A cylindrical distribution with heavy-tailed linear part

Abstract

A cylindrical distribution whose linear part models heavy-tailedness is proposed. The conditional distribution of the linear variable given the circular variable is a generalized Pareto-type distribution. Therefore, it may not have any conditional moments; however, the mode and median have closed-form expressions. The circular marginal distribution is a wrapped Cauchy distribution, and the conditional distribution of the circular variable given the linear variable belongs to a family of symmetric distributions. These properties allow its application to cylindrical data, whose linear observations may take large values and whose circular observations are symmetric. As illustrative examples, the proposed distribution is fitted to two data sets, and the results are compared with those by other cylindrical distributions that cannot model heavy-tailedness for the linear parts.

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7

References

  1. Aakala, T., Shimatani, K., Abe, T., Kubota, Y., & Kuuluvainen, T. (2016). Crown asymmetry in high latitude forests: Disentangling the directional effects of tree competition and solar radiation. Oikos, 125, 1035–1043.

    Article  Google Scholar 

  2. Abe, T., & Ley, C. (2017). A tractable, parsimonious and flexible model for cylindrical data, with applications. Econometrics and Statistics, 4, 91–104.

    MathSciNet  Article  Google Scholar 

  3. Abe, T., & Pewsey, A. (2011). Sine-skewed circular distributions. Statistical Papers, 52, 683–707.

    MathSciNet  Article  Google Scholar 

  4. Abe, T., & Shimatani, I. K. (2018). Cylindrical distributions and their applications to biological data. In C. Ley & T. Verdebout (Eds.), Applied directional statistics: Modern methods and case studies. Boca Raton: Chapman and Hall/CRC.

    Google Scholar 

  5. Banerjee, A., Dhillon, I. S., Ghosh, J., & Sra, S. (2005). Clustering on the unit hypersphere using von Mises–Fisher distributions. Journal of Machine Learning Research, 6, 1345–1382.

    MathSciNet  MATH  Google Scholar 

  6. Batschelet, E. (1981). Circular statistics in biology. London: Academic Press.

    Google Scholar 

  7. Crovella, M. E., & Taqqu, M. S. (1999). Estimating the heavy tail index from scaling properties. Methodology and Computing in Applied Probability, 1, 55–79.

    MathSciNet  Article  Google Scholar 

  8. Fisher, N. I. (1993). Statistical analysis of circular data. Cambridge: Cambridge University Press.

    Google Scholar 

  9. Fisher, N. I., & Lee, A. J. (1992). Regression models for an angular response. Biometrics, 48, 665–677.

    MathSciNet  Article  Google Scholar 

  10. Foss, S., Korshunov, D., & Zachary, S. (2011). An introduction to heavy-tailed and subexponential distributions. New York: Springer.

    Google Scholar 

  11. Gradshteyn, I. S., & Ryzhik, I. M. (2007). Table of integrals, series, and products (7th ed.). Burlington: Academic Press.

    Google Scholar 

  12. Johnson, R. A., & Wehrly, T. E. (1977). Measures and models for angular correlation and angular-linear correlation. Journal of the Royal Statistical Society: Series B, 39, 222–229.

    MathSciNet  MATH  Google Scholar 

  13. Johnson, R. A., & Wehrly, T. E. (1978). Some angular-linear distributions and related regression models. Journal of the American Statistical Association, 73, 602–606.

    MathSciNet  Article  Google Scholar 

  14. Kareiva, P. M., & Shigesada, N. (1983). Analyzing insect movement as a correlated random walk. Oecologia, 56, 234–238.

    Article  Google Scholar 

  15. Kato, S., & Shimizu, K. (2008). Dependent models for observations which include angular ones. Journal of Statistical Planning and Inference, 138, 3538–3549.

    MathSciNet  Article  Google Scholar 

  16. Lagona, F., Picone, M., & Maruotti, A. (2015). A hidden Markov model for the analysis of cylindrical time series. Environmetrics, 26, 534–544.

    MathSciNet  Article  Google Scholar 

  17. Ley, C., & Verdebout, T. (2017). Modern directional statistics. Boca Raton: CRC Press.

    Google Scholar 

  18. Mardia, K. V. (1976). Linear-circular correlation coefficients and rhythmometry. Biometrika, 63, 403–405.

    Article  Google Scholar 

  19. Mardia, K. V., & Jupp, P. E. (2000). Directional statistics. Chichester: Wiley.

    Google Scholar 

  20. Mardia, K. V., & Sutton, T. W. (1978). A model for cylindrical variables with applications. Journal of the Royal Statistical Society: Series B, 40, 229–233.

