betaFeedback

WorksMathematics

Testing the Pareto law through a Stein-Laplace characterization

Abstract

A Stein–Laplace characterization yields two closed-form goodness-of-fit statistics for the Pareto type-I family with unknown shape and known lower endpoint. The signed integral statistic TnT_n has a standard normal limit after studentization, but explicit non-Pareto alternatives show that it is not omnibus consistent. The integrated squared discrepancy SnS_n yields a consistent test against every fixed non-Pareto alternative on the specified support with a finite, positive logarithmic moment. We derive the weighted chi-square null limit of nSnnS_n and give conditions for an asymptotically valid parametric-bootstrap and plug-in covariance-operator method. The latter uses Nyström eigenvalue approximation to avoid repeated sample generation and parameter estimation. A large Monte Carlo comparison study shows competitive performance relative to existing procedures in the literature. In the paired study, the two calibrations have similar empirical power, with slightly higher small-sample rejection rates for the bootstrap in some settings. In the timing experiment, covariance-operator calibration is approximately 2828–3737 times faster at n=200n=200. An exploratory threshold-sensitivity analysis of Danish fire-insurance losses illustrates the dependence of Pareto goodness-of-fit conclusions on the specified tail region.

Introduction

Pareto models describe heavy-tailed data through a simple distributional form above a lower endpoint. Introduced by Pareto [32] to describe wealth, they are also used in economics [35], finance [34], insurance [15], and environmental science [20]. For a Pareto type-I random variable X∼P(α)X \sim\mathcal{P}(\alpha), where α>0\alpha> 0 is a shape parameter, the distribution function and probability density function are

F(x;α):={1−1xα,x≥10,otherwise,f(x;α):={αxα+1,x≥10,otherwise,α>0.(1)F(x;\alpha) := \begin{cases} 1-\frac{1}{x^\alpha}, & x \ge1 \\ 0, & \text{otherwise}, \end{cases} \qquad f(x;\alpha) := \begin{cases} \frac{\alpha}{x^{\alpha+1}}, & x \ge1 \\ 0, & \text{otherwise}, \end{cases} \qquad\alpha> 0. \tag*{(1)}

Let X1,…,XnX_1,\ldots,X_n be independent and identically distributed (iid) random variables with an unknown distribution function FF supported on [1,∞)[1,\infty). We test the composite null hypothesis

H0:F∈{P(α):α>0}H_0:F\in\{\mathcal{P}(\alpha):\alpha>0\}

for some unknown shape parameter α>0\alpha>0, against the alternative

H1:F∉{P(α):α>0},H_1:F\notin\{\mathcal{P}(\alpha):\alpha>0\},

where the decision is based on the data X1,…,XnX_1,\ldots,X_n. For a different positive known lower threshold xm>0x_m>0, the observations should first be divided by xmx_m.

Existing Pareto goodness-of-fit procedures use empirical distribution functions [10], probability plots [9], or empirical characteristic functions [27]. Characterization-based tests begin with an identity that distinguishes the Pareto family from other distributions; see [29, 7].

Stein’s method, originally developed for normal approximation [36], characterizes probability distributions through operator identities. Using this approach, Betsch and Ebner [6] derived fixed-point characterizations of continuous univariate distributions. For the Pareto type-I distribution, Avhad et al. [3] established a Stein-type characterization and proposed goodness-of-fit tests based on integral-type and Cramér–von Mises-type statistics for both complete and right-censored samples. More recently, Bhati and Khandelwal [8] combined the Laplace transform with the Stein-based fixed-point characterization of Betsch and Ebner [6]. They obtained a characterization of the Pareto type-I distribution and proposed supremum, integral, and L2L^2-type test statistics based on its empirical counterpart.

In the present paper, we use exponential test functions within the Stein density approach to obtain a Laplace-transform like characterization of the Pareto type-I distribution and construct two test statistics. Integrating the signed empirical discrepancy yields TnT_n, a simple moment-type diagnostic, whereas integrating its square yields SnS_n. Squaring before integration prevents cancellation between discrepancies at different transform arguments. This distinction has consequences for consistency of the procedures. The test based on TnT_n is blind against certain non-Pareto alternatives, whereas the test based on SnS_n is consistent against every fixed non-Pareto alternative satisfying the logarithmic moment condition stated below.

