Lorenz zonoid measures to compare predictive accuracy

In this paper, a novel approach for model comparison is presented. The standard linear regression model framework is considered and extended through the employment of the Lorenz zonoid tool. By means of the Lorenz zonoid, which in the univariate case corresponds to the classical Gini coefficient, a twofold aim is achieved. On the one hand the contribution of any additional independent variable introduced into the linear model is determined and, on the other hand, a relative measure of the variability characterising the underlying response variable is pro- posed. The results is a Lorenz zonoids-based dependence measure which mimics the partial correlation coefficient by simplifying the interpretation, especially when dealing with phenomena affected by high variability.