The Korean Journal of Applied Statistics (2016) DOI: http://dx.doi.org/10.5351/KJAS.2016.29.4.

741
29(4), 741752

A numerical study of adjusted parameter estimation in

normal inverse Gaussian distribution

Jeongyoen Yoona Seongjoo Songa,1

a
Department of Statistics, Korea University

(Received April 11, 2016; Revised May 16, 2016; Accepted May 17, 2016)

Abstract
Numerous studies have shown that normal inverse Gaussian (NIG) distribution adequately ts the empirical
return distribution of nancial securities. The estimation of parameters can also be done relatively easily,
which makes the NIG distribution more useful in nancial markets. The maximum likelihood estimation
and the method of moments estimation are easy to implement; however, we may encounter a problem in
practice when a relationship among the moments is violated. In this paper, we investigate this problem in
the parameter estimation and try to nd a simple solution through simulations. We examine the eect of
our adjusted estimation method with real data: daily log returns of KOSPI, S&P500, FTSE and HANG
SENG. We also checked the performance of our method by computing the value at risk of daily log return
data. The results show that our method improves the stability of parameter estimation, while it retains a
comparable performance in goodness-of-t.

Keywords: normal inverse Gaussian distribution, feasible domain, parameter estimation, Value at Risk

1.

.
.
. 1973 Black Scholes -
(Black Scholes, 1973)
.
(Clark, 1973; Fielitz Smith, 1972; Mandelbrot, 1963; ).
1976 Merton (Merton, 1976), 1976
Cox Ross (Cox Ross, 1976) 1987 Hull White
(Hull White, 1987) . 1990
This research was supported by the Basic Science Research Program through the National Research
Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (NRF-2013
R1A1A3012819).
1 Corresponding author: Department of Statistics, Korea University, 145 Anam-ro, Seongbuk-Gu, Seoul

02841, Korea. E-mail: sjsong@korea.ac.kr

742 Jeongyoen Yoon, Seongjoo Song

. -
generalized
hyperbolic(GH) (Barndor-Nielsen, 1977), variance Gamma(VG) (Madan Seneta, 1990), normal
inverse Gaussian(NIG) (Barndor-Nielsen, 1997), CGMY (Carr , 2002) .
-
. Prause (1997) hyperbolic NIG
Value at Risk(VaR) hyperbolic NIG
. Geman (2002) CGMY SPX
, Eriksson (2009) NIG S&P500
.
(MLE)
(MME) , MME
. MME
. Figueroa-Lopez (2011) NIG

, Ghysels Wang (2014) NIG VG, generalized skewed t
.
, Ghysels Wang (2014)
NIG
. NIG ,
. ,
VaR NIG .

2. Normal Inverse Gaussian

Barndor-Nielsen (1997) NIG

, .

e+(x) ( )
fN IG (x) = K1 2 + (x )2 , x R, (2.1)
2 + (x )2

K1 index 1 3 Bessel, 2 2 .
, . .
NIG Inverse Gaussian(IG) ,
a > 0 b > 0 IG(a, b)
( )
a 3 1 ( 2 1 2 )
fIG (x) = exp(ab)x 2 exp a x + b x , x > 0
2 2

. a = 1, b = 2 2 IG W ,

Xt = 2 It + WIt ,

X , , , = 0 NIG ,
(2.1) .
 A numerical study of adjusted parameter estimation in normal inverse Gaussian distribution 743

Table 3.1. Summary statistics of daliy log returns for KOSPI, S&P500, FTSE and HANG SENG
KOSPI S&P500 FTSE HANG SENG
Mean 0.0003928 0.0001369 0.0000328 0.0001701
Standard deviation 0.0149 0.0126 0.0121 0.0148
Skewness 0.5187 0.1988 0.1575 0.0061
Excess kurtosis 6.1761 8.7912 6.9078 9.0775
Max 0.1128 0.1096 0.0938 0.1341
Min 0.1281 0.0947 0.0926 0.1358

NIG (maximum likelihood estimate; MLE)

(moment method estimator; MME) , MLE
. MME , , (excess kurtosis)
(2.2) .
3 , K, S, V , M .

3 S V 2 V 3
= , = 2 + 2, = , = 2 , =M . (2.2)
V (3K 5S )
2 3 + 2

Ghysels Wang (2014)

{( ) }
DN IG = K, S 2 : 3K > 5S 2

NIG (feasible domain) . (2.1) >

0, > 0, < < 3K 5S 2
. DN IG
MLE MME , 3K 5S 2
0
. 0
,
,
.

