Skip to main content

Estimating standard errors in regular vine copula models

Abstract

We describe a new algorithm for the computation of the score function and observed information in regular vine (R-vine) copula models. R-vine copulas are constructed hierarchically from bivariate copulas as building blocks only, and the algorithm exploits this hierarchical nature for subsequent computation of log-likelihood derivatives. This allows to routinely estimate standard errors of parameter estimates, and overcomes reliability and accuracy issues associated with numerical differentiation in multidimensional models. Results obtained using the proposed methods are discussed in the context of the asymptotic efficiency of different estimation methods and of an application to exchange rate data.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2

Notes

  1. 1.

    We use the adaptive integration routines supplied by Steven G. Johnson and Balasubramanian Narasimhan in the cubature package available on CRAN which are based on Genz and Malik (1980) and Berntsen et al. (1991).

References

  1. Aas K, Czado C, Frigessi A, Bakken H (2009) Pair-copula construction of multiple dependence. Insur Math Econ 44:182–198

    MathSciNet  Article  MATH  Google Scholar 

  2. Bates D, Maechler M (2012) Matrix: sparse and dense matrix classes and methods. http://CRAN.R-project.org/package=Matrix. R package version 1.0-6

  3. Bedford T, Cooke R (2001) Probability density decomposition for conditionally dependent random variables modeled by vines. Ann Math Artif Intell 32:245–268

    MathSciNet  Article  Google Scholar 

  4. Bedford T, Cooke R (2002) Vines—a new graphical model for dependent random variables. Ann Stat 30:1031–1068

    MathSciNet  Article  MATH  Google Scholar 

  5. Berntsen J, Espelid TO, Genz A (1991) An adaptive algorithm for the approximate calculation of multiple integrals. ACM Trans Math Soft 17(4):437–451

    MathSciNet  Article  MATH  Google Scholar 

  6. Czado C (2010) Pair-copula constructions of multivariate copulas. In: Jaworski P, Durante F, Härdle WK, Rychlik T (eds) Copula theory and its applications. Lecture notes in statistics, vol 198, pp 93–109. Springer, Berlin

  7. Czado C, Schepsmeier U, Min A (2012) Maximum likelihood estimation of mixed C-vines with application to exchange rates. Stat Model 12(3):229–255

    Article  Google Scholar 

  8. Dißmann J (2010) Statistical inference for regular vines and application. Diploma Thesis. Technische Universität München, Germany, Faculty of Mathematics

  9. Dißmann J, Brechmann E, Czado C, Kurowicka D (2013) Selecting and estimating regular vine copulae and application to financial returns. Comput Stat Data Anal 59:52–69

    Article  Google Scholar 

  10. Genest C, Ghoudi K, Rivest LP (1995) A semiparametric estimation procedure of dependence parameters in multivariate families of distributions. Biometrika 82:543–552

    Google Scholar 

  11. Genz AC, Malik AA (1980) An adaptive algorithm for numeric integration over an n-dimensional rectangular region. J Comput Appl Math 6(4):295–302

    Article  MATH  Google Scholar 

  12. Higham NJ (2002) Computing the nearest correlation matrix—a problem from finance. IMA J Numer Anal 22:329–343

    MathSciNet  Article  MATH  Google Scholar 

  13. Hobæk Haff I (2013) Parameter estimation for pair-copula constructions. Bernoulli 19(2):462–491

    MathSciNet  Article  MATH  Google Scholar 

  14. Isserlis L (1918) On a formula for the product-moment coefficient of any order of a normal frequency distribution in any number of variables. Biometrika 12(1/2):134–139

    Article  Google Scholar 

  15. Joe H (1996) Families of m-variate distributions with given margins and m(m-1)/2 bivariate dependence parameters. In: Rüschendorf L, Schweizer B, Taylor MD (eds) Distributions with fixed marginals and related topics, vol 28. Inst. Math. Statist, Hayward, pp 120–141

    Chapter  Google Scholar 

  16. Joe H, Xu JJ (1996) The estimation method of inference functions of margins for multivariate models. Technical report 166, Department of Statistics, University of British Columbia

  17. Kurowicka D, Cooke R (2006) Uncertainty analysis with high dimensional dependence modeling. Wiley series in probability and statistics. Wiley, Chichester

