Abstract
Weissman extrapolation methodology for estimating extreme quantiles from heavy-tailed distributions is based on two estimators: an order statistic to estimate an intermediate quantile and an estimator of the tail-index. The common practice is to select the same intermediate sequence for both estimators. In this work, we show how an adapted choice of two different intermediate sequences leads to a reduction of the asymptotic bias associated with the resulting refined Weissman estimator. The asymptotic normality of the latter estimator is established and a data-driven method is introduced for the practical selection of the intermediate sequences. Our approach is compared to the Weissman estimator and to six bias reduced estimators of extreme quantiles on a large scale simulation study. It appears that the refined Weissman estimator outperforms its competitors in a wide variety of situations, especially in the challenging high bias cases. Finally, an illustration on an actuarial real data set is provided.
Similar content being viewed by others
Code availability
Code is available at https://github.com/michael-allouche/refined-weissman.git. The Secura Belgian reinsurance data set is available in the package CASdatasets of the R software (Dutang and Charpentier 2020).
References
Alves, M.I.F., Gomes, M.I., de Haan, L.: A new class of semi-parametric estimators of the second order parameter. Port. Math. 60(2), 193–214 (2003a)
Alves, M.I.F., de Haan, L., Lin, T.: Estimation of the parameter controlling the speed of convergence in extreme value theory. Math. Methods Statist. 12(2), 155–176 (2003b)
Beirlant, J., Dierckx, G., Guillou, A.: Estimation of the extreme-value index and generalized quantile plots. Bernoulli 11(6), 949–970 (2005)
Beirlant, J., Goegebeur, Y., Segers, J., Teugels, J.: Statistics of Extremes: Theory and Applications. Wiley (2004)
Beran, J., Schell, D., Stehlík, M.: The harmonic moment tail index estimator: asymptotic distribution and robustness. Ann. Inst. Stat. Math. 66(1), 193–220 (2014)
Bingham, N.H., Goldie, C.M., Teugels, J.L.: Regular variation. Cambridge University Press (1989)
Brilhante, M.F., Gomes, M.I., Pestana, D.: A simple generalisation of the Hill estimator. Comput. Stat. Data Anal. 57(1), 518–535 (2013)
Caeiro, F., Gomes, M.I., Pestana, D.: Direct reduction of bias of the classical Hill estimator. Revstat Stat. J. 3(2), 113–136 (2005)
Cai, J.J., Einmahl, J.H., de Haan, L., Zhou, C. (2015). Estimation of the marginal expected shortfall: the mean when a related variable is extreme. J. Roy. Stat. Soc. Ser. B, 417–442
Ciuperca, G., Mercadier, C.: Semi-parametric estimation for heavy tailed distributions. Extremes 13, 55–87 (2010)
Daouia, A., Gardes, L., Girard, S., Lekina, A.: Kernel estimators of extreme level curves. Test 20(2), 311–333 (2011)
Daouia, A., Gijbels, I., Stupfler, G.: Extremiles: a new perspective on asymmetric least squares. J. Am. Stat. Assoc. 114(527), 1366–1381 (2019)
Daouia, A., Girard, S., Stupfler, G.: Estimation of tail risk based on extreme expectiles. J. Roy. Stat. Soc. B 80, 262–292 (2018)
Daouia, A., Girard, S., Stupfler, G.: Extreme M-quantiles as risk measures: From L1 to Lp optimization. Bernoulli 25, 264–309 (2019)
Daouia, A., Girard, S., Stupfler, G.: Tail expectile process and risk assessment. Bernoulli 26(1), 531–556 (2020)
de Haan, L., Ferreira, A.: Extreme value theory: an introduction. Springer Science and Business Media (2007)
de Haan, L., Peng, L.: Comparison of tail index estimators. Stat. Neerl. 52(1), 60–70 (1998)
de Wet, T., Goegebeur, Y., Munch, M.R.: Asymptotically unbiased estimation of the second order tail parameter. Stat. Probab. Lett. 82, 565–573 (2012)
