Abstract
A number of applications presume that asset returns are normally distributed, even though they are widely known to be skewed leptokurtic and fat-tailed and excess kurtosis. This leads to the underestimation or overestimation of the true value-at-risk (VaR). This study utilizes a composite trapezoid rule, a numerical integral method, for estimating quantiles on the skewed generalized t distribution (SGT) which permits returns innovation to flexibly treat skewness, leptokurtosis and fat tails. Daily spot prices of the thirteen stock indices in North America, Europe and Asia provide data for examining the one-day-ahead VaR forecasting performance of the GARCH model with normal, student’s t and SGT distributions. Empirical results indicate that the SGT provides a good fit to the empirical distribution of the log-returns followed by student’s t and normal distributions. Moreover, for all confidence levels, all models tend to underestimate real market risk. Furthermore, the GARCH-based model, with SGT distributional setting, generates the most conservative VaR forecasts followed by student’s t and normal distributions for a long position. Consequently, it appears reasonable to conclude that, from the viewpoint of accuracy, the influence of both skewness and fat-tails effects (SGT) is more important than only the effect of fat-tails (student’s t) on VaR estimates in stock markets for a long position.
Similar content being viewed by others
Notes
The general consensus regarding volatility forecasting in most of the literature is that generalized autoregressive conditional heteroskedasticity (GARCH) models. This study thus considers the applicability of the GARCH(1,1) model in modeling VaRs.
The standardized SGT distribution, which has zero mean and unit variance, was checked by Mathematica software and another analogous standardized SGT distribution was proposed by Bali and Theodossiou (2007).
See Faires and Burden (2003) for more details.
The parameters are estimated by QMLE (Quasi maximum likelihood estimation; QMLE) and the BFGS optimization algorithm, using the econometric package of WinRATS 6.1.
LRN for GARCH-T model follows the χ2(1) distribution with one degree of freedom. On the other hand, LRN for GARCH-SGT model follows the χ2(3) distribution with three degree of freedom.
LRT for GARCH-SGT model follows the χ2(2) distribution with two degree of freedom. LRN and LRT are the log-likelihood ratio test statistics and are specified as follows: LR = −2(LRr − LRu) ~ χ2(m), where LRr and LRu are, respectively, the maximum value of the log-likelihood values under the null hypothesis of the restricted model and the alternative hypothesis of the unrestricted model, and m is the number of the restricted parameters in the restricted model. For example, LRN for GARCH-SGT model could be used to test the null hypothesis that log-returns are normally distributed against the alternative hypothesis that they are SGT distributed. The null hypothesis for testing normality is \( {\text{H}}_{0} :\kappa = 2,\lambda = 0\,{\text{and}}\,{\text{n}} \to \infty \) and the alternative hypothesis is \( {\text{H}}_{1} :\kappa \in {\text{R}}^{ + } ,{\text{n}} > 2\,{\text{and}}\left| \lambda \right| < 1 \). Restate, LRN = −2(LRr − LRu) ~ χ2(3) where LRr and LRu are respectively the maximum value of the log-likelihood values under the null hypothesis of restricted model (GARCH-N model) and the alternative hypothesis of unrestricted model (GARCH-SGT model), and m is the number of the restricted parameters in the restricted model(\( \kappa = 2,\lambda = 0\,{\text{and}}\,n \to \infty \)) and equal to 3 in this case.