    MATH  Google Scholar 

  21. Maruotti, A. (2016). Analyzing longitudinal circular data by projected normal models: A semi-parametric approach based on finite mixture models. Environmental and Ecological Statistics, 23, 257–277.

    MathSciNet  Article  Google Scholar 

  22. Nuñez-Antonio, G., & Gutiérrez-Peña, E. (2014). A Bayesian model for longitudinal circular data based on the projected normal distribution. Computational Statistics and Data Analysis, 71, 506–519.

    MathSciNet  Article  Google Scholar 

  23. Pearson, K. (1905). The problem of the random walk. Nature, 72, 294.

    Article  Google Scholar 

  24. Pewsey, A. (2002). Testing circular symmetry. Canadian Journal of Statistics, 30, 591–600.

    MathSciNet  Article  Google Scholar 

  25. Rivest, L. P., Duchesne, T., Nicosia, A., & Fortin, D. (2016). A general angular regression model for the analysis of data on animal movement in ecology. Journal of the Royal Statistical Society: Series C, 65, 445–463.

    MathSciNet  Article  Google Scholar 

  26. Shimizu, K., & Iida, K. (2002). Pearson type VII distributions on spheres. Communications in Statistics-Theory and Methods, 31, 513–526.

    MathSciNet  Article  Google Scholar 

  27. Wang, F., & Gelfand, A. E. (2014). Modeling space and space-time directional data using projected Gaussian processes. Journal of the American Statistical Association, 109, 1565–1580.

    MathSciNet  Article  Google Scholar 

  28. Wang, F., Gelfand, A. E., & Jona-Lasinio, G. (2015). Joint spatio-temporal analysis of a linear and a directional variable: Space-time modeling of wave heights and wave directions in the Adriatic Sea. Statistica Sinica, 25, 25–39.

    MathSciNet  MATH  Google Scholar 

Download references

Acknowledgements

Tomoaki Imoto was supported in part by JSPS KAKENHI Grant Number 18K13459. Toshihiro Abe was supported in part by Nanzan University of Pache Research Subsidy I-A-1 for the 2018 academic year.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Tomoaki Imoto.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Appendix

A Generalized Gamma mixture

It is assumed that \((X, \varTheta )'|V \sim\) AL\((V^{-1}, \delta , \mu , \kappa )\) and \(V \sim\) GGa\(\left( \alpha , \beta , 1/\delta \right)\). Then the unconditional density \(g(x, \theta )\) of \((X, \varTheta )'\) is derived as follows:

$$\begin{aligned} g(x, \theta )= & {} \int _0^{\infty } f_{\mathrm{Wei}}(x, \theta |v)f_\mathrm{GGa}(v) \mathrm{d}v\\= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2\pi \delta ^2} \frac{\beta ^{\alpha }}{\varGamma (\alpha )}x^{1/\delta -1} \int _{0}^{\infty } v^{\alpha /\delta + 1/\delta - 1} e^{ -v^{1/\delta } \left[ \beta + x^{1/\delta } \left\{ 1- \kappa \cos (\theta - \mu ) \right\} \right] } \mathrm{d}v. \end{aligned}$$

The substitution \(v=w^{\delta }\) yields

$$\begin{aligned} g(x, \theta )= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2\pi \delta ^2} \frac{\beta ^{\alpha }}{\varGamma (\alpha )}x^{1/\delta -1} \int _{0}^{\infty } w^{\alpha - \delta + 1} e^{ -w \left[ \beta + x^{1/\delta } \left\{ 1- \kappa \cos (\theta - \mu ) \right\} \right] } \delta w^{\delta -1}\mathrm{d}w\\= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2\pi \delta } \frac{\beta ^{\alpha }}{\varGamma (\alpha )}x^{1/\delta -1} \int _{0}^{\infty } w^{\alpha } e^{ -w \left[ \beta + x^{1/\delta } \left\{ 1- \kappa \cos (\theta - \mu ) \right\} \right] } \mathrm{d}w\\= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2\pi \delta } \frac{\beta ^{\alpha }}{\varGamma (\alpha )}x^{1/\delta -1} \frac{\varGamma (\alpha +1)}{\left[ \beta + x^{1/\delta } \left\{ 1- \kappa \cos (\theta - \mu ) \right\} \right] ^{\alpha +1}}\\= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2\pi \delta } \frac{\alpha }{\beta } x^{1/\delta -1} \left[ 1 + \frac{x^{1/\delta }}{\beta } \left\{ 1- \kappa \cos (\theta - \mu ) \right\} \right] ^{-(\alpha +1)}. \end{aligned}$$

Here, by replacing \(\alpha =\delta /\tau\) and \(\beta =\delta \sigma ^{1/\delta }/\tau\), it follows that \(g(x, \theta )\) is equal to the proposed density (2).

