<r>If you have a multivariate distribution it consist of a copula and marginal distributions so yes you are implicitly using a copula model. I have - here on Wilmott.com - given the following answers as to why using copulas. Maybe it also interesting for you so I repeat part of it here: As already n...

<r>Thank you both.I have researched it a bit further and it seems like it belongs to the family of generalised hyperbolic distributions (GH) see: <URL url="http://en.wikipedia.org/wiki/Generalised_hyperbolic_distribution"><LINK_TEXT text="http://en.wikipedia.org/wiki/Generalise ... stribution">http:...

A random variable X is normal distributed. I'm looking for the name and form for the distribution 1/X if it even has a name. RegardsJonas

<t>If you have ML estimate for each family of distributions then use AIC (as a rule of thumb) for distinguish between them. I know this is not a direct answer to your question but maybe Matlab can give you AIC for each fit and then you just have to choose the distribution family which has the lowest...

See also http://www.wilmott.com/messageview.cfm? ... =25426Hope it helps/Jonas

<t>To see the difference between the frank copula and the gaussian try to enhance the parameter of the frank copula. To get more intuition remember to look at the distribution function - not the copula in isolation (which in itself is a distribution function). I have attach a picture showing 4 2-dim...

<r>Do you want to calculate probability, density or maybe simulate from a Gaussian copula? Or is it really a multivariate probability distribution you are interested in and not the copula par se? See also <URL url="http://www.wilmott.com/messageview.cfm?catid=8&threadid=25426"><LINK_TEXT text="h...

<r>Do you have to implement the gaussian copula yourself? Otherwise there is an open source statistical software package called R that has an implementation for the Gaussian, t and certain Archimedean copulas. This library is relatively easy to use. If R isn’t installed, download R from <URL url="ht...

<t>Are there any standards in the in the industry with regard to managing option risk? Are all the greeks in use or only some and are they aggregated to a single risk number - for example VaR? Any pointers to relevant literature would be appreciated. I’m especial interested in literature about the p...

<t>One of my mathematics professors argued that mathematics is a sub branch of philosophy and as such should be placed under a philosophy department. I tend to agree. The reason why it is placed under the science department is its close historical connection – as a toolbox – to physics and other “ha...

ntruvant: why not use the method proposed by Peter Glossner? From my point of view it seems to be a fairly acceptable technique.

<t>Thanks for your reply!Have you tried other distributional assumption than log-normal for the analytical Lorenz curve? I’m thinking different distributions for “defaulted” and “all” obligors – maybe a beta distribution for the defaulted obligors? I have also implemented the model and it is now wor...

<t>That’s also the article I have been working with. My problem is that the estimates I’m getting for the s.e. are way off even when I use a million simulated observations. The lower estimates are zero for all rating scores which is not realistic. I suspect that it is optimization routines not worki...

<t>I’m trying to estimate PD based on a Lorenz curve approach. The starting point is a rating system which to each obligor assigns a rating score (m) between 0 and 100. That is: PD(m) := L’(Fa(m))*PDmeanWhere L’ is the differentiated Lorenz curve and PDmean is the mean default rate for all rating sc...

<t>Thanks for answering I’m sorry to say that I don’t completely understand your answer. Both series, after standardising, are white noises (not Gaussian) and their dependence structure has a strong time element. For example the correlation coefficient clearly changes over time. With respect to the ...