References
Ané T (2006) An analysis of the flexibility of asymmetric power GARCH models. Comput Stat Data Anal 51:1293–1311
Angelidis T, Benos A, Degiannakis S (2004) The use of GARCH models in VaR estimation. Stat Methodol 1:105–128
Angelidis T, Benos A, Degiannakis S (2007) A robust VaR model under different time periods and weighting schemes. Rev Quant Finance Account 28:187–201
Arora RK, Das H, Jain PK (2009) Stock returns and volatility: evidence from select emerging markets. Rev Pacific Basin Financ Markets Policies 12(4):567–592
Baillie R, de Gennaro R (1990) Stock returns and volatility. J Financ Quant Anal 25:203–214
Baixauli JS, Alvarez S (2006) Evaluating effects of excess kurtosis on VaR estimates: evidence for international stock indices. Rev Quant Finance Account 27:27–46
Bali TG, Theodossiou P (2007) A conditional-SGT-VaR approach with alternative GARCH models. Ann Oper Res 151:241–267
Billio M, Pelizon L (2000) Value-at-risk: a multivariate switching regime approach. J Empir Finance 7:531–554
Bollerslev T (1986) Generalized autoregressive conditional heteroskedasticity. J Econom 31:307–327
Bollerslev T, Chou RY, Kroner KF (1992) ARCH modeling in finance: a review of the theory and empirical evidence. J Econom 52:5–59
Brooks C, Persand G (2003) The effect of asymmetries on stock index return Value-at-Risk estimates. J Risk Finance 4:29–42
Cheong CW (2008) Heavy-tailed value-at-risk analysis for Malaysian stock exchange. Physica A 387:4285–4298
Engle RF, Manganelli S (2004) CAViaR: conditional autoregressive Value at Risk by regression quantiles. J Bus Econ Stat 22:367–381
Faires JD, Burden R (2003) Numerical methods, 3rd edn. Tomson Learning, Pacific Grove
Fama E (1965) The behavior of stock market prices. J Bus 38:34–105
Giot P, Laurent S (2003) Value-at-Risk for long and short trading positions. J Appl Econom 18:641–664
Huang YC, Lin BJ (2004) Value-at-Risk analysis for Taiwan stock index futures: fat tails and conditional asymmetries in return innovations. Rev Quant Finance Account 22:79–95
Jansen D, de Vries C (1991) On the frequency of large stock returns: putting booms and busts into perspective. Rev Econ Stat 73:18–24
Jarque CM, Bera AK (1987) A test for normality of observations and regression residuals. Int Stat Rev 55:163–172
Jorion P (2000) Value at Risk: the new benchmark for managing financial risk. McGraw-Hill, New York
Kearns P, Pagan A (1997) Estimating the density tail index for financial times series. Rev Econ Stat 79:171–175
Kupiec P (1995) Techniques for verifying the accuracy of risk management models. J Deriv 3:73–84
Lee MC, Su JB, Liu HC (2008) Value-at-risk in US stock indices with skewed generalized error distribution. Appl Financ Econ Lett 4:425–431
Lin CH, Chien CC, Chen SW (2006) Incorporating the time-varying tail-fatness into the historical simulation method for portfolio Value-at-Risk. Rev Pacific Basin Financ Markets Polices 9(2):257–274
Lopez JA (1999) Methods for evaluating value-at-risk estimates. Federal Reserve Bank San Francisco Econ Rev 2:3–17
Loretan M, Phillips P (1994) Testing the covariance structure of heavy-tailed time series. J Empir Finance 1:211–248
Mandelbrot B (1963) The variation of certain speculative prices. J Bus 36:394–419
So MKP, Yu PLH (2006) Empirical analysis of GARCH models in value at risk estimation. Int Financ Markets Inst Money 16:180–197
Tapia RA, Thompson JR (1978) Nonparametric probability density estimation. Johns Hopkins University Press, Baltimore (Section 2.5.)
Theodossiou P (1998) Financial data and the skewed generalized t distribution. Manag Sci 44:1650–1661
Author information
Authors and Affiliations
Corresponding author
Appendix 1: Derivation of the standardized SGT
Appendix 1: Derivation of the standardized SGT
The probability density function for the Skewed Generalized t (SGT) distribution introduced by Theodossiou (1998) is represented as follows:
where \( \theta^{\prime} = \frac{1}{{{\text{S}}(\lambda )}}\left( {\frac{\kappa }{{{\text{n}} - 2}}} \right)^{{\frac{1}{\kappa }}} {\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{{ - \frac{1}{2}}} ,{\text{S}}(\lambda ) = \sqrt {1 + 3\lambda^{2} - 4{\text{A}}^{2} \lambda^{2} } ,\) \( {\text{A}} = {\text{B}}\left( {\frac{2}{\kappa },\frac{{{\text{n}} - 1}}{\kappa }} \right){\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{{ - \frac{1}{2}}} {\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{{ - \frac{1}{2}}} , \) \( {\text{C}}^{\prime } = \frac{\kappa }{2}\frac{{{\text{S}}(\lambda )}}{\sigma }{\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{{ - \frac{3}{2}}} \).