    Book  Google Scholar 

  18. Mardia KV, Marshall RJ (1984) Maximum likelihood estimation of models for residual covariance in spatial regression. Biometrika 71(1):135–146

    MathSciNet  Article  MATH  Google Scholar 

  19. McCullough BD (1998) Assessing the reliability of statistical software: part I. Am Stat 52(4):358–366

    Google Scholar 

  20. McCullough BD (1999) Assessing the reliability of statistical software: part II. Am Stat 53(2):149–159

    Google Scholar 

  21. Morales-Nápoles O, Cooke RM, Kurowicka D (2010) About the number of vines and regular vines on n nodes (preprint, personal communication)

  22. Panagiotelis A, Czado C, Joe H (2012) Pair copula constructions for multivariate discrete data. J Am Stat Assoc 107(499):1063–1072

    MathSciNet  Article  MATH  Google Scholar 

  23. Schepsmeier U (2010) Maximum likelihood estimation of C-vine pair-copula constructions on bivariate copulas from different families. Diploma thesis, Center of Mathematical Sciences, Munich University of Technology, Garching bei München

  24. Schepsmeier U, Stöber J (2013) Derivatives and fisher information of bivariate copulas. Stat Papers. doi:10.1007/s00362-013-0498-x

  25. Smith M, Min A, Almeida C, Czado C (2010) Modelling longitudinal data using a pair-copula decomposition of serial dependence. J Am Stat Assoc 105(492):1467–1479

    MathSciNet  Article  Google Scholar 

  26. Smith MD (2007) Invariance theorems for Fisher information. Commun Stat Theory Methods 36(12):2213–2222

    Article  MATH  Google Scholar 

  27. Stöber J, Czado C (2013) Detecting regime switches in the dependence structure of high dimensional financial data. Comput Stat Data Anal (in press)

  28. Stöber J, Joe H, Czado C (2013) Simplified pair copula constructions—limitations and extensions. J Multivar Anal 119:101–118

    Google Scholar 

Download references

Acknowledgments

Numerical calculations were performed on a Linux cluster supported by DFG grant INST 95/919-1 FUGG. The second author gratefully acknowledges the support of the TUM Graduate School’s International School of Applied Mathematics, the first author is supported by TUM’s TopMath program and a research stipend provided by Allianz Deutschland AG.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Jakob Stöber.

Appendices

Appendix 1: Algorithm for the calculation of second derivatives

In Sect. 4 we introduced the seven possible cases of dependence which can occur during the calculation of the second log-likelihood derivative. In the following, we illustrate these cases in detail. In case 1 we determine

$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \gamma )\right) \right) \nonumber \\&\quad = \frac{\partial _1 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \gamma )\right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \gamma )\right) } \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ) \right) \nonumber \\&\qquad \times \left( \frac{\partial }{\partial \gamma } F_{V \vert \mathbf Z }(v \vert \mathbf z , \gamma ) \right) \nonumber \\&\quad \quad - \left( \frac{\partial }{\partial \theta } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \gamma )\right) \right) \right) \nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \gamma )\right) \right) \right) , \end{aligned}$$
(24)

for case 2

$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) \right) \nonumber \\&\quad = \frac{\partial _1 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) } \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ) \right) \nonumber \\&\qquad \times \left( \frac{\partial }{\partial \gamma } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\qquad - \left( \frac{\partial }{\partial \theta } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) \right) \right) \nonumber \\&\qquad \times \left( \frac{\partial }{\partial \gamma } \ln (\left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) \right) \right) \nonumber \\&\qquad + \frac{\partial _2 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) } \times \left( \frac{\partial }{\partial \theta } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\qquad \times \left( \frac{\partial }{\partial \gamma } F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\qquad + \frac{ \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) } \times \left( \frac{\partial ^2}{\partial \theta \partial \gamma } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) ,\nonumber \\ \end{aligned}$$
(25)