B Modality

The first derivatives of the logarithm of the density (2) are given by

$$\begin{aligned} \begin{array}{l} \displaystyle \frac{\partial \log f_{\mathrm{GPar}}(x, \theta )}{\partial x} = \frac{1 - \delta }{ \delta x} - \frac{ x^{1/\delta -1} (\delta + \tau ) \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} }{ \delta \left[ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} \right] }, \\ \\ \displaystyle \frac{\partial \log f_{\mathrm{GPar}}(x, \theta )}{\partial \theta } = - \frac{ \kappa x^{1/\delta } (\delta + \tau ) \sin (\theta - \mu ) }{ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} }. \end{array} \end{aligned}$$

When \(\delta >1\), as there is no stationary point on \((0,\infty )\times [0,2\pi )\), and \(\displaystyle \lim _{x \rightarrow 0} f_{\mathrm{GPar}}(x,\theta ) = \infty\), the density (2) is unimodal and the mode is taken on the line \(x=0\).

When \(\delta <1\), the stationary points of \(\log f_{\mathrm{GPar}}(x, \theta )\) on \((0, \infty )\times [0, 2\pi )\) are

$$\begin{aligned} (x,\theta )=\left( \sigma \left\{ \frac{1-\delta }{(1-\kappa )(1+\tau )} \right\} ^{\delta }, \mu \right) =(x_1,\theta _1) \end{aligned}$$

and

$$\begin{aligned} (x,\theta )=\left( \sigma \left\{ \frac{1-\delta }{(1+\kappa )(1+\tau )} \right\} ^{\delta }, \mu +\pi \ (\mathrm{mod}\ 2\pi ) \right) =(x_2,\theta _2). \end{aligned}$$

The second derivatives of the logarithm of the density (2) are given by

$$\begin{aligned} \frac{\partial ^2 \log f_{\mathrm{GPar}}(x, \theta )}{\partial x^2}= & {} -\frac{1-\delta }{\delta ^2 x^2} \Biggl ( \delta + \frac{ x^{1/\delta } (\delta + \tau ) \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} }{ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} }\\&-\,\frac{ x^{2/\delta } \tau (\delta + \tau ) \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} ^2 }{ (1-\delta )\left[ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} \right] ^2 } \Biggl ), \\ \frac{\partial ^2 \log f_{\mathrm{GPar}}(x, \theta )}{\partial \theta ^2}= & {} - \frac{ \kappa x^{1/\delta } (\delta + \tau ) \cos (\theta - \mu ) }{ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} }\\&+\, \frac{ \kappa ^2 x^{2/\delta } \tau (\delta + \tau ) \sin ^2 (\theta - \mu ) }{ \left[ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} \right] ^2},\\ \frac{\partial ^2 \log f_{\mathrm{GPar}}(x, \theta )}{\partial x \partial \theta }= & {} - \frac{\kappa x^{1/\delta -1} (\delta + \tau ) \sin (\theta - \mu )}{ \delta \left[ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} \right] }\\&+\, \frac{ \kappa x^{2/\delta -1} \tau (\delta + \tau ) \sin (\theta - \mu )\left\{ 1-\kappa \cos (\theta -\mu )\right\} }{ \delta \left[ \sigma ^{1/\delta } \delta + \tau x^{1/\delta } \left\{ 1 - \kappa \cos (\theta - \mu ) \right\} \right] ^2}, \\ \end{aligned}$$

and thus, the Hessian matrices at \((x,\theta )=(x_1,\theta _1)\) and at \((x,\theta )=(x_2,\theta _2)\) are given by

$$\begin{aligned} H_1= \left( \begin{array}{cc} -\frac{(1-\delta )(1+\tau )}{\delta (\delta +\tau ) x_1^2} &{} 0 \\ 0 &{} -\frac{\kappa (1-\delta )}{1-\kappa } \end{array}\right) \quad \text{ and }\quad H_2= \left( \begin{array}{cc} -\frac{(1-\delta )(1+\tau )}{\delta (\delta +\tau ) x_2^2} &{} 0 \\ 0 &{} \frac{\kappa (1-\delta )}{1+\kappa } \end{array}\right) , \end{aligned}$$

respectively. As \(\sigma >0\), \(\tau >0\), and \(0 \le \kappa <1\), the matrix \(H_1\) is negative-definite and the matrix \(H_2\) is positive-definite. Therefore, the density (2) is unimodal, and the mode is taken at \((x, \theta ) = (x_1,\theta _1)\).

