝜈 = 5

(6) 00√5
fu (u) = (979699 6573926 X1010 + 1012312171860 8X1010 u2 + 650898019296X1010 u
π(180 + u2 )1
+ 32 358903872X108 u6 + 131556506 X108 u8 + 31 79616X10 u10 + 11115 36X10 u12
+ 2153X106 u1 + 293 X10 u16 + 2500u18 + u20 )
(12)
The tails of the n-fold self-convolution of a Student’s t-distribution maintain the character of the original t-distribution. For large t,
the n-fold ν = 5 pdf goes as u−6 . Thus, n-day returns will maintain the fat tails of the distribution of the daily returns, which is
known to be described well by a Student’s t-distribution. This is in agreement with the analysis of Bouchaud and Potters (2003, p.
33). Fig. 6 shows the shapes of the convoluted probability density functions for different n (=1, 2, 4, 6, 10 & 15) fold self
convolution with υ = 5. The tails of the distributions show a u−6 behavior that is characteristic of a Student’s t-distribution with
υ = 5 degrees of freedom.
4.2 Infinite Divisibility & Self-decomposability
Relating student’s t distributions to the Lévy processes, the crucial role are paid the properties of infinite divisibility or
self-decomposability (Grigelionis 2013; Sato 1999). Grosswald (1976) proved that the standard student’s t-distribution of any
degree of freedom is infinitely divisible, by deriving the following formula:
∞ gυ (u)
K υ−1 (x) = xK υ (x) ∫0 du, υ ≥ −1, x > 0 (13)
x2 + u

where, g υ (u) = 2[π2 x(Jυ2 (√x) + Yυ2 (√x))]−1 , x > 0,

Jυ (x) and Yυ (x) are the Bessel functions of the first kind and second kind, respectively while Jurek (2001) proved that the

5
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

general student t-distribution is infinite divisible(ID) and self-decomposable(SD) by using Lemma 1 and Lemma 3 3. That means,
the class of all self-decomposable characteristic (or probability distributions) are infinitely divisible,

i.e., ∀(n ≥ 1) ∃ (ψn ), ∀(u ∈ ℝ) s.t. ψ(u) = [ψn (u)]n (14)

The characteristic function of every infinitely divisible distribution can be represented in a very special form, the
Lé vy–Khintchineformula:

1
ψμ (u) = eΦ(u) = exp (iub − cu2 + ∫ℝ (eiux − 1 − iux1{|x|>1} ) υ(dx)) (15)
2

where the coefficient b ∈ ℝ, c ≥ 0 and the L´evy measure υ(dx) which satisfies υ(dx) = 0 and ∫ℝ (x 2 ⋀ 1) υ(dx) < ∞ are
unique.
The Lé vy–Khintchine representation of the characteristic function of the general ST(υ, σ, μ) can be obtained from the results of
υ
Halgreen (2005) by choosing α = |β| = 0, σ > 0 and λ = − < 0 as follows;
2

ψ(u) = exp (iuμ + ∫ℝ (eiux − 1 − iux1{|x|>1} ) g(x)dx), (16)

1 ∞ e−|x|√2ydy
With g(x) = |x| ∫0
π2 y(J2υ⁄ (σ√2y)+ Y2υ⁄ (σ√2y))
2 2

4.3 Lévy- Student Processes

Student t-distribution is infinitely divisible and therefore generate a Lévy processes L = (L(t))t≥0 , such that the distribution of
L(1), has symmetric ST(υ, σ, μ) distribution (for simplicity and without loss of generality, we consider t = 1 with Student
marginal. However, as ST distributions are not closed under convolution as above and Nadarajah & Dey (2005), the marginals of
these Lévy processes are Student distributed only for one time horizon t 0. On other time horizons, the marginal are not Student
distributed. Their distribution can be derived analytically only for some special cases (Heyde & Leonenko 2005; Cufaro 2007).
Apart from these cases, they have to be derived numerically in quite an involved manner (Barndorff & Shephard 2004). For this
reason, only few research papers have covered Lévy Student processes, so far.
The law of L(t) is determined by the law of L(1), which is ID. The independent and stationary of the increments of the Lévy
process leads to the cumulant transform ϰ(u) = log(ψ(u)) is given by;
ϰL(t) (u) = t ϰL(1) (u) u ∈ ℝ. (17)
For υ > 1 and E[L(t)] = tμ, t ≥ 0, the process can be split into L(t) = tμ t ≥ 1, with + L0 (t), E[L0 (t)]
= 0. By Eq. (16),
the characteristic function of the random variable L0 (t), t ≥ 0 with the characteristic function ψ(u), of the process at time t and
ψL0(t) (u) = (ψL0(1) (u))t , u ∈ ℝ. (18)
As the characteristic function of a random variable equals its Fourier transformation up to some constant factors, the inverse
Fourier transformation reproduces the density from the characteristic function. Let φL(1) be the characteristic function of the
Student distribution, the PDF of L0 (t) is given;