and case 3 yields

$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln (c_{U,V\vert \mathbf Z }(F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ))) \nonumber \\&\quad = \frac{\partial _1 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }\nonumber \\&\qquad \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ) \right) \left( \frac{\partial }{\partial \gamma } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\quad \quad - \left( \frac{\partial }{\partial \theta } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) \right) \right) \nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \gamma } \ln (\left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) \right) \right) \nonumber \\&\quad \quad + \frac{\partial _1 \partial _1 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }\nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ) \right) \left( \frac{\partial }{\partial \gamma } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\quad \quad + \frac{\partial _1 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }\nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \theta } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) \left( \frac{\partial }{\partial \gamma } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\quad \quad + \frac{\partial _1 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }\nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ) \right) \left( \frac{\partial }{\partial \gamma } F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\quad \quad + \frac{\partial _2 \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }\nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \theta } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) \left( \frac{\partial }{\partial \gamma } F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\quad \quad + \frac{ \partial _1 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) } \times \left( \frac{\partial ^2}{\partial \theta \partial \gamma } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) \nonumber \\&\qquad + \frac{ \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) }{c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta , \gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma )\right) } \times \left( \frac{\partial ^2}{\partial \theta \partial \gamma } F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta , \gamma ) \right) .\nonumber \\ \end{aligned}$$
(26)

Similarly, we have for case 4 that

$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \gamma \right) \right) \nonumber \\&\quad = \left( \frac{\partial }{\partial \theta } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \gamma \right) \right) \right) \nonumber \\&\quad \quad \times \frac{- \partial _\gamma c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \gamma \right) } \nonumber \\&\quad \quad + \frac{\partial _\gamma \partial _1 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \gamma \right) } \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta ) \right) , \end{aligned}$$
(27)

and

$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v \vert \mathbf z , \theta )\vert \gamma \right) \right) \nonumber \\&\quad = \left( \frac{\partial }{\partial \theta } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) \right) \right) \nonumber \\&\quad \quad \times \frac{- \partial _\gamma c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) } \nonumber \\&\quad \quad + \frac{\partial _\gamma \partial _1 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) } \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta ) \right) \nonumber \\&\quad \quad + \frac{\partial _\gamma \partial _2 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ), F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta )\vert \gamma \right) } \times \left( \frac{\partial }{\partial \theta } F_{V \vert \mathbf Z }(v\vert \mathbf z , \theta ) \right) ,\quad \quad \end{aligned}$$
(28)

for the fifth case.Finally,

$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) \right) \nonumber \\&\quad = \frac{\partial _1 \partial _1 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) }\nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \theta } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta ,\gamma ) \right) \times \left( \frac{\partial }{\partial \gamma } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta ,\gamma ) \right) \nonumber \\&\quad \quad - \left( \frac{\partial }{\partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) \right) \right) \nonumber \\&\quad \quad \times \left( \frac{\partial }{\partial \theta } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) \right) \right) \nonumber \\&\quad \quad + \frac{\partial _1 c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z , \theta ,\gamma ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \right) }\nonumber \\&\quad \quad \times \left( \frac{\partial ^2}{\partial \gamma \partial \theta } F_{U \vert \mathbf Z }(u\vert \mathbf z , \theta ,\gamma ) \right) ,\end{aligned}$$
(29)
$$\begin{aligned}&\frac{\partial ^2}{\partial \theta \partial \gamma } \ln \left( c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \vert \theta ,\gamma \right) \right) \nonumber \\&\quad = \frac{\partial _\theta \partial _\gamma c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \vert \theta ,\gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \theta ,\gamma \right) } \nonumber \\&\quad \quad - \frac{\partial _\theta c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u ), F_{V \vert \mathbf Z }(v \vert \mathbf z ) \vert \theta ,\gamma \right) \times \partial _\gamma c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z ), F_{V \vert \mathbf Z }(v) \vert \theta ,\gamma \right) }{ c_{U,V\vert \mathbf Z }\left( F_{U \vert \mathbf Z }(u \vert \mathbf z ), F_{V \vert \mathbf Z }(v \vert \mathbf z )\vert \theta ,\gamma \right) ^2}.\nonumber \\ \end{aligned}$$
(30)

