betaFeedback

WorksMathematics

Statistics of the Shepp pp-product

Abstract

In 1962, Shepp famously discovered a product of normal random variables that preserves normality. The Shepp product, which takes the form XY/X2+Y2XY/\sqrt{X^2 + Y^2}, has since been thoroughly studied and has found numerous connections to other areas of statistics. Among other things, it has an extension to nn normal variables, gives a multiplicative analogue of the central limit theorem, and applies rather unexpectedly to genomics as a test statistic for alignment-free sequence analysis. The Shepp product is evidently the p=2p =2 special case of XY/Xp+YppXY/\sqrt[p]{X^p + Y^p} that we call the Shepp pp-product. We will show that the Shepp pp-product, particularly when p=1p = 1 and p=∞p = \infty (the latter in a limiting sense), is no less fascinating and applicable than the original p=2p = 2 case. Just as the Shepp 22-product preserves normal distributions, the Shepp 11-product preserves Cauchy distributions, and the Shepp ∞\infty-product preserves exponential distributions. In fact, the converse is also true in an appropriate sense, allowing us to characterize the Cauchy, normal, and exponential distributions as the unique distributions preserved by the Shepp pp-product for p=1,2,∞p =1,2,\infty, respectively. We will study the multiplicative analogue of infinite divisibility with respect to the Shepp pp-product, establish an asymptotic theory for the Shepp pp-product of nn i.i.d.\ random variables, and estimate the rates of convergence in Kolmogorov distance. Some of our results are new even for the p=2p=2 case. We will also discuss new applications of the Shepp pp-product in statistics, computational biology, and statistical physics. In particular, our new test statistic for exponentiality based on the Shepp ∞\infty-product is more powerful, both theoretically and empirically, than the rule-of-thumb Kolmogorov–Smirnov test based on the relative efficiency.

Introduction

It is well-known that the straightforward product XYXY of two independent normal variables XX and YY is not normally distributed. On the other hand, if X∼N(0,σ12)X\sim N(0,\sigma_1^2) and Y∼N(0,σ22)Y\sim N(0,\sigma_2^2) are independent, then a product of the form

Z=XYX2+Y2(1)Z=\frac{XY}{\sqrt{X^2+Y^2}} \tag*{(1)}

satisfies Z∼N(0,σ2)Z\sim N(0,\sigma^2) with

1σ=1σ1+1σ2,\frac{1}{\sigma}=\frac{1}{\sigma_1}+\frac{1}{\sigma_2},

a beautiful result first discovered by Shepp [37] and that has since found numerous different proofs [3, 27, 30]. Shepp’s product has many fascinating properties: a converse holds in an appropriate sense [25 Theorem 1]; it may be extended to nn independent Xi∼N(0,σi2)X_i\sim N(0,\sigma_i^2), i=1,…,ni=1,\ldots,n,

S(X1,…,Xn):=X1⋯Xn∑Xi12⋯Xin−12∼N(0,σ2),1σ=1σ1+⋯+1σn,(2)S(X_1,\ldots,X_n):=\frac{X_1\cdots X_n}{\sqrt{\sum X_{i_1}^2\cdots X_{i_{n-1}}^2}}\sim N(0,\sigma^2),\qquad\frac{1}{\sigma}=\frac{1}{\sigma_1}+\cdots+\frac{1}{\sigma_n}, \tag*{(2)}

where the sum in the denominator is over all 1≤i1<i2<⋯<in−1≤n1\le i_1<i_2<\cdots<i_{n-1}\le n; and in addition it satisfies a central limit theorem: For any i.i.d. random variables XiX_i, i∈Ni\in\mathbb{N}, i.e., not necessarily normal, nS(X1,…,Xn)nS(X_1,\ldots,X_n) is asymptotically normally distributed as n→∞n\to\infty [28].

Shepp’s product also has various applications, and not just in statistics, but in areas like computational biology and statistical physics too. We briefly review some of these as background for our

2020 Mathematics Subject Classification. 62E20, 60E07, 62E10, 62P10, 62P35.

Key words and phrases. Shepp product, stable distributions, extreme order statistics, characterization of distributions, multiplicative central limit theorem, rates of convergence. proposed applications in this article. In statistics, the Shepp product SS may be used to construct scale-invariant normality tests [42]: For i.i.d. X1,…,XnX_1,\ldots,X_n, let UnU_n and VnV_n be the UU- and VV-statistical empirical distribution functions associated with SS, i.e.,

