Title: Sparse nonparametric dynamic graphical models
Authors: Fabrizio Poggioni - University La Sapienza (Italy) [presenting]
Mauro Bernardi - University of Padua (Italy)
Lea Petrella - Sapienza University of Rome (Italy)
Abstract: A sparse nonparametric dynamic graphical model is proposed for financial applications in which we employ a semiparametric multiple quantile model with CAViaR specification to describe the marginal distributions and a LASSO-penalized Gaussian copula-VAR model to describe the multivariate distribution of financial returns as a sparse dynamic model. In order to use the multiple quantile models as marginal distributions the estimated quantile functions must be invertible, in this way we can get the marginal CDFs from the estimated multiple quantiles. It is therefore necessary to guarantee the monotonicity of the estimated quantiles and, consequently, the absence of crossing. We contribute to the topic of quantile crossing for semiparametric models by defining a non-crossing parametric space for multiple quantile CAViaR models. Furthermore, we find computationally convenient to include the defined non-crossing necessary conditions as linear constraints to the multiple quantile estimation problem. Finally, we present an empirical application of the proposed methodology.