Detecting Outliers in Simultaneous Linear Models

Abstract

The robustness literature provides many solutions to the problem of detecting and handling outliers. In this paper we study how to detect ‘influential points’ in a simultaneous linear model. The model describes the Dutch economy during the post-war period. Outliers in both explanatory and the dependent variables are located by using regression diagnostics. The (interactive) computerprogram lists influential points by methods due to Belsley, Kuh and Welsch. The estimation method is modified two stage least squares. Finally simultaneous predictions are presented and compared to the 2SLS predictions and the realisations.