3.

3.1.

2001 1 2015 5 KOSPI, S&P500, FTSE, HANG SENG

. FTSE
HANG SENG . Table
3.1 Figure 3.1 .
.
Table 3.2 NIG .
KOSPI MLE MME
. MLE
MME .
744 Jeongyoen Yoon, Seongjoo Song

Figure 3.1. Empirical distribution of daliy log returns for KOPSI, S&P500, FTSE and HANG SENG.

Table 3.2. Estimated parameters and log likelihood for KOSPI, S&P500, FTSE and HANG SENG
Index Method Log likelihood
MLE 49.6684 6.3323 0.0109 0.0018 10274.59
KOSPI
MME 48.6216 6.0371 0.0107 0.0017 10274.34
MLE 50.1853 6.1679 0.0078 0.0011 11218.61
S&P500
MME 46.4398 1.8037 0.0074 0.0004 11211.32
MLE 55.3589 4.7324 0.0081 0.0007 11674.25
FTSE
MME 54.4665 1.8891 0.0081 0.0003 11672.15
MLE 46.4754 2.6826 0.0102 0.0007 10437.27
HANG SENG
MME 38.3717 0.0451 0.0086 0.0002 10424.48
MLE = maximum likelihood estimate, MME = moment method estimator.

Figure 3.2. Comparison of the garphs in KOSPI; the tted NIG distribution, Normal distribution and Empirical
distribution (NIG = normal inverse Gaussian, MLE = maximum likelihood estimate, MME = moment method
estimator).

Figure 3.2 KOSPI MLE MME NIG

. KOSPI NIG
. S&P500, FTSE, HANG SENG
.
Kolmogorov-Smirnov(KS) Anderson-Darling
(AD) . Table 3.3 KS, AD p-value
 A numerical study of adjusted parameter estimation in normal inverse Gaussian distribution 745

Table 3.3. Test statistics of Kolmogorov-Smirnov test and Anderson-Darling test (p-value in parentheses)
Kolmogorov-Smirnov test Anderson-Darling test
Index
MLE MME MLE MME
0.0142 0.0145 0.7253 0.6391
KOSPI
(0.6635) (0.6405) (0.7994) (0.7322)
0.0142 0.0203 0.6044 1.1262
S&P500
(0.6513) (0.2243) (0.7056) (0.1102)
0.0181 0.0226 0.6812 0.1681
FTSE
(0.3352) (0.1243) (0.7649) (0.4315)
0.0187 0.0259 1.2625 2.4292
HANG SENG
(0.3118) (0.0569) (0.2447) (0.0541)
MLE = maximum likelihood estimate, MME = moment method estimator.

. 0.05
NIG .

3.2.

2 . 2001 1 2015 5
, KOSPI 3568, S&P500 3623, FTSE 3747, HANG
SENG 3599. 250 1
250 KOSPI 3319, S&P500 3374, FTSE 3498, HANG
SENG 3350 . DN IG MLE
MME KOSPI 114, S&P500 495, FTSE 26, HANG
SENG 138 . DN IG (,
3K 5S 2 0 ), MLE MME

. Figure 3.3 KOSPI .
. S&P500, FTSE, HANG SENG
.
DN IG 3K
5S 2 ,
. 90% 3K 5S 2 1
, 0 1 . 250
KOSPI 3K 5S 2
, NIG 1,000 MLE
. 0.0006, 0.0001, 0.023, 0.116
3K 5S 2 0.00172. 1,000 1,000 MLE
. 3K 5S 2 MLE
3K 5S 2 , MME (2.2) 3K 5S 2
. , 0.2
0.2 1 0.1 (RMSE)
Figure 3.4 . Figure 3.4 0.5 0.6
746 Jeongyoen Yoon, Seongjoo Song

Figure 3.3. Empirical distributions of estimated parameters in case of KOSPI (MLE = maximum likelihood
estimate, MME = moment method estimator).

Figure 3.4. Root Mean Square Errors (RMSE) of parameters with values of from 0.2 to 1 by 0.1.