Un(t)=2n(n−1)∑1≤i<j≤nI{S(Xi,Xj)<t/2},Vn(t)=1n2∑i,j=1nI{S(Xi,Xj)<t/2},U_n(t)=\frac{2}{n(n-1)}\sum_{1\le i<j\le n}\mathbb{I}_{\{S(X_i,X_j)<t/2\}},\qquad V_n(t)=\frac{1}{n^2}\sum_{i,j=1}^{n}\mathbb{I}_{\{S(X_i,X_j)<t/2\}},

and let FnF_n denote the empirical distribution function. Then the integral and Kolmogorov statistics may be defined by

IUn:=∫R(Vn(t)−Fn(t)) dFn(t),DUn:=sup⁡t∈R∣Un(t)−Fn(t)∣IU_n:=\int_{\mathbb{R}}\left(V_n(t)-F_n(t)\right)\,dF_n(t),\qquad DU_n:=\sup_{t\in\mathbb{R}}\left|U_n(t)-F_n(t)\right|

respectively. In particular, IUnIU_n is scale-free, unlike the Kolmogorov–Smirnov test. The asymptotic behaviors and the power of these test statistics may be found in [42].

In genomics, a statistic based on Shepp product is used to test the difference between two cDNA sequences in alignment-free sequence analysis [31]:

D2S=∑w∈AkXwYwXw2+Yw2.D_2^S=\sum_{w\in A^k}\frac{X_wY_w}{\sqrt{X_w^2+Y_w^2}}.

Here, XwX_w and YwY_w are centralized count variables measuring the frequency of word ww in the two sequences, and AkA^k denotes the set of words of length kk in the alphabet AA. Under some mild conditions, D2SD_2^S converges to the normal distribution under the null hypothesis. This has been extended to the nn-variate Shepp product (2) for multiple sequence comparison and is used to conduct extensive simulations [33]. We refer interested readers to [40] for a review of the role of Shepp’s product in alignment-free sequence analysis.

The Shepp product is also a special case of a Wald statistic under a singular hypothesis [13, 6]. Suppose we have estimators θ^1\hat{\theta}_1 and θ^2\hat{\theta}_2 for unknown parameters θ1\theta_1 and θ2\theta_2 respectively. We are interested in testing H0:θ1θ2=0H_0:\theta_1\theta_2=0 under the singular condition θ1=θ2=0\theta_1=\theta_2=0. If we have n((θ^1,θ^2)−(θ1,θ2))→dN(0,I2)\sqrt{n}((\hat{\theta}_1,\hat{\theta}_2)-(\theta_1,\theta_2))\xrightarrow{d}N(0,I_2), a Wald statistic can be constructed to test H0H_0. Surprisingly, it converges to the Shepp product given θ1=θ2=0\theta_1=\theta_2=0:

Wald statistic:=nθ^1θ^2θ^12+θ^22→dXYX2+Y2,X,Y→iidN(0,1).\text{Wald statistic}:=\sqrt{n}\frac{\hat{\theta}_1\hat{\theta}_2}{\sqrt{\hat{\theta}_1^2+\hat{\theta}_2^2}}\xrightarrow{d}\frac{XY}{\sqrt{X^2+Y^2}},\qquad X,Y\xrightarrow{\mathrm{iid}}N(0,1).

Because of the singular hypothesis, the asymptotic distribution of the Wald statistic is no longer N(0,1)N(0,1) but N(0,14)N(0,\frac{1}{4}), and, indeed, this follows from the distribution theory of the Shepp product. A consequence is that the standard Wald test is conservative when the true parameters are zero.

The goal of our article is to show that the Shepp pp-product, defined by

Sp(X,Y):={XY(Xp+Yp)1/p,1≤p<∞,1max⁡(1/X,1/Y),p=∞,(3)S_p(X,Y):= \begin{cases} \frac{XY}{(X^p+Y^p)^{1/p}}, & 1\le p<\infty,\\[6pt] \frac{1}{\max(1/X,1/Y)}, & p=\infty, \end{cases} \tag*{(3)}

is no less fascinating and useful than Shepp’s original product in (1), which corresponds to the case p=2p=2 and will henceforth be called the Shepp 2-product. We should perhaps mention that while it is the point-of-view undertaken in this article, it is unusual for the expression in (1) to be regarded as a product of random variables. Nevertheless, this perspective is convenient as we may easily show that the Shepp pp-product in (3) is associative for all p∈[1,∞]p\in[1,\infty] (it is obviously commutative); this in turn allows us to unambiguously define a Shepp pp-product of nn random variables Sp(X1,…,Xn)S_p(X_1,\ldots,X_n), which agrees with (2) when p=2p=2. We will study the Shepp pp-product with an emphasis on its asymptotic behavior and discuss several applications, particularly for the cases p=1p = 1 and ∞\infty. Observe that for p∈[1,∞)p \in[1,\infty),