Dekkers, A.L.M., Einmahl, J.H.J., de Haan, L.: A moment estimator for the index of an extreme-value distribution. Ann. Stat. 17, 1833–1855 (1989)
Deme, E., Gardes, L., Girard, S.: On the estimation of the second order parameter for heavy-tailed distributions. Revstat Stat. J. 11, 277–299 (2013)
Drees, H., de Haan, L., Resnick, S.: How to make a Hill plot. Ann. Stat. 28, 254–274 (2000)
Dutang, C., Charpentier, A.: CASdatasets: Insurance datasets. R package version 1.0-11. (2020)
El Methni, J., Gardes, L., Girard, S., Guillou, A.: Estimation of extreme quantiles from heavy and light tailed distributions. J. Stat. Plan. Inference 142(10), 2735–2747 (2012)
El Methni, J., Stupfler, G.: Extreme versions of Wang risk measures and their estimation for heavy-tailed distributions. Stat. Sin. 27, 907–930 (2017)
El Methni, J., Stupfler, G.: Improved estimators of extreme Wang distortion risk measures for very heavy-tailed distributions. Econometrics Stat. 6, 129–148 (2018)
Goegebeur, Y., Beirlant, J., de Wet, T.: Kernel estimators for the second order parameter in extreme value statistics. J. Stat. Plan. Inference 140, 2632–2652 (2010)
Goegebeur, Y., Guillou, A., Schorgen, A.: Nonparametric regression estimation of conditional tails: the random covariate case. Statistics 48(4), 732–755 (2014)
Gomes, M.I., Brilhante, M.F., Caeiro, F., Pestana, D.: A new partially reduced-bias mean-of-order p class of extreme value index estimators. Comput. Stat. Data Anal. 82, 223–237 (2015)
Gomes, M.I., Brilhante, M.F., Pestana, D.: New reduced-bias estimators of a positive extreme value index. Commun. Stat. Simul. Comput. 45(3), 833–862 (2016)
Gomes, M.I., Caeiro, F., Figueiredo, F., Henriques-Rodrigues, L., Pestana, D.: Reduced-bias and partially reduced-bias mean-of-order-p value-at-risk estimation: a Monte-Carlo comparison and an application. J. Stat. Comput. Simul. 90(10), 1735–1752 (2020a)
Gomes, M.I., Caeiro, F., Figueiredo, F., Henriques-Rodrigues, L., Pestana, D.: Corrected-Hill versus partially reduced-bias value-at-risk estimation. Commun. Stat. Simul. Comput. 49(4), 867–885 (2020b)
Gomes, M.I., de Haan, L., Peng, L.: Semi-parametric estimation of the second order parameter in statistics of extremes. Extremes 5, 387–414 (2002)
Gomes, M.I., Pestana, D.: A sturdy reduced-bias extreme quantile (VaR) estimator. J. Am. Stat. Assoc. 102(477), 280–292 (2007)
Hall, P.: On some simple estimates of an exponent of regular variation. J. Roy. Stat. Soc. B 44(1), 37–42 (1982)
Hall, P., Welsh, A.W.: Adaptive estimates of parameters of regular variation. Ann. Stat. 13, 331–341 (1985)
Hill, B.M.: A simple general approach to inference about the tail of a distribution. Ann. Stat. 3, 1163–1174 (1975)
Kazama, S., Sato, A., Kawagoe, S.: Evaluating the cost of flood damage based on changes in extreme rainfall in Japan. Sustain. Sci. 4(61). (2009). https://doi.org/10.1007/s11625-008-0064-y
Kratz, M., Resnick, S.I.: The QQ-estimator and heavy tails. Stoch. Model. 12(4), 699–724 (1996)
Manjunath, B.G., Caeiro, F.: evt0: Mean of order p, peaks over random threshold Hill and high quantile estimates. R package version 1.1-3 (2013)
Paulauskas, V., Vaiciulis, M.: On an improvement of Hill and some other estimators. Lith. Math. J. 53(3), 336–355 (2013)
Peng, L.: Asymptotic unbiased estimators for the extreme value index. Stat. Probab. Lett. 38, 107–115 (1998)
Pickands, J.: Statistical inference using extreme order statistics. Ann. Stat. 3(1), 119–131 (1975)
Resnick, S., Stărică, C.: Smoothing the Hill estimator. Adv. Appl. Probab. 29, 271–293 (1997)
Rootzén, H., Tajvidi, N.: Extreme value statistics and wind storm losses: a case study. Scand. Actuar. J. 1, 70–94 (1997)
