Abstract
The tail index is an important parameter in the whole of extreme value theory. In this article, we consider the estimation of the tail index in the presence of a random covariate, where the conditional distribution of the variable of interest is of Pareto-type. More precisely, we use a logarithmic function to link the tail index to the nonlinear predictor induced by covariates, which forms the nonparametric tail index regression models. To estimate the unknown function, we develop an estimation procedure via a local likelihood method. Consistency and asymptotic normality of the estimated functions are established. Subsequently, these theoretical results are illustrated through simulated and real datasets.
Similar content being viewed by others
References
Beirlant J, Goegebeur Y (2003) Regression with response distributions of Pareto-type. Comput Stat Data Anal 42:595–619
Beirlant J, Goegebeur Y (2004) Local polynomial maximum likelihood estimation for Pareto-type distributions. J Multivar Anal 89:97–118
Beirlant J, Dierckx G, Goegebeur Y, Matthys G (1999) Tail index estimation and an exponential regression model. Extremes 2:177–200
Cai Z, Fan J, Li R (2000) Efficient estimation and inferences for varying-coefficient models. J Am Stat Assoc 95:888–902
Chavez-Demoulin V, Davison AC (2005) Generalized additive modelling of sample extremes. J R Stat Soc Ser C 54:207–222
Chernozhukov V, Du S (2006) Extremal quantiles and value-at-risk. Technical Report 07-01, Department of Economics, Massachusetts Institute of Technology
Coles S (2001) An introduction to statistical modeling of extreme values. Springer, London
Csörgö S, Viharos L (1997) Asymptotic normality of least-squares estimators of tail indices. Bernoulli 3:351–370
Csörgö S, Deheuvels P, Mason D (1985) Kernel estimates of the tail index of a distribution. Ann Stat 13:1050–1077
Daouia A, Gardes L, Girard S, Lekina A (2011) Kernel estimators of extreme level curves. Test 20:311–333
Daouia A, Gardes L, Girard S (2013) On kernel smoothing for extremal quantile regression. Bernoulli 19:2557–2589
Davison AC, Ramesh NI (2000) Local likelihood smoothing of sample extremes. J R Stat Soc Ser B 62:191–208
Davison AC, Smith RL (1990) Models for exceedances over high thresholds. J R Stat Soc Ser B 52:393–442
de Haan L, Ferreira A (2006) Extreme value theory: an introduction. Springer, New York
Dekkers ALM, Einmahl JHJ, de Haan L (1989) A moment estimators for the index of an extreme-value distribution. Ann Stat 17:1833–1855
Drees H (1998) On smooth statistical tail functionals. Scand J Stat 25:187–210
Einmahl JHJ, de Haan L, Zhou C (2016) Statistics of heteroscedastic extremes. J R Stat Soc Ser B 78:31–51
Fan J, Gijbels I (1996) Local polynomial modelling and its applications. Chapman Hall, New York
Fan J, Heckman NE, Wand MP (1995) Local polynomial kernel regression for generalized linear models and quasi-likelihood functions. J Am Stat Assoc 90:141–150
Fan J, Gijbels I, King M (1997) Local likelihood and local partial likelihood in hazard regression. Ann Stat 25:1661–1690
Fletcher R, Reeves CM (1964) Function minimization by conjugate gradients. Comput J 7:149–154
Gardes L (2015) A general estimator for the extreme value index: application to conditional and heteroscedastic extremes. Extremes 18:479–510
Gardes L (2018) Tail dimension reduction for extreme quantile estimation. Extremes 21:57–95
Gardes L, Girard S (2008) A moving window approach for nonparametric estimation of the conditional tail index. J Multivar Anal 99:2368–2388
Gardes L, Girard S (2010) Conditional extremes from heavy-tailed distributions: an application to the estimation of extreme rainfall return levels. Extremes 13:177–204
Gardes L, Girard S (2012) Functional kernel estimators of large conditional quantiles. Electron J Stat 6:1715–1744
Gardes L, Stupfler G (2014) Estimation of the conditional tail index using a smoothed local Hill estimator. Extremes 17:45–75
Gardes L, Stupfler G (2019) An integrated functional Weissman estimator for conditional extreme quantiles. REVSTAT Stat J 17(1):109–144
Goegebeur Y, Guillou A, Schorgen A (2014a) Nonparametric regression estimation of conditional tails: the random covariate case. Statistics 48:732–755
Goegebeur Y, Guillou A, Osmann M (2014b) A local moment type estimator for the extreme value index in regression with random covariates. Can J Stat 42:487–507
Goegebeur Y, Guillou A, Stupfler G (2015) Uniform asymptotic properties of a nonparametric regression estimator of conditional tails. Annales de l’Institut Henri Poincar\(\acute{e}\) (B): Probability and Statistics 51:1190–1213
Goldie CM, Smith RL (1987) Slow variation with remainder: a survey of the thoery and its applications. Q J Math Oxf J 38:45–71