∞
fL(t) (x) = ∫−∞ eiux ψL0(t) (u) du
t
v Kv (√vσ|u|) v
∞
= ∫−∞ eiux (eiuμ 21−2 2
v (√υσ|u|)2 ) du
Γ( )
2

υ⁄ )
2t(1− 2 ∞
= υ
πΓt ( )
∫0 cos(ux)eiuμt (√υσ|u|)υt⁄2 K t υ (√υσ|u|)du . (19)
2 2

When t = 1 the expression (19) can be exactly calculated and coincides with the PDF (03) of a ST(υ, σ, μ). Furthermore, the
principal drawback for not being stable is in the subsequent definition of the Lévy–Student process. In fact the CF of the Student

3
Lemma 3: Let 𝐿(𝑡)be a levy process with having the strong Markov and scaling (the process 𝐿(𝑡) has the scaling property if
𝑑
for each 0<c<1 there exists a constant h(c) such that 𝐿(𝑐𝑡) = ℎ(𝑐)𝐿(𝑡)) properties. For any independent random variable
𝑑
≥ 0, there is 𝐿(𝑡 + ) − 𝐿( ) = 𝐿(𝑡). If is SD and the ℎ(𝑐) is homeomorphism of unit interval, then 𝐿( ) is SD.

6
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

Lévy process φL0(t) (u) coincides with (06) only for only for t = 1, while for t ≠ 1 it is no more the CF of a Student law
(Petronia el al. 2006).
The integral on the right hand side of (18) can be computed numerically for specific values of υ. The results are shown in Fig. 6,
where the densities of the convolution semi-group are represented for values of t varying from 1 to 3, with the parameters of
υ = 3, μ = 0 and unit value of σ. The integral is difficult to evaluate for even υ. Therefore, it is better to see the behavior of
Student Lévy process when υ = 3 situation due to results in section 3 (value of degree of freedom in both indexes are close to 3).
5. Discussion
The Student t-distribution has strong reason to be regarded as an alternative model of first choice particularly when the benchmark
normal or Black Scholes, model is found to be inadequate. It implies an Inverse gamma distribution for the marginal distribution
of the squared volatility.
Student distributions are infinitely divisible and self-decomposable; there exist Lévy processes with Student marginal’s for a
certain point in time. The densities of the Lévy Student processes for any point in time are given by integrals (18) which can be
solved analytically only for few special cases. There are some problem solving the problem numerically due to slow convergence
and heavy tails of the distributions. Therefore, a model can be developed for a specific values of υ to price the option. This
distributional property can be exploited to identify possible dynamics of the volatility process and hence the evolution of the asset
price process can be derived

