Abstract
The distribution of linear combinations of independent Gumbel random variables is of great interest for modeling risk and extremes in the most different areas of application. In this paper we develop near-exact approximations for the distribution of linear combination of independent Gumbel random variables based on a shifted generalized near-integer gamma distribution and on the distribution of the difference of two independent generalized integer gamma distributions. These near-exact distributions are computationally appealing and numerical studies confirm their accuracy, as assessed by a proximity measure used in related studies. We illustrate the proposed approximations on applied problems in networks engineering, computational biology, and flood risk management.
Similar content being viewed by others
References
Albrecher, H., Constantinescu, C., Loisel, S.: Explicit ruin formulas for models with dependence among risks. Ins. Math. Econ. 48, 265–270 (2011)
Aldous, D., Shepp, L.: The least variable phase-type distribution is Erlang. Stoch. Mod. 3, 467–473 (1987)
Amari, S.V., Misra, R.B.: Closed-form expressions for distribution of sum of exponential random variables. IEEE Trans. Rel. 46, 519–522 (1997)
Antal, T., Sylos Labini, F., Vasilyev, N.L., Baryshev, Y.V.: Galaxy-distribution and extreme-value statistics. Europhy. Lett. 58, 59001 (2009)
Arnold, B.C., Balakrishnan, N., Nagaraja, H.N.: Records. Wiley, New York (1998)
Bailey, T.L., Gribskov, M.: Score distributions for simultaneous matching to multiple motifs. J. Comput. Bio. 4, 45–59 (1997)
Balakrishnan, N., Ahsanullah, M., Chan, P.S.: Relations for single and product moments of record values from Gumbel distribution. Stat. Prob. Lett. 15, 223–227 (1992)
Berry, A.: The accuracy of the Gaussian approximation to the sum of independent variates. Trans. Am. Math. Soc. 49, 122–136 (1941)
Burda, M., Harding, M., Hausman, J.: A Poisson mixture model of discrete choice. J. Econom. 166, 184–203 (2012)
Castillo, E., Hadi, A.S., Balakrishnan, N., Sarabia, J.M.: Extreme Value and Related Models with Applications in Engineering and Science. Wiley, Hoboken (2005)
Cetinkaya, C., Kanodia, V., Knightly, E.W.: Scalable services via egress admission control. IEEE Trans. Multimed. 3, 69–81 (2001)
Coelho, C.A.: The generalized integer Gamma distribution—a basis for distributions in multivariate statistics. J. Mult. Anal. 64, 86–102 (1998)
Coelho, C.A.: The generalized near-integer Gamma distribution: A basis for ‘near-exact’ approximations to the distribution of statistics which are the product of an odd number of independent Beta random variables. J. Mult. Anal. 89, 191–218 (2004)
Coelho, C.A., Arnold, B.C., Marques, F.J.: The distribution of the product of powers of independent uniform random variables. J. Mult. Anal. 113, 19–36 (2013)
Coelho, C.A., Marques, F.J.: Near-exact distributions for the independence and sphericity likelihood ratio test statistics. J. Mult. Anal. 101, 583–593 (2010)
Coelho, C.A., Marques, F.J.: Near-exact distributions for the likelihood ratio test statistic to test equality of several variance-covariance matrices in elliptically contoured distributions. Comput. Stat. 27, 627–659 (2012)
Coelho, C.A., Mexia, J.T.: Product and Ratio of Generalized Gamma-Ratio Random Variables: Exact and Near-exact Distributions-Applications. Lambert Academic Publishing AG & Co, Saarbrücken (2010)
Domingos, P., Hulten, G.: A general framework for mining massive data streams. J. Comput. Graph. Stat. 12, 945–949 (2003)
