Exact minimax strategies for predictive density estimation, data compression and model selection

The Bayes procedure with uniform prior on location (and log-scale) parameters is shown to be exact minimax optimal for location and scale families in problems of universal data compression, predictive density estimation and model selection.