When \(\delta =1\), as there is no stationary point, and \(f_\mathrm{GPar}(0, \theta _1)=f_{\mathrm{GPar}}(0, \theta _2)\) and

$$\begin{aligned} f_{\mathrm{GPar}}(0,\theta _1)-f_\mathrm{GPar}(x,\theta _2)=\frac{\sqrt{1-\kappa ^2}}{2 \pi \sigma } \left[ 1-\left[ \frac{1}{1+\tau x\left\{ 1-\kappa \cos (\theta _2-\mu )/\sigma \right\} } \right] ^{1/\tau +1}\right] >0 \end{aligned}$$

for arbitrary \(\theta _1\), \(\theta _2\), and \(x>0\), the density (2) is unimodal, and the mode is taken on the line \(x=0\).

C Calculations about marginal distributions

The marginal density \(f_{\varTheta }(\theta )\) of \(\varTheta\) is obtained as follows:

$$\begin{aligned} f_{\varTheta }(\theta )= & {} \int _{0}^{\infty } f_{\mathrm{GPar}}(x,\theta ) \mathrm{d}x \\= & {} \frac{ \sqrt{1-\kappa ^2} }{2\pi \sigma \delta } \int _{0}^{\infty } \left( \frac{x}{\sigma } \right) ^{1/\delta -1} \left[ 1+\frac{\tau }{\delta } \left( \frac{x}{\sigma } \right) ^{1/\delta } \left\{ 1 - \kappa \cos (\theta -\mu ) \right\} \right] ^{-(\delta /\tau +1)} \mathrm{d}x. \end{aligned}$$

The substitution \(x=\sigma (\delta y/\tau )^{\delta }\) yields

$$\begin{aligned} f_{\varTheta }(\theta ) = \frac{ \sqrt{1 - \kappa ^2} }{2\pi } \int _{0}^{\infty } \left[ 1+y \left\{ 1 - \kappa \cos (\theta -\mu ) \right\} \right] ^{-(\delta /\tau +1)} \mathrm{d}y = \frac{ \sqrt{1 - \kappa ^2} }{ 2 \pi \{ 1-\kappa \cos (\theta - \mu ) \} } \end{aligned}$$

by the integral formula \(\int _0^{\infty }(1+\alpha y)^{-(\beta +1)}\mathrm{d}y=1/(\alpha \beta )\) for \(\alpha , \beta >0\).

As the density (2) is expressed by

$$\begin{aligned} f_{\mathrm{GPar}}(x, \theta )= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2 \pi \sigma \delta } \left( \frac{x}{\sigma } \right) ^{1/\delta -1} \left\{ 1 + \frac{\tau }{\delta } \left( \frac{x}{\sigma } \right) ^{1/\delta } \right\} ^{-(\delta /\tau +1)} \\&\times \, \left\{ 1 - \frac{ \kappa \tau (x/\sigma )^{1/\delta } }{ \delta + \tau (x/\sigma )^{1/\delta } } \cos (\theta -\mu ) \right\} ^{-(\delta /\tau +1)} \end{aligned}$$

and the combination of Equations 9.112 and 9.134.2 in Gradshteyn and Ryzhik (2007) yields

$$\begin{aligned} \frac{1}{2\pi }\int _0^{2\pi }\frac{\mathrm{d}t}{(1-z \cos t)^p} ={}_2F_1\left( \frac{p}{2},\frac{p+1}{2};1;z^2 \right) , \quad |z|<1, \end{aligned}$$

the marginal density \(f_X(x)\) of X is

$$\begin{aligned} f_{X}(x)= & {} \int _{0}^{2\pi } f_{\mathrm{GPar}}(x,\theta ) \mathrm{d}\theta \\= & {} \frac{ \sqrt{1-\kappa ^2} }{ 2 \pi \sigma \delta } \left( \frac{x}{\sigma } \right) ^{1/\delta -1} \left\{ 1 + \frac{\tau }{\delta } \left( \frac{x}{\sigma } \right) ^{1/\delta } \right\} ^{-(\delta /\tau +1)}\\&\times \, \int _{0}^{2\pi }\left\{ 1 - \frac{ \kappa \tau (x/\sigma )^{1/\delta } }{ \delta + \tau (x/\sigma )^{1/\delta } } \cos (\theta -\mu ) \right\} ^{-(\delta /\tau +1)} \mathrm{d}\theta \\= & {} \frac{\sqrt{1-\kappa ^2}}{\sigma \delta } \left( \frac{x}{\sigma } \right) ^{1/\delta - 1} \left\{ 1 + \frac{\tau }{\delta } \left( \frac{x}{\sigma } \right) ^{1/\delta } \right\} ^{-(\delta /\tau + 1)} \\&\times \, {}_2F_1 \left( \frac{\delta /\tau + 1}{2}, \frac{\delta /\tau }{2}+1; 1; \left( \frac{\kappa (x /\sigma )^{1/\delta }}{\delta /\tau + (x/\sigma )^{1/\delta }} \right) ^2 \right) . \end{aligned}$$

