In this talk I will consider the problem of developing prediction or forecasting functions that are based on maximizing association or agreement measures, both for univariate and multivariate responses, and for a multi-dimensional predictor or feature. Particular focus will be the concordance correlation coefficient proposed by Lin in a Biometrics article in 1989. Prediction functions will be compared with more commonly-used predictors. We consider finite-sample and asymptotic properties of the predictors, and demonstrate their utility in some practical and applied settings, including risk assessments in biomedical, environmental, and ecological settings.