A studentized version of TnT_n has a standard normal limiting null distribution, permitting direct asymptotic calibration without resampling. In contrast, the limiting null distribution of nSnnS_n is an infinite weighted sum of independent χ12\chi_1^2 random variables. The weights are the eigenvalues of the covariance operator of the limiting Gaussian process and depend on the unknown shape parameter α0\alpha_0. We consider two methods for obtaining critical values. The first uses a parametric bootstrap, with the shape parameter re-estimated in each bootstrap sample. The second replaces α0\alpha_0 by α^n\widehat{\alpha}_n in the covariance operator, numerically approximates its eigenvalues, and uses the resulting weighted chi-square distribution to obtain approximate critical values. We compare the computational cost and finite-sample performance of these approaches in a simulation study. In the settings considered, the plug-in procedure is computationally faster, particularly for larger samples, whereas the bootstrap generally yields slightly higher empirical power for small samples.

The remainder of the article is organized as follows. Section 2 presents the characterization and constructs the test statistics. Section 3 derives their null limits, identifies the alternatives detected consistently by TnT_n, and establishes consistency and calibration results for SnS_n. § 4 compares empirical size and power with those of existing procedures. § 5 compares the two calibrations in terms of rejection rates and computation time. § 6 examines threshold sensitivity using Danish fire-insurance losses. § 7 discusses the implications and directions for further research.

New goodness-of-fit tests for the Pareto law

A Stein–Laplace characterization

We specialize the density approach to Stein’s method ([22, 21]) to the Pareto family ([6 Section 3]). Exponential test functions yield the following characterization.

Theorem 1. Let α>0\alpha> 0 and let XX be a positive random variable. Then XX has the Pareto type-I distribution P(α)\mathcal{P}(\alpha) if and only if, for every t>0t > 0,

αexp⁡(−t)−E[(t+α+1X)exp⁡(−tX)]=0.(2)\alpha\exp(-t) - \mathbb{E}\left[\left(t + \frac{\alpha+ 1}{X}\right) \exp(-tX)\right] = 0. \tag*{(2)}

Proof. Integration by parts proves the forward implication. For X∼P(α)X \sim\mathcal{P}(\alpha),

E[(t+α+1X)e−tX]=−∫1∞ddx{αe−txx−α−1}dx=αe−t.\mathbb{E}\left[\left(t + \frac{\alpha+ 1}{X}\right)e^{-tX}\right] = -\int_{1}^{\infty} \frac{\mathrm{d}}{\mathrm{d}x}\left\{\alpha e^{-tx}x^{-\alpha-1}\right\}\mathrm{d}x = \alpha e^{-t}.

For the converse, we identify the Laplace transform. Let φ(t):=E[e−tX]\varphi(t) := \mathbb{E}[e^{-tX}] and suppose (2) holds. The nonnegative terms inside the expectation in (2) are integrable. Tonelli’s theorem therefore gives

tφ(t)+(α+1)∫t∞φ(s) ds=αe−t.t\varphi(t) + (\alpha+ 1)\int_{t}^{\infty}\varphi(s)\,\mathrm{d}s = \alpha e^{-t}.

The Laplace transform φ\varphi is differentiable for t>0t > 0, since the function x↦xe−txx \mapsto xe^{-tx} is bounded for each such tt. Differentiation gives tφ′(t)−αφ(t)=−αe−tt\varphi'(t) - \alpha\varphi(t) = -\alpha e^{-t}. Multiplying by t−α−1t^{-\alpha-1} and integrating to infinity, using 0≤φ(t)≤10 \leq\varphi(t) \leq1, gives

φ(t)=αtα∫t∞u−α−1e−u du=∫1∞e−txαx−α−1 dx.\varphi(t) = \alpha t^\alpha\int_{t}^{\infty}u^{-\alpha-1}e^{-u}\,\mathrm{d}u = \int_{1}^{\infty}e^{-tx}\alpha x^{-\alpha-1}\,\mathrm{d}x.

The uniqueness theorem for Laplace transforms ([38 Chapter II]) identifies the law of XX.

The boundary term in Theorem 1 comes from the lower endpoint of the Pareto support. To see its relation to the density approach, write pα(x):=αx−α−11{x≥1}p_\alpha(x) := \alpha x^{-\alpha-1}\mathbb{1}_{\{x\geq1\}}, where 1A\mathbb{1}_A denotes the indicator of a set or event AA. On the interior of the support, the standard density-Stein operator ([6 Section 3]) acts on a test function gg as

(Aαg)(x):=g′(x)−α+1xg(x).(\mathcal{A}_\alpha g)(x) := g'(x) - \frac{\alpha+1}{x}g(x).

For suitable gg, integration by parts gives

∫1∞{g′(x)pα(x)+g(x)pα′(x)} dx=−αg(1).\int_{1}^{\infty}\{g'(x)p_\alpha(x) + g(x)p_\alpha'(x)\}\,\mathrm{d}x = -\alpha g(1).

Choosing the characterizing function class {gt(x)=e−tx:t>0}\{g_t(x) = e^{-tx}: t > 0\} yields (2).