As

$$\begin{aligned}&{}_2F_1 \left( \frac{\delta /\tau + 1}{2}, \frac{\delta /\tau }{2}+1; 1; \left( \frac{\kappa (x/\sigma )^{1/\delta }}{\delta /\tau + (x/\sigma )^{1/\delta }} \right) ^2 \right) \\&\quad = 1 + \sum _{k=1}^{\infty } \frac{\{\delta /(2\tau ) + k - 1/2\} \cdots \{\delta /(2\tau ) + 1/2\}}{ \{\delta /\tau + (x/\sigma )^{1/\delta }\}^k } \frac{\{\delta /(2\tau ) + k - 1\} \cdots \{\delta /(2\tau ) + 1\}}{ \{\delta /\tau + (x/\sigma )^{1/\delta }\}^k } \frac{ \{\kappa (x/\sigma )^{1/\delta }\}^{2k} }{(k!)^2} \\&\quad \rightarrow \sum _{k=0}^{\infty } \frac{ \{\kappa (x/\sigma )^{1/\delta }\}^{2k} }{2^{2k} (k!)^2} = I_0 \left( \kappa \left( \frac{x}{\sigma } \right) ^{1/\delta } \right) ,\quad \text{ as }\quad \tau \rightarrow 0, \end{aligned}$$

it is confirmed that

$$\begin{aligned} f_X(x) \rightarrow \frac{\sqrt{1-\kappa ^2}}{\sigma \delta } \left( \frac{x}{\sigma } \right) ^{1/\delta - 1} e^{ - (x/\sigma )^{1/\delta } } I_0 \left( \kappa \left( \frac{x}{\sigma } \right) ^{1/\delta } \right) , \quad \text{ as }\quad \tau \rightarrow 0. \end{aligned}$$

D Joint moments

It is assumed that \((X, \varTheta )'|V \sim\) AL\((V^{-1}, \delta , \mu , \kappa )\) and \(V \sim\) GGa\(\left( \alpha , \beta , 1/\delta \right)\). Then, as \(\mathrm{E} \left[ X^p e^{i q \varTheta } | V=v \right] =v^{-p} A_{p, q} e^{i q \mu }\) and the unconditional distribution of \((X, \varTheta )'\) is the proposed distribution (2), the joint moments are obtained by

$$\begin{aligned} \mathrm{E} \left[ X^p e^{i q \varTheta } \right]= & {} \int _0^{\infty } \mathrm{E} \left[ X^p e^{i q \varTheta } | V=v \right] f_{\mathrm{GGa}}(v) \mathrm{d}v \\= & {} A_{p, q} e^{i q \mu } \int _0^{\infty } \frac{\beta ^{\alpha }}{\varGamma (\alpha ) \delta } v^{\alpha /\delta - p -1} e^{-\beta v^{1/\delta }} \mathrm{d}v \\= & {} A_{p, q} e^{i q \mu } \int _0^{\infty } \frac{\beta ^{\alpha }}{\varGamma (\alpha )} w^{\alpha -p \delta -1} e^{-\beta w} \mathrm{d}w \\= & {} A_{p, q} \beta ^{p\delta }\frac{\varGamma \left( \alpha - p \delta \right) }{\varGamma (\alpha )} e^{i q \mu },\quad \alpha -p\delta >0. \end{aligned}$$

In the third equation, the substitution \(v=w^{\delta }\) was made. By replacing \(\alpha =\delta /\tau\) and \(\beta =\delta \sigma ^{1/\delta }/\tau\), the above joint moments yield (9).

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Imoto, T., Shimizu, K. & Abe, T. A cylindrical distribution with heavy-tailed linear part. Jpn J Stat Data Sci 2, 129–154 (2019). https://doi.org/10.1007/s42081-019-00031-5

Download citation

Keywords

  • Earthquake
  • Generalized Pareto distribution
  • Heavy-tailed distribution
  • Wrapped Cauchy distribution

Mathematics Subject Classification

  • 60E05
  • 62H11