Appendix 2: Calculation of the covariance matrix in the Gaussian case

While analytical results on the Fisher information for the multivariate normal distribution are well known (Mardia and Marshall 1984) we will now illustrate how the matrices \(\varvec{\mathcal{K}}_{\theta }\) and \(\varvec{\mathcal{J}}_{\theta }\) (Eqs. 7 and 8) can be calculated. We consider a 3-dimensional Gaussian distribution

$$\begin{aligned} \begin{pmatrix} X_1\\ X_2\\ X_3\\ \end{pmatrix} \sim N_3(\varvec{0},\varvec{\varSigma }), \ \ \ \varvec{\varSigma } = \begin{pmatrix} 1 &{}\quad \rho _{12} &{}\quad \rho _{13} \\ \rho _{12} &{}\quad 1 &{}\quad \rho _{23} \\ \rho _{13} &{}\quad \rho _{23} &{}\quad 1 \\ \end{pmatrix}, \end{aligned}$$

with density \(f_{123}\) and corresponding copula \(c_{123}\). Exampli gratia, we show the computation for the entry \((2,1)\) in \(\varvec{\mathcal{K}}_{\theta }\) in detail. The other entries in \(\varvec{\mathcal{K}}_{\theta }\) and \(\varvec{\mathcal{J}}_{\theta }\) are obtained similarly. The first step is to calculate the following integral:

$$\begin{aligned}&\int \limits _{[0,1]^{3}} \left( \frac{\partial }{\partial \rho _{12}} \ln (c_{12}(u_1,u_2|\rho _{12}))\right) \nonumber \\&\quad \times \left( \frac{\partial }{\partial \rho _{23}} \ln (c_{23}(u_2,u_3|\rho _{23}))\right) c_{123}(u_1,u_2,u_3) du_1du_2 du_3, \end{aligned}$$
(31)

where \(c_{12}\) and \(c_{23}\) are the corresponding copulas to the bivariate marginal distributions \(f_{12}\) and \(f_{23}\), respectively. Since the integral is independent of the univariate marginal distributions, we can compute it using standard normal margins (see Smith 2007):

$$\begin{aligned}&\int \limits _\mathbb{R ^3} \left( \frac{\partial }{\partial \rho _{12}} \ln (f_{12}(x_1,x_2|\rho _{12}))\right) \nonumber \\&\quad \times \left( \frac{\partial }{\partial \rho _{23}} \ln (f_{23}(x_2,x_3|\rho _{23}))\right) f_{123}(x_1,x_2,x_3) dx_1dx_2dx_3, \end{aligned}$$
(32)

where \(f_{12}\) and \(f_{23}\) are the according bivariate normal distributions. The 3-dimensional and bivariate normal densities in (31) and (32) can be expressed as

$$\begin{aligned}&f_{123}(x_1,x_2,x_3) = \frac{\sqrt{2}}{{\pi }^{3/2}\sqrt{2\,\rho _{13}\rho _{12}\rho _{23}-{\rho _{13}}^{2}-{\rho _{12}}^{2}+1-{\rho _{23}}^{2}}} \nonumber \\&\quad \times \exp \Bigg \{-\frac{1}{2}{\frac{-{x_{{1}}}^{2}\!-\!{x_{{2}}}^{2}\!-\!{x_{{3}}}^{2} \!+\!{x_{{1}}}^{2}{\rho _{23}}^{2} \!+\!{x_{{2}}}^{2}{\rho _{13}}^{2} \!+\!{x_{{3}}}^{2}{\rho _{12}}^{2} \!+\!2\,x_{{1}}x_{{2}}\rho _{12} \!+\!2\,x_{{1}}x_{{3}}\rho _{13} }{-2\,\rho _{13}\rho _{12}\rho _{23}\!+\!{\rho _{13}}^{2} \!+\!{\rho _{12}}^{2}\!-\!1\!+\!{\rho _{23}}^{2}}}\Bigg \}\nonumber \\&\quad \times \exp \Bigg \{\frac{ +2\,x_{{2}}x_{{3}}\rho _{23} -2\,x_{{1}}x_{{2}}\rho _{13}\rho _{23} -2\,x_{{1}}x_{{3}}\rho _{12}\rho _{23} -2\,x_{{2}}x_{{3}}\rho _{13}\rho _{12}}{-2\,\rho _{13}\rho _{12}\rho _{23} +{\rho _{13}}^{2}+{\rho _{12}}^{2}-1+{\rho _{23}}^{2}}\Bigg \}, \end{aligned}$$
(33)

and

$$\begin{aligned} f_{12}(x_1,x_2) = \frac{1}{2\pi }{\frac{1}{\sqrt{1-{\rho _{12}}^{2}}}} \,{\exp \left\{ -\frac{1}{2}\,{\frac{-{x_{{1}}}^{2}+2\,x_{{1}}x_{{2}}\rho _{12} -{x_{{2}}}^{2}}{ \left( -1+\rho _{12} \right) \left( \rho _{12}+1 \right) }}\right\} }. \end{aligned}$$
(34)