0.3 0.4 0.5

0.5 . .
-MLE( -MME) .
3K 5S 2
. 0.5 ,
0.053, 0.116 3K 5S 2 0.0917
 A numerical study of adjusted parameter estimation in normal inverse Gaussian distribution 747

Table 3.4. Means, standard deviations and coecient of variations of estimated parameters for scenario 1 (3K
5S 2 = 0.0917) and scenario 2 (3K 5S 2 = 6.0917)
Scenario 1 Scenario 2

8974.21 8465.372 0.0576 0.0189 133.57 6.564 0.013 0.00003
MLE (176159.8) (176135.7) (0.0325) (0.0357) (27.841) (11.131) (0.002) (0.001)
[1962.956] [2080.662] [56.592] [188.312] [21.593] [169.561] [17.735] [2830.9]
10118.51 9609.66 0.0575 0.0189 133.74 6.564 0.013 0.00005
MME (201098.1) (201074.2) (0.0326) (0.0357) (27.886) (11.538) (0.002) (0.001)
[1987.429] [2092.418] [56.6017] [188.309] [20.851] [173.3] [18.488] [2372.13]
418.213 64.2934 0.0396 0.0055 134.83 7.151 0.013 0.00008
-MLE (40.318) (44.987) (0.0039) (0.004) (29.394) (11.485) (0.002) (0.001)
[9.641] [69.971] [9.912] [74.078] [21.8] [160.61] [17.511] [1239.4]
418.338 64.313 0.0396 0.0055 134.91 7.164 0.013 0.00008
-MME (40.318) (44.998) (0.0039) (0.004) (28.539) (11.897) (0.002) (0.001)
[9.634] [69.968] [9.917] [74.076] [21.155] [166.072] [18.274] [1336.2]
True values 1060.756 379.4156 0.08638 0.03248 121.6831 5.7126 0.01212 0.00003
MLE = maximum likelihood estimate, MME = moment method estimator.

2.053, 0.116 3K 5S 2 6.0917 .

Table 3.4 1000 MLE MME
, . 1 -MLE -MME
. .
2 . , MLE MME
.

3.3.

250 -MLE -MME

. Figure 3.5 KOSPI . MLE MME
, MME 3K 5S 2
. MLE
. -MLE -MME (y)
. S&P500,
FTSE, HANG SENG .
Barndor-Nielsen (1985) Hyperbolic Shape Trian-
gle . Hyperbolic Shape Triangle NIG , ,
.
1
= , = .

1+ 2 2

, 0 || < < 1
. 1 0
. 0 .
748 Jeongyoen Yoon, Seongjoo Song

Figure 3.5. MLE, MME, -MLE, and -MME with KOPSI daily log returns from January 2001 to May 2015
(MLE = maximum likelihood estimate, MME = moment method estimator).

Figure 3.6 KOSPI Hyperbolic Shape Tri-

angle. -MLE and -MME Triangle
. Hyperbolic Shape Triangle
.

3.4. VaR

VaR
. Kim Song (2011) NIG VG
VaR NIG VaR .
VaR VaR .
 A numerical study of adjusted parameter estimation in normal inverse Gaussian distribution 749

Figure 3.6. Estimated values of MLE, MME, -MLE and -MME in Hyperbolic Shape Triangle; KOSPI (MLE =
maximum likelihood estimate, MME = moment method estimator).

(exponential weighted moving average method; EWMA) VaR

. EWMA VaR ,
, .

t2 = t1
2
+ (1 )rt2 ,

t1
2
rt2 Riskmet-
rics 0.94 . EWMA 250
VaR , NIG
250 NIG VaR .
KOSPI 3318, S&P500 3373, FTSE 3497, HANG SENG 3349 95%
VaR . VaR (Kupiec, 1995),
Table 3.5 . VaR ,
5% KOSPI 165.9 S&P500 168.65,
FTSE 174.85, HANG SNEG 167.45.
95% VaR NIG EWMA KOSPI -
750 Jeongyoen Yoon, Seongjoo Song

Table 3.5. Backtesting results for the 95% Value at Risk (VaR): the number of times that the actual loss exceeds
VaR and p-values of Kupiec test
KOSPI S&P500 FTSE HANG SENG
# reject Kupiec # reject Kupiec # reject Kupiec # reject Kupiec
EWMA 200 0.0089 191 0.082 200 0.055 178 0.403
-MLE 180 0.275 190 0.096 174 0.654 167 0.978
-MME 173 0.585 191 0.082 177 0.862 173 0.656
EWMA = exponential weighted moving average method, MLE = maximum likelihood estimate,
MME = moment method estimator.