Esseen, C.-G.: Fourier analysis of distribution functions, a mathematical study of the Laplace–Gaussian law. Acta Math. 77, 1–125 (1945)
Gil-Pelaez, J.: Note on the inversion theorem. Biometrika 38, 481–482 (1951)
Gumbel, E.J.: The return period of flood flows. Ann. Math. Stat. 12, 163–190 (1941)
Grilo, L.M., Coelho, C.A.: Development and study of two near-exact approximations to the distribution of the product of an odd number of independent Beta random variables. J. Stat. Plann. Infer. 5, 1560–1575 (2007)
Hwang, H.-K.: On convergence rates in the central limit theorems for combinatorial structures. Euro. J. Combin. 19, 329–343 (1998)
Hosking, J.R.M., Wallis, J.R., Wood, E.F.: Estimation of the generalized extreme-value distribution by the method of probability-weighted moments. Technometrics 27, 251–261 (1985)
Keich, U., Nagarajan, N.: A fast and numerically robust method for exact multinomial goodness-of-fit test. J. Comput. Graph. Stat. 15, 779–802 (2006)
Loaiciga, H.A., Leipnik, R.B.: Analysis of extreme hydrology events with Gumbel distributions: marginal and additive cases. Stoch. Environ. Res. Risk Ass. 13, 251–259 (1999)
Loève, M.: Probability Theory, 4th edn. Springer, New York (1977)
Marques, F.J.: On the product of independent Generalized Gamma random variables. Discussion Paper 19–2012, CMA-FCT-Universidade Nova de Lisboa (2012)
Marques, F.J., Coelho, C.A.: Near-exact distributions for the sphericity likelihood ratio test statistic. J. Stat. Plann. Infer. 138, 726–741 (2008)
Marques, F.J., Coelho, C.A., Arnold, B.C.: A general near-exact distribution theory for the most common likelihood ratio test statistics used in multivariate analysis. Test 20, 180–203 (2011)
Müller, P., Quintana, F.: Nonparametric Bayesian data analysis. Stat. Sci. 19, 95–110 (2004)
Nadarajah, S.: Exact distribution of the linear combination of \(p\) Gumbel random variables. Int. J. Comput. Math. 85, 1355–1362 (2008)
Nadarajah, S., Kotz, S.: Comments on scalable services via egress admission control. IEEE Trans. Multimed. 10, 160–161 (2008)
Natural Environment Research Council: Flood Studies Report, Vol. 4, London (1975)
O’Cinneide, C.A.: Characterization of phase-type distributions. Stoch. Models 6, 1–57 (1990)
Sandve, G.K., Drabløs, F.: A survey of motif discovery methods in an integrated framework. Biol. Dir. 1, 11 (2006)
Tiago de Oliveira, J.: Decision results for the parameters of the extreme value (Gumbel) distribution based on the mean and the standard deviation. Trabajos de Estadística y de Investigación Operativa 14, 61–81 (1963)
Wang, J.Z.: Selection of the \(K\) largest order statistics for the domain of attraction of the Gumbel distribution. J. Am. Stat. Assoc. 90, 1055–1061 (1995)
Acknowledgments
We thank the Editor, the Associate Editor, and two Reviewers for their careful reading and constructive comments. We also thank Anthony Davison, Barry Arnold, and Vanda de Carvalho for helpful discussions and recommendations on an earlier version of this paper. Part of this work was developed during an academic visit of F. Marques and C. Coelho to the Ecole Polytechnique Fédérale de Lausanne where M. de Carvalho was a Post-Doctoral Fellow. This work was partially supported by the Fundação para a Ciência e a Tecnologia (Portuguese Foundation for Science and Technology) through the project PEst-OE/MAT/UI0297/2014 (Centro de Matemática e Aplicações), and under the Fondecyt Project 11121186.
Author information
Authors and Affiliations
Corresponding author
Electronic supplementary material
Below is the link to the electronic supplementary material.