Further, the derivatives needed in Eq. (32) are

$$\begin{aligned} \frac{\partial }{\partial \rho _{12}} \ln (f_{12}(x_1,x_2|\rho _{12})) = -{\frac{ {\rho _{12}}^{3}-x_{{1}}x_{{2}}{\rho _{12}}^{2}+{x_{{2}}}^{2}\rho _{12} -\rho _{12}+{x_{{1}}}^{2}\rho _{12}-x_{{1}}x_{{2}} }{ \left( -1+\rho _{12} \right) ^{2} \left( \rho _{12}+1 \right) ^{2}} },\nonumber \\ \end{aligned}$$
(35)
figured
figuree

and

$$\begin{aligned} \frac{\partial }{\partial \rho _{23}} \ln (f_{23}(x_2,x_3|\rho _{23})) \!=\! -{\frac{ {\rho _{23}}^{3}\!-\!x_{{2}}x_{{3}}{\rho _{23}}^{2}\!+\!{x_{{3}}}^{2}\rho _{23} \!-\!\rho _{23}\!+\!{x_{{2}}}^{2}\rho _{23}-x_{{2}}x_{{3}} }{ \left( -1+\rho _{23} \right) ^{2} \left( \rho _{23}\!+\!1 \right) ^{2}} }.\qquad \end{aligned}$$
(36)

Using (33), (35) and (36) in (32) we get

$$\begin{aligned} (32)&= \int \limits _\mathbb{R ^3} {\frac{ {\rho _{12}}^{3}-x_{{1}}x_{{2}}{\rho _{12}}^{2}+{x_{{2}}}^{2}\rho _{12} -\rho _{12}+{x_{{1}}}^{2}\rho _{12}-x_{{1}}x_{{2}} }{ \left( -1+\rho _{12} \right) ^{2} \left( \rho _{12}+1 \right) ^{2}} } \nonumber \\&\times {\frac{ {\rho _{23}}^{3}-x_{{2}}x_{{3}}{\rho _{23}}^{2}+{x_{{3}}}^{2}\rho _{23} -\rho _{23}+{x_{{2}}}^{2}\rho _{23}-x_{{2}}x_{{3}} }{ \left( -1+\rho _{23} \right) ^{2} \left( \rho _{23}+1 \right) ^{2}} }\nonumber \\&\times f_{123}(x_1,x_2,x_3) dx_1dx_2dx_3. \end{aligned}$$
(37)

The integral (37) can be solved using well known results on product moments of multivariate normal distributions (see Isserlis 1918).

$$\begin{aligned} (37)&= \frac{ \rho _{23}{\rho _{12}}^{3}+{\rho _{12}}^{3}{\rho _{23}}^{3} -3\,{\rho _{23}}^{2}{\rho _{12}}^{2}\rho _{13}-{\rho _{12}}^{2}\rho _{13} +2\,\rho _{23}{\rho _{13}}^{2}\rho _{12} }{ \left( \rho _{23}+1 \right) ^{2} \left( -1+\rho _{23} \right) ^{2} \left( -1+{\rho _{12}}^{2} \right) ^{2} } \nonumber \\&\quad +\frac{-\rho _{12}\rho _{23}+{\rho _{23}}^{3} \rho _{12}+\rho _{13}-{\rho _{23}}^{2}\rho _{13}}{ \left( \rho _{23}+1 \right) ^{2} \left( -1+\rho _{23} \right) ^{2} \left( -1+{\rho _{12}}^{2} \right) ^{2} } \end{aligned}$$
(38)