MME NIG , FTSE HANG SENG -MLE NIG

. 3K 5S 2 MLE
MME KOSPI 3.44%, S&P500 14.67%, FTSE 0.74%,
HANG SENG 4.12%. 95% VaR
, MLE KOSPI 5.18%, S&P500 6.01%, FTSE 5.04%
HANG SENG 4.95% , MME KOSPI 5.06%, S&P500 6.04%, FTSE 4.93%
HANG SENG 4.94% . .

4.

NIG
. 3K 5S 2 0 0
. 3K 5S 2
max{3K 5S 2 , } 0.5 .
, NIG 3K 5S 2 0 -MLE -MME
MLE MME . 3K 5S 2
0 MLE MME, -MLE -MME .
2001 1 2015 5 KOSPI, S&P500, FTSE, HANG SENG
-MLE -MME . Hyperbolic Shape Triangle
, -MLE -MME .
VaR -MLE -MME NIG
.
NIG
. ,
.
.

References

Barndor-Nielsen, O. E. (1977). Exponentially decreasing distributions for the logarithm of particle size.
In Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences,
353, 401419.
Barndor-Nielsen, O. E. (1997). Processes of normal inverse gaussian type, Finance and Stochastics, 2,
4168.
 A numerical study of adjusted parameter estimation in normal inverse Gaussian distribution 751

Barndor-Nielsen, O. E., Blsild, P., Jensen, J. L., and Srensen, M. (1985). The Fascination of Sand,
Springer, 5787.
Black, F. and Scholes, M. (1973). The pricing of options and corporate liabilities, The Journal of Political
Economy, 2, 637654.
Carr, P., Geman, H., Madan, D., and Yor, M. (2002). The ne structure of asset returns: an empirical
investigation, Journal of Business, 75, 305333.
Clark, P. K. (1973). A subordinated stochastic process model with nite variance for speculative prices,
Econometrica, 41, 135155.
Cox, J. and Ross, S. (1976). The valuation of options for alternative stochastic processes, Journal of
Financial Economics, 3, 145166.
Eriksson, A., Ghysels, E., and Wang, F. (2009). The normal inverse Gaussian distribution and the pricing
of derivatives, The Journal of Derivatives, 16, 2337.
Fielitz, B. D. and Smith, E. W. (1972). Asymmetric stable distributions of stock price changes, Journal of
the American Statistical Association, 67, 813814.
Figueroa-Lopez, J. E., Lancette, S. R., Lee, K., and Mi, Y. (2011). Estimation of NIG and VG models for
high frequency nancial data, Handbook of Modeling High-Frequency Data in Finance, John Wiley &
Sons, 326.
Geman, H. (2002). Pure jump Levy processes for asset price modeling, Journal of Banking and Finance,
26, 12971316.
Ghysels, E. and Wang, F. (2014). Moment-implied densities: properties and applications, Journal of Busi-
ness & Economic Statistics, 32, 88111.
Hull, J. and White, A. (1987). The pricing of options on assets with stochastic volatilities, The Journal of
Finance, 42, 281300.
Kim, T. and Song, S. (2011). Value-at-Risk Estimation using NIG and VG Distribution, Journal of the
Korean Data Analysis Society, 13, 17751788.
Kupiec, P. H. (1995). Techniques for verifying the accuracy of risk measurement models, The Journal of
Derivatives, 3, 7384.
Madan, D. and Seneta, E. (1990). The VG model for share market returns, Journal of Business, 63,
511524.
Mandelbrot, B. (1963). The variation of certain speculative prices, Journal of Business, 36, 394419.
Merton, R. C. (1976). Option pricing when underlying stock returns are discontinuous, Journal of Financial
Economics, 3, 125144.
Prause, K. (1997). Modelling nancial data using generalized hyperbolic distributions, FDM Preprint, 48,
University of Freiburg.
752 Jeongyoen Yoon, Seongjoo Song

Normal inverse Gaussian

a a,1
a

(2016 4 11 , 2016 5 16 , 2016 5 17 )

normal inverse Gaussian(NIG)
5 ,
. NIG
. KOSPI,
S&P500, FTSE HANG SENG VaR
. ,
VaR
.

: normal inverse Gaussian, , , Value at Risk

2013 ()
(NRF-2013R1A1A3012819).
1 : (02841) 145, . E-mail: sjsong@korea.ac.kr