I'm doing some numerical experiments based on copulas but have trouble with simulation from a multivariate mixture copula. Though there are efficient simulation algorithms for multivariate Gauss copula, Student-t copula and Clayton copula, I have not found simulation algorithms for the mixture models. Any efficient algorithm or source code?
Hi Zhe, in this case the inverse transform sampling is used. The idea is to generate sapmles using a uniform distribution and them to use the inversetransform :
ttps://en.wikipedia.org/wiki/Inverse_transform_sampling
The drawback is that the reciprocal of the cumulative distribution function must be known.
I am not quite sure what you are simulating, but one simple answer is that the weighted average of x * copula_density_1 + (1-x)* copula_density_2 is a valid copula density. So if you choose how many points of each copula density to simulate, and use the simulation algorithms you have for each one, and add them together, then you are simulating a mixture. You can numerically calculate the copula based on the simulated density. This may be too naive an answer.
Hi,
You may have a look to this recent paper:
.
A new time-varying optimal copula model identifying the dependence across markets.
By: Liu, Bing-Yue; Ji, Qiang; Fan, Ying. Quantitative Finance. Mar2017, Vol. 17 Issue 3, p437-453. 17p. DOI: 10.1080/14697688.2016.1205208.
Sujets: Distribution (Probability theory); Economic models; Computers in economics; Copula functions; Time-varying systems
best regards
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Algorithms