Title: Graphical modelling of multivariate panel data models
Authors: Celia Gil-Bermejo - Universidad Complutense de Madrid (Spain) [presenting]
Antonio Jesus Sanchez Fuentes - Pablo de Olavide University (Spain)
Jorge Onrubia - Universidad Complutense de Madrid - Instituto Complutense de Estudios Internacionales (ICEI-UCM) (Spain)
Abstract: A new approach is proposed to both test the existence of causal relationships between variables in a panel data environment using a VAR model and determine one final causality path excluding those relationships which are redundant. One of the main novelties is that we extend the number of relevant variables, mostly limited to two/three variables when using panel data. Once we set the dependence criteria (in our case, the concept of Granger causality), we apply the PC algorithm in order to debug potential indirect relationships between variables. This algorithm uses an iterative process where different conditional tests between each pair of variables are carried out. Thanks to these individual measures, we construct one synthetic measure for the whole sample. Finally, once the causality path between all the possible combinations of variables has been established, we draw it using causal maps. These figures provide a visual guide which makes explicit complex interlinked relationships. Moreover, this approach helps us to analyse the determining factors influencing these relationships.