Since \(\left( \rho _{23}+1 \right) ^{2} \left( -1+\rho _{23} \right) ^{2} = (1-\rho _{23}^2)^2\) we can simplify Eq. (38) to

$$\begin{aligned} (31) = (38)&= \frac{(\rho _{13}-\rho _{12}\rho _{23})(1-\rho _{12}^2)(1-\rho _{23}^2) + 2\rho _{12}\rho _{23}(\rho _{13}-\rho _{12}\rho _{23})^2 }{ (1-\rho _{12}^2)^2(1-\rho _{23}^2)^2 } \\&= \frac{\rho _{13}-\rho _{12} \rho _{23}}{(1-\rho _{12}^2)(1-\rho _{23}^2)} +2\rho _{12}\rho _{23}\frac{(\rho _{13}-\rho _{12}\rho _{23})^2}{(1-\rho _{12}^2)^2(1-\rho _{23}^2)^2} \\&= \frac{k_{12}}{(1-\rho _{12}^2)(1-\rho _{23}^2)}, \end{aligned}$$

with

$$\begin{aligned} k_{12} = (\rho _{13}-\rho _{12}\rho _{23})\left( 1+2\rho _{12}\rho _{23} \frac{\rho _{13}-\rho _{12}\rho _{23}}{(1-\rho _{12}^2)(1-\rho _{23}^2)} \right) . \end{aligned}$$

For the computation of terms corresponding to parameter \(\rho _{13|2}\), note that

$$\begin{aligned} \rho _{13|2} = \frac{\rho _{13} - \rho _{12}\rho _{23}}{\sqrt{(1-\rho _{12}^2)(1-\rho _{23}^2)}}, \end{aligned}$$

and

$$\begin{aligned} \rho _{13} = \rho _{13|2}\sqrt{(1-\rho _{12}^2)(1-\rho _{23}^2)} + \rho _{12}\rho _{23}, \end{aligned}$$

which means that (33), (35) and (36) can easily be re-parametrized.

The final matrices are

$$\begin{aligned} \varvec{\mathcal{K}}_{\theta }&= \begin{pmatrix} \frac{1+\rho _{12}^2}{(1-\rho _{12}^2)^2} &{} \frac{k_{12}}{(1-\rho _{12}^2)(1-\rho _{23}^2)} &{} 0 \\ \frac{k_{12}}{(1-\rho _{12}^2)(1-\rho _{23}^2)} &{} \frac{1+\rho _{23}^2}{(1-\rho _{23}^2)^2} &{} 0 \\ 0 &{} 0 &{} \frac{1+\rho _{13|2}}{(\rho _{13|2}^2-1)^2} \\ \end{pmatrix},\\ \varvec{\mathcal{J}}_{\theta }&= \begin{pmatrix} \frac{1+\rho _{12}^2}{(1-\rho _{12}^2)^2} &{} 0 &{} 0 \\ 0 &{} \frac{1+\rho _{23}^2}{(1-\rho _{23}^2)^2} &{} 0 \\ \frac{\rho _{13|2}\rho _{12}}{(\rho _{12}^2-1)(\rho _{13|2}^2-1)} &{} \frac{\rho _{13|2}\rho _{23}}{(\rho _{23}^2-1)(\rho _{13|2}^2-1)} &{} \frac{1+\rho _{13|2}}{(\rho _{13|2}^2-1)^2} \end{pmatrix}. \end{aligned}$$

Appendix 3: Selected model for the exchange rate data

To obtain marginally uniformly distributed copula data on \([0,1]^8\), we conduct a pre-analysis as described in (Schepsmeier (2010), Chapter 5). AR(1)-GARCH(1,1) models (Table 3) are selected for the marginal time series, and the resulting standardized residuals are transformed using the non-parametric rank transformation (see Genest et al. 1995). We could also employ the probability integral transformation based on the parametric error distributions (IFM, Joe and Xu 1996) but since we are only interested in dependence properties here, we choose the non-parametric alternative which is more robust with respect to misspecification of marginal error distributions.

Table 3 Parameters of the AR(1)-GARCH(1,1) models

The R-vine describing the exchange rate data set is specified by the structure matrix \(M\), the copula family matrix \(\mathcal B \) and the estimated copula parameter matrix \(\hat{\varvec{\theta }}^{MLE}\). For simplicity, we use the following abbreviations: 1 = AUD (Australian dollar), 2 = JPY (Japanese yen), 3 = BRL (Brazilian real), 4 = CAD (Canadian dollar), 5 = EUR (Euro), 6 = CHF (Swiss frank), 7 = INR (Indian rupee) and 8 = GBP (British pound).

