Title: Group transformation models
Authors: Christian Gourieroux - University of Toronto and CREST (Canada) [presenting]
Alain Monfort - ENSAE Paris (France)
Jean-Michel Zakoian - CREST (France)
Abstract: Semi-parametric transformation models relate the endogenous variables to the errors through a parametric transformation depending on observed explanatory variables without specifying the error distribution. It is shown that in such model, called group transformation model (GTM), any pseudo maximum likelihood (PML) estimation approach provides consistent estimators of the sensitivity parameters of the explanatory variables, whenever artificial intercept parameters are introduced at appropriate places. This modelling principle with the associated PML estimation method is illustrated by several examples of application to multivariate ARCH models, qualitative models, Loss-Given-Default, omitted heterogeneity, seasonal adjustment, peer effect, or directional statistics.