Abstract
In many situations, statistical forecasts are useful insofar as they are a trustworthy guide to downstream rational decision making. In this talk, we ask what properties high dimensional forecasts must have in order for downstream decision makers to be incentivized to best respond to them, as if they were correct --- and whether we can efficiently make such forecasts in high dimensional strategic settings. As applications, we discuss new algorithms for learning in games with large action spaces, and new prior-free results in sequential principal agent games.