The pair-copula families in the application were chosen from the elliptical copulas Gauss and Student’s \(t\) copula, the Archimedean Clayton, Gumbel, Frank and Joe copula, and their rotated versions. The selection is done using to AIC/BIC.

$$\begin{aligned} M&= \begin{pmatrix} 8\\ 7 &{}\quad 7 \\ 2 &{}\quad 2 &{}\quad 6 \\ 3 &{}\quad 3 &{}\quad 2 &{}\quad 5 \\ 6 &{}\quad 4 &{}\quad 3 &{}\quad 2 &{}\quad 4 \\ 4 &{}\quad 1 &{}\quad 4 &{}\quad 3 &{}\quad 2 &{}\quad 3 \\ 1 &{}\quad 5 &{}\quad 1 &{}\quad 4 &{}\quad 3 &{}\quad 2 &{}\quad 2 \\ 5 &{}\quad 6 &{}\quad 5 &{}\quad 1 &{}\quad 1 &{}\quad 1 &{}\quad 1 &{}\quad 1\\ \end{pmatrix}\end{aligned}$$
(39)
$$\begin{aligned} \hat{\varvec{\theta }}^{MLE}&= \begin{pmatrix} \, &{}\quad &{}\quad &{}\quad &{}\quad &{} \quad &{}\quad 11.85 &{}\quad 8.96 \\ &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 7.74 \\ &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 3.76 \\ &{}\quad -0.70 &{}\quad -0.09 &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 9.97 \\ &{}\quad -0.88 &{}\quad -1.44 &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 8.41 \\ 0.07 &{}\quad -1.10 &{}\quad -0.73 &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 7.46 \\ 0.26 &{}\quad -1.23 &{}\quad -1.17 &{}\quad 1.13 &{}\quad 1.08 &{}\quad 0.63 &{}\quad &{}\quad \\ 0.72 &{}\quad 0.55 &{}\quad 0.88 &{}\quad 0.63 &{}\quad 0.54 &{}\quad 0.48 &{}\quad 0.3 &{}\quad \\ \end{pmatrix}\end{aligned}$$
(40)
$$\begin{aligned} \mathcal B&= \begin{pmatrix} \, &{} \\ \text{ Indep. }&{} \\ \text{ Indep. } &{} \text{ Indep. }\\ \text{ Indep. } &{} \text{ Frank } &{} \text{ Gauss } \\ \text{ Indep. } &{} \text{ Frank } &{} \text{ Frank } &{} \text{ Indep. } \\ \text{ Gauss } &{} \text{ r. } \text{ Joe } &{} \text{ Frank } &{} \text{ Indep. } &{} \text{ Indep. } \\ \text{ Student's } \text{-t } &{} \text{ r. } \text{ Gumbel } &{} \text{ r. } \text{ Gumbel } &{} \text{ Gumbel } &{} \text{ Frank } &{} \text{ Frank } \\ \text{ Student's } \text{ t } &{} \text{ Student's } \text{ t } &{} \text{ Student's } \text{ t } &{} \text{ Student's } \text{ t } &{} \text{ Student's } \text{ t } &{} \text{ Student's } \text{ t } &{} \text{ Gauss }\\ \end{pmatrix}\nonumber \\ \end{aligned}$$
(41)

The standard errors corresponding to the parameters in (40) are

$$\begin{aligned} {\varvec{SE}}_n^{MLE} = \begin{pmatrix} \, &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 5.00 &{}\quad 2.54 \\ &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 1.95 \\ &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 0.59 \\ &{}\quad 0.19 &{}\quad 0.03 &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 3.04 \\ &{}\quad 0.19 &{}\quad 0.20 &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 2.40 \\ 0.03 &{}\quad 0.03 &{}\quad 0.19 &{}\quad &{}\quad &{}\quad &{}\quad &{}\quad 2.01 \\ 0.03 &{}\quad 0.03 &{}\quad 0.03 &{}\quad 0.02 &{}\quad 0.19 &{}\quad 0.19 &{}\quad &{}\quad \\ 0.01 &{}\quad 0.02 &{}\quad 0.01 &{}\quad 0.02 &{}\quad 0.02 &{}\quad 0.02 &{}\quad 0.03 &{}\quad \\ \end{pmatrix}. \end{aligned}$$
(42)

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Stöber, J., Schepsmeier, U. Estimating standard errors in regular vine copula models. Comput Stat 28, 2679–2707 (2013). https://doi.org/10.1007/s00180-013-0423-8

Download citation

Keywords

  • Copula
  • Exchange rates
  • R-vine
  • Standard errors