The timing of elections is flexible in many countries. We study this optimization by first creating a Bayesian learning model of a mean-reverting political support process. We then explore optimal electoral timing, modelling it as a renewable American option with interacting waiting and stopping values. Inter alia, we show that the expected longevity is a convex, then concave, function of the support. Finally, we calibrate our model to the post-1945 Labour-Tory U.K. rivalry. Our story quite well explains when the elections were called. We also show that election options approximately double the expected time in power in the current streak.