Schultze, J., Steinebach, J.: On least squares estimates of an exponential tail coefficient. Stat. Risk Model. 14(4), 353–372 (1996)
Stupfler, G.: On a relationship between randomly and non-randomly thresholded empirical average excesses for heavy tails. Extremes 22, 749–769 (2019)
Weissman, I.: Estimation of parameters and large quantiles based on the k largest observations. J. Am. Stat. Assoc. 73(364), 812–815 (1978)
Acknowledgements
The authors would like to thank the two referees and the Associate Editor for their valuable suggestions, which have significantly improved the paper. The authors would also like to thank Frederico Caeiro for providing the R code associated with Algorithm 4.1 and Algorithm 4.2 from Gomes et al. (2020a). This work is supported by the French National Research Agency (ANR) in the framework of the Investissements d’Avenir Program (ANR-15-IDEX-02). S. Girard also acknowledges the support of the Chair Stress Test, Risk Management and Financial Steering, led by the French Ecole Polytechnique and its Foundation and sponsored by BNP Paribas.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
A. Appendix: proofs
Proof of Theorem 1
It follows the same lines as the one of de Haan and Ferreira (2007, Theorem 4.3.8). Let us consider the expansion
where we have introduced:
Let us first focus on \(T_{0,n}\). From de Haan and Ferreira (2007, Theorem 2.3.9), it follows from the second-order condition (5) that, for any \(\varepsilon\), \(\delta >0\), there exists \(t_0>1\) such that for all \(t\ge t_0\) and \(x\ge 1\),
where \(A_0\) is asymptotically equivalent to A. Letting \(x=d_n\) and \(t=n/k_n\) then yields
or equivalently,
where \(|R_n|\le d_n^{\rho +\delta }\). Now, writing \(|A_0|(t)=t^\rho \ell (t)\), where \(\ell\) is a slowly-varying function, it follows,
as \(n\rightarrow \infty\). As a consequence, we obtain
and letting \(\varepsilon \rightarrow 0\) yields
Second, under the assumption \(\sqrt{k'_n}(\hat{\gamma }_n(k'_n)-\gamma ) {\mathop {\longrightarrow }\limits ^{d}}\mathcal{N}(\lambda \mu , \sigma ^2)\) as \(\sqrt{k'_n} A(n/k_n')\rightarrow \lambda\), we get
where \(\xi _n{\mathop {\longrightarrow }\limits ^{d}}\mathcal{N}(0,1)\). Recalling that \((\log d_n)/\sqrt{k'_n} \rightarrow 0\) as \(n\rightarrow \infty\), the following first order expansion holds
In particular, \(d_n^{\hat{\gamma }_n(k'_n)-\gamma }{\mathop {\longrightarrow }\limits ^{\mathbb {P}}}1\) and therefore,
where \(\xi '_n{\mathop {\longrightarrow }\limits ^{d}}\mathcal{N}(0,1)\), from de Haan and Ferreira (2007, Theorem 2.2.1). Third, it immediately follows from (22) that
Finally, in view of (21) and recalling that \(\sqrt{k'_n}A_0(n/k'_n)\rightarrow \lambda\), one has
Collecting (21), (23), (24), and (25) yields
since \(T_{0,n}=1+o(T_{3,n})\). Besides, assumptions \((k'_n/k_n)^\rho /(\log d_n)\rightarrow c\ge 0\) and \(d_n\rightarrow \infty\) as \(n\rightarrow \infty\) imply
and the result is proved. \(\blacksquare\)
Proof of Corollary 1
Under assumption (7), Eq. (26) in the proof of Theorem 1 can be simplified as
since \(\ell\) is asymptotically constant. Let us moreover note that
and therefore
which proves the result. \(\blacksquare\)
Proof of Lemma 1
(i) Letting \(f(x)=-\rho (\log x)/(1-x^\rho )\) for all \(x\ge 1\) and \(\rho <0\), from (9) one has \(k_n^\star = \mu ^{1/\rho } k_n (f(d_n))^{1/\rho }\) with \(d_n=k_n/(n\alpha _n)\ge 1\). First, routine calculations give:
for all \(d_n\ge 1\). As a conclusion, \({\partial k_n^\star }/{\partial k_n}\ge 0\) which proves that \(k_n^\star\) is an increasing function of \(k_n\). Second, it is easily shown that f is increasing and \(f(1)=1\), leading to \(f(d_n)\ge 1\) and thus \(k_n^\star \le \mu ^{1/\rho }k_n\).