Hall P (1982) On some simple estimates of an exponent of regular variation. J R Stat Soc Ser B 44:37–42
Hall P, Tajvidi N (2000) Nonparametric analysis of temporal trend when fitting parametric models to extreme-value data. Stat Sci 15:153–167
Hill BM (1975) A simple general approach to inference about the tail of a distribution. Ann Stat 3:1163–1174
Lehmann EL, Casella G (1998) Theory of point estimation. Springer, New York
Pagan AR, Schwert GW (1990a) Testing for covariance stationarity in stock market data. Econ Lett 33:165–170
Pagan AR, Schwert GW (1990b) Alternative models for conditional stock volatility. J Econ 45:267–290
Quintos C, Fan Z, Phillips PCB (2001) Structural change tests in tail behaviour and the Asian crisis. Rev Econ Stud 68:633–663
Smith RL (1987) Estimating tails of probability distributions. Ann Stat 15:1174–1207
Smith RL (1989) Extreme value analysis of environmental time series: an application to trend detectionin ground-level ozone. Stat Sci 4:367–393
Stupfler G (2013) A moment estimator for the conditional extreme-value index. Electron J Stat 7:2298–2343
Stupfler G (2016) Estimating the conditional extreme-value index under random right-censoring. J Multivar Anal 144:1–24
Tibshirani R, Hastie T (1987) Local likelihood estimation. J Am Stat Assoc 82:559–567
Wang H, Li D (2013) Estimation of extreme conditional quantiles through power transformation. J Am Stat Assoc 108:1062–1074
Wang H, Tsai C (2009) Tail index regression. J Am Stat Assoc 104:1232–1240
Acknowledgements
We are very grateful to the editor, associate editor and anonymous reviewers for their constructive and helpful comments which greatly improved the quality of the article. Ma thanks the Natural Science Foundation of Ningxia (No. 2019AAC03130), First-Class Disciplines Foundation of Ningxia (No. NXYLXK2017B09), General Scientific Research Project of North Minzu University (No. 2018SXKY01) and the funding support from The Key Project of North Minzu University under Grant (No. ZDZX201804). Huang acknowledges the funding support from National Natural Science Foundation of China (Nos. 11371318 and 11871425), Zhejiang Provincial Natural Science Foundation (Nos. LY17A010016 and LY18A010005) and the Fundamental Research Funds for the Central Universities.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 Lemmas and their proofs
The following Lemmas will be used in the proofs of Theorem 1 and 2.
Lemma 1
Under Conditions (C.3), (C.6), (C.7) and (C.8), we have
and
for any interior point x of \(\mathcal {X}\), as \(n\rightarrow \infty \).
Proof
By Condition (C.7) and (2.1), it can be shown that
for any y, as \(n\rightarrow \infty \). Therefore, by Conditions (C.3), (C.6) and (C.8), we have
Similarly, we can also obtain (5.2). \(\square \)
Lemma 2
Suppose that Conditions (C.5) and (C.8) are satisfied, then
Proof
Under the above Conditions and (2.2) in effect, the proof of the Lemma 2 can immediately be completed from the results that was obtained by Theorem 1 of Wang and Tsai (2009). We here skip the detail. \(\square \)
1.2 Proofs of the main results
Proof of Theorem 3.1
In order to establish the existence and consistency of \(\hat{\varvec{\beta }}\), we adapt the proof of Theorem 5.1 in Chapter 6 of Lehmann and Casella (1998). However, before this, we need the following definitions. Define \(\varvec{\theta }=S(\varvec{\beta }-\varvec{\beta }^{0})\), \(\hat{\varvec{\theta }}=S(\hat{\varvec{\beta }}-\varvec{\beta }^{0})\), and \(\mathbf{U}_i=S^{-1}\mathbf{X}_i=\big (1,\frac{X_i-x}{h},\ldots ,(\frac{X_i-x}{h})^p\big )^{T}\). Then, we obtain
Thus the problem is equivalent to showing that there exists a solution \(\hat{\varvec{\theta }}\) to the likelihood equation
such that \(\hat{\varvec{\theta }}{\mathop {\rightarrow }\limits ^{P}}\mathbf 0 ,~~ \mathrm {as}~ n\rightarrow \infty \).
Denoted by \(Q_\epsilon \) the sphere centered at origin with radius \(\epsilon \). We only need to show that for any sufficiently small \(\epsilon \), as \(n\rightarrow \infty \), the probability tends to 1 that
at all points \(\varvec{\theta }\) on the surface of \(Q_\epsilon \). Therefore, \(L_n(\varvec{\theta })\) has a local maximum in the interior of \(Q_\epsilon \). Because at a local maximum the likelihood equation (5.3) must be satisfied, it will follow that for any \(\epsilon >0\), with probability tending to 1, the likelihood equation (5.3) has a solution \(\hat{\varvec{\theta }}(\epsilon )\) within \(Q_\epsilon \). So, let \(\hat{\varvec{\theta }}\) be the closest root to 0, then we have \({\mathbb {P}}\{\Vert \hat{\varvec{\theta }}\Vert ^{2}\le \epsilon \}\rightarrow 1, ~~\mathrm {as}~ n\rightarrow \infty \). It implies that \(S(\hat{\varvec{\beta }}-\varvec{\beta }^{0}){\mathop {\rightarrow }\limits ^{P}}\mathbf 0 , ~~\mathrm {as}~ n\rightarrow \infty \).
