A robust conditional approximation of marginal tail probabilities.

The aim of this contribution is to derive a robust approximate conditional procedure used to eliminate nuisance parameters in regression and scale models. Unlike the approximations to exact conditional solutions based on the likelihood function and on the maximum likelihood estimator, the robust conditional approximation of marginal tail probabilities does not suffer from lack of robustness to model misspecification. To assess the performance of the proposed robust conditional procedure the results of sensitivity analyses are discussed.