Appendix
Appendix
1.1 Appendix 1: Results and definitions on distributions of interest
Part I: The GIG and GNIG distributions
Let \(X_j\overset{\text {ind.}}{\sim }\text {Gamma}(r_j,\lambda _j)\) with shape parameters \(r_j \in {\mathbb {N}}\) and rate parameters \(\lambda _j\in {{\mathbb {R}}_+^*}\), all different, for \(j=1,\ldots ,\ell \). The GIG distribution of depth \(\ell \in {\mathbb {N}}\), introduced by Coelho (1998), is defined as the distribution of \(Y = \sum _{j=1}^\ell X_j\), and we denote this by \(Y \sim \text {GIG}(\varvec{r},\varvec{\lambda },\ell )\), for \(\varvec{r} = (r_1,\ldots ,r_\ell )\) and \(\varvec{\lambda } = (\lambda _1,\ldots ,\lambda _\ell ).\) The density and distribution functions of \(Y\) are
and
where \(y > 0,\,K=\prod _{i=1}^p \lambda _i^{r_i}\),
and the \(c_{j,k}\) are given in (11)–(13) in Coelho (1998). The GNIG distribution of depth \((\ell +1) \in {\mathbb {N}}\), introduced by Coelho (2004), is defined as the distribution of \(Y^{\star } = X^{\star } + \sum _{j=1}^{\ell } X_j\), where \(X^{\star }\) is independent of \(\sum _{j=1}^{\ell } X_j\), and \(X^{\star } \sim \text {Gamma}(\rho \), \(l\)), with \(\rho \in {{\mathbb {R}}_+^*} \backslash {\mathbb {N}}\). We denote this by \(Y^{\star } \sim \text {GNIG}(\varvec{r}^{\star }, \varvec{\lambda }^{\star }, \ell +1)\), where \(\varvec{r}^{\star }=(\varvec{r},\rho )\) and \(\varvec{\lambda }^{\star } = (\varvec{\lambda }, l)\), and the corresponding density and distribution functions are
and
for \(y > 0\) and where \(c^*_{j,k}=(c_{j,k}\lambda _j^k)/\varGamma (k)\); in the above expressions \(_1F_1(\cdot )\) denotes the Kummer confluent hypergeometric function.
The random variable \(X^*=X+\theta \) is a shifted Gamma distribution with rate \(\lambda \in {{\mathbb {R}}_+^*}\), shape \(r \in {{\mathbb {R}}_+^*}\), and shift \(\theta \in {\mathbb {R}}\), if \(X\sim \text {Gamma}(r,\lambda )\), and we denote this by \(X^*\sim \text {SGamma}(r,\lambda ,\theta )\); the shifted GIG and GNIG distributions are analogously defined and denoted by \(\mathrm{SGIG}(\varvec{r},\varvec{\lambda },\ell ,\theta )\) and \(\mathrm{SGNIG}(\varvec{r}^{\star },\varvec{\lambda }^{\star },\ell +1,\theta )\).
Part II: The DGIG distribution and the sum (and the difference) of a DGIG random variable with an independent Gamma random variable
Let \(X_1\sim \text {GIG}(\varvec{r}_1,\varvec{\lambda }_1,p_1)\), with \(\varvec{r}_1=(r_{11},\dots ,r_{1p_1})\) and \(\varvec{\lambda }_1=(\lambda _{11},\dots ,\lambda _{1p_1})\), and \(X_2\sim \text {GIG}(\varvec{r}_2,\varvec{\lambda }_2,p_2)\), with \(\varvec{r}_2=(r_{21},\dots ,r_{2p_2})\) and \(\varvec{\lambda }_2=(\lambda _{21},\dots ,\lambda _{2p_2})\) be two independent random variables with GIG distributions. Let us then consider the random variable \(Y=X_1-X_2\). \(Y\) has a DGIG distribution whose density and distribution functions are given by (2.12) and (2.15) in Coelho and Mexia (2010), and we denote this by \(Y\sim \text {DGIG}(\varvec{r}_1,\varvec{r}_2,\varvec{\lambda }_1,\varvec{\lambda }_2,p_1,p_2)\). The shifted SDGIG distribution, with shift \(\theta \in {\mathbb {R}}\), is denoted by \(Y\sim \mathrm{SDGIG}(\varvec{r}_1,\varvec{r}_2,\varvec{\lambda }_1,\varvec{\lambda }_2,p_1,p_2,\theta )\). Next we obtain results on the distribution of the sum (and the difference) of a DGIG with an independent Gamma random variable; these results are relevant for our third near-exact distribution. One useful way to look at the distribution of \(Y\) is to see it as a particular mixture of integer Gamma or Erlang distributions. Indeed, after some rearrangements the density and distribution functions of \(Y\) may be respectively written as