Now we show under assumptions that (5.4) holds. Using Taylor expansion of \(L_n(\varvec{\theta })\) around 0, we have that
where \(\varvec{\theta }^{*}\) lying between \(\mathbf 0 \) and \(\varvec{\theta }\),
and
Firstly, by the Strong Law of Large Numbers, we get
Next, we have
In the last step, we just use a change of variables. By Taylor series expansion and Condition (C.8), the above integral equals
where \(x^{*}\) lying between x and \(x+uh\). By Condition (C.2) together with (5.6) and Lemma 2, we obtain that
Again, by Lemma 1, Conditions (C.1), (C.2) and (C.4), the foregoing expression can be expressed as
Therefore, with probability tending to 1,
Secondly, similar to (5.6), we can get
where
by Conditions (C.2), (C.4) and (C.6). Note that \(\Lambda _{0}\) is defined in (3.1). Again, from Lemmas 1 and 2, it follows that
Hence for all \(\varvec{\theta }\in Q_\epsilon \),
with probability tending to 1, where \(\lambda \) is the smallest eigenvalue of \(\Lambda _{0}\).
Finally, using the same technical as was used in the proof of \(L'_n(\mathbf 0 )\), together with Conditions (C.2), (C.6), (C.8), and Lemma 2, we can show that
for some constant \(C>0\). Thus, substituting (5.7), (5.11) and (5.12) into (5.5), with probability tending to 1, we obtain that
for all \(\varvec{\theta }\in Q_\epsilon \), when \(\epsilon \) is small enough. Hence the proof of Theorem 3.1 is completed. \(\square \)
Proof of Theorem 3.2
By Taylor expansion and Condition (C.5), we have
From (5.10), we conclude that
On the other hand, to establish asymptotic normality, it remains to calculate the mean and variance of \(L'_{n}(0)\), and verify the Lyapounov condition.
By using a Taylor expansion, we get that
Next, we evaluate \({\mathbb {E}}{L'_{n}(0)}\), which is given by
The right-hand side of the above expression can be written as
For \(E_{n1}\), similar to the proof of Theorem 1, we have
Again, by using Taylor’ expansion and noting Conditions (C.2), (C.6), (C.8) and Lemma 2, \(E_{n1}\) can be written as
where \(x^{*}\) is lying between x and \(x+uh\). It follows from Lemma 1 that
For \(E_{n2}\), Conditions (C.2), (C.3), (C.6), (C.8), Lemma 2 and (5.14) implies that
Then under Lemma 1, we have the following result
Putting (5.15) together with (5.16) and (5.17) yields
For the covariance term of \(L'_{n}(0)\),
where
and
By the result of \({\mathbb {E}}L'_{n}(0)\), we have that \(\frac{n}{n_0}\varvec{r}_n =\psi (x)\frac{m^{(p+1)}(x)}{(p+1)!}h^{p+1}\int u^{p+1}\varvec{u}K(u)du+o(1)\). It means that \(\frac{n}{n_0^{2}}\varvec{r}_n\varvec{r}_n^{T}\rightarrow 0\), as \(n\rightarrow \infty \).
In addition, we also can show that
where \(\Lambda _1\) is defined in (3.3). So it follows from (5.18), (5.19), Lemmas 1 and 2 that
On the other hand, to establish the asymptotic normality for \(L'_{n}(0)\), it is necessary to show that for any constant vector \(\varvec{d}\ne 0\), as \(n\rightarrow \infty \) ,
The left-hand side of (5.21) can be written as
where \(W_{i}=\big (1-\exp (\mathbf{X}_{i}^{T}\varvec{\beta }^0)\log (Y_i/u_n)\big ){\varvec{d}}^{T}{} \mathbf{U}_i\). Next, it is sufficient to verify the Lyapounov condition:
Under Conditions (C.2) and (C.6), the left-hand side of (5.22) is bounded by
and thus the Lyapounov condition hold for the (5.21). By using (5.13) and the slutsky theorem, we have
where \(\mathscr {B}(x)\) is defined in (3.6). We complete the proof of Theorem 3.2. \(\square \)
Rights and permissions
About this article
Cite this article
Ma, Y., Wei, B. & Huang, W. A nonparametric estimator for the conditional tail index of Pareto-type distributions. Metrika 83, 17–44 (2020). https://doi.org/10.1007/s00184-019-00723-8
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00184-019-00723-8