References
Aas, K & Haff, I, H 2006, ‘The generalized hyperbolic skew Student's t-distribution’, Financial Econometrics, vol. 4, no. 2. pp. 275-309.
Bailey, R W 1994, ‘Polar generation of random variates with the t-distribution’, Mathematics of Computation, vol. 62, No. 206, pp. 779-781.
Barndorff, N, Haldeen, C 1977 ‘Infinite divisibility of the Hyperbolic and generalized inverse Gaussian distributions’, Z. Wahrscheinlichkeitsth,
vol. 38, pp. 309-312.
Barndorff, N, O, E, Shephard, N 2004, Financial volatility: stochastic volatility and lévy based models, Cambridge University Press, Cambridge.
Bouchaud, J, P & Potters, M 2003, Theory of Financial Risk and Derivative Pricing, Cambridge University Press, Cambridge.
Cufaro, P, N 2007, ‘Mixtures in non-stable Lévy processes’, Journal of Physics A: Mathematical and Theoretical, vol. 40, pp. 2227-2250.
Fisher, R A 1925, ‘Applications of Student's distribution’, Metron, vol.5, pp. 90-104.
Fisher, R A 1931, ‘Introduction to table of Hh functions’. In: British Association Mathematical Tables, vol. 1, pp. 26-35.
Grigelionis, B 2013, Student’s t-distribution and related stochastic processes, Springer, London.
Grosswald, E 1976, ‘The Student t-distribution of any degree of freedom is infinity divisible’, Z. Wahrscheinlichkeitstheor.verw.Geb, vol. 36, pp.
103-109.
Heyde, C, C, Leonenko, N, N 2005, ‘Student processes’, Advances in Applied Probability, vol. 37, no. 2, pp. 342-365.
Jurek, Z, J 2001, ‘Remarks on the self-decomposability and new examples’, Demonstration Mathematica XXXIV, vol. 2, pp. 241-250.
Kumari, S, N & Tan A 2013, ‘Asset return distribution and its application to gold future index’, GSTF Journal of Mathematics, Statistics and
Operation Research, vol. 2, no. 1, pp. 72-76.
Nadarajah, S & Dey, D, K 2005, ‘Convolutions of the t-distribution’, Computers & Mathematics with Applications, vol. 49, pp. 715-721.
Petronia, N, C, Martinob, S, D, Sienab, S, D, Illuminati, F 2006, ‘Lévy–Student processes for a stochastic model of beam halos’, Nuclear
Instrument and Methods in Physics Research A, vol. 561, pp. 237-243.
Sato, K 1999, Levy Processes and Infinitely Divisible Distributions, Cambridge University Press, Cambridge.
Student, 1980, ‘The probable error of a mean’, Biometrika, vol. 6, no. 1, pp. 1-25.

Sandya Nilmini Kumari is a PhD student in Univeristi Brunei Darussalam. She completed her first degree in BSc. in Statistics at University of
Sri Jewardenapura, Sri Lanka and afterward she has obtained MSc. in Applied Statistics in University of Peradeniya, Sri Lanka. She is working as
a lecturer in Uva Wellassa University, Sri Lanka.
Abby Tan is a Senior Lecturer in Universiti Brunei Darussalam. He obtained his first degree in Mathematics with first class honours at University
of Manchester Institute of Science and Technology (UMIST). He later obtained his PhD from The University of Manchester. His research area is
mostly in stochastic volatility of option prices.

7
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

Appendix

Appendix A Appendix B

Appendix C

0.05
Probability_Density
0.4 4.77

0.5 0.821

1
0.3

30
0.2 mean : 0.05
N 0,1 variance : 1.88243
skewness : 2.08737
excess kurtosis : 6.25449

0.1

x
6 4 2 2 4 6

Fig.2. Structural behavior of the ST(υ, σ, μ) distribution with

Fig.1. Plot of the standard normal density and student
different parameter values
t-densities with 0.5,1,3 and 30 degrees of freedom

(a).Sampling distribution from (b).Sampling distribution from (c).Sampling distribution from (d).Sampling distribution
Method I Method II Method III from Method IV
Fig. 3. The results of above four methods, (which each one has ST distribution with 4 degrees of freedom) are compared with the
normal distribution

8
Mathematical Theory and Modeling www.iiste.org
ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.3, No.10, 2013

Fig.4. First 10 moments of the Standard ST Distribution

Probability_Density
Probability_Density
120
SmoothKernel
35 SmoothKernel
100
30
normal
25 normal 80

20 60
StudentTDist
StudentTDist
15
40
10

5 20

Log_Return Log_Return
0.05 0.05 0.10 0.03 0.02 0.01 0.01 0.02

(a).Gold Returns (b).S&P 500 Returns

Fig. 5. Densities Estimation for return, from Smooth Kernel, Student T and Normal Distribution

Probability_Density
Fig.4. First 10 moments of the Standard ST Distribution

n 1
0.8
n 2
0.6 n 4
n 6
0.4
n 10
n 15
0.2

u
10 5 0 5 10

Fig.6. Shapes of the convoluted PDFs for different n (=1, 2, 4, Fig. 7. Convolution Semi-group Densities
6, 10 & 15) with υ = 5

Table 1: Parameter estimates and goodness of fit-tests for the gold future index & S&P 500 index
Index Model υ μ σ AD 𝛘𝟐 Log-Likeli-hood
Gold Gaussian - 0.00063 0.0139 14.386 176.12 6642
(0.000) (0.000)
Student T 3.45 0.00044 0.0103 0.4712 47.14 6886
(0.841) (0.243)
S&P 500 Gaussian - 0.00005 0.0053 20.145 196.45 5614
(0.000) (0.000)
Student T 2.89 0.00023 0.0032 0.6532 32.15 5941
(0.521) (0.154)