Research Seminar on April, 4: Vladimir Pyrlik (CERGE-EI)

Topic: Shrinkage in Estimating High Dimensional Copulas 
Date & Time: April, 4 at 15.00
Place: 3A Kantemirovskaya st., room: 339

International Laboratory of Game Theory and Decision Making  invites you to participate in a research seminar.

Date & Time: April, 2019 (Thursday) 15.00 - 16.20
Place: 3A Kantemirovskaya st., room 339
Topic: Shrinkage in Estimating High Dimensional Copulas (joint with Stanislav Anatolyev
Speaker: Vladimir Pyrlik (CERGE-EI)
Language: English


Abstract:
As a combination of separate marginal distributions and a dependence structure, copulas have proved a convenient framework to synthesize joint distributions, including in high dimensions. Currently, the high dimensional settings analyzed with copulas contain at most several hundred variables, as higher dimensionality appears too demanding for either model selection or estimation. However, the practical application in many fields demands higher dimensionality. In our paper, we propose to adopt recently developed techniques of large covariance matrices estimation for the task of copulas estimation, with the dimensionality of the data going well beyond that studied in the literature. We apply the large covariance matrices shrinkage estimators of Ledoit and Wolf (2004, 2017) to estimate some types of copulas in high dimensions. We consider Gaussian and Student's t copulas as well as their skewed versions and take the dimensionality of the data up to thousands of variables. A simulation study shows that the shrinkage estimation of the large matrix parameters of the copulas significantly outperforms the traditional estimators, including Kendall's rank correlation-based estimator and sample correlation matrix of pseudo-observations.

Everyone interested is invited!

Follow this link to our calendar: https://tinyurl.com/spbecon