1Sp(X1,…,Xn)p=∑i=1n1Xipand1S∞(X1,…,Xn)=max⁡i=1,…,n(1Xi),(4)\frac{1}{S_p(X_1,\ldots,X_n)^p}=\sum_{i=1}^{n}\frac{1}{X_i^p}\quad\text{and}\quad\frac{1}{S_\infty(X_1,\ldots,X_n)}=\max_{i=1,\ldots,n}\left(\frac{1}{X_i}\right), \tag*{(4)}

which connects Shepp pp-product to sum of reciprocals for finite pp, and extreme order statistics for p=∞p = \infty. The latter may be analyzed using standard distribution theory, but the former is not well-studied — reciprocals are typically heavy-tailed, and their i.i.d. sum does not exhibit the usual asymptotic normality with n−1/2n^{-1/2}-scaling. Our Shepp pp-product opens a door for the study of such sums using the theory of stable distribution for finite pp, and the theory of extreme order statistics for p=∞p = \infty. We summarize some of our key findings and several new applications below.

Characterization of Cauchy, normal, and exponential distributions

References

  1. [1]S. Ascher. A survey of tests for exponentiality. Comm. Statist. Theory Methods, 19(5):1811–1825, 1990.DOI
  2. [2]N. Bansal, G. G. Hamedani, E. S. Key, H. Volkmer, H. Zhang, and J. Behboodian. Some characterizations of the normal distribution. Statist. Probab. Lett., 42(4):393–400, 1999.DOI
  3. [3]L. Baringhaus, N. Henze, and D. Morgenstern. Some elementary proofs of the normality of XY/(X² + Y²)¹ᐟ² when X and Y are normal. Comput. Math. Appl., 15(11):943–944, 1988.
  4. [4]P. J. Bickel and D. A. Freedman. Some asymptotic theory for the bootstrap. Ann. Statist., 9(6):1196–1217, 1981.DOI
  5. [5]L. H. Y. Chen, L. Goldstein, and Q.-M. Shao. Normal approximation by Stein’s method. Probability and its Applications (New York). Springer, Heidelberg, 2011.DOI
  6. [6]M. Drton and H. Xiao. Wald tests of singular hypotheses. Bernoulli, 22(1):38–59, 2016.
  7. [7]M. Falk. Rates of uniform convergence of extreme order statistics. Ann. Inst. Statist. Math., 38(2):245–262, 1986.DOI
  8. [8]P. Forster, L. Forster, C. Renfrew, and M. Forster. Phylogenetic network analysis of sars-cov-2 genomes. Proceedings of the National Academy of Sciences, 117(17):9241–9243, 2020.DOI
  9. [9]M. R. Fréchet. Sur la loi de probabilité de l’écart maximum. Ann. Soc. Polon. Math., 6:93–116, 1927.
  10. [10]J. Galambos. The asymptotic theory of extreme order statistics. John Wiley & Sons, New York-Chichester-Brisbane, 1978. Wiley Series in Probability and Mathematical Statistics.
  11. [11]J. Galambos and I. Simonelli. Comments on a recent limit theorem of Quine: “A central limit theorem for self-normalized products of random variables” [Statist. Probab. Lett. 43 (1999), no. 2, 137–143; MR1693273 (2000j:60033)] by M. P. Quine. Statist. Probab. Lett., 63(1):89–95, 2003.DOI
  12. [12]J. Galambos and I. Simonelli. Characterizations of probability distributions by properties of products of random variables. J. Appl. Statist. Sci., 13(1):1–10, 2004.
  13. [13]G. F. V. Glonek. On the behaviour of Wald statistics for the disjunction of two regular hypotheses. J. Roy. Statist. Soc. Ser. B, 55(3):749–755, 1993.DOI
  14. [14]B. V. Gnedenko and A. N. Kolmogorov. Limit distributions for sums of independent random variables. Addison-Wesley Mathematics Series. Addison-Wesley, Cambridge, MA, 1954. Translated and annotated by K. L. Chung. With an Appendix by J. L. Doob. MR:0062975. Zbl:0056.36001.
  15. [15]P. Hall. Two-sided bounds on the rate of convergence to a stable law. Z. Wahrsch. Verw. Gebiete, 57(3):349–364, 1981.DOI
  16. [16]G. G. Hamedani, H. Volkmer, and J. Wesolowski. Characterization problems related to the Shepp property. Comm. Statist. Theory Methods, 36(5-8):1049–1057, 2007.DOI
  17. [17]W. Hoeffding. A class of statistics with asymptotically normal distribution. Ann. Math. Statistics, 19:293–325, 1948.
  18. [18]O. Johnson and R. Samworth. Central limit theorem and convergence to stable laws in Mallows distance. Bernoulli, 11(5):829–845, 2005.arxiv.org/abs/math/0406218
