When predicting rankings of a target variable as accurately as possible, what should be used to judge prediction models?

When predicting rankings of a target variable, using the Gini coefficient is particularly appropriate because it is specifically designed to assess the discriminative power of a model. The Gini coefficient measures the inequality among values of a frequency distribution, making it valuable in determining how well a prediction model can distinguish between different classes or outcomes.

In the context of ranking predictions, the Gini coefficient can indicate how effectively the model ranks the positive instances higher than the negative ones. A higher Gini coefficient implies better discriminatory ability, thus enhancing the model's utility in applications where the ranking of results is crucial. This makes it the preferred choice for evaluating models focused on ranking rather than just binary decisions or point estimates.

The KS statistic, while useful in assessing model performance, is primarily aimed at measuring the maximum separation between the cumulative distribution of predicted probabilities for the positive and negative classes, rather than focusing specifically on ranking. Average squared error is a measure that assesses the average of the squares of the errors or deviations from the actual values, which is more suited for regression problems and does not directly relate to ranking efficacy. Misclassification refers to incorrectly predicted outcomes, and while it helps in assessing accuracy, it does not provide insights into how well the model ranks predictions in terms of their order.