Two ways to aggregate the discrepancy

Let X1,X2,…,XnX_1, X_2, \ldots, X_n be iid with distribution function FF on [1,∞)[1,\infty). We estimate the shape parameter α\alpha by the Pareto maximum likelihood estimator (MLE) α^n:=n/∑j=1nlog⁡Xj\widehat{\alpha}_n := n/\sum_{j=1}^{n}\log X_j, for details see Johnson et al. [19].

Remark 1. For alternatives that may be identical to 1 almost surely, we set α^n=1\widehat{\alpha}_n=1 on the event ∑jlog⁡Xj=0\sum_j\log X_j=0 and define the statistics using this convention. If E⁡F[log⁡X1]>0\operatorname{E}_F[\log X_1]>0, then F({1})<1F(\{1\})<1 and this event occurs only finitely often almost surely. This does not affect any limit below. Under the null hypothesis, it has probability zero.

We replace the expectation in (2) by the sample average and substitute α^n\widehat{\alpha}_n for α\alpha, which leads to the empirical discrepancy

Dn(t):=α^ne−t−1n∑j=1n(t+α^n+1Xj)e−tXj,t>0.(3)D_n(t):=\widehat{\alpha}_n e^{-t}-\frac{1}{n}\sum_{j=1}^{n}\left(t+\frac{\widehat{\alpha}_n+1}{X_j}\right)e^{-tX_j},\qquad t>0. \tag*{(3)}

Integrating gives

Tn:=∫0∞Dn(t) dt=α^n−(α^n+2)1n∑j=1nXj−2.(4)T_n:=\int_{0}^{\infty}D_n(t)\,\mathrm{d}t=\widehat{\alpha}_n-(\widehat{\alpha}_n+2)\frac{1}{n}\sum_{j=1}^{n}X_j^{-2}. \tag*{(4)}

This statistic reduces the transform identity to a single moment restriction. To retain discrepancies across all values of tt, we square before integrating and obtain