and
where, for \({j\!=\!1,\dots ,p_1;k\!=\!1,\dots ,r_{1j};i\!=\!0,\dots ,k-1}\),
and, for \({j=1,\dots ,p_2;k=1,\dots ,r_{2j};i=0,\dots ,k-1}\),
with
and \(c_{jk}\,{(j=1,\dots ,p_1;k=1,\dots ,r_{1j})}\) given by (2.9)–(2.11) in Coelho and Mexia (2010), with \(p\) replaced by \(p_1\) and \(r_j\) replaced by \(r_{1j}\) and \(d_{jk}\,(j\!=\!1,\dots ,p_2;k\!=\!1,\dots ,r_{2j})\) defined in a similar manner, replacing \(p_1\) by \(p_2\) and \(r_{1j}\) by \(r_{2j}\), and where, for \(y\ge 0\),
and
are respectively the density and distribution functions of \(Y_{jki}\sim \text {Gamma}(k-i,\lambda _{1j})\), while \(f^{}_{Y^*_{jki}}(\,\cdot \,)\) and \(F^{}_{Y^*_{jki}}(\,\cdot \,)\) are the density and distribution functions of \(Y^*_{jki}\sim \text {Gamma} (k-i,\lambda _{2j})\).
The weights \(p^{}_{jki}\) and \(p^*_{jki}\) verify the relation
Let now \(W\sim \text {Gamma}(\rho ,\lambda )\), where \(\rho \) is a positive non-integer real, be independent of \(Y\). We will consider the random variables \(Z_1=Y+W\) and \(Z_2=Y-W\) and derive their distribution functions. The distribution function of \(Z_1\), will be given by
which, for \({z\ge 0}\), using the notation introduced above for the GNIG distribution function, with \(\varvec{r}^\star =(k-i,\rho )\) and \(\varvec{\lambda }_1^\star =(\lambda _{1j},\lambda )\), may be written as
while for \(z< 0\) we have
We thus have
Concerning \(Z_2=Y-W\) we have, for \(z<0\), using the notation introduced above for the GNIG distribution function, with \(\varvec{r}^\star =(k-i,\rho )\) and \(\varvec{\lambda }_2^\star =(\lambda _{2j},\lambda )\),
while for \(z\ge 0\) we have
so that
It remains now to obtain the distribution function of random variables of the type of \(Z^*=W-Y^*\), where \(W\sim \text {Gamma}(\rho ,\lambda )\) and \(Y^*\sim \text {Gamma}(r,\lambda _1)\), where \(\rho ,\lambda _1\) and \(\lambda _2\) are positive reals and \(r\) is a positive integer. The distribution function of \(Z^*\) is given by
which for \(z\ge 0\), using the expression in (24) for the distribution function of an integer Gamma or Erlang distribution, yields
while for \(z<0\) it yields
and as such
1.2 Appendix 2: Representation of a logarithmized Gamma distribution as an infinite sum of shifted exponential distributions
If \(X\sim \mathrm{Gamma}(r,\lambda )\) its \(h\)th moment is given by
Then, the random variable \(Y=-\log \, X\) has what we call a logarithmized Gamma distribution and its characteristic function may be obtained from (27) in the following way
Using the equality
we have
Hence \(\varPhi _Y\) is also the characteristic function of an infinite sum of independent shifted Exponential distributions. This shows that a logarithmized Gamma random variable may be represented as an infinite sum of independent shifted Exponential random variables.
Rights and permissions
About this article
Cite this article
Marques, F.J., Coelho, C.A. & de Carvalho, M. On the distribution of linear combinations of independent Gumbel random variables. Stat Comput 25, 683–701 (2015). https://doi.org/10.1007/s11222-014-9453-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11222-014-9453-5