(ii) is a consequence of \(f(x) \sim -\rho \log x\) as \(x\rightarrow \infty\).
(iii) Remark that assumption \(\log (n\alpha _n)/\log (k_n)\rightarrow c' \le 0\) implies that \(\log d_n\sim (1-c') \log k_n\) as \(n\rightarrow \infty\). The conclusion follows. \(\blacksquare\)
Proof of Corollary 2
It is sufficient to prove that assumptions (i) and (iii) of Theorem 1 hold true. First, Lemma 1(ii) entails that \(k_n^\star /k_n\sim \tau (\log d_n)^{1/\rho }\) as \(n\rightarrow \infty\). Besides, from (7), we have \({A(n/k_n^\star )}/{A(n/k_n)} \sim (k_n^\star /k_n)^{-\rho },\) so that
as \(n\rightarrow \infty\) in view of the first part of (10). Assumption (i) of Theorem 1 thus holds true with \(\lambda = \lambda ' \tau ^{1/2-\rho }\). Second,
as \(n\rightarrow \infty\) in view of the second part of (10). Third,
as \(n\rightarrow \infty\). Collecting (27) and (28) proves that assumption (iii) of Theorem 1 thus holds true with \(c=\tau ^\rho\). \(\blacksquare\)
Proof of Lemma 2
Recall that, from the proof of Lemma 1(iii), \(\log d_n\sim (1-c') \log k_n\) as \(n\rightarrow \infty\). Let us then observe that
and consequently
Besides, since \(\hat{\rho }_n\) is a consistent estimator of \(\rho\) and \(\mu (\cdot )\) is continuous, it follows that \((-\hat{\rho }_n \mu (\hat{\rho }_n))^{1/\hat{\rho }_n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}(-\rho \mu (\rho ))^{1/\rho }\) and therefore
in view of Lemma 1(iii). Remarking that the right hand side term tends to infinity in probability, one immediately has
and the result is proved. \(\blacksquare\)
Proof of Corollary 3
The first step is to prove that
To this end, recall that \(\log (n\alpha _n)/\log (k_n)\rightarrow c'\le 0\) implies that \(\log d_n\sim (1-c') \log k_n\) as \(n\rightarrow \infty\) and therefore condition (10) of Corollary 2 is fulfilled under the assumptions of Corollary 3. Besides, recalling that \(k_n/n\rightarrow 0\) as \(n\rightarrow \infty\), Lemma 2 entails that \(\hat{k}_n^\star {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\infty\) and \(\hat{k}_n^\star /n {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\). Therefore, for n large enough, \(\hat{k}_n^\star < n\) almost surely. Besides, for all \(m_n\in \{1,\dots ,n\}\),
By Stupfler (2019, Lemma 8), since \(\hat{k}_n^\star \in \{1,\dots ,n-1\}\) and \(\hat{k}_n^\star {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\infty\), it is enough to show that the desired convergence (29) holds with \(\hat{q}_n(\alpha _n,k_n,\hat{k}_n^\star )\) replaced by its de-conditioned version \(\hat{q}_n(\alpha _n,k_n,m_n)\). This is a direct consequence of Corollary 2. The second and final step consists in replacing \(\hat{k}_n^\star\) by its non random version \(k_n^\star\) in the rate of convergence of (29). This can be achieved using Lemma 2 and Slutsky’s lemma. \(\blacksquare\)
B. Appendix: tables
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Allouche, M., El Methni, J. & Girard, S. A refined Weissman estimator for extreme quantiles. Extremes 26, 545–572 (2023). https://doi.org/10.1007/s10687-022-00452-8
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10687-022-00452-8