References

  1. [1]Allison, J., Milošević, B., Obradović, M., and Smuts, M. (2022). Distribution-free goodness-of-fit tests for the Pareto distribution based on a characterization. Computational Statistics, 37(1):403–418.DOI
  2. [3]Avhad, G. V., Kattumannil, S. K., and Lahiri, A. (2026). Goodness-of-fit tests for Pareto type-I distribution with complete and censored data. Journal of Statistical Computation and Simulation, 96(13):3322–3351.
  3. [4]Bader, B., Yan, J., and Zhang, X. (2018). Automated threshold selection for extreme value analysis via ordered goodness-of-fit tests with adjustment for false discovery rate. The Annals of Applied Statistics, 12(1):310–329.
  4. [5]Beran, R. (1997). Diagnosing bootstrap success. Annals of the Institute of Statistical Mathematics, 49(1):1–24.DOI
  5. [6]Betsch, S. and Ebner, B. (2021). Fixed point characterizations of continuous univariate probability distributions and their applications. Annals of the Institute of Statistical Mathematics, 73:31–59.
  6. [7]Bhati, D., Batsidis, A., and Khandelwal, S. (2025). New goodness of fit tests for the Pareto distribution using Stein’s characterization for uncensored and random right censored data. Communications in Statistics-Theory and Methods, 54(24):7865–7889.DOI
  7. [8]Bhati, D. and Khandelwal, S. (2026). Laplace transform driven Stein-type goodness-of-fit tests for Pareto distribution. arXiv preprint arXiv:2604.22486.arxiv.org/abs/2604.22486
  8. [9]Castillo, E., Hadi, A. S., Balakrishnan, N., and Sarabia, J. M. (2004). Extreme Value and Related Models with Applications in Engineering and Science. New Jersey: John Wiley & Sons, Hoboken, NJ.
  9. [10]Choulakian, V. and Stephens, M. A. (2001). Goodness-of-fit tests for the generalized Pareto distribution. Technometrics, 43(4):478–484.DOI
  10. [11]Clauset, A., Shalizi, C. R., and Newman, M. E. J. (2009). Power-law distributions in empirical data. SIAM Review, 51(4):661–703.
  11. [12]Drees, H., Janßen, A., Resnick, S. I., and Wang, T. (2020). On a minimum distance procedure for threshold selection in tail analysis. SIAM Journal on Mathematics of Data Science, 2(1):75–102.DOI
  12. [13]Duan, J. and Wang, W. (2014). Effective Dynamics of Stochastic Partial Differential Equations. Academic Press, Boston.
  13. [14]Ebner, B., Jiménez-Gamero, M. D., and Milošević, B. (2025). Efficient eigenvalue approximation in covariance operators via Rayleigh–Ritz with statistical applications. Statistical Papers, 66(6):135.DOI
  14. [15]Embrechts, P., Klüppelberg, C., and Mikosch, T. (1997). Modelling Extremal Events: for Insurance and Finance. Springer, Berlin, Heidelberg.
  15. [16]Grama, I. and Spokoiny, V. (2008). Statistics of extremes by oracle estimation. The Annals of Statistics, 36(4):1619–1648.
  16. [17]Henze, N. (2024). Asymptotic stochastics. An introduction with a view towards statistics, volume 10 of Math. Study Resour. Berlin: Springer.DOI
  17. [18]Janssen, A. (2000). Global power functions of goodness of fit tests. The Annals of Statistics, 28(1):239–253.DOI
  18. [19]Johnson, N. L., Kotz, S., and Balakrishnan, N. (1994). Continuous Univariate Distributions, volume 1. John Wiley & Sons, New York, 2 edition.
  19. [20]Katz, R. W., Parlange, M. B., and Naveau, P. (2002). Statistics of extremes in hydrology. Advances in Water Resources, 25(8-12):1287–1304.DOI
  20. [21]Ley, C., Reinert, G., and Swan, Y. (2017). Stein’s method for comparison of univariate distributions. Probability Surveys, 14:1–52.arxiv.org/abs/1408.2998
  21. [22]Ley, C. and Swan, Y. (2013). Stein’s density approach and information inequalities. Electronic Communications in Probability, 18:1–14. Paper No. 7.arxiv.org/abs/1210.3921
  22. [23]McNeil, A. J. (1997). Estimating the tails of loss severity distributions using extreme value theory. ASTIN Bulletin: The Journal of the IAA, 27(1):117–137.DOI
  23. [24]Meintanis, S. G. (2009). A unified approach of testing for discrete and continuous Pareto laws. Statistical Papers, 50:569–580.
  24. [25]Meintanis, S. G., Jiménez-Gamero, M. D., and Alba-Fernández, V. (2014). A class of goodness-of-fit tests based on transformation. Communications in Statistics-Theory and Methods, 43(8):1708–1735.DOI
  25. [26]Murphy, C., Tawn, J. A., and Varty, Z. (2025). Automated threshold selection and associated inference uncertainty for univariate extremes. Technometrics, 67(2):215–224.arxiv.org/abs/2310.17999
  26. [27]Ndwandwe, L., Allison, J., Smuts, M., and Visagie, I. (2023a). On a new class of tests for the Pareto distribution using Fourier methods. Stat, 12(1):e566.arxiv.org/abs/2301.09385
  27. [28]Ndwandwe, L., Allison, J. S., Santana, L., and Visagie, I. J. H. (2023b). Testing for the Pareto type I distribution: a comparative study. METRON, 81(2):215–256.arxiv.org/abs/2211.10088
  28. [29]Ngatchou-Wandji, J., Nombebe, T., Santana, L., and Allison, J. (2024). On classes of consistent tests for the Type I Pareto distribution based on a characterization involving order statistics. Statistics, 58(3):521–551.
  29. [30]Nyström, E. J. (1930). Über die praktische Auflösung von Integralgleichungen mit Anwendungen auf Randwertaufgaben. Acta Mathematica, 54:185–204.DOI
  30. [31]Obradović, M., Jovanović, M., and Milošević, B. (2015). Goodness-of-fit tests for Pareto distribution based on a characterization and their asymptotics. Statistics, 49(5):1026–1041.DOI
  31. [32]Pareto, V. (1896). Cours d’économie politique: professé à l’Université de Lausanne, volume 1. F. Rouge.
  32. [33]Pfaff, B., McNeil, A., Stephenson, A., and Zivot, E. (2018). evir: Extreme Values in R. R package version 1.7-4.
  33. [34]Rydberg, T. H. (2000). Realistic statistical modelling of financial data. International Statistical Review, 68(3):233–258.DOI
  34. [35]Simon, H. A. (1955). On a class of skew distribution functions. Biometrika, 42(3/4):425–440.DOI
  35. [36]Stein, C. (1972). A bound for the error in the normal approximation to the distribution of a sum of dependent random variables. In Le Cam, L. M., Neyman, J., and Scott, E. L., editors, Proceedings of the Sixth Berkeley Symposium on Mathematical Statistics and Probability, volume 2, pages 583–602, Berkeley. University of California Press.DOI
  36. [37]Van der Vaart, A. W. (1998). Asymptotic Statistics, volume 3 of Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, Cambridge.
  37. [38]Widder, D. V. (1941). The Laplace Transform. Princeton University Press, Princeton, NJ.

Paper details

vlnc:W_9bkpfk2g
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT 6 Astra

View statement
Looking for related papers…
Contents