  19. [19]D. Li and T. D. Rogers. Asymptotic behavior for iterated functions of random variables. Ann. Appl. Probab., 9(4):1175–1201, 1999.DOI
  20. [20]E. Lukacs. Characteristic functions. Griffin’s Statistical Monographs & Courses, No. 5. Hafner Publishing Co., New York, 1960.
  21. [21]J. Mijnheer. On the rate of convergence to a stable limit law. II. Litovsk. Mat. Sb., 26(3):482–487, 1986.DOI
  22. [22]Y. Nikitin. Asymptotic efficiency of nonparametric tests. Cambridge University Press, Cambridge, 1995.
  23. [23]Y. Y. Nikitin. Large deviations of U-empirical Kolmogorov-Smirnov tests and their efficiency. J. Nonparametr. Stat., 22(5-6):649–668, 2010.arxiv.org/abs/0906.0428
  24. [24]Y. Y. Nikitin. Tests based on characterizations, and their efficiencies: a survey. Acta Comment. Univ. Tartu. Math., 21(1):3–24, 2017.DOI
  25. [25]S. Y. Novak. A new characterization of the normal law. Statist. Probab. Lett., 77(1):95–98, 2007.DOI
  26. [26]N. S. Pillai and X.-L. Meng. An unexpected encounter with Cauchy and Lévy. Ann. Statist., 44(5):2089–2097, 2016.arxiv.org/abs/1505.01957
  27. [27]M. P. Quine. A result of Shepp. Appl. Math. Lett., 7(6):33–34, 1994.DOI
  28. [28]M. P. Quine. A central limit theorem for self-normalized products of random variables. Statist. Probab. Lett., 43(2):137–143, 1999.DOI
  29. [29]S. T. Rachev. The Monge-Kantorovich problem on mass transfer and its applications in stochastics. Teor. Veroyatnost. i Primen., 29(4):625–653, 1984.DOI
  30. [30]J. G. Reid. Normal functions of normal random variables. Comput. Math. Appl., 14(3):157–160, 1987.DOI
  31. [31]G. Reinert, D. Chew, F. Sun, and M. S. Waterman. Alignment-free sequence comparison (I): Statistics and power. J. Comput. Biol., 16(12):1615–1634, 2009.DOI
  32. [32]R.-D. Reiss. Approximate distributions of order statistics. Springer Series in Statistics. Springer-Verlag, New York, 1989. With applications to nonparametric statistics.DOI
  33. [33]J. Ren, K. Song, F. Sun, M. Deng, and G. Reinert. Multiple alignment-free sequence comparison. Bioinformatics, 29(21):2690–2698, 2013.DOI
  34. [34]G. Samorodnitsky and M. S. Taqqu. Stable non-Gaussian random processes. Stochastic Modeling. Chapman & Hall, New York, 1994. Stochastic models with infinite variance.
  35. [35]R. L. Schilling and L. Partzsch. Brownian motion. De Gruyter, Berlin, 2012. An introduction to stochastic processes, With a chapter on simulation by Björn Böttcher.
  36. [36]E. Seneta. Regularly varying functions. Lecture Notes in Mathematics, Vol. 508. Springer-Verlag, Berlin-New York, 1976.DOI
  37. [37]L. Shepp. Problem 62-9, normal functions of normal random variables. SIAM Rev. Soc. Ind. Appl. Math., 4(3):255–256, 1962.
  38. [38]B. W. Silverman. Convergence of a class of empirical distribution functions of dependent random variables. Ann. Probab., 11(3):745–751, 1983.DOI
  39. [39]R. L. Smith. Uniform rates of convergence in extreme-value theory. Adv. in Appl. Probab., 14(3):600–622, 1982.DOI
  40. [40]K. Song, J. Ren, G. Reinert, M. Deng, M. S. Waterman, and S. Fengzhu. New developments of alignment-free sequence comparison: measures, statistics and next-generation sequencing. Brief. Bioinform., 15(3):343–353, 2013.DOI
  41. [41]C. Villani. Topics in optimal transportation, volume 58. American Mathematical Soc., 2021.
  42. [42]K. Y. Volkova and Y. Y. Nikitin. On the asymptotic efficiency of normality tests based on the Shepp property. Vestnik St. Petersburg Univ. Math., 42(4):256–261, 2009.DOI
  43. [43]W. Weibull. A statistical distribution function of wide applicability. J. Appl. Mech., 18:293–297, 1951.DOI

Paper details

vlnc:W_w4g4h2e6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 19

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents