This paper is now published in Econometrica.
We introduce random evolving lotteries to study preference for non-instrumental information. Each period, the agent enjoys a flow payoff from holding a lottery that will resolve at the terminal date. We provide a representation theorem for non-separable risk consumption preferences and use it to characterize information seeking and its opposite, information aversion. To address applications, we characterize peak-trough utilities that aggregate trajectories of flow utilities linearly but, in addition, put weight on the best (peak) and worst (trough) lotteries along each path. We identify conditions for the ostrich effect, decision makers’ tendency to prefer information after good news to information after bad news. Our model permits savoring (enjoying the gradual arrival of good and sudden arrival of bad news) and dread (disliking the gradual arrival of bad and sudden arrival of good news) and a preference for skewed information