ON THE APPLICATION OF ROBUST, REGRESSION-BASED DIAGNOSTICS TO MODELS OF CONDITIONAL MEANS AND CONDITIONAL VARIANCES

A strategy is proposed for applying a class of robust, regression-based diagnostics to nonlinear models of conditional means and conditional variances for cross-section or time-series data. The distinguishing feature of the current approach, which builds on already popular residual-based procedures, is that no auxiliary assumptions are imposed at any testing stage. Consequently, the statistics are guaranteed to have the correct asymptotic size under the relevant null hypothesis. Several new, regression-based conditional mean and conditional variance diagnostics are proposed. The case of incompletely specified dynamics in time-series models is explicitly covered. © 1991.