The expected value of x is \( \mu = {\text{E}}({\text{x}}) = 2\lambda {\text{A}}\sigma /{\text{S}}(\lambda ) \) and the variance of x is \( \text{var} ({\text{x}}) = {\text{E}}({\text{x}}^{2} ) - \mu^{2} = \sigma^{2} \). For simplicity, the derivation of parameters, C, θ and δ, for the standardized SGT distribution is accomplished using the transformed random variable ε = (x−μ)/σ, which has a mean of E(ε) = 0 and a variance of var(ε) = 1. The random x thus can be expressed as x = μ + σε and let \( \delta = \mu /\sigma = 2\lambda {\text{A}}/{\text{S}}(\lambda ) \), therefore \( {\text{dx}}/{\text{d}}\varepsilon = \sigma , \) \( {\text{x}}/\sigma = \varepsilon + \mu /\sigma = \varepsilon + \delta \) and \( {\text{sign}}({\text{x}}) = {\text{sign}}(\mu + \sigma \varepsilon ) = {\text{sign}}\left( {\sigma \left( {\varepsilon + \mu /\sigma } \right)} \right) = {\text{sign}}\left( {\varepsilon + \mu /\sigma } \right) = {\text{sign}}\left( {\varepsilon + \delta } \right) \).
Substitute the above expressions and Eq. 16 into the following equation.
where \( {\text{C}}^{\prime } = \frac{\kappa }{2}\frac{{{\text{S}}(\lambda )}}{\sigma }{\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{{ - \frac{3}{2}}} , \) \( \theta^{\prime } = \frac{1}{{{\text{S}}(\lambda )}}\left( {\frac{\kappa }{{{\text{n}} - 2}}} \right)^{{\frac{1}{\kappa }}} {\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{{ - \frac{1}{2}}} , \)
Letting \( \theta^{\kappa } = \frac{{({\text{n}} - 2)\theta^{\prime \kappa } }}{\kappa } = \left[ {\frac{1}{{{\text{S}}(\lambda )}}{\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{{ - \frac{1}{2}}} } \right]^{\kappa } \) and substituting the expression into Eq. 17 gives the equation which is the standardized SGT distribution.
where \( \theta = \frac{1}{{{\text{S}}(\lambda )}}{\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{{ - \frac{1}{2}}} ,{\text{S}}(\lambda ) = \sqrt {1 + 3\lambda^{2} - 4{\text{A}}^{2} \lambda^{2} } , \) \( {\text{A}} = {\text{B}}\left( {\frac{2}{\kappa },\frac{{{\text{n}} - 1}}{\kappa }} \right){\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{ - 0.5} {\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{ - 0.5} , \) \( \delta = \frac{{2\lambda {\text{A}}}}{{{\text{S}}(\lambda )}}, \) \( {\text{C}} = \frac{{\kappa {\text{S}}(\lambda )}}{2}{\text{B}}\left( {\frac{3}{\kappa },\frac{{{\text{n}} - 2}}{\kappa }} \right)^{\frac{1}{2}} {\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{{ - \frac{3}{2}}} = \frac{\kappa }{2\theta }{\text{B}}\left( {\frac{1}{\kappa },\frac{\text{n}}{\kappa }} \right)^{ - 1} . \)
Particularly, the SGT distribution generates the student’ t distribution for λ = 0 and κ = 2. Using the recurrence identities of Beta function, \( {\text{B}}\left( {{\text{a}},{\text{b}}} \right) = {\text{B}}\left( {{\text{b}},{\text{a}}} \right) \), \( {\text{B}}\left( {{\text{a}} + 1,{\text{b}}} \right) = {\text{a}} \cdot {\text{B}}\left( {{\text{a}},{\text{b}}} \right)/\left( {{\text{a}} + {\text{b}}} \right), \) and \( {\text{B}}\left( {{\text{a}} - 1,{\text{b}}} \right) = \frac{{{\text{a}} + {\text{b}} - 1}}{{{\text{a}} - 1}} \cdot {\text{B}}\left( {{\text{a}},{\text{b}}} \right), \) we can obtain \( {\text{B}}\left( {\frac{3}{2},\frac{{{\text{n}} - 2}}{2}} \right) = \frac{1}{{{\text{n}} - 2}}{\text{B}}\left( {\frac{1}{2},\frac{\text{n}}{2}} \right) \).
Therefore \( \delta = 2\lambda {\text{A}}/{\text{S}}(\lambda ) = 0,{\text{S}}(\lambda ) = \sqrt {1 + 3\lambda^{2} - 4{\text{A}}^{2} \lambda^{2} } = 1, \) and
The probability density function for the standardized student’ t distribution thus can be represented as follows.
Rights and permissions
About this article
Cite this article
Lee, CF., Su, JB. Alternative statistical distributions for estimating value-at-risk: theory and evidence. Rev Quant Finan Acc 39, 309–331 (2012). https://doi.org/10.1007/s11156-011-